The Illusion of Control Why Financial Crises Happen, and What We Can (And Can't) Do About It

The Illusion of Control
Y8019-Danielsson.indb i 2/22/22 10:54 AM

This page intentionally left blank
The Illusion
of Control
Why Financial Crises
Happen, and What We Can
(and Can’t) Do about It
Jón Daníelsson
New Haven & London
Y8019-Danielsson.indb iii 2/22/22 10:54 AM

Published with assistance from the Louis Stern Memorial Fund.
Copyright © 2022 by Jón Daníelsson.

All rights reserved.
This book may not be reproduced, in whole or in part, including illustrations,
in any form (beyond that copying permitted by Sections 107 and 108 of the U.S.
Copyright Law and except by reviewers for the public press), without written
permission from the publishers.
Yale University Press books may be purchased in quantity for educational,

business, or promotional use. For information, please e-mail sales.press@yale
.edu (U.S. office) or [email protected] (U.K. office).
Set in Galliard type by Newgen North America.

Printed in the United States of America.
Library of Congress Control Number: 2021948855

isbn 978-0-300-23481-7 (hardcover : alk. paper)
A catalogue record for this book is available from the British Library.
This paper meets the requirements of ansi/niso z39.48-1992 (Permanence of

Paper).
10 9 8 7 6 5 4 3 2 1
Y8019-Danielsson.indb iv 2/22/22 10:54 AM

Contents
Acknowledgments vii
1 Riding the Tiger 1
2 Systemic Risk 6
3 Groundhog Day 26
4 The Risk Panopticon 51
5 The Myth of the Riskometer 72
6 Ideas Matter: Risk and Uncertainty 94
7 Endogenous Risk 109
8 If You Can’t Take the Risk, Change Riskometers 128
9 The Goldilocks Challenge 147
10 The Risk Theater 168
11 The Uniformity, Efficiency, and Stability Trilemma 189
12 All about BoB: Robots and the Future of Risk 205
Y8019-Danielsson.indb v 2/22/22 10:54 AM

vi CONTENTS
13 The Path Not to Take 223
14 What to Do? 240
Notes 255
Bibliography 259
Index 267
Y8019-Danielsson.indb vi 2/22/22 10:54 AM

Acknowledgments
Several friends and colleagues made invaluable contributions to the

book. I borrowed many arguments from joint work with Charles Good-
hart, Kevin James, Robert Macrae, Andreas Uthemann, Marcela Valen-
zuela, Ilknur Zer, and Jean-Pierre Zigrand.
I had the privilege of consulting several experts, including Peter An-
drews, formerly of the FCA, Jon Frost at the Bank for International Set-
tlements, risk system expert Rupert Goodwin, fund manager Jacqueline
Li, Eric Morrison from the FCA, physicist Donal O’Connell, and three
lawyers: Gestur Jonsson, Hafliði Kristján Lárusson, and Eva Micheler.
I was fortunate to employ several excellent London School of Econom-
ics (LSE) students to assist me on background research for the book:
Sophia Chang, Jia Rong Fan, and Morgane Fouche. Our manager at the
Systemic Risk Centre, Ann Law, was very helpful in getting the book
under way. The many drawings in the book were made by two excellent
artists, Lukas Bischoff and Ricardo Galvao.
Without all of these people, this book would not have seen the light
of day.
I am grateful to the Economic and Social Research Council (UK), grant
number: ES/K002309/1 and ES/R009724/1, and the Engineering and
Physical Sciences Research Council (UK), grant number EP/P031730/1,
for their support.
vii
Y8019-Danielsson.indb vii 2/22/22 10:54 AM

The Illusion of Control
Y8019-Danielsson.indb ix 2/22/22 10:54 AM

1
Riding the Tiger
The man who didn’t trust the models saved the world.
Figure 1. Credit: Copyright © Ricardo Galvão.
Almost every economic outcome we care about is long term.

Pensions, the environment, crises, real estate, education, you name it,
what matters is what happens years and decades hence. Day-to-day fluc-
tuations don’t matter to most of us—the short run isn’t very important.
So it stands to reason that the way we manage our financial lives should
emphasize the long run. But by and large, it doesn’t. We are good at man-
aging today’s risk but at the expense of ignoring the promises and threats
of the future. That is the illusion of control.
A quick test. What do you think are the riskiest years of the past cen-
tury? Covid in 2020? The global crisis in 2008? The Great Depression in
the 1930s? Let’s ask the financial markets and their go-to measurement of
1
Y8019-Danielsson.indb 1 2/22/22 10:54 AM

2 RIDING THE TIGER
risk, volatility. The accompanying drawing depicts the annual stock mar-
ket risk in the United States, and it confirms the usual suspects (Figure 1).
But these are not the riskiest years by a long shot. In 1962 and 1983 we
were almost hit by the ultimate tail event, even though the financial mar-
kets remained calm. The Cuban missile crisis nearly brought the United
States and the Soviet Union to blows in 1962, only for the Soviets to back
down at the last minute. Even more interesting is 1983 because that is
when we almost got into a nuclear war, even if we didn’t know that until
much later.
What happened was that the premier of the Soviet Union, Yuri An-
dropov, got it in his head that the United States was planning to launch
a preemptive nuclear strike. He instructed his spies to find evidence sup-
porting his suspicion, and KGB agents everywhere went into overdrive,
looking for that evidence. Careers depended on it. If you have a choice
of a juicy posting in Washington or being literally sent to Siberia (as the
KGB rep in Novosibirsk) of course you find proof. A prime example of
confirmation bias. We believe something terrible will happen and find
grounds to support that, even if there is no truth to it. In 1983 the early
warning models of the Soviet Union detected a nuclear attack. The man
on watch that night, Stanislav Petrov, didn’t trust the signal and decided
unilaterally not to launch a counterattack. The man who didn’t trust the
models saved the world. The Soviet investigators subsequently confirmed
he was right. The false alarm came about because of a rare alignment of
sunlight on high-altitude clouds above North Dakota and the Molniya
orbits of the detection satellites. Colonel Petrov died in 2017, by that time
globally recognized for having saved humanity.
The way we measure financial risk today, with what I call the risk-
ometer, has much in common with Andropov’s early warning systems.
Both rely on imperfect models and inaccurate measurements to make cru-
cial decisions. While high-altitude clouds bedeviled the Soviet’s models,
the problem for today’s riskometer arises from its emphasis on the recent
past and short-term risk. The reason is simple. That is the easiest risk to
measure, as the modelers have plenty of data.
The problem is that short-term risk isn’t all that important, not for
investors and especially not for the financial authorities. For them, what
matters is systemic risk, the chance of spectacular financial crises, like the
one we suffered in 2008. Long-term threats, like systemic crises or our

RIDING THE TIGER 3
retirement not being as comfortable as hoped for, are easy to under-

stand, at least conceptually. Take crises. Banks have too much money and
can’t find productive investments, so they start lending to increasingly
low-quality borrowers, often in real estate. In the beginning it looks like
magic. Money flows in, developers build houses, and everybody feels rich,
encouraging more lending and more building in a happy, virtuous cycle.
Then it all comes crashing down.
What are we to do? Regulation, of course, using the Panopticon. An
idea dating to the eighteenth-century English philosopher Jeremy Ben-
tham, who proposed regulating society by setting up posts to observe
human activity. Does it work? In traffic certainly. If no police or cameras
monitor speeding, I suspect many will be tempted to drive too fast. The
chance of getting caught keeps the roads relatively safe. Surely we can
also employ the Panopticon to regulate finance and so prevent crises and
all the horrible losses. We do use it, but it doesn’t work all that well. The
reason has to do with the interplay of two complex topics, the difficulty
of measuring risk and human ingenuity.
Unlike temperature or prices, risk cannot be directly observed. Instead,
it has to be inferred by how prices have moved in the past. That requires a
model. And as the statistician George Box had it, “All models are wrong,
but some are useful.”1 There are many models for measuring risk, they all
disagree with each other, and there is no obvious way to tell which is the
most accurate. Even then, all the riskometers capture is the short term
because that is where the data is.
And then we have human behavior. Hyman Minsky observed forty
years ago that stability is destabilizing. If we think the world is safe, we
want to take on more risk, which eventually creates its own instability.
And since the time between decisions and bad things happening can be
years or decades, it is hard to control risk. As my coauthor Charles Good-
hart put it, “Any observed statistical regularity will tend to collapse once
pressure is placed upon it for control purposes.”2 What Goodhart’s law
tells us is that when the risk managers start controlling risk, we tend to
react in a way that makes the risk measurements incorrect.
I have been working on the human dimension to risk for some time
and, along with coauthors, have come up with a helpful classification
scheme: exogenous and endogenous risk. Endogenous risk captures
what results from the interactions of the human beings that make up the

4 RIDING THE TIGER
financial system. The opposite of endogenous is exogenous. When an

asteroid hit the Gulf of Mexico sixty-five million years ago, wiping out
the dinosaurs, that was an exogenous shock. The dinosaurs certainly did
nothing to cause their demise.
By viewing the financial system through the lens of endogenous risk
we learn quite a bit about what works and what doesn’t, often in surpris-
ing ways. At the root of many a financial crisis and large losses is a vicious
feedback between well-meaning rules and the self-preservation instincts
of market participants. But as such forces of fragility only show them-
selves at the worst possible moment, we don’t know how dangerous they
are until it’s too late. It is difficult, to the point of impossible, to measure
endogenous risk. Why? Because all the riskometers can capture is the less
important exogenous risk. And that, in turn, explains why it is so easy to
manipulate risk measurements. Suppose my job is to manage the risk of
some portfolio. One day my boss calls me into her office: “Jón, risk is too
high, and you have to make it lower. However, our traders are doing a
fantastic job, and I want to keep the nice returns. Now go and play with
your computers and make it happen. And don’t get caught make sure you
comply with all regulations.” Easy. The UBS bank went bankrupt in 2008
because it deliberately chose to measure risk incorrectly, a common prob-
lem in that crisis. The banks are good at manipulating the riskometers,
thinking they are fooling the regulators, but, at the end of the day, they’re
just deceiving themselves.
Endogenous risk is at the heart of the Goldilocks challenge facing the
financial authorities. We want financial regulations to be just right. Not
too strict, so all economic activity is strangled. Not so relaxed that we
get too many crises and lose all our money. We need a balance, and that
is not easy. The regulators focus on what went wrong last time around,
while the next crisis will happen in a completely different place. Precisely
what happened in the years before 2008 when the central banks were con-
cerned only with inflation, missing all the warning signs of the pending
crisis. And the reaction to Covid-19 in 2020 was shaped by the lessons of
2008. Incorrectly, as the financial impact of the virus was very different.
Regulation by the rearview mirror.
There are broader lessons from the dichotomy between exogenous
and endogenous risk, and one is a particular fallacy of composition. The

RIDING THE TIGER 5
financial authorities aim to get the best out of the financial system by
making all the financial institutions prudent, like Volvos, the world’s
safest car. Nobody makes crazy investments, and everybody follows the
rules. Surely, then, investors will enjoy stable returns, banks won’t fail,
and financial crises will not happen. Will turning the banks into Volvos
make us safe? No. It perversely makes crises even more likely, since it re-
duces the shock-absorbing capacity of the system.
Is artificial intelligence, AI, the proper response to the challenges of
endogenous risk? It depends. AI will increase efficiency, allowing us to
eliminate a lot of tedious risk management and compliance jobs. Finan-
cial services will become cheaper and more reliable, and regulations will
be better enforced. So what can go wrong? Imagine the Bank of En-
gland bot—BoB—is put in charge of financial stability. He talks to his
counterparts in regulated banks, passing on information and enforcing
compliance. It sounds like the perfect way to control the financial system.
Except we have to trust BoB to make the right decisions. And, unlike
humans, we don’t know how he reasons and decides. So what will happen
when BoB runs into some problem he has never seen before? A human
being can draw on their accumulated experience and the canon of human
knowledge. AI will not be able to do that. Meanwhile, it is pretty easy for
hostile agents to take advantage of BoB. He has to look everywhere. A
hostile agent—some trader or terrorist or nation state or criminal—only
has to find one weakness to exploit. And they can do that in complete
secrecy until it is too late for BoB to do anything about it. BoB cannot
win regardless of the state of technology.
These observations all sound a bit pessimistic, but that is not the impres-
sion I want to leave. The financial system is highly resilient and by and large
does a good job—most public comments to the contrary. However, the
way we deal with the system today takes us in the wrong direction. So, what
to do? Embrace diversity, the most potent force of financial stability and
good investment performance. The more different the financial institutions
that make up the system are and the more the authorities embrace that very
diversity, the more stable the system becomes and the better it performs.
To the benefit of us all. What gets in the way is self-interest and politics.
The decision makers are antidiversity, preferring uniform ways of doing
business to protect their profits and jobs. All that is needed is political will.

2
Systemic Risk
The financial crisis was a pest epidemic, spreading with raging speed from
house to house.
—Stephan Skalweit
Figure 2. Credit: Lukas Bischoff/IllustrationX.
Leendert Pieter de Neufville was only twenty-one in 1751 when

he founded his Amsterdam bank, De Neufville Brothers. His big chance
came a few years later with the Seven Years’ War, what Americans call
the French and Indian War. Amsterdam was then the most sophisticated
financial center in the world, leading in financial innovation and having
the most creative bankers and the highest amount of speculative capital.
The Dutch guilder was the US dollar of its day, the reserve currency that

SYSTEMIC RISK 7
facilitated trade throughout Europe and beyond. Leendert took full ad-
vantage, building up one of the world’s wealthiest and most prestigious
banks by financing the Prussian side of the war. He lived well, furnishing
his house only with the finest-quality objects and owning an excellent
collection of paintings—but not a single book.
The way he made his money was thoroughly modern: rapid, irrespon-
sible financial innovation in the form of acceptance loans, not all that
different from the financial instruments so damaging in the 2008 crisis.
Cheap short-term borrowing was used to make long-term loans at high
interest rates, involving long chains of obligations, spanning multiple
banks and countries.
The key to all the profit was borrowed money. For every twenty-three
guilders he lent to Prussia, De Neufville supplied one and borrowed
twenty-two, secured by commodities. Highly profitable in good times,
but it didn’t take much for things to go wrong, as they did, spectacularly,
when the war ended in 1763, culminating in the first modern financial cri-
sis.1 Commodity prices crashed when the war ended because the farmers
could now finally start producing again, making all the commodity-based
collateral behind De Neufville’s acceptance loans worth little. Investors
got spooked and decided not to roll over these short-term loans—they
all went on strike, just like their successors in 2007. As De Neufville didn’t
have enough ready cash to repay his creditors and keep his bank alive, he
had to sell his vast holdings of commodities. However, that only caused
prices to fall further, a process known as a fire sale. Falling prices induce
speculators to sell, and when they sell, prices fall further, in a vicious loop.
It did not take long for De Neufville to default.
Thus was born the first global systemic crisis, described at the time as
a pest epidemic. It spread with raging speed from house to house. The
city of Hamburg was hit hard, and on 4 August its mayor wrote to the
mayor of Amsterdam asking him to bail out De Neufville. Amsterdam
refused. Fortunately, the crisis turned out to be relatively short-lived, and
Amsterdam, Hamburg, and the other major centers recovered quickly.
Berlin suffered badly because of how its emperor, Friedrich II, reacted.
He imposed a payment standstill and bailouts, violating the contracts that
allowed funds to flow from Amsterdam to Berlin and causing the bankers

8 SYSTEMIC RISK
to mistrust the Prussian authorities. The money stopped flowing, and a

severe recession ensued. Not the only time a government’s response to a
crisis made things worse.
The 1763 crisis is what might now be classified as a V-shaped crisis,
except for Berlin, which suffered a U crisis. In a V crisis we see a sharp
temporary drop in economic activity, but as nothing fundamental is de-
stroyed the recovery is similarly quick. The Asian crisis in 1998 was also of
the V variety while the Great Depression and the 2008 crises, given their
sharp crash and slow recovery, were U shaped. What about Covid-19? The
jury is still out, but it seems closer to K, quick winners and losers.
De Neufville’s creditors tried to claw their money back but with limited
success. He was forced to auction off some of his paintings in 1765, in-
cluding The Milkmaid by Johannes Vermeer for 560 guilders, now one of
the finest attractions of the Rijksmuseum in Amsterdam. Leendert man-
aged to retain most of his houses, including one at Herengracht 70–72 in
Amsterdam, a very desirable address then as now.
What happened in 1763 was a systemic financial crisis, and the chance
of that happening is systemic financial risk. For the remainder of the book
I will omit the word “financial” and just call it systemic risk. Seventeen
sixty-three was the first modern global financial crisis because it was not
a crisis caused by war or crop failure but rather by shadow banking and
the extensive use of sophisticated financial instruments allowing risk to
hide and spread by efficient and interconnected financial centers. We have
experienced many systemic crises since, and it is remarkable how similar
they all are. Even though the world in 1763 was very different from that
in 2008, the crisis dynamics were eerily similar: shadow banking creating
deep vulnerabilities that became visible only when it was too late. Crises
are all fundamentally the same. They differ only in details.
My definition of systemic risk is the chance that the financial system
will fail spectacularly in doing what it is supposed to. The global au-
thorities, in the form of the International Monetary Fund, the Bank for
International Settlements, and the Financial Stability Board, prefer a nar-
rower definition: “The disruption to the flow of financial services that is
(i) caused by an impairment of all or parts of the financial system; and
(ii) has the potential to have serious negative consequences for the real
economy.”2 The key here is the two words “financial system.” It has to

SYSTEMIC RISK 9
be either the root cause or the amplifier for a crisis to be called systemic.
Since the financial system is so fundamental to the economy, most eco-
nomic crises are also systemic crises. What about Covid-19? It certainly hit
the economy, shaving 10 percent to 20 percent off most countries’ GDP
in the second quarter of 2020. But the financial system was a bystander,
not the cause, and it made things neither better nor worse. The Covid-19
crisis was not systemic.
The focus of systemic risk is not on any individual financial institution;
instead, it is on the financial system in its entirety and how it affects the
real economy. The failure of a bank, or even a banking crisis, is not neces-
sarily systemic. We need a connection between the financial system and
the real economy.
Systemic crises are costly, easily 10 percent of GDP or more, so for the
United States in the trillions of dollars. Fortunately, they are not frequent,
and I think most people can expect to suffer such a crisis at most once
in their lifetime. If one takes the relatively loose definition used in the
IMF crises database, then maintained by Luc Laeven and Fabián Valencia,
we find that the typical OECD country suffers a systemic crisis once in
forty-three years on average. While the one in forty-three years is the his-
torical average, it is hotly debated whether the future will be as tranquil.
A lot of commentators maintain that increased complexity and intercon-
nectedness make financial crises more frequent. The United States, and
especially the United Kingdom, are more crisis prone, the UK enduring
a systemic crisis once every seventeen years. The last one was in 2008, so
2025 is the due date for the next one if it arrives on schedule. If anything,
the once-in-forty-three-years figure is an overestimate, as the database
includes relatively nonextreme events like Black Monday in October 1987
and the Long Term Capital Management crisis in 1998. If we exclude all
the mildest crises, we get about one systemic crisis per typical lifetime.
The reason for the once-in-a-lifetime frequency is that the most severe
crises happen only after we forget the last one. Crises change behavior,
and those who come of age during one will be affected by it for the rest
of their lives. We had to wait for the twenty-year-olds of 1929 to retire
in the 1970s for the seeds of the next crisis to be sown. It then took a
quarter-century for the seeds to bear fruit, culminating in the events of
autumn 2008. Politics and lobbying push toward a high-risk, high-return

10 SYSTEMIC RISK
financial system, and when memories of the previous crisis fade there is
little pushback.
Considering the once-in-a-lifetime frequency of systemic crises, the
term “systemic crisis” is as overused today as it was underused before
2008. Most of this usage is imprecise and contradictory, and one often
gets the impression that commentators are talking only about the last cri-
sis or financial scandal when they use the phrase “systemic crisis.”
The financial system is vulnerable to many types of shocks, some com-
ing from outside the financial system, like Covid-19, and others generated
by the system itself. Some shocks are idiosyncratic, affecting only a single
institution or asset, while others impact the financial system as a whole
plus the real economy. The small shocks often arrive from outside the sys-
tem; all the large ones are created by the interaction of the human beings
who make up the financial system. There may be an outside trigger, but
the real damage is caused by the system turning on itself.
If I had to describe a textbook financial crisis, it would go something
like this: Financial institutions have too much money to lend. When they
run out of high-quality borrowers, they start making increasingly low-
quality loans, often in real estate. In the beginning it all looks ingenious.
Developers borrow money to build new houses, which stimulates pros-
perity and demand for homes. Property prices go up. Everybody feels
wealthier and more optimistic, thereby encouraging more lending and
more building in a happy, virtuous cycle. This caused many a crisis, like
the savings-and-loan debacle in the United States in the 1980s and the
Spanish crisis in 2010.
Eventually, when the little boy yells, “The Emperor has no clothes!”
people realize that all the prosperity was built on sand (Figure 2).3 There
is no strong underlying economy, and it all comes crashing to the ground;
the virtuous feedback loop becomes vicious. Prices fall, developers fail,
the banks lose money, the economy contracts, prices fall more—the same
fire sale process De Neufville endured. The precrisis rise in prices is much
slower than the fall: prices go up the escalator and down the lift (or eleva-
tor if in America).
Most crises follow the textbook. The one in 1914 was atypical, which
is why it is my favorite. It was triggered by the assassination of Archduke
Franz Ferdinand of Austria on 28 June 1914, leading to posturing among the

SYSTEMIC RISK 11
great powers of Europe amid raised expectations of an impending war. This

anticipation created worries that financial institutions would experience
difficulties in having cross-border loans repaid—after all, if two countries
are at war enforcing contracts across borders would naturally be difficult.
Surprisingly, even after the war started, remittance between the Central
Powers and the Allies continued via neutral countries, notably Switzer-
land, which helped propel it into the ranks of major financial centers.
The immediate consequence of the assassination of the archduke was
difficulty in clearing trades. When one buys and sells stocks, it takes time
for ownership to be transferred to the buyer and money to the seller, a
process called clearing. In 1914 clearing took two or three weeks, and if
one was buying stocks across borders, the securities were sent one way
and the money the other way by post. Given all the modern databases
and fast computers today, one might think clearing nowadays is instanta-
neous. But even now clearing can take days and even weeks. Slow clearing
was at the heart of the Gamestop and Robinhood crisis and conspiracy
theories in January 2021.
Any disruption to the clearing process is costly. Suppose you sell Google
stock. You are legally required to deliver the Google stock and you have
a legal right to receive cash at the same time, that is, the trade clears. If
all you are doing is selling Google, a short-term disruption to the trading
process might not be that problematic. However, most people have plans
for the money they received from the sale. Perhaps you want to use it to
buy Microsoft, and to pay for it you need the money from selling Google.
If, for some reason, you don’t receive that money, you may default on
the obligation to pay for Microsoft, even if you acted prudently and in
good faith.
The 1914 crisis first affected the stock exchanges in Berlin, Paris, and
Vienna. To protect their local banks the authorities in those cities decided
to suspend the clearing process, giving the dust time to settle. If every-
body is trading locally that should not be a huge problem; at least nobody
defaults because they cannot deliver cash. However, it is problematic if
London remains open when Paris is closed. London-based banks send
cash and securities to banks in Paris, but the reciprocate does not happen.
One might think this is an issue only for the sophisticated London banks
doing business with the continent. Not so. The interconnectedness of the

12 SYSTEMIC RISK
financial system meant that everybody was vulnerable. Even banks that
thought they were sensible, like conservative rural banks trading only
with major London banks and judiciously avoiding taking on too much
risk, were nevertheless immediately affected.
First, those dealing with the continent got into difficulty, as they had
to deliver to the continental trading partners but did not receive cash
from their European counterparts. Then their business partners got into
trouble, and in short order those dealing with them faced difficulty. In
the financial system everybody is exposed to everybody, whether they
like it or not. We might think we are prudent by not doing business with
those who like a lot of risk. However, if the people we are dealing with
are exposed to risk, so are we.
A good example of the network effect, why financial crises are often
said to be like a contagious disease, is as follows. Patient Zero infects the
people around her, who in turn infect the people around them, and in
short order everybody is sick. Similarly, the vulnerability quickly spreads
from Bank Zero to all the other banks. The financial system is vulnerable
not only because of risk but also—and even more so—because of how
everybody is connected to everybody else.
London was at the center of the network in 1914, a place it had by
then already occupied for a century and a half, after eclipsing Amsterdam.
Made in 1774, a drawing shows the network system of the European fi-
nancial system, with London in the center (Figure 3). We have been mak-
ing network diagrams ever since, and after the crisis in 2008 the creation
of network diagrams became a veritable industry. Such diagrams never
made much sense to me, as all they show is that everybody is connected to
everybody else, which we know anyway. The players change, but the net-
work is always there, the source of simultaneous danger and prosperity.
The network makes the economy efficient, creating wealth, good jobs,
and ample cheap goods, but it also transmits crises, both financial and
medical. The plague in the 1300s was transmitted by trading networks,
just as Covid-19 was in 2020.
If the 1914 crisis had been purely financial, nobody would really have
cared besides a handful of people involved with the financial system. But
it never is. The financial system directs resources for the real economy,
and if it is not doing its job companies can’t borrow and invest. They can’t

SYSTEMIC RISK 13
Figure 3. The financial system network in 1774. Source: M. Postlethwayt,

The Universal Dictionary of Trade and Commerce, 4th ed. (London, 1774),
vol 1. Arbitrage.
pay suppliers and salaries, the very essence of a systemic crisis. Economic
activity grinds to a halt, and everybody suffers. That is the reason finance
is so heavily scrutinized and regulated, and why the banks always get
these obnoxious bailouts when they misbehave.
Once under way, the 1914 crisis followed predictable patterns. Market
participants became very cautious, wanting the safest asset: gold. Banks
deposited their excess funds to the safety of the Bank of England instead
of lending them out, so that the crisis quickly spread from the City of

14 SYSTEMIC RISK
London to the real economy as commercial credit evaporated. In the be-

ginning the Bank of England and the leading financial institutions in the
City of London tried to sort it out by themselves. However, things got
worse, and it ultimately became the Treasury that made the most critical
decisions. It is always like that. A serious crisis response needs the legiti-
macy of the political leadership, something an institution like the Bank of
England, led by nonelected bureaucrats, simply cannot deliver.
Finally, both London’s and New York’s stock exchanges closed on
31 July and remained closed until January 1915. Why would one want to
close the stock exchange? To prevent a vicious mark-to-market feedback
loop. A company is solvent if the value of its assets exceeds the value of
its debt. If the price of its assets falls, the company may find itself in dif-
ficulty or even insolvency because debt does not fall along with the value
of the assets. To protect itself, the company may then sell its holdings.
That, however, makes prices fall even more, putting even more compa-
nies into difficulty, who will then also sell, and a vicious mark-to-market
feedback loop is born. Soon everybody seems to be moving in lockstep,
doing exactly the same thing, all selling and all buying the same assets.
The resulting extreme volatility is the outward manifestation of a crisis
in the making.
When the stock exchange closed, prices froze and the feedback loop
was halted. Provided confidence has recovered by the time the market
reopens, a crisis is averted. The same thinking is behind the circuit break-
ers on modern exchanges, which, if prices fall too much, halt trading
to give market participants time to make sense of what is happening.
The Chinese wanted to prevent the worst when they closed the Shanghai
market on 24 January 2020 for six trading days when its Covid-19 crisis
broke out. The closing might have prevented the appearance of excessive
volatility, but the market still opened down 8 percent.
The 1914 crisis showed that it does not take all that much for panic to
happen when confidence evaporates. The very same efficiency that was
so beneficial before a shock is a curse after it amplifies the initial shock
into something much worse. Even rumors that a large institution might
fail could cause panic. In 1914 it was clear that some banks were heavily
exposed to cross-border lending and would fail in case of war. That was

SYSTEMIC RISK 15
enough to trigger the crisis. People did not need the actual event of a war
to make this happen; the mere expectation of one was sufficient.
The financial authorities in 1914 did the right thing, effecting a massive
intervention, saving the City of London from the worst and, along with
it, the real economy. By comparison, events in the 2008 and Covid-19
crises were quite mild. However, the ultimate consequence for the City
of London was severe. It was the most important financial center in the
world in 1914, but by 1918 was eclipsed by New York, which still retains
the crown.
There are two main ingredients in the typical crisis story: excessive le-
verage and an interconnected financial system; in other words, too much
risk and a dangerous network structure. But we don’t need extreme risk-
taking for a crisis to happen. The 1914 crisis did not happen because of
too much risk. No, instead, the forces of instability preyed on the very
sophisticated financial system we depend on to give us efficient financial
intermediation and leverage, helping the economy grow—but only in
good times. That fantastic financial system also acts as the catalyst for
financial crises. And that means it is all about politics.
It’s the Politics, Stupid

When Bill Clinton was running for president in 1992, his cam-
paign’s unofficial theme was “It’s the economy, stupid.”4 Indeed, the
main reason he defeated George Bush Sr. was the economic recession of
the previous year. Today, it is the other way around—when explaining
the market, one might say, “It’s the politics, stupid.” Today’s primary
source of financial risk is not excessive risk-taking; it is quantitative easing
and low interest rates, Brexit and Donald Trump and the South China
Sea, Russia, Ukraine, Qatar, and Italy. The most severe financial crisis of
the world today, that in Venezuela, is caused by politics. Politics strongly
shape the reactions to Covid-19.
Government policies drive the financial markets. The financial markets
react to central banks’ interest rate decisions and quantitative easing.
When Covid-19 hit in early 2020 and the financial markets came under
stress, the central banks were expected to step up to the plate and save

16 SYSTEMIC RISK
the market. Not surprisingly, many commentators question why we si-

multaneously permit all the excesses in the financial system and also allow
it to exercise the Greenspan put every time something bad happens. The
moniker “Greenspan put” is named after the former chairman of the Fed-
eral Reserve, Alan Greenspan, because every time the financial markets
got the hiccups, the Federal Reserve was ready with a liquidity injection.
If the Fed bails us out there is no reason to behave reasonably: profits
are private, while losses are socialized. The consequence was that almost
everybody was blind to the hidden risk being created before 2008; those
times were even labeled the Great Moderation. The socialization of losses
and privatization of profits continue to anger activists on the left and the
right and is one of the main drivers of populism.
Governments can be a direct trigger of systemic risk, as in 1914. The
leading cause of the European sovereign debt crisis in the 2010s was not
bank misbehavior but government profligacy. The subprime mortgages
that were so instrumental in the 2008 crisis were created because succes-
sive US governments wanted to encourage homeownership for its poor-
est and riskiest citizens. However, the link between today’s politics and
the instability of tomorrow is not clear except to a handful of people who
closely monitor the relationship between the two. The very government
policies that culminated in instability may have been launched with the
best intentions, and their impact on systemic risk is often indirect and
counterintuitive.
As a consequence, some commentators have made robust remarks on
the origins of systemic risk. The late Fischer Black, of Black-Scholes5
fame, stated, “When you hear the government talking about systemic
risk, hold on to your wallet! It means that they want you to pay more
taxes for more regulations, which are likely to create systemic risk by
interfering with private contracting. . . . In sum, when you think about
systemic risks, you’ll be close to the truth if you think of the government
as causing them rather than protecting us from them.”6
Do the policy makers recognize that governments also cause systemic
risk? No. A few years ago I was at a large central bank conference in a Eu-
ropean capital discussing how the European authorities should respond
to the European sovereign debt crisis, then raging. In the closing panel,
leading policy makers debated what should be measured, controlled, and

SYSTEMIC RISK 17
eliminated, all the standard things like excessive risk-taking, opaque finan-
cial instruments, and perverse incentives. In the Q&A session I asked the
panel whether it was possible that the governments also caused systemic
risk. The chairman of the panel shouted at me, “No, the governments are
the solution, not the problem.”
Systemic crises are serious. Hundreds of millions of people suffered
in the Great Depression, driving the politics that caused World War II.
People lost their jobs, their savings, and even their lives. Populism thrived.
No wonder we would like to prevent them. The first lecture of my LSE
course Global Financial Systems is about systemic risk. About twenty
minutes into it I ask the students the following question: “Those of you
who would want to live in a country without any systemic crises, please
raise your hands.” In a typical year about eighty out of one hundred stu-
dents will do so. If I reverse the question and ask them if they want to live
in a country with systemic crises, nobody raises their hands. If we were to
ask most politicians, journalists, regulators, and pundits the same ques-
tion, I suspect most would agree with the students.
It is straightforward to prevent systemic crises: both Cuba and North
Korea manage it quite well. Get rid of the financial system, and voilà, no
more systemic crises. Easy, but the costs are unacceptable. Who wants
to live in Cuba or North Korea? We need the financial system to take
risk. With risk comes failure, an essential part of a healthy economy; an
economy without failures does not take enough risk and does not grow
fast enough. While it is easy to regulate risk out of the financial system,
it is not easy to do so without killing economic growth at the same time.
On the other hand, too many bankruptcies and crises are also a sign
that something is wrong. The financial system could be too unstable,
with excessive risk, perhaps too much corruption, or money channeled
to unproductive uses for political reasons. There is a balance between
having too many and too few crises. We need to encourage the financial
industry to take enough risk so the economy grows while also preventing
too much risk from causing systemic crises. A classic risk–return trade-
off, and one that is not trivial.
Economic crises afflict poor, less developed countries more frequently
and severely than their rich and more developed counterparts, not least
because developing countries are more likely to suffer from bad economic

18 SYSTEMIC RISK
management and volatile politics. They depend on a small number of

exports, often commodities, and are hence vulnerable to the whims of
the global markets. It is hard to raise money in these countries, as the
wealthy export their capital, forcing governments and other borrowers to
raise funds in foreign currency from global financial centers. No surprise
that sovereign debt, foreign exchange, and general economic crises are
endemic in less-developed countries.
It is not so with systemic risk: the very lack of development protects
them. Countries with underdeveloped financial systems, where credit
cards and bank accounts are rare and everybody uses cash, are mostly safe
from systemic risk. If the banks in such a country fail, most people can
shrug it off. If my bank in London fails today, my debit card stops work-
ing, so I can no longer buy lunch. All my money is tied up in my bank
accounts, and if I no longer have access to them there is nothing I can
buy. Not only will I starve to death, but all the vendors selling to me will
also not get any business. In developed countries the safety of the banks
is of primary importance. The more sophisticated the financial system
becomes, the more vulnerable it is to disruptions. Systemic risk is the
disease of the affluent.
We Had Forgotten History by 2008

Everybody thought we had solved the problem of crises by 2007.
Why? Two reasons. The first is that systemic crises are quite rare, so no-
body was worried. If you haven’t experienced one in your lifetime, you
will not expect one to be around the corner either. The second reason is
hubris. The experts had come to believe that modern risk management
systems and regulations were sophisticated and robust. Policy makers
and academics and practitioners all thought crises were associated with
less-developed economies, countries with Weak financial regulations, and
plenty of corruption and abuse. The developed Western economies had
found the magical way of preventing crises by combining efficient finan-
cial regulations and atomistic risk management. Hubris.
Atomistic risk management means that if we use sophisticated risk
management techniques to manage all the small risks, they will not grow

SYSTEMIC RISK 19
into big risks—voilà, crises are eliminated. That logic is fatally flawed. It is
true that the whole of the financial system is made up of everything that’s
going on within it, all the way down to individual transactions. We can
know all about the atoms that make up the human body and be experts
in biology, but all that knowledge tells us nothing about what makes a
person tick. It’s the same in finance. We simply don’t know how to add
up all the individual micro risks to get the risk in a portfolio, bank, or the
system. If we think we do, all we end up with is a false sense of safety.
We thought risk had magically disappeared by 2007 because all the
micro risks were under control. What we missed was that all the sophis-
ticated risk management techniques accomplished was to increase sys-
temic risk. The reason is simple. The financial system is, for all practical
purposes, infinitely complex, so no matter how intensively we study the
system and how hard we try to control it, we still can focus only on a
tiny part of it—systemic risk emerges precisely where we are not looking.
Paradoxically, the sophisticated risk management techniques increased
the complexity of the financial system, thereby creating new avenues for
crises to emerge.
Take US subprime mortgages, an important contributor to the 2008
crisis. The value of American subprime mortgages was about $1.3 trillion
in March 2007. In normal conditions only a fraction of those borrowers
were likely to default. In a disaster scenario, if half the borrowers default
and only half of the creditors’ money were recovered, total losses would
be $325 billion. While that figure sounds like a lot, it is quite small com-
pared to the overall size of the American financial market. Right before
the crisis in 2007 the outstanding volume of bonds in the United States
was $32 trillion, or twice the GDP, while the total value of the stock mar-
ket was $20 trillion, so a $325 billion loss is about 1.6 percent of the stock
market. A loss of that magnitude on the US Standard and Poor’s 500
index has happened on 1,274 days since 1929, or 5.5 percent of the time.
The actual subprime losses turned out to be much smaller.
How can a worst-case potential subprime loss of $325 billion cause so
much damage when the stock market can suffer losses of $325 billion
multiple times without batting an eye? The reason is that stock market
losses are visible and expected. The mortgage losses were opaque; the

20 SYSTEMIC RISK
risk was hidden, and it caught almost everybody by surprise. I say almost
because some did anticipate it, like the heroes of Michael Lewis’s book
The Big Short.
Two thousand eight was not the first time a financial crisis played out
this way. Financial crises tend to be quite similar to each other, and the
crises of 1763 and 1914 have plenty in common with 2008. Why then did
2008 happen? Hubris and forgotten history. While the policy makers of a
hundred years ago were alive to the danger of systemic risk, their twenty-
first-century counterparts assumed the problem away. I recall being in
central bank conferences in the early 2000s and hearing most speakers
argue that the central banks should focus only on inflation. Financial sta-
bility was not something the central banks should be concerned with
because it is impure, sullying the pristine reputation of monetary policy,
so crucial for price stability. Precisely why the financial authorities were
caught by complete surprise in 2008. The Bank of England, deciding in
the early 2000s that all that mattered to it was monetary policy, proceeded
to close down divisions focusing on financial stability and regulations, so
when the crisis in 2008 started it was desperately short of expertise. The
then governor of the Bank of England, Mervyn King, said in August 2007
after the crisis was already under way and Northern Rock was failing,
“Our banking system is much more resilient than in the past. . . . The
growth of securitization has reduced that fragility significantly.” It wasn’t
only the central banks. Just about the only academic institution con-
cerned with systemic risk was the London School of Economics. Credit
goes to Charles Goodhart, who constantly reminded us that systemic risk
was important and worth studying. Still, it was not good from a career
perspective. I recall getting a referee report from a top journal on one of
my papers on crises in 2003, rejected as “irrelevant because the problem
of crises has been solved.”
The reaction to the Covid virus has its origins in the fumbling response
to the crisis in 2008. Having been caught short in 2008, the policy au-
thorities in 2020 responded with vigor, aided by strong political sup-
port and determined to prevent a repeat of 2008. Policy making is often
like that. Underreaction followed by a rearview-mirror-guided over-
reaction. Unfortunately, though crises are all fundamentally the same,

SYSTEMIC RISK 21
the details differ, and regulations target the details much more than the
fundamentals.
Covid-19 offers a fantastic example of the trade-off between safety and
growth: Do we lock the economy down to prevent the virus from spread-
ing but at the expense of very significant economic damage, or keep
everything open and hope for herd immunity? The answer to that ques-
tion shows why we need political leadership; the authorities concerned
with health tend to prefer lockdown, and the economic authorities to
keep things open. It is the job of the prime minister or president to arbi-
trate and decide on the best course of action. Directing the fight against
Covid-19 is not a job that can be delegated to officials.
We have the same debate in the financial system: Do we seek risk and so
deregulate the system, hoping for more growth, or do we nail the system
down to prevent crises? Both camps have their adherents, and it is the job
of the political leadership to decide.
The Covid-19 and 2008 crises have a lot in common. Like most, they
comprise a four-step process. Willful dismissal of the threats before the
crisis event, followed by a weak initial reaction. When things get so bad
they can’t be ignored, we get overreaction and eventually, as we gain
knowledge and experience, an uneasy balance between safety and growth.
But the differences the two crises are more important. When the Covid-19
was under way I wrote a piece with three of my LSE colleagues, Robert
Macrae, Dimitri Vayanos, and Jean-Pierre Zigrand, titled “The Corona-
virus Crisis Is No 2008,” in which we argued that these two crises were
quite different. The 2008 crisis originated within the financial system,
caused by its willingness to take risk, aided by the willful ignorance of the
dangers and excessive complexity. Covid-19 came from nowhere, certainly
not the financial system. While the policy authorities reacted strongly to
Covid-19, finance was a small part of that, and we haven’t seen any finan-
cial crisis to come out of Covid-19.
So Covid-19 impacted the financial markets, but it cannot be called
systemic in a financial sense. What Covid-19 does do for us is to provide
a handy framework for thinking about financial regulations, what works
and what doesn’t work. I’ll return to that theme once I have erected the
necessary scaffolding to make the argument. But as a brief preview, recent

22 SYSTEMIC RISK
empirical evidence suggests that how the financial authorities reacted to

Covid-19 has sharply increased moral hazard, that is, the financial mar-
kets’ expectation that the central banks are ready, willing, and able to step
in and bail them out. They, as always, will react by taking yet more risk.
Who Is to Blame? Banksters? Politicians?

Regulators? Nobody?
When a systemic crisis happens, someone must be to blame. Is it
the “banksters” who took too much risk and abused the financial system?
Is it the regulators who were asleep on the watch? The politicians who
cheered the banks on and protected them from scrutiny? The households
that borrowed too much and saved too little? Someone made the deci-
sions that led to the crisis. Should they be put in jail?
Take the example of Spain, going through a huge boom after it joined
the euro, fueled by the influx of very cheap money. While some banks
were careful, others were not, especially the savings banks, the cajas. The
lending boom seemed beyond anyone’s control from about 2005, and
everybody enjoyed the party. Who is responsible? The designers of the
euro? The European Central Bank? The Spanish government that joined
the euro? The international banks that lent to Spain? The central bank
of Spain? The Spanish commercial banks? The borrowers? If we try hard
enough, we can always find someone to blame. As Cardinal Richelieu
said, “Give me six lines written by the most honest man, and I will find
something there to hang him.”7
Still, it is not that easy to find someone to blame if we can’t use the
methods of Cardinal Richelieu. My native country, Iceland, was the coun-
try worst hit by the global crisis in the autumn of 2008 and since then
has been trying very hard to find someone to blame. The government
convened a special court to prosecute the then prime minister for allow-
ing the crisis to happen, eventually convicting him of negligence—to
my mind, a miscarriage of justice. Imagine, if we were able to prosecute
politicians for incompetence, there wouldn’t be many left.
The Icelanders had better luck in going after the bankers, some of
whom have spent years in jail. But it took an extraordinary effort, a special
prosecution service, and financial police with considerable resources and

SYSTEMIC RISK 23
very strong political backing, all helped by questionable actions like long
solitary confinement for those accused but not yet convicted. In the end,
the bankers were convicted not for causing the crisis but for minor mis-
conduct relating to individual transactions: just like the US authorities
got Al Capone for tax evasion, not for being a mafia boss. The regulators
and most of the political leadership got off scot-free, even if they are as
much at fault as the bankers. The Central Bank of Iceland, which bore
so much blame for allowing the crisis to happen, fired the governors but
promoted everybody else. So, ironically, the career bureaucrats who were
responsible benefited.
The Icelandic case is an extreme form of what we have seen in other
countries. The prosecutors can go after bankers for specific misconduct,
some abuse of office, but they cannot convict anyone for causing a crisis.
Spain sent its former finance minister and IMF chief Rodrigo Rato to jail
for four years and six months for embezzlement. He was not convicted
for the failure of his bank (Bankia), rescued at huge expense by the Span-
ish taxpayer. He was not punished even for the losses suffered by two
hundred thousand savers whom he had persuaded to buy subordinated
Bankia bonds right before it failed. No, he misused his corporate credit
cards. One might wonder what the Spanish regulator was doing all along,
not only condoning but also actively encouraging such abuse of unso-
phisticated investors. Why aren’t they prosecuted?
The former CEO of Barclays, John Varley, is the only CEO of a global
bank to face charges because of his conduct in the financial crisis. How-
ever, that was only because of how he tried to save his bank, not for
getting it in trouble in the first place. The United States has sent at least
thirty-five bankers to jail for crimes related to the financial crisis, most
relating to small amounts of money at small banks for personal gain. One
person, known as Fabulous Fab—real name Fabrice Tourre—was con-
victed of crimes relating to structured credit products while working as a
junior employee at Goldman Sachs. He and everybody else convicted in
the United States were small-fry.
Even when there is clear abuse we cannot find anyone to blame. The
best case is LIBOR, the standard benchmark interest rate used to decide
on interest rates charged on mortgages and loans worldwide. It was sur-
prisingly easy to manipulate LIBOR, as it was based on an average of

24 SYSTEMIC RISK
banks’ estimates of market interest rates, and the employee making the
submission in each bank had some leeway in the number she came up
with. Is it 5.25 percent or 5.26 percent? A derivative trader who knows
which of these two numbers is to be submitted has more than an even
chance of profiting, and it is alleged that employees across banks col-
luded in their submissions, guaranteeing profits. The banks maintain it
was all the fault of rogue junior employees, and nobody higher up had an
inkling of the abuse. Still, they didn’t question the profits from manipu-
lating LIBOR. Did the regulators know? While strenuously denied, many
observers allege that banks deliberately posted excessively low numbers
during the 2008 crisis with the acquiescence of the regulators in order to
lower funding costs for banks in difficulty.
The manipulation of LIBOR is very costly for lenders and borrowers,
and one might think the abuse would be severely punished. Not so. Em-
ployees in several banks were found to have been manipulating LIBOR,
and some banks have admitted to doing exactly that. Exactly one per-
son has, at the time of writing, been punished: a junior UBS and then
Citigroup employee who was sent to jail by the British authorities for
eleven years. No senior manager, financial institution, or regulator has
been prosecuted or convicted for the LIBOR abuse.
The reason is that it is really difficult to find anyone guilty in a court of
law. The prosecutors in the United Kingdom and the United States have
tried but mostly failed. The bankers might be responsible for causing a
crisis in the court of public opinion, but the Icelandic bankers, like their
counterparts in other countries, were careful not to break the law. Stupid-
ity is not a crime. Greed is not a crime. Legally manipulating the rules
is not a crime. Recklessness was not a crime in 2008, even though it has
become a criminal offense in the United Kingdom since.
The bankers, regulators, and politicians cannot be convicted of a crime
that does not exist, and we cannot (or should not) change the law retro-
actively. When we do a postmortem on financial crises, we see that the
bankers excessively expanded their banks, taking on too much risk, not
properly scrutinizing lending decisions, and ignoring liquidity risk. The
regulators looked the other way, and the politicians cheered the excesses.
It may be greedy, incompetent, arrogant, and immoral, but not, so long
as the letter of the law is complied with, illegal.

SYSTEMIC RISK 25
All we have is specific misconduct, and even then it is really hard to

prove. It involves complicated financial transactions understood only by a
handful of experts and difficult to explain in a court of law to nonexpert
judges and juries. Especially since conviction hinges on a fine interpreta-
tion of rules, the burden of proof is on the prosecutors, and it is not easy
to find people guilty beyond a reasonable doubt.
There is another, more insidious problem in finding banks or bankers
guilty, and that is the problem of too big to fail, illustrated by an example
from the world’s second largest bank, HSBC. In the years before 2010
it failed to monitor transactions involving drug traffickers in its Mexican
subsidiary. Because these transactions were in US dollars they were sub-
ject to clearing in New York and hence to US law. The US government
refused to prosecute HSBC because, as explained by Eric Holder, the US
attorney general at the time, “I am concerned that the size of some of
these [financial] institutions becomes so large that it does become dif-
ficult for us to prosecute them.”8 In other words, if the US government
had decided to go after HSBC it might have triggered its failure, and, be-
cause HSBC is a systemically important financial institution, such failure
might result in a systemic crisis. The largest banks have a get-out-of-jail-
free card. What the US authorities ended up doing is to fine the banks,
and in the aftermath of the 2008 crisis banks have paid over $320 billion
in fines without admitting to any misconduct. The fines have become yet
another revenue source for the government, a tax by another name, borne
by the banks’ clients in the form of higher fees. Other countries haven’t
even gone that far.
Systemic risk is the chance that the financial system does not do its job,
perhaps that a major financial crisis causes an economic recession. It hap-
pens because of the typical dilemma ever-present in the financial system.
We want economic growth, and that necessitates risk. With risk comes
the chance of failure and crisis. It is hard to prevent systemic risk because
it emerges in the most obscure parts of the financial system, exploiting
unknown vulnerabilities, making the policy makers’ job difficult—but
certainly not impossible. We know quite a lot about why crises occur and
how to prevent them. Yet they are all too familiar.

3
Groundhog Day
To Stop the Duke, go for Gold.

—Poster from 1832
In the movie Groundhog Day a TV weatherman played by Bill

Murray is covering the annual Groundhog Day event in Punxsutawney,
Pennsylvania. What was meant to be a one-day assignment turns into an
infinite loop, the same day being repeated endlessly until he finds a way
out. It is the same with banking crises. They happen again and again, and
the financial authorities cannot make them go away. I don’t think they
can either. Even Bill Murray’s solution, virtue, will not work.
26

GROUNDHOG DAY 27
The inevitability of crises gives me one of my favorite exam questions:

“We know what causes banking crises and how to prevent them, and still
they happen all the time. Why?” There are many ways to answer. The glib
reply is “greed and fear,” but perhaps the simplest is that the financial
system has the annoying habit of misdirection, telling us it is okay to take
more risk when we really should not.
We need banks and all the other parts of the financial system, as with-
out them we wouldn’t have anything resembling the modern economy.
The problem is that financial institutions like to take on too much risk,
and bankruptcies and crises are the inevitable outcome. As so often hap-
pens, the very thing that makes the economy grow also can destroy it. We
need a balance between stability and growth, and that balance is tough
to achieve.
It is hard to defend banks. They are a symbol of greed. Of the privatiza-
tion of profits and socialization of losses. Bankers are highly paid, arro-
gant, all too quick to blame the less fortunate for not being as successful
as they are, and lacking in empathy. That makes banks an obvious target
for political activists, and we have seen many attempts by anticapitalist
protesters like Occupy Wall Street to exhort people to go and take their
money out of banks for political reasons—a politically motivated bank
run, an event in which people queue up to take their money out of a bank.
If the activists were to get enough people to run the banks, they would
succeed, and the banks, or even the entire banking system, would col-
lapse. Occupy Wall Street activists, with no luck whatsoever, hope that
that would destroy the capitalist system. It is hard to motivate enough
people to deliberately cause a bank run. The only case I know of in
which such a politically motivated bank run was successful occurred in
1832 when the prime minister of the United Kingdom, the Duke of Wel-
lington, was resisting any changes to how members of Parliament were
selected. The so-called rotten boroughs were popular with the political
leadership. The problem was that entire regions were effectively disen-
franchised, and the demand for reform was strong. Still, the duke resisted
until the launching of a successful campaign under the banner “To Stop
the Duke, go for Gold,” which plastered posters all over the country urg-
ing people to take their paper money to the Bank of England and convert

28 GROUNDHOG DAY
it to gold. Since it had only about 40 percent of the gold necessary to

cover all outstanding paper money, there was a real chance the Bank of
England would run out of gold and, as a consequence, go bankrupt,
while the government quite possibly would default also. Enough people
queued up, and the government was forced to pass the Reform Act.
Do We Need Banks?
Banks are essential for the economy. They provide financial in-
termediation—the channeling of funds from one person or entity to an-
other across time and space. Banks reallocate resources, diversify risk,
allow us to build up pensions for old age, and enable companies to make
multidecade investments. At one end, a large number of savers put small
amounts of money into the banks, expecting their money to be safe and
available on demand. At the other end, the banks make a small number
of large thirty-year loans to companies building factories. Banks are also
dangerous and exploitative. They take advantage of their clients, fail, and
cause financial crises. The response of society is to enjoy the benefits of
the banks while also regulating them heavily.
Easy in theory, hard in practice. If we don’t want banks to fail, they can
make only safe loans, and that means lending only to risk-free govern-
ments—if we can find any. If the banks are too safe, the cost of thirty-
year loans would be too high while interest on bank deposits would be
too low, so people would not save and companies would not borrow.
The factories would not get built and the economy would not grow. We
could not borrow money to buy a car or a house, nor save for old age.
The argument is reminiscent of that made by the Peruvian economist
Hernando de Soto in his book The Mystery of Capital, that a system of
property rights and enforceable legal contracts is a necessary condition
for economic prosperity.
The result is one of the dilemmas we find so often in the financial sys-
tem. If we make the banks too safe, we throttle economic growth. If we
are too eager and unshackle them, the banks fail. Thus we are continu-
ally debating the trade-off between safety and growth. After the Great
Depression we heavily regulated the banks, then deregulated them after
the Bretton Woods system collapsed in 1973, and turned back to regula-

GROUNDHOG DAY 29
tions after the 2008 crisis. There are early indications that the regulatory
pendulum has now started to swing back, with banking regulations to be
relaxed yet again. That process was already under way when Covid-19 hit
in 2020, accelerating the process.
There is nothing unique about the safety or growth debate in the bank-
ing system. It applies to most areas of the public domain, like speed lim-
its, and today it is most strongly manifested in the debate over how to
respond to Covid-19. Do we lock everything down to halt the virus but at
the expense of killing the economy?
Banks are inherently fragile, but it is hard to address that fragility. Sup-
pose a chocolate maker is in difficulty. If the business’s debts exceed its
assets, it is insolvent and will be shut down. Most likely, someone else will
buy the factories and continue production. And if not, competitors will
happily step in. The disruption to society is minimal; shareholders and
perhaps employees lose out but not many others.
Banks are different. They can fail even when prudently run and, con-
versely, remain solvent so long as they are trusted. If I believe my bank is
well managed and properly regulated, I will keep my money in a bank so
it can continue operating. If, however, I lose faith in the bank and take
my money out, that, by itself, can cause the bank to fail, even if it is pru-
dently run and solvent.
A bank run can happen at a bank even if there is nothing wrong with
it. All we need is for depositors to get worried, then a bank run becomes
a self-fulfilling prophecy. I remember seeing a news item on CNN con-
cerning a bank about to be closed down. The journalist made a mistake,
using an image of a different bank. That is all it took for the other bank
to be hit by a run.
Why are banks fragile? Two main reasons. The first is bank runs. If a
sufficient number of depositors want their money back, the bank cannot
fulfill all those requests because most of its assets are tied up in long-
term loans. I can take my money out of the bank anytime I want, but my
bank cannot and should not be able to call in its thirty-year loans when
it pleases.
My favorite description of a bank run comes from the 1946 movie
It’s a Wonderful Life, featuring James Stewart, who plays a banker faced
with a bank run during the Great Depression. In the following scene, he

30 GROUNDHOG DAY
addresses the angry crowd demanding their money back. Though the
transcript doesn’t do it justice, you can (at the time of writing) find it on
YouTube: “No, but you, you . . . you’re thinking of this place all wrong.
As if I had the money back in a safe. The money’s not here. Your money’s
in Joe’s house . . . right next to yours. And in the Kennedy house, and
Mrs. Macklin’s house, and a hundred others. Why, you’re lending them
the money to build, and then, they’re going to pay it back to you as best
they can. Now what are you going to do? Foreclose on them?”1
When a bank is hit with a run it may fail. It is not insolvent as it has
more assets than liabilities, but it is illiquid—it cannot convert its assets
into cash on demand. It is a little like a situation in which I suddenly
need a large amount of money, far more than what I have in my bank
account. While I own a house in London, it will take time to sell it,
and I cannot satisfy an immediate demand for a large amount of cash. If
depositors become worried about their bank’s solvency—perhaps it has
made too many bad loans—depositors will want their money back. If
enough depositors agree, they will queue up in front of the bank to get
the money—the classical definition of a bank run.
The second bank fragility arises because of how banks create money.
Every country in the world uses fiat money: money that is the creation
of the central bank as an agent of the government and underpinned by
its stability and reputation. Money takes many forms. The central banks
create the monetary base, M0, consisting of money held on account
with the central bank plus the total amount of physical money: notes and
coins. That is, however, only a fraction of the overall amount of money
in the system, the reason being we have a fractional reserve system where
the banking system creates money. Suppose the reserve requirement is
1 percent. If I deposit €100 into a bank account, the bank has to hold on
to €1 (reserve requirement) and can lend out €99. I still own my €100 and
can spend it whenever I like, but now the borrower has €99, and she can
spend it also whenever she likes: together we have €199 ready cash, known
as M1. If the borrower then leaves the money in her bank account, her
bank can lend out 99 percent of that, €98, and that can, in turn, be further
lent out, etc., etc. This is the most basic way the banking system creates
money. So what are the amounts?

GROUNDHOG DAY 31
In the eurozone in August 2018 base money was €3.2 trillion, but the
amount of physical money in circulation plus demand deposits was M1
(€8.1 trillion), M2 (€11.1 trillion) further adds savings accounts, and M3
(€12.0 trillion) is large amounts of money that are locked for a period of
time deposits, institutional money market funds, short-term repurchases,
and other, similar assets. Every euro created by the European Central
Bank becomes €3.4 when the banks are done with it. Even that captures
but a fraction of the money in the system because as we go about our
daily economic life we constantly create new money simply by borrow-
ing and lending. Nobody knows how much money is in the system, and
certainly nobody controls it. That is why it is so hard to control inflation.
The amount of money has a direct impact on the fortunes of the econ-
omy. If the economy is growing rapidly, the supply of money needs to
grow with it to prevent deflation. If, however, the supply of money col-
lapses, there is not enough money to maintain economic activity, and
the economy goes into recession. Now you see why it is dangerous when
banks are clamoring for liquidity in a crisis. They take all the higher forms
of money, M1, M2, M3, and beyond and convert them into M0. The sup-
ply of money is collapsing, taking the economy down with it.
These two vulnerabilities, bank runs and the nature of money, come
together in banking crises. A bank run can lead to cascading failures be-
cause depositors might view a single bank’s bankruptcy as a symptom of
system-wide difficulties. Depositors have limited information about the
quality of banks’ assets and may feel that if hidden problems have been
allowed to fester in one bank, they might also be present in other banks.
They see bankers as incompetent and greedy and the regulators as bun-
gling and even corrupt.
The best-known example of bank runs causing a systemic crisis hap-
pened during the Great Depression of 1929 to 1933. The United States
lost over a third of its banks to bankruptcies, and people preferred to store
their money under their mattresses rather than keeping it in banks—the
backdrop to It’s a Wonderful Life. The panic originated in 1931 with the
Bank of United States (which specialized in immigrant clients) mislead-
ingly and deliberately implying by its name that it was government-
owned. In early 1931 rumors spread that it might be in trouble. Initially,

32 GROUNDHOG DAY
the local authority—the New York Federal Reserve Bank—tried to ar-

range a rescue package but did not manage to put it together. The fail-
ure of the Bank of United States marked a profound change in public
sentiment toward banks. A real estate bubble in Chicago collapsed soon
after, in May 1931, and thirty Chicago banks defaulted. Depositors, who
were unable to tell whether a bank was good or not, began pulling their
cash indiscriminately out of all banks, good and bad, triggering multiple
bank runs.
Such withdrawals caused the supply of money to collapse. Each M0
dollar in 1928 was amplified into $6.5 by the financial system. By 1933 the
corresponding number was $4 and continued to fall for a decade after-
ward. That meant that while the amount of M0 increased in the Depres-
sion, M2 fell, from $46.4 billion to $32.2 billion, with devastating con-
sequences for the real economy. Milton Friedman and Anna Schwartz,
in their seminal book A Monetary History of the United States: 1867–1960
(1963), argued that the collapse in the supply of money was a leading
cause of the Great Depression. The reason we got all the liquidity created
in the crisis in 2008 is owing to the lesson from the Great Depression and
the arguments in Friedman’s and Schwartz’s book.
A vicious cycle ensued. All banks felt the need to protect themselves,
and desperate bankers called in loans, which the borrowers did not have
the time or resources to repay. With economic prospects poor, capital
investment ceased, and the surviving banks became even more conser-
vative. Servicing debts became harder because prices and incomes fell
significantly while debts remained at the same dollar amount. If this all
sounds familiar, it is not unlike what happened (but on a smaller scale) in
the aftermath of the 2008 crisis.
Fortunately, banking crises are not all that frequent, contrary to what
many think. Figure 5 shows the year when every banking crisis in the
world starts from 1800.2 Two thousand eight has the highest number of
new banking crises at twenty-two, followed by 1931 with eighteen. Still,
since the number of countries in 2008 is much higher now than in 1931,
a much higher fraction of countries in the world suffered a banking crisis
in the Great Depression. The most dangerous decade is the 1990s, with
seventy-four new banking crises.

GROUNDHOG DAY 33
Figure 5. Number of banking crises in each year from 1800 to 2017.

Credit: Lukas Bischoff/IllustrationX.
Banking crises are not pleasant. Growth turns negative, bankruptcies

abound, unemployment skyrockets, and general misery spreads. While the
immediate cause of banking crises is the interplay between bank runs and
the fractional reserve banking system, that is only the technical endgame.
The fundamental causes are deeper, and the vulnerabilities are created
years, even decades, before the actual crises hit the newspaper headlines.
The origin is usually a stable economy that we perceive as having low
risk. Since risk is seen as low, everybody thinks borrowing to invest or
consume is a fantastic idea. Banks are highly profitable, and all that invest-
ment and consumption make the economy grow. As if by magic, we are
always right. The investors must be really smart, getting into the market
at the right time, and the bankers are brilliant. But eventually the only
thing keeping the growth alive is borrowing. As time passes it becomes
harder and harder to find productive investments, and in the endgame
the money goes to speculative real estate and the like. And that cannot
go on forever: a crash is inevitable. Most crises are as simple as that. Sup-
pose instead we had played it safe and not borrowed. Then the economy
would not have grown, and we did the right thing, this time by playing it
safe. We are right either way, and our choices are self-validating, at least
in the short run.
There are exceptions. Some crises, but not many, are caused by cor-
ruption and outright theft. I can only think of three, and I keep asking

34 GROUNDHOG DAY
people if they know better examples. Please get in touch if you know of
any. Two are from Latin America: Venezuela in 1994 and the Dominican
Republic in 2003, in both cases involving banks that were so large as to
be systemically important. The banks did not record deposits as liabilities;
instead, the insiders (senior management and owners) looted the banks
from within by stealing their assets. Because of the banks’ systemic im-
portance, the central bank felt it was necessary to compensate depositors.
At a high cost: the macroeconomy’s destabilization. In Venezuela the
rogue bank paid high interest rates, forcing other banks to do the same,
thereby weakening the entire banking system. More recently about a bil-
lion dollars disappeared over two days from three Moldovan banks that
had been taken over by unknown owners in 2012. The money was sent to
British- and Hong Kong–based companies with equally opaque owner-
ship. The Moldovan government was forced to bail out the banks at the
hefty cost of 15 percent of GDP.
Other crises have been caused by external events like wars, natural di-
sasters, or a change in the political system. Even viruses. Some have their
roots in disastrous economic policies, as in Venezuela today. However,
these origins of banking crises are uncommon and, since they are caused
by outside forces, hard to prevent. Even if the banks are safe, the crisis
storm can be so strong that it destroys everything in its wake, leaving little
for financial policy to do except to mitigate the worst.
The most common type of banking crisis is caused simply when banks
make too many bad loans. There is no reason such a crisis couldn’t be pre-
vented—in theory. There are plenty of signs, including excessive buildup
of credit, a boom in real estate, relaxed lending standards, and economic
well-being that is at odds with the economy’s fundamentals. But it is very
hard to do something about it.
Banking crises remind me of the old Road Runner cartoons in which
Wile E. Coyote is chasing the Road Runner and, on occasion, Coyote
runs off a cliff. He keeps running for a few meters until he looks down
and sees there is nothing but air beneath him. Bankers do that too (Fig-
ure 6). Like Wile E. we believe everything is okay until it is too late,
and there’s nothing we can do except manage the crisis. The reason was
voiced best by the former CEO of Citigroup, Chuck Prince, who, when

GROUNDHOG DAY 35
Figure 6. Running off a cliff. Credit: Copyright © Ricardo Galvão.
asked before the 2008 crisis why nobody stopped all the excess, said, “As
long as the music is playing, you’ve got to get up and dance.”3
Booms precede most banking crises. Everybody enjoys the benefits.
The economy is growing. Everybody feels richer. The politicians, policy
makers, and bankers must be geniuses. The financial system tells us what
we want to hear: we are doing the right thing, and we are really, really
smart. All agree until it all goes horribly wrong. Some know better. Ex-
perts in the central banks, inquisitive journalists, and the bankers who
write the crazy loans. But it is not in their interest to issue a warning. If
they do, they risk being denounced, losing their income, and even being
subjected to prosecution. So nobody speaks out, and the party continues.
The former head of the Federal Reserve wrote in the 1950s that the Fed’s
most important job is “to take away the punch bowl just as the party gets
going.”4 It is really hard to do so.
One of the main causes of banking crises is financial liberalization. It
can be enormously valuable for a country to become a global financial
center. The City of London and Amsterdam were the key to the success
of the British and Dutch empires and their prosperity ever since. On the
face of it, liberalization can seem like a sensible idea. Enjoy the fruits of a
financial system that is open for business, welcoming companies from all
over the world. In some countries the attraction is to unshackle lethargic

36 GROUNDHOG DAY
domestic banks and encourage them to invest in the domestic economy,

especially the small- and medium-sized enterprises that are the drivers of
job creation and economic growth. Elsewhere, the appeal is all the high-
paying jobs that come with international finance.
By converting a country into an international financial hub, vast riches
await. We only have to look at the wealthiest country in the world, Lux-
embourg, which got rid of its inefficient mining and industrial base and
replaced it with finance, very successfully and with lightning speed, in
1973. Singapore has even gone so far as to give the central bank a mandate
to develop the financial sector, which so far has been quite good for Sin-
gapore. Except it isn’t that easy. Many a country has tried financial liber-
alization, and most experiments have ended up in tears. Cyprus, Iceland,
and Thailand are among the many cautionary tales. The reason is that we
face one of those dilemmas we find everywhere in finance: the benefit is
the same as the vulnerability.
Some countries can sustain advanced financial systems for centuries.
The three cities that were the global financial centers in the sixteenth
century have kept their status to this day: Amsterdam, London, and
Hamburg (if we allow for Hamburg losing its financial industry to Berlin
and eventually to Frankfurt after it joined the German Empire in 1871).
Something in these countries’ DNA allows them to make a success out
of it. They have specialized schools focused on finance, a stable political
leadership, a robust legal system that understands finance, government
agencies familiar with regulating high finance, and banks with experience
operating in lightly regulated environments. It confirms what my sociolo-
gist colleagues like to stress: culture matters.
Large financial systems are certainly beneficial to those lucky enough
to have them. Dutch merchants’ success in the seventeenth century and
the British industrialization in the nineteenth century would not have
been possible without a robust financial system backing them up. Finance
is the source of the highest-paying jobs around and creates a lot of tax
revenue. The UK is an example: the House of Commons found in 2017
that the gross value added by the financial system to the UK economy is
£124.2 billion a year, 7.2 percent of the total. The financial sector provides
1.1 million jobs. Even better, the UK has a large trade surplus in finance:
exports of financial services were £55.5 billion and imports only £11.7 bil-

GROUNDHOG DAY 37
lion. The financial sector contributed £71.4 billion in the tax take, 11.5 per-
cent of government revenue.
No wonder many countries aspire to do the same. That said, it is not as
easy as it sounds, and the execution has to be just right. A common mis-
take is to reduce oversight and activity restrictions but maintain implicit
or explicit government guarantees such as deposit insurance. This creates
a nasty moral hazard problem because it can enable financial institutions
to borrow cheaply and use the money for high-risk activities, all implic-
itly or explicitly underwritten by the taxpayer. This is what happened in
many Asian countries, such as Thailand and South Korea, that got into a
crisis in 1998. Their banks borrowed abroad, and because of government
guarantees and lax oversight the banks did not care whom they lent the
money to. Not surprisingly, it did not end well. It is much better to learn
from the one country that made a recent success by becoming a financial
center, Luxembourg. I was once in a panel discussion there, along with
the governor of the central bank and other experts. When the governor
learned I was from Iceland, he snickered and said that Icelanders forgot
the first lesson of creating an offshore financial center: protect the coun-
try from financial institutions’ failures.
The typical outcome is that following liberalization, banks overexpand,
artificially inflating asset prices, creating positive feedback loops among
bank lending, market prices, and profits. Meanwhile, the banks lack ex-
perience in managing risk and are disdainful of risk management, seeing
it as a loss-maker that gets in the way of making money. The regulators
are similarly ill-prepared, focusing on the positive outcomes while missing
the signs of excessive risk because they have never been there before. And
the political leadership always keeps the regulators on a short leash, not
allowing anything to spoil the party. Meanwhile, government policies are
accommodating, interest rates are low, and governments cut taxes and
increase expenditure because the economy is booming.
The financial crises in Iceland and Ireland in 2008 had precisely these
roots. Even countries with a culture of a liberalized, competitive financial
system can make the same mistakes, like the United States in the Sav-
ings & Loan (S&L) crisis. Mortgages were the raison d’être of the S&L
banks—banks like that used to be common in many countries, often
under the name of savings banks or something similar. The crisis started

38 GROUNDHOG DAY
when the sleepy S&Ls found it difficult to cope with all the financial
turmoil in the 1970s. Inflation was high and increasing, interest rates did
not keep up, and banks that focused on collecting deposits and making
mortgages suffered increasingly large losses. The financial authorities de-
cided to deregulate the industry with the view that the S&Ls would grow
their way out of trouble; at the time, deregulation was de rigueur in the
United States. The intention was to allow the S&Ls to expand into the
parts of banking services previously closed to them.
The financial authorities allowed the S&Ls to use lenient accounting
rules, eliminated restrictions on the minimum numbers of stockholders,
and reduced the level of oversight. However, crucial to the story, the
government continued to provide deposit insurance, guaranteeing that
depositors would get their money back if the S&Ls went bust. Many
S&Ls took advantage, often falling into the hands of rogue bankers. The
best known was Charles Keating, who got into the business because “I
know the business inside out, and I always felt that an S&L, if they’d relax
the rules, was the biggest moneymaker in the world.”5 Initially, his S&L,
Lincoln, grew rapidly, but its investments, not least in real estate and
junk bonds, turned out not to be as good as he’d hoped for. Eventually,
Lincoln was closed down, with the cost to taxpayers exceeding $3 bil-
lion. Keating, for his troubles, lost all of his wealth and spent 4½ years
in prison. Eventually, the total cost of resolving all the failed S&Ls was
$160 billion, including $132 billion from taxpayers.
After the United States lost so many of its banks in the Great Depres-
sion, it took Franklin Roosevelt’s election as president to change things.
He pushed Congress to pass the 1933 Banking Act in June 1933, establish-
ing the Federal Deposit Insurance Corporation (FDIC), which provided
insurance coverage for deposits up to $2,500. This proved quite effective,
and the bank runs stopped. Europe had its share of bank runs in the Great
Depression, not as many as the United States, but, unlike those in the US,
the most serious were deliberately caused by the French government. It
all started when the German and Austrian governments announced their
intention to create a customs union in March 1931. The French did not
like that very much and forced their banks to run the Austrian banks—in
effect, run Austria. France could do that because it had been deliberately
undervaluing its currency for quite some time and, as a consequence,
had the biggest gold reserves in the world. The largest bank in Austria,

GROUNDHOG DAY 39
Credit-Anstalt, consequently defaulted, and the crisis spread to other

Austrian banks and soon after to German and Hungarian banks. The
European bank failures triggered by the French government were a sig-
nificant contributor to the subsequent political instability and depression
in Europe. While France succeeded in its immediate political objective,
that is, preventing the customs union, it paid a heavy price both politi-
cally and economically.
So, what to do about bank runs? Deposit insurance as eloquently mod-
eled by Douglas Diamond and Philip Dybvig. Their model concludes
that so long as depositors believe they will be protected they will not run
their bank; a credible deposit insurance scheme will consequently never
actually need to pay out. In the United States today, the FDIC continues
to provide deposit insurance to depositors up to a limit, now $250,000,
payable on the day a bank is closed. I once saw that in action while visit-
ing a friend in Houston, Texas. He was working for a company whose
bank went bust. On the same day the bank was closed, he got a phone call
saying he could collect his deposit insurance check. We went to the bank,
queued up for half an hour, and got a check for the full amount. Amaz-
ingly efficient. This efficiency and the public recognition of it have been
very beneficial to the financial stability of the United States. If Europe
had such an efficient system, some of the drama in Europe’s 2008 crisis
would have been averted, and the Northern Rock run would not have
embarrassed the Bank of England and the British government.
In the good old days (or at least the textbook version of the good old
days) a bank was an institution that collected deposits that were used
to make loans. The money came from what is known in the jargon as
retail clients, you and I. Banks today are just as likely to borrow from
other financial institutions, what is known as wholesale funding. That
creates new sorts of vulnerabilities because wholesale lenders are much
more likely to be aware of pending difficulties and can be very quick in
pulling their money out. This was the root of the last bank run in the
United Kingdom, Northern Rock in 2007—thereby embarrassing the
British authorities who had forgotten the lessons of their previous bank
run (Overend & Gurney in 1866).
The run was caused by the cleverness of Northern Rock management
which had captured about a third of the British mortgage market by us-
ing an ingenious business model. In the old-school model of banking,

40 GROUNDHOG DAY
Figure 7. Northern Rock’s business model. Credit: Lukas Bischoff/

IllustrationX.
people deposit money in banks that then make mortgages. Not Northern
Rock. It found a new model (Figure 7). Suppose Northern Rock bor-
rowed £100 million for three months from the wholesale markets, using
the money to make one thousand mortgages. It then bundled the mort-
gages into a structured credit product that it sold to investors and used
the proceeds to repay the three-month £100 million loan. Highly profit-
able. But there was hidden liquidity risk.
All was fine so long as Northern Rock was able to sell the mortgages—
if it could not, the bank would be forced to default on the initial three-
month loan, which eventually happened in the summer of 2007. It wasn’t
Northern Rock’s fault when the credit markets worldwide froze; it was
said that investors went on strike, just like their Amsterdam ancestors
back in 1763. Nobody wanted to buy structured credit products, and the
first victim was Northern Rock. It took a few months, but its impend-
ing demise was known to everybody in the credit markets except, as it
appears, the bank’s regulators: the Financial Services Authority and the
Bank of England. Eventually, the authorities tried to resolve the crisis
behind the scenes.
The Bank of England then made an unfortunate mistake, announcing
in October 2007 that Northern Rock was in difficulty and receiving sup-
port from the Bank of England. The Bank’s decision makers seem to have
believed that the public would find the announcement reassuring—all is
fine, no reason to panic, we know what we are doing and are here to pro-

GROUNDHOG DAY 41
tect you. It didn’t quite work out that way, and the following day people
around the United Kingdom queued up to get their money out of North-
ern Rock—the first bank run in Britain for a century and a half. The Bank
of England should have known better. The same mistakes happened with
the Reconstruction Finance Corporation in the Great Depression and
many times before and since. We have to trust the banks, regulators, and
the government. Unfortunately, by the time the crisis comes along, the
trust has long evaporated. Not surprisingly, Northern Rock’s retail clients
called the bluff.
Northern Rock went through two waves of bank runs, first by the so-
phisticated wholesale investors in July 2007 and then by unsophisticated
retail investors in October. The wholesale run shows that the financial
markets had a much better understanding of the bank’s problems than
either the supervisor or the general public. One reason for the October
run was that the British deposit insurance scheme was relatively weak, and
depositors felt they had no choice but to queue up and take their money
out. After the first £2,000 of deposits, the scheme protected only 90 per-
cent of savings of up to £33,000—guaranteeing a maximum payout of
£31,700. Even worse, it would take several months to get the money. This
so-called coinsurance was intended to incentivize depositors to monitor
the banks. Well, given that the Financial Services Authority missed the
problems at Northern Rock, it would be surprising if ordinary deposi-
tors with less information could do better. The only sensible strategy for
depositors was to run the bank. Contrast this with the American deposit
insurance scheme and my friend’s experience in Houston. The Northern
Rock run would not have happened there.
With the benefit of hindsight, it is clear that Northern Rock’s failure
was inevitable given time. It remained only solvent so long as it could
borrow without limits. Because there were no limits on borrowing,
Northern Rock and most other banks in the world behaved in a way that
ensured that credit would dry up. The failure of Northern Rock was a
self-fulfilling prophecy.
In response, the British authorities overreacted, announcing unlimited
deposit insurance to reassure bank clients and prevent contagious runs
across the United Kingdom. This demonstrates the problem created when
the authorities have inadequate policies in place and then are confronted

42 GROUNDHOG DAY
by a crisis. The United Kingdom went from having a deposit insurance

scheme that was too weak to one that was too strong. Underreaction, fol-
lowed by overreaction, is a common problem in policy making.
The unlimited guarantee soon came to haunt the authorities, aided by
the fact that the internet speeds up the bank run process—even worse,
making them when the regulators are not at work. In 2007 and 2008
an Icelandic bank (Icesave) facing difficulties borrowing from the capital
markets hit on the ingenious idea of opening high-interest accounts in
the United Kingdom. By promising interest rates considerably above the
prevailing market rates, Icesave saw money flooding in. Because the Brit-
ish government provided unlimited deposit insurance, nobody was too
concerned. Eventually, when the crisis hit its maximum intensity at the
end of September 2008, depositors did get worried and ran the Icesave
bank. Because it was an internet bank, the bank run was not only much
faster than the classical ones (in which people have to physically queue up
in front of a bank), it also inconveniently took place over a weekend when
the supervisors were off duty.
The British authorities were not the only ones who forgot the histori-
cal bank run lessons. The Troika—the European Commission, the Eu-
ropean Central Bank, and the International Monetary Fund—made the
most elementary of mistakes when dealing with the Cypriot crisis in 2013.
Most European crisis countries got into trouble for typical reasons: Ire-
land, Spain, and Iceland from bank overexpansion, Greece and Portugal
due to misbehaving governments. Cyprus was different. It was a victim of
two unique policy mistakes.
First, the Cypriot banks’ business model was unusual, as they didn’t
borrow from the wholesale markets. It all started when the Cypriots,
looking for an industry, realized that becoming an offshore banking cen-
ter was a marvelous idea. If it works for Singapore, why not Cyprus? They
started to collect deposits from wealthy foreign individuals, primarily
Russians, becoming an offshore haven inside the European Union. The
Cypriot banks then had a quandary: what to do with all the money? In a
typical banking model, banks lend to households and companies and the
government. But there is only so much lending Cyprus can absorb, so the
Cypriot banks decided to invest in the sovereign bonds of the European
country next to them, Greece. It seemed like a good idea. The prevailing
attitude among the financial markets and the authorities was that Euro-

GROUNDHOG DAY 43
pean sovereign debt was risk free, that somehow it was all insured by the
collective. This rather unfortunate misunderstanding was reinforced by
law in a European Union directive stipulating that sovereign debt was,
and is, recorded as risk free in banks’ accounts. The European Commis-
sion told the Cypriot banks that they were to consider Greek government
bonds safe on their books, and that is where they put their money. Think
about it for a second. Sovereign debt is, in fact, risky, and governments
are free to default on it, but the banks buying the debt are obliged to
consider the debt entirely safe. Talk about mixed signals.
When the crisis started, 37 percent of the Cypriot banks’ deposits be-
longed to foreigners, 80 percent of whom were Russians. Then the Greek
sovereign debt crisis happened, and in the second bailout, in 2011, the
owners of Greek sovereign debt were made to suffer a 50 percent haircut.
At that moment it became clear that the Cypriot banks were the walking
dead, so a slow-running but accelerating bank run ensued. As so often is
the case, those in the know took their money out first. It didn’t happen
quickly: Greece defaulted in early summer 2011, and it wasn’t until early
spring 2012 that the Cypriot banks finally started to run out of money.
The Cypriot government refused to recognize the problem at first, and
the Troika did not press the issue. A crisis was clearly about to happen,
and nobody did anything. Not the Cypriot government, not the IMF,
not the European authorities. Still, that is not the dumbest mistake they
made, as a dumber one was to follow.
Now it became really interesting. The typical way to resolve a failing
bank is to let junior creditors take the first loss. However, the Cypriot
banks didn’t have any bondholders—all they had was depositors. That
left the government in a quandary. Whom to hit for the money? Un-
der European regulations, depositors with less than €100,000 in a bank
are fully insured. In the crisis meeting with the Troika and Ecofin, the
Cypriot authorities maintained they did not want to excessively hurt the
foreign depositors, so key to the country’s business model, one from
which the political and regulatory leadership personally profited. Instead,
they opted to hit all depositors with a 6.75 percent tax, including insured
deposits.
Why would they do such a thing? So that offshore clients would not
suffer, as otherwise they might leave and never come back. Why would the
European authorities agree? The best explanation is that the emergency

44 GROUNDHOG DAY
meeting was held on 25 March 2013, and by four in the morning, when
the decision was made, the policy makers were really tired. When the
press release eventually came a few hours later, it was pointed out that
insured depositors throughout Europe would no longer think they were
fully protected, so everybody would rush to run the banks when the next
crisis came along. Not surprisingly, the authorities quickly backtracked.
It was perhaps the dumbest policy blunder in the whole crisis. Given that
everybody knew for almost a year that the banks were failing and that
the importance of confidence in preventing bank runs is well known, it is
incomprehensible that no authority, neither Cypriot, European, nor the
IMF, prepared for what was an entirely foreseeable and avoidable crisis.
Banks and governments have a symbiotic relationship. Banks are always
encouraged (and usually required) to buy government debt. Banks are
also the primary source of financing for small and medium-sized compa-
nies, the main driver of economic prosperity, and a source of significant
tax revenues. In good times this relationship is virtuous. Profitable bank-
ing and increasing risk go hand in hand with growth and rising govern-
ment revenues. This is when governments should pay down their sover-
eign debt, the one saving grace of the Irish and Icelandic economic policy
before their 2008 crisis. If only Britain and the United States had done
the same thing.
This virtuous cycle can quickly turn vicious, and it can be the fault of
the banks, the governments, or both. A crisis in the banking sector will
put government finances under considerable strain: directly because of
the cost of providing bailouts and indirectly since banks slow down their
financing of the economy. If things get especially bad, a banking crisis can
culminate in a sovereign debt crisis, where the government cannot meet
its obligations. That is what happened in Ireland and nearly in Spain and
Iceland too.
Similarly, a government facing financial difficulties can cause problems
for the banking system. It may need to increase taxes or spend less, which
then slows down the economy. A more direct channel is when banks
own a lot of government debt. If the government’s credit rating gets
downgraded, the banks’ holdings of sovereign debt are immediately and
adversely affected, so their riskiness goes up. Governments know this, and
most have a law saying that government debt is risk free. However, that

GROUNDHOG DAY 45
has the unfortunate side effect of subsidizing government debt at the

expense of lending to productive private sector companies, further slow-
ing growth. If it goes too far and the government defaults, it can take the
banks down with it, precisely what happened in Cyprus and is likely to
happen in certain European countries, such as Italy.
This process is known as the bank–government doom loop, a major
cause of the European crisis in the past few years. While there is no easy
way to deal with the bank–government doom loop, it can be mitigated
by reducing banks’ holdings of their own government debt. Banks need
to hold some safe assets, but there is no reason Italian banks should not
hold their safe assets in the form of German government bonds, except
the Italian government needs someone to buy its own bonds. There-
fore, it is unfortunate that the opposite has been happening in Europe;
banks are buying more of their government debt, not least because of the
way the European Central Bank has to do quantitative easing. When the
Bank of England does quantitative easing, it can directly buy gilts, British
government bonds. It is the same with the Federal Reserve, the Bank of
Japan, and almost every other central bank. It is not so with the European
Central Bank. Until recently it was prevented from buying government
bonds by its rules, so if it wanted to do quantitative easing it had to give
money to other entities, which then bought government bonds. And
those entities are the banks. So, Italy’s banks were obliged to buy Italian
government bonds and use those as collateral to borrow from the Euro-
pean Central Bank.
How serious is this problem? The think tank Bruegel collects bank
holdings of sovereign debt for some countries. Only 4 percent of US
sovereign debt is held by its banks, while the corresponding numbers for
Germany and Italy are 23 percent and 20 percent, respectively. Not a seri-
ous problem for Germany because of its low debt, but not so for Italy be-
cause of its very high debt levels. A sovereign debt crisis in Italy will most
likely trigger its banks’ failure in a typical bank–sovereign doom loop.
Crises are costly. The government spends large amounts of money
dealing with them, and they affect economic growth. The first type of
cost is termed the direct impact and can be more or less accurately identi-
fied, while the longer-term economic consequences, the indirect impacts,
are much harder to measure. A common way to put a number to it is to

46 GROUNDHOG DAY
Figure 8. UK cost of crisis scenarios. Credit: Lukas Bischoff/IllustrationX.
estimate output losses by extrapolating from the trend of real GDP be-
fore a crisis and calculating the difference between the trend prediction
and actual outcomes. We can see the direct and indirect costs in the IMF
crisis database. The costliest crisis was Indonesia’s in 1998, with direct
costs of 57 percent of GDP, and the highest indirect costs happened in
the Finnish 1991 crisis, at 75 percent of GDP. However, while the direct
impacts are fairly accurately measured, the indirect impacts are likely to be
overstated. The reason is that economic growth tends to be abnormally
high in the years before the crisis because of the financial excesses that
culminated in the crisis. That is why many commentators on the 2008
crisis overstate their case. The real GDP in the United Kingdom is pre-
sented on the accompanying graph (Figure 8). If one directly extrapolates
from economic growth before the crisis, the indirect cost would be £212
billion. However, if a more prudent policy path would have led to lower
growth, there actually might have been a gain from the combined precri-
sis excess and postcrisis decline. There is no right way to identify which
outcome is more likely.
The IMF crisis database was initially developed by the World Bank, and
some of the early developers of the database, including Patrick Honohan
and Daniela Klingebiel, published an interesting paper focusing on the
fiscal costs of banking crises and how the various government responses
contribute to the cost or resolution. They find that if countries do not
extend some policies of unlimited deposit guarantees, open-ended liquid-
ity support, repeated recapitalizations, debtor bailouts, and regulatory

GROUNDHOG DAY 47
forbearance, the fiscal costs of resolution would be around 1 percent of

GDP on average, one-tenth of the actual costs. However, if governments
employ all of these approaches the fiscal costs will be six times larger. We
like to waste money on resolving financial crises. The primary damage to
the economy brought on by a crisis may not be caused by the actual crisis
event but by inappropriate government responses. It is interesting that
Patrick Honohan later became the governor of the Bank of Ireland, fac-
ing very high fiscal costs (caused by the sort of mistakes identified in his
study) by the Irish authorities before he took office.
It is also hard to estimate the cost of inaction. One of the most contro-
versial decisions in the 2008 crisis was allowing Lehman Brothers to fail.
Some commentators maintain that this was a colossal mistake, one that
set in motion the subsequent damaging events. Others argue that bail-
ing out Lehman would only have delayed the inevitable. Both arguments
have considerable merit (I lean toward the latter). But a silly interpreta-
tion of Lehman’s failure is often expressed by European politicians who
like to blame it for all the ills that came after, including the European
crisis and, in particular, the Greek bailout. Lehman haunts us to this day.
One reason for all the Covid-19 bailouts is that the consensus in the policy
makers’ circle is that Lehman’s failure is at the root of all the horrors that
happened afterward. If only Lehman had been bailed out.
The late 1980s and early 1990s were a time of banking crises. While the
most spectacular was the blowup of the Japanese banking system in 1992,
the Scandinavian crisis is more interesting. Three Scandinavian countries,
Sweden, Norway, and Finland (well, Finland is not Scandinavian if you
go by strict geography, but it identifies as such), suffered a severe banking
crisis in the early 1990s. The reason for it was that these countries liber-
alized their financial markets and implemented expansionary macroeco-
nomic policies in the late 1980s, including removing caps on lending and
interest rates and encouraging more competition, all while reducing the
level of oversight. The result was a huge lending boom in the late 1980s,
followed by a crash in the early 1990s. While the chain of events is depress-
ingly familiar, what is interesting is the way the Scandinavians (especially
the Swedes) dealt with the crisis.
The ideal way for governments to resolve a banking crisis is the good
bank/bad bank method, perfect in theory, hard in practice. Split up a

48 GROUNDHOG DAY
failing bank, separating the dodgy assets into one institution (the bad
bank) while keeping most of its operations and solid assets in the good
bank. The bad bank becomes an asset management firm; the government
aims to sell the good bank but hold on to the bad assets until they expire.
If the dodgy assets are valued cheaply during the crisis when the bad bank
is established, as they tend to be, the government potentially can make
significant profits, an argument often used to justify this approach to tax-
payers. However, if the bank was insolvent, the bad bank must, by defini-
tion, have a negative value, so a profit for the government is not the ex-
pected outcome. Realistically, the taxpayers should expect to lose money,
but hopefully the benefits from having a well-functioning bank replacing a
failing bank outweigh the expected loss. The Swedes eventually kept their
crisis costs to 4.3 percent of GDP, a fraction of what was initially feared.
It is difficult to implement such ruthless efficiency because all the spe-
cial interests pull it in different directions—everybody wants a bailout.
Most governments find it hard to follow the Swedish example, and two
factors determine whether they can. The first is how much faith we have
in the government’s ability to implement sensible policies that benefit
us in the long run. Do we trust the government to do the right thing
or see it as incompetent or corrupt, its every initiative to be resisted, no
matter how sensible they are. Second is the degree of democracy, that is,
whether the government has to take the will of the people into account.
In democracies where the trust in government is low, like Italy, it is dif-
ficult to deal with banking crises, while the Swedish government was able
to make difficult decisions in the 1990s because the Swedes trusted their
government.
The trade-off can be easily depicted (Figure 9). The financial authori-
ties need to be competent, determined, and given sufficient political cover
to deal efficiently with a financial crisis. Muddling through and wishing
the problem would go away are more common, the end result often be-
ing zombie banks. That is when banks are insolvent, but the governments
cannot close them down, like parts of the Italian banking system today.
The term “zombie bank” was first used when discussing the S&L crisis in
the 1980s, but it became widespread when referring to the Japanese crisis
in the 1990s. Instead of shutting down or recapitalizing its failed banks,

GROUNDHOG DAY 49
Figure 9. Ability to execute policy. Credit: Lukas Bischoff/IllustrationX.
the Japanese government kept them on life support. Banks were allowed
to—even expected to—keep nonperforming loans on their books as if
they were performing, that is, lending to failed borrowers, just so they
could repay old loans: a tactic known as evergreening. This steadily weak-
ened the banks and denied credit to the better-performing parts of the
private sector. The Japanese authorities’ inability to effectively deal with
their banking crisis is one of the main reasons the Japanese economy has
been stagnating ever since.
The contrast between the Swedish and the Japanese crises shows what
the authorities should do: prevent the emergence of zombie banks at all
costs; restructure or shut down failed banks as quickly as possible; and
let the banks’ creditors and shareholders take the losses. To put it simply,
the two main objectives of banking crisis resolution are for the banks to
continue serving the real economy and minimize the cost to the taxpayer.
That is what the Swedes did better than anybody and that explains why
their resolution of the 1992 crisis remains the gold standard.
In the 2008 crisis and the subsequent European crisis, some govern-
ments, including those of the United States, Switzerland, and the United
Kingdom, actively tried to prevent the emergence of zombie banks.
Other European countries, like France, Italy, and Germany, have found
that more difficult, lobbying in the G20 for lenient regulations. That re-
flects their banks’ weakness, that is, why they are becoming ever more un-
competitive. Continental European banks are in retreat, American banks
in ascendance. The worst problems are in Italy and other Mediterranean

50 GROUNDHOG DAY
countries, which have a large zombie banking sector they find difficult to
deal with.
Banking crises are not complicated. They have plagued us for centuries,
and we know why they happen and how to prevent them. Still, they are
depressingly frequent. Even when there are clear signs of excesses that in
all likelihood will lead to a crisis, there is little willingness to do something
about it. We enjoy the party too much. What we can do is monitor the
banks more closely and force them to measure and manage risk prop-
erly. A vast edifice has been created for precisely this purpose. Still, crises
happen.

4
The Risk Panopticon
Quis custodiet ipsos custodes?
A few years ago, when taking a London black cab, I saw the cab-
bie reading one of my favorite books on risk, Peter Bernstein’s Against the
Gods. We started chatting, and I mentioned the book I was then writing
(what you are now reading). He then told me a great story about why it
is difficult to control risk. He had just made a trip to Canary Wharf (the
London financial district) and got stopped in a security check. My driver
was really surprised when he saw that the only cars being searched were
London black cabs. Upon asking the security guard why he got the an-
swer: “Because I have a family.” The point is that the most unlikely type
of terrorist in London is a driver of a black cab, so if one wants to avoid
51

52 THE RISK PANOPTICON
an unpleasant encounter with a terrorist when tasked with randomly scru-

tinizing vehicles, the safest choice is indeed black cabs.
This anecdote shows the difficulty facing regulators. While it is easy to
identify the bad behavior we want to see modified or stopped, we need
human beings to implement the regulations. They will have their own
abilities and biases and interests, which, more likely than not, will frus-
trate the objectives of the regulations. We need oversight—Quis custodiet
ipsos custodes? 1—ensuring that the person implementing regulations does
what is expected of her, instead of what is easiest or safest or most profit-
able. If you hire a security guard to randomly check cars entering Canary
Wharf, you also need to employ someone to ensure that he does not pick
only the safest of cars to investigate. Otherwise, it just ends up being a
regulation for peace of mind, for the appearance but not the reality of
safety—security theater.
Some regulations work better than others. Take traffic. We drive on the
left side of the road in Britain and the right side in France, and we have no
problem switching sides when emerging from the Channel Tunnel. We
stop at red, hesitate on amber and go on green. We mostly stick to the
speed limit, and if we misbehave an army of computers and police keep
us in line. It works quite well if we judge it by the rapid fall of traffic ac-
cidents. While we complain about getting a ticket, I suspect most people
are quite happy that other drivers are kept in check. We mostly agree with
and support what the traffic regulators are trying to do. What about fi-
nancial system? Isn’t regulating finance just like regulating traffic? There
is plenty of abusive behavior and many crises.
Here I need to distinguish between the acts of regulating and supervis-
ing. In common usage, we often use these interchangeably, which is usu-
ally perfectly fine, but sometimes it helps to separate the two. Regulations
are the rules, supervision the enforcement of them. It is much easier to
create good regulations than good supervision. I was once in a car driv-
ing on the two-lane motorway between New Delhi and Agra in India.
The Indians drive on the left side of the road, as we do in Britain, and
most of the traffic laws are quite similar. It is different with supervision.
On that motorway I saw people driving in the wrong direction, various
animals wandering around, people having a picnic, and the occasional
market stall. This is something I have never seen on a British motorway.

THE RISK PANOPTICON 53
Okay, I once played football on the M25 motorway when stuck because
of an accident. While many things could account for the difference, I sus-
pect a major one is the intensity of supervision, the enforcement of traffic
laws. India and Britain have similar regulations (traffic laws) but different
supervision (enforcement of traffic laws). Without efficient enforcement,
the regulations become meaningless.
Financial regulations are broadly similar in most countries in the world.
They tend to take the lead from G20, the group of the world’s largest
economies that represents about 90 percent of global GDP; for example,
the Basel banking regulations, securities regulations, and insurance regu-
lations. So, regulations are broadly similar. It’s different with supervision.
Countries interpret the regulations differently, many authorities do a de-
cent job, some are incompetent, others have been captured, and they are
often too underfunded to do the job properly. What is really needed for
effective supervision is the risk Panopticon.
The intellectual origin of the risk Panopticon is the work of the
seventeenth-century English philosopher Jeremy Bentham. It all started
when he visited his brother, who was working in Russia. What interested
Jeremy was that his brother had come up with an ingenious way to man-
age his unskilled employees. He set up observation posts in the middle of
his workshops, allowing inspectors to monitor the workers without being
seen. Jeremy thought this a splendid idea and proposed using it for other
human activities such as schools, factories, and even hospitals. It is his
proposal for prisons that got the most traction, the Panopticon, from the
Greek παν (“all”) and οπτικος (“seeing”). If you find yourself in London,
you can go see Jeremy Bentham’s body, still on display at the University
College London.
In the Panopticon the prison cells are located in buildings that sur-
round a central monitoring tower, allowing the prison guards to observe
the prisoners without being seen themselves. Because the prisoners never
know if they are being observed, they assume they are and thus behave
properly, allowing a small number of prison guards to keep watch over
many prisoners. Not too different from all the CCTVs installed in our
cities. The only remaining job is to prevent the prison guards from abus-
ing the convicts, accomplished by allowing the public to inspect the jails
any time they want.

I once saw the power of the Panopticon in action while driving from
Spain to Portugal. The traffic laws are the same on both sides of the bor-
der, and the speed limit on the motorway was 120km/h. On the Spanish
side everybody drove at 120 or below. As soon as they crossed the border,
their speed increased to an average of 150. The reason is simple: the Span-
ish traffic police had unmarked cars checking speed while the Portuguese
police did not. The experiment’s simplicity rules out any other factors; it
involves the same drivers and the same cars. They did not start disregard-
ing their personal safety or acquire better cars upon crossing the border.
The only thing that changed their behavior was the chance of getting a
ticket on the Spanish side.
The one area where the Panopticon view of regulation has been par-
ticularly successful is in public transport. Most bus systems are based on
users’ contactless cards, so one is left with two options: either buy a ticket
and travel legally or commit fraud and hope to get off the bus without
being caught. What determines one’s choice? Morality and individual at-
titude to risk. Getting caught can be painful. Just ask Jonathan Burrows,
a lapsed fund manager at BlackRock, caught in November 2013 going
through a London Underground ticket gate, not having paid the right
amount. He admitted to regularly traveling without the right ticket, and
the train company estimated the amount of money he had dodged at
£43,000. After he was caught he did not inform his employer as he was
supposed to. The relevant financial regulator, the Financial Conduct Au-
thority (FCA), took a dim view and declared him “not fit and proper,”
ending his twenty-year career in finance. He did not need to avoid the
£21.50-a-day fee, as, by all accounts, he was well compensated by Black-
Rock and owned two mansions worth £4 million.
Burrows was just exploiting a loophole. His local station, Stonegate,
an hour and twenty-two minutes from his final stop in the City of Lon-
don, is rural and has no ticket barriers. So he took advantage, tapping his
Oyster Card only when arriving. This meant that he paid only the price
of the Underground, not the full fare from Stonegate to the London Un-
derground stop. So why would the FCA care about a simple fare fraud?
Because Burrows was in a position of trust at BlackRock, so how could
he be trusted with other peoples’ money if he so casually committed
petty fraud? What I find surprising is, why bother? I have my Oyster Card

checked a few times a year, and the fine if caught does not seem to be
worth the benefit of cheating, not to mention losing a glittering career as
a BlackRock fund manager.
Modern financial regulations very much fit into this Benthamite view
of the world. Financial institutions release large amounts of information
to the public authorities in the full knowledge that the authorities can
look only at a small fraction of all that information. But the banks don’t
know where the authorities are looking, and, even more important, how
information released today will be used in future investigations. The reg-
ulators can inspect the banks without being seen. The problem is that
there can be too much of a good thing, including too much data. So, how
well does the panopticon work for the financial supervisors? It depends
on the objective. It might help the micro regulators focused on individual
banks, protecting individual bank clients. However, what it misses is the
risk that arises from market participants’ interactions in times of stress—
most important, systemic risk.
We Used to Execute Failed Bankers

and Still Suffered Crises
Long before we had much in the form of economic regulations,
we had banking regulations. They are embedded even in major religions:
Christians banned usury until the fifteenth century and Muslims to this
date. Financial regulations in many European countries have religious un-
dertones, such as restricting or banning various trading strategies because
they might be construed as gambling. There are two reasons banking has
always been so regulated. First, banks are a good source of government
revenue, so governments like to keep them on a tight leash. Second, ef-
ficient and uninterrupted banking services are essential to all economies.
Many economic historians will tell you that the establishment of the Bank
of England in 1694 was key to the emergence of London as the world’s fi-
nancial center by the eighteenth century, underpinning the United King-
dom’s status as the world’s greatest economy of the era.
Bank failures impose costs on society far exceeding the private costs
of the banks’ owners, employees, and counterparties, so it stands to rea-
son banks should be regulated. But it is not an easy job, much harder

than regulating other human endeavors, like traffic. We have recently seen
plenty of banking crises and abuse, and it is not possible to make a cred-
ible case that the banking system, unlike traffic, is better behaved now
than it was in the past.
Why is regulating banks so much harder than regulating traffic? The
problem is that the financial system is one of the most complex of all hu-
man constructs, and the financial entities have strong incentives to mis-
behave in a way that is undetected. The private incentives of bankers are
not well aligned with the interest of society at large. There is little down-
side for bank employees working with other peoples’ money (dismissal,
at worst), but they can enjoy a significant upside in terms of high salaries
and bonuses. Hence bankers have an incentive to take more risks than
desired by either their employers, clients, shareholders, or society. Banks
have significant advantages over their clients. They sell sophisticated fi-
nancial products to people who probably have only a very rudimentary
knowledge of finance, perhaps not even understanding basic percentages
or present value calculations. An OECD study found that 40 percent of
the population does not understand diversification, and only 27 percent
are able to both calculate simple interest and recognize the added benefit
of compounding over five years.
Given a chance, banks certainly don’t hesitate to abuse their clients:
witness the Wells Fargo account fraud scandal. The abuse of clients and
financial crises happen all too frequently. What can we do? Punish the
bankers severely if they misbehave? Claw back their bonuses or even put
them in jail? Perhaps even reintroduce the pillory, appropriately on the
square in front of the Bank of England. What better place to punish all
the failed bankers?
We did try more strict punishments in the olden days. In the thirteenth
century one of the major financial centers in Europe was Barcelona, as
told by Meir Kohn in his 1999 article “Early Deposit Banking.” Banks
were tightly regulated. They were forced to have substantial capital and
were required to pay cash within twenty-four hours of demand. If banks
failed, the owners got into serious trouble, both with the Almighty and
the city authorities. Francesch Castello was beheaded in front of his bank
in 1360, yet even the threat of decapitation did not stop his fellow bankers
from misbehaving.

We aren’t going to reintroduce the pillory, not to mention decapita-

tion, and it is not easy to put bankers in jail. What we have is compensa-
tion—money. The received wisdom maintains that what motivates bank-
ers is money. Threaten to take the money away, and they will behave
prudently. The reality is much more nuanced. I don’t think the absolute
level of compensation is what matters most to people; relative compensa-
tion is much more important. If I get a $1,000 bonus and my buddy gets
$500, I am happy, but if I get a million dollars and he gets three million,
I’ll be sad. It doesn’t have to be money; power and authority motivate
just as well. We can certainly have abuse and crises without bonuses or
excessive compensation, as in the Japanese crisis of 1992.
Might clawbacks work better, in which bankers have to give up com-
pensation that is paid out only in the future? One could ask the senior
bankers behind the Wells Fargo phantom account scandal. John Stumpf,
the CEO, will give up $69 million, while the person directly responsible,
Carrie Tolstedt, the head of community banking, will lose $67 million.
This might sound like a lot, but their regular annual compensation (not
subject to clawbacks) was in the millions of dollars. I am skeptical that
clawbacks will deter many bankers, despite its being a useful outlet for
moral outrage.
Punishing the bankers is predicated on actually finding misbehavior, like
fraud or abuse. But it’s much harder to discipline someone for making a
bad investment, even one that leads to a financial crisis. The reason is that
someone whose job it is to make investment decisions is, by definition, em-
ployed to take risk, and with risk comes the chance of failure. There are two
problems with punishing someone for making bad investments. The first is
that they most likely will get large bonuses if they succeed, regardless of how
dumb the investment was in the first place, while if they fail, at worst they
lose their bonus and their job—the upside is much better than the down-
side. The second reason is that if we hire someone to take risk, we must
accept failure, and if we punish them for failed investments, they may end
up becoming too risk averse. Not so fast, you may think, we can simply mea-
sure the amount of risk they take and reward or punish them accordingly.
Well, to do that we have to be able to measure the risk accurately. It is next
to impossible to measure financial risk precisely, and it may well be impos-
sible to discriminate between unlucky, incompetent, and corrupt bankers.

Given how the financial system is set up and what drives its cycles,
banks can and will end up taking too much risk. Individual bankers take
excessive risk again and again, and we will go through repeated boom-to-
bust cycles. It is really hard to effectively prevent the worst outcomes with
regulations, but there is a right way and the wrong way to regulate banks.
Punishing bankers or even executing them as the regulators did to poor
Francesch Castello is unlikely to work. We need something else.
Bank Capital
The primary tool for controlling how banks behave is bank capi-
tal. Bank capital has two purposes. The first is a buffer so that if a bank gets
into difficulty, it has some reserves to draw on before hitting bankruptcy.
The main reason for capital is, however, limits to leverage. The more capi-
tal a bank has to hold, for the same amount of assets, the less levered it is,
and hence safer. The concept of bank capital is quite confusing, especially
for those trained in accounting, economics, or the law. The most com-
mon usage follows Adam Smith, who defines capital in The Wealth of Na-
tions (1776) as “that part of a man’s stock which he expects to afford him
revenue.” To Karl Marx, in Das Kapital, capital is more nefarious: wealth
that is used to create more wealth, something that exists only because of
economic exchange or the circulation of money. Modern usage follows
both Smith and Marx and is often quite contradictory. We have concepts
like capitalization—the market value of a corporation—and economists
talk about capital as one of the two main inputs in production, the other
being labor. One can also find capital to be the net market value of a firm
after all its debts have been subtracted.
A few years ago the university lecturers’ union in the United Kingdom
was protesting in front of the main entrance of the London School of
Economics against the evils of capitalism. As I walked through the door,
crossing the picket line, the protesters asked me if I was a capitalist, as if
that were a horrible thing. I answered, “Yes” and added, “So are you.”
They disliked that and demanded an explanation. I responded by ask-
ing them if they belonged to the university pension fund. They said yes.
Well, the pension fund owns stock in companies, and the definition of
a capitalist, according to Karl Marx, is someone who owns the means

of production. Therefore, anyone who owns stock in a company, which

every member of our pension fund does, is a capitalist. The protesters did
not like that reply.
When thinking about bank capital, especially for those trained in eco-
nomics or accounting, you should forget everything you know about
capital and start anew. Bank capital is an artificial accounting construct
and has only a vague relationship with the term’s most common usage,
as discussed by Smith and Marx. When I lecture on capital to my stu-
dents, I like to define it as things that the banking regulators have cho-
sen to call capital. Bank capital is composed of two components: com-
mon equity, that is, assets minus liabilities and financial instruments
that are not quite common equity but help meet the objectives of the
regulators—equity-like.
At the end of 2019 the world’s largest bank, JP Morgan, had total assets
worth $2.7 trillion, while its capital was $169 billion. The ratio of capital
to total assets is called leverage ratio, and regulations require the leverage
ratio to be at least 3 percent. JP Morgan beats that comfortably at 6.3 per-
cent. Some of the bank’s assets, such as loans to governments, are quite
safe, but others, like mortgages and loans and medium-sized enterprises
(MSEs), are risky, hence regulators have come up with the concept of
risk-weighted assets: the higher the risk, the higher the weight. JP Mor-
gan’s risk-weighted assets are $1.5 trillion.
The $64,000 question is how much capital a bank should have. One
might think it would be a good idea to make capital really high, but there
is a downside. The more capital a bank has to hold, the less it can lend
and the more expensive those loans are to borrowers. And since banks are
a major source of funding to companies, especially the all-important small
and medium sized enterprises (SMEs), the end result is less investment
and less economic growth. The balance is delicate.
Some observers, like Anat Admati and Martin Hellwig, maintain that
capital should be much higher than it is now because the costs from the
inevitable bank failures outweigh the benefits a high-risk banking system
provides to society. The banks respond by arguing that this underesti-
mates the costs to society because the cost of bank lending will become
much higher and interest paid on savings much lower—both borrowers
and savers lose out, and economic growth will suffer. Regardless of which

is right, this is the proper debate to have because it frames the discussion
in terms of what we want from banks, not in a more narrow sense of just
wanting banks to be safe.
Before I leave capital, I want to make two points, one technical and one
philosophical (if you want more details, you are welcome to download
my slides on regulations and capital).2 The technical point is that capital
is made up of multiple parts. At the most basic level, we have common
equity. We then have additional bits that are called Tier 2 and Tier 1 capi-
tal. Add to that a buffer for banks in case they run into difficulty, called
the capital conservation buffer, and another for countries in trouble, the
countercyclical capital buffer. Finally, a special buffer for systemically im-
portant banks. I used all of this buffer complexity to good effect in my
exam in 2020, asking which part of capital requirements would be relaxed
to help banks so they could help with the Covid-19 recovery.
The philosophical point is that even though it is often said to be a
buffer that protects when things go sour, that is only partially true. The
reason is that banks are required to hold minimum capital, which they are
unable to reduce in bad times. As noted by Charles Goodhart, a capital
buffer that can’t be used isn’t much of a buffer: “The weary traveler . . .
arrives at the railway station late at night, and, to his delight, sees a taxi
there who could take him to his distant destination. He hails the taxi, but
the taxi driver replies that he cannot take him, since local bylaws require
that there must always be one taxi standing ready at the station.”3
Schrödinger’s Bank
Financial regulations used to be something each country set for
itself. There were some agreements between countries but not many in-
ternational standards. This was fine so long as the global financial system
remained fragmented or heavily regulated, as it was under the post–World
War II Bretton Woods system, when banks were by and large confined to
their home countries and not allowed to operate across borders.
This state of affairs changed when the Bretton Woods system collapsed
in 1972, bringing a philosophical change in attitudes toward global fi-
nance: the Washington Consensus. From then on, national frontiers con-
tinued to open up to global finance. Banks could operate across borders,

and money moved freely between countries. Financial regulations did

not keep up. Even after the borders started opening, regulations were
still national, and the regulators didn’t talk much to their counterparts
in other countries. This arrangement was not sustainable, and a series of
crises brought home the need for global coordination.
Failures don’t have to be very large to have severe repercussions, as
when the bankruptcy of the eightieth-largest bank in Germany, Bankhaus
Herstatt, in 1974, caused serious disruption to global financial markets.
The Herstatt family has a long history in banking, dating back to Jowan
David Herstatt in 1727. After the collapse of the Bretton Woods system,
Bankhaus Herstatt was an active player in the foreign exchange market,
speculating on the exchange rate between the German mark and the US
dollar. It was not all that successful, and losses mounted. Eventually, the
bank was forced into liquidation on 26 June 1974.
Interestingly, when Herstatt failed, it was not closed down in every
jurisdiction simultaneously. The German authorities closed the domestic
operations, but the bank continued operating abroad because the Ger-
man authorities did not inform their foreign counterparts of the closure.
Herstatt was half dead and half alive, like Schrödinger’s cat. The day Her-
statt failed, it had been buying a large number of German marks, receiv-
ing them in Germany, and due to pay dollars for them in New York. It
failed after receiving the German marks and before delivering the dollars.
The reason was the time zone difference between Frankfurt and New
York. It received marks at the end of the German business day and was to
pay out dollars at the end of the New York business day, six hours later.
The eventual losses to its creditors were $1.3 billion, the single largest be-
ing Moscow Norodny at $365 million. While $1.3 billion was undoubtedly
a more serious number in 1974 than it is today, it is still a relatively small
amount.
It was damaging because of how it occurred. The German regulators
kept their foreign counterparts in the dark about the seriousness of Her-
statt’s risk. They had been intensively monitoring the bank for at least
three years before 1974 and knew a collapse was imminent. Why did they
keep it a secret? We can only speculate, but I think there are two main
reasons. First, their mandate was to protect Germany, its processes, laws,
and reputation, which meant mistrusting foreigners. Second, they were

embarrassed that a bank on their watch would fail. Whatever the reason,
the German financial regulatory system was reformed soon after.
When the German bank Wirecard failed in the summer of 2020, the cir-
cumstances were eerily similar to those of Herstatt half a century earlier.
The German regulators of Wirecard had been aware of some problems
but failed to act. The head German regulator said in a testimony to the
German parliament, Bundestag, that the German regulatory system func-
tioned well in normal circumstances but failed in a crisis.
The Herstatt bankruptcy demonstrated a weakness of the settlement
system, until then not recognized, and served as a wake-up call to the
global financial authorities. They might want to be purely domestic,
managing their banks with minimum international coordination, but the
banks don’t play ball. They are international—and that requires interna-
tional coordination.
It wasn’t really until the case of Banco Ambrosiano eight years later
that the case was forcefully made. Ambrosiano was the largest private
banking group in Italy, with operations in fifteen countries. At the center
of the bank’s failure was its chairman, Roberto Calvi, called God’s banker
by the Italian press due to his close association with the Holy See. Calvi
was determined to transform his bank from a relatively small regional
bank with strong religious overtones into a major international financial
institution. One of his initial steps was to form a holding company in
Luxembourg, which was not subject to Italy’s banking regulations.
Calvi’s problems began in 1978 when the Bank of Italy conducted an
extensive audit of his financial empire, noting unorthodox operations in-
volving $1.2 billion in unsecured borrowings. Ambrosiano was buying up
its stock by using dollars, an illegal operation according to Italian bank-
ing regulations. Ambrosiano collapsed for quite predictable reasons: the
Italian lira fell relative to the dollar, so the lira amount of its liabilities
sharply increased, while the value of its Italian assets stayed constant in lira
terms. Calvi was sentenced to four years in jail but released pending ap-
peal. He fled Italy but was found hanging under the Blackfriars Bridge in
London in 1982. To this day, it’s not know whether he committed suicide
or was killed, a topic that continues to arouse controversy. I cross Black-
friars Bridge on my way to work every day and have frequently spared a
thought for poor Calvi. The failure of Banco Ambrosiano left more than

two hundred international financial institutions with large losses, threat-

ening the stability of the entire international banking system. Ultimately,
this triggered major changes in international regulations.
The Ambrosiano collapse turned out not to be the last chapter in the
reform saga, as the failure of the Bank for Commerce and Credit Interna-
tional (BCCI) showed. BCCI was registered in Luxembourg with head
offices in Karachi and London and branches throughout the world. It
was closed down by the London authorities in July 1991 because of wide-
spread fraud, and it is now believed that BCCI’s financial statements had
been falsified from its establishment in 1972. BCCI failed to record de-
posits as liabilities and created fictitious loans that generated substantial
profits, using depositors’ money to fund their own proprietary trading
activities and covering up the resulting losses with more fictitious loans.
One reason BCCI got away with it was that it used a complex structure
consisting of a holding company, incorporated in Luxembourg, and two
principal subsidiaries in the Cayman Islands and Luxembourg. Even be-
fore these problems surfaced in 1990 supervisors and commercial bankers
were wary of BCCI because of its rapid growth and opaque corporate
structure. Yet while BCCI was sometimes mentioned in the press “chiefly
for the mystery that surrounded it,” market participants generally saw
BCCI as a bank that lost money through incompetence rather than fraud.
Concerns about fraud led eventually to discussions between BCCI’s then
current and previous auditors (Price Waterhouse and Ernst & Young), the
banking supervisors, and BCCI’s shareholders. After BCCI failed, the liq-
uidators, Deloitte & Touche, filed a lawsuit against the auditors, settling
for $175 million in 1998. BCCI creditors also attempted to sue the Bank
of England as BCCI’s regulators.
I draw two main lessons from the case of BCCI. The first is that it is
necessary to ensure that banks are supervised everywhere they operate
and that neither the regulators in their home country nor the countries
where they operate should be able to avoid regulating them. The second
lesson is more interesting because it nicely demonstrates the reputation
risk of central banks that conduct banking supervision. The reputation of
the Bank of England took a hit, and that was worrying because the bank
needs its high reputation to be able to credibly implement monetary pol-
icy. The BCCI scandal motivated the separation of banking supervision

from the Bank of England in 1997, creating the Financial Services Au-
thority (FSA). And since Britain is generally seen as a leader in regulatory
methodology, many countries followed suit.
Was that a good idea? The global crisis in 2008 showed what can happen
if the central banks are not in charge of supervision: they lack oversight,
don’t know what is happening on the ground, don’t have the necessary
expertise, and can refuse responsibility. The British authorities eventually
split the FSA into the independent Financial Conduct Authority (FCA)
and the Prudential Regulation Authority (PRA), which was remerged
with the Bank of England. Other countries are now following suit. I am
curious to see what the British authorities will do after the next crisis.
Will the PRA be remerged with the FCA or will the FCA rejoin the Bank
of England?
The Rules from Basel

In the years after World War II banks were mostly prevented from
operating outside their home countries. As the restrictions were gradually
lifted in the 1960s, banks started to operate internationally, and the need
for global coordination was recognized. In response, the main financial
centers at the time, those making up the G10 group (Belgium, Canada,
France, Italy, Japan, Germany, Sweden, the Netherlands, the United
Kingdom, and the United States), together with Luxembourg, set up the
Basel Committee on Banking Supervision, or Basel committee in short,
as a new international organization tasked with designing international
banking regulations in 1974. Charles Goodhart wrote a history of the
committee.
Each country had two or more members on the committee, typically
one from the regulator and one from the central bank. But why Lux-
embourg? you may ask. It is undoubtedly a financial center but hardly
belongs in the G10. The reason is that a Luxembourgish banker, Albert
Dondelinger, was instrumental in setting up the committee, and it would
have been churlish to exclude him and Luxembourg. However, since
Luxembourg did not have a central bank (it used the Belgian franc as
currency, so the National Bank of Belgium was its central bank), it got
only one seat.

The Basel committee is hosted at but is distinct from the Bank for
International Settlements, whose head office is in Basel, Switzerland. It
does not possess any formal powers, acting instead as a vehicle for seeking
agreement on common standards but leaving implementation to member
countries. The Basel committee’s limited membership used to be a source
of frustration for the rest of the world, as many countries have felt they had
no choice but to implement Basel-style regulations without the ability to
influence them. This aggravation has now been remedied, as following the
2008 crisis the committee expanded to include twenty-eight jurisdictions.
The Basel committee reports to the G20 group of countries, as do
other parts of the international financial architecture. The most impor-
tant part of the committee’s work is the Basel Capital Accords, a set of
rules for determining how much capital a bank should have. The first
Basel Accord, now referred to as Basel I, was decided in 1988 and imple-
mented in 1992. The successor, Basel II, was proposed in the late 1990s
and partially implemented from 2008.
Why did the Basel committee decide to harmonize bank capital rules?
It goes back to the early 1980s, when bank capital ratios of major banks in
Europe and the United States were perhaps 8 percent to 10 percent, while
the Japanese banks operated with much lower capital, around 4 percent,
giving them a competitive advantage. The lower the capital ratio, the
lower the cost of lending. Not surprisingly, the Japanese banks became
a significant presence in the European and American corporate lending
markets, elbowing out the local banks. Understandably upset, the Euro-
pean and American banks pushed for their capital ratios to be lowered to
match those of the Japanese. The regulators didn’t like that very much
and instead forced the Japanese to raise their capital standards to the Eu-
ropean and American levels.
When I spent a few months at the Bank of Japan in 2000, some of my
Japanese colleagues called Basel I the anti-Japanese conspiracy, blaming it
for the collapse of the Japanese banking system in the early 1990s and the
subsequent recession. There is a lot of truth in that. Having to double the
capital ratio within a few years proved quite difficult and certainly contrib-
uted to the Japanese banking crisis.
Basel I succeeded in achieving its intended purposes, raising capital
levels when they were low and trending down. Still, the accord wasn’t

perfect, focusing on credit risk with crude risk weights. If a bank lent
$1 billion to an AAA-rated firm, like Apple, it had to hold $80 million in
capital. If it lent the same amount to an OECD government, like Turkey,
Japan, Mexico, or Greece, no capital was needed. It is easy to see why
governments liked this arrangement. It made loans to companies and in-
dividuals more expensive than loans to governments, subsidizing govern-
ment borrowing at the private sector’s expense. Financial repression, the
economists call it.
Long before the 2008 crisis, and anticipating some of the things that
would eventually go wrong, the Basel committee embarked on a revision
of Basel I in the 1990s, what came to be known as Basel II. It did away
with the zero-risk weighting for the OECD government debt, but the
European Union didn’t like this move very much and passed a directive
saying that all member states’ sovereign debt was risk free. While subsidiz-
ing government borrowing in Europe, the directive also helped Greece to
borrow itself into default and was the leading cause of the Cypriot crisis
in 2012 (see chapter 3). While Basel II was announced at the turn of the
century, the extensive lobbying that followed delayed its implementation
until 2008. This meant that Basel II reflected the regulatory concerns and
technological developments of the mid-1990s, so it was already out of
date by the time of implementation, a common problem in international
financial regulations.
I was always of two minds about Basel II. It was a big step toward
making risk management and especially financial regulations scientific.
But what bothers me most is the belief that risk can be measured and
controlled accurately. While that might be a reasonable (or at least accept-
able) assumption in the internal management of individual risks, it is a big
leap to apply it to risk control for an entire financial institution, not to
mention the entire financial system. The Basel II proposals inspired me to
write my favorite paper, eventually published in 2002 under the title “The
Emperor Has No Clothes: Limits to Risk Modeling,” where I argued that
risk was not all that well measured.
When Basel II was announced I got together with several LSE col-
leagues, including Charles Goodhart and Hyun Song Shin, to write an
official comment, titled An Academic Critique to Basel II, in response to
a call for comments on the initial Basel II proposals. We certainly were

skeptical: “Heavy reliance on credit rating agencies for the standard ap-
proach to credit risk is misguided as they have been shown to provide
conflicting and inconsistent forecasts of individual clients’ creditworthi-
ness. They are unregulated, and the quality of their risk estimates is largely
unobservable.” One of the leading causes of the 2008 crisis was structured
credit products composed of subprime mortgages that could not have
been created without the credit-rating agencies. As it turned out, the
ratings were abysmal, demonstrating the folly of relying on the rating
agencies for financial regulations: “Statistical models used for forecasting
risk have been proven to give inconsistent and biased forecasts. The Ba-
sel committee has chosen poor quality measures of risk when better risk
measures are available.”
When the Basel II regulations were initially proposed in the mid-1990s
there was little understanding of the underlying risk-measurement tech-
niques’ reliability. By the time the proposals came out, that had changed.
The low-quality methods chosen by the Basel committee significantly
contributed to the regulators’ and the banks’ improper appreciation of
financial risk before 2007: “The proposed regulations fail to consider the
fact that risk is endogenous. Value-at-Risk can destabilize and induce
crashes when they would not otherwise occur. Financial regulation is in-
herently procyclical. Our view is that this set of proposals will, overall,
exacerbate this tendency significantly. Insofar as the purpose of financial
regulation is to reduce the likelihood of systemic crises, these proposals
will actually tend to negate, not promote this useful purpose.”
The statement describes exactly what happened in the years before
2008. As banks were implementing Basel II in the early 2000s, the risk
management methodologies they were required to implement told them
that risk was low and therefore it was perfectly acceptable, even expected,
to take more risk. By not recognizing the consequent risk, the Basel II
regulations amplified the financial cycle, helping to create the right condi-
tions for the 2008 crisis.
Regulating a Malevolent Entity

The tallest skyscraper in the world was built in Taipei in 2004.
While the appropriately named Taipei 101 is no longer the world’s tallest,

it is the most interesting from the perspective of risk. Every tall build-
ing has to contend with the elements—in Taipei 101 earthquakes and
typhoons—so risk management is critical to prevent a catastrophe. The
devices used to manage the risk are usually hidden, but not in Taipei 101.
It uses a 728-ton gold-painted orb that is open to the public as a counter-
weight that swings like a pendulum at the top of the building. A risk man-
agement system that has become a tourist attraction in its own right. The
engineers even hired the Sanrio Company, the creators of Hello Kitty, to
create the “Damper Babies,” cute figurines representing the risk manage-
ment system.
Taipei 101 was tested in August 2015 when it was hit by Typhoon
Soudelor, with winds of 145 mph. The golden orb swung more than a
meter from its regular position, but nothing happened to the building,
a monument to the engineers’ calculations. If structural engineers can
create risk management systems that protect tall buildings like Taipei 101,
why can’t the financial engineers working in the regulatory agencies pro-
tect us from financial crises? Because of a crucial difference between the
risks in these two disciplines. The reason the job of the civil engineers is
relatively straightforward is that they can ignore the human element. If
they calculate that a wall one meter thick is expected to collapse every five
hundred years, nature couldn’t care less. Risk is exogenous.
In finance, nature is not neutral, it is malevolent, and risk is endog-
enous. The reason is that all rules and regulations change behavior and
outcomes. Human beings, being human, don’t just naively comply—they
change their behavior in response. Immediately after a regulator comes
up with rules, perhaps for determining bank capital, the bankers look for
a way around the rules. They will try their best to make the capital appear
to be very high to the outside world while making it as low as possible in
practice. The technical name for this is capital structure arbitrage. Many
of the banks that failed in 2008 had some of the highest levels of capital
going, but that capital turned out to be illusionary. There is always a
cat-and-mouse game going on between the authorities and the banks.
While the bankers may pore over stacks of regulations to comply, they are
much more enthusiastic about finding loopholes. Regulations are inher-
ently backward looking and change at a glacial pace, giving fast-moving
and forward-looking bankers ample room to look for places to take risk

where the authorities are not looking. Because the financial system is al-
most infinitely complex, it is technically impossible to regulate but a tiny
part of it, leaving plenty of room for misbehavior.
Regulations may end up causing profitable activities to move to the
shadow parallel banking system or abroad. A good example is Regulation
Q in the United States from the 1960s and 1970s, which limited the inter-
est banks could pay on deposits, the idea being that high interest rates
were inflationary (this erroneous notion has recently resurfaced in Tur-
key). Regulation Q simply caused deposits to move to the parallel bank-
ing system—the money market funds—where market interest rates could
be paid. Regulation Q was abolished a long time ago, but the money
market mutual funds continue to flourish and have become a significant
contributor to systemic risk in the United States. Financial activities can
also shift abroad, as when the Eurodollar market first came into being in
the 1950s, when the Soviet Union’s oil revenue—all in US dollars—was
deposited outside the United States for fear of being frozen by US regula-
tors. This resulted in a vast offshore pool of dollars outside the control of
US authorities, primarily in Europe, hence the term “Eurodollar,” help-
ing London become a world-leading financial center again.
When we regulate the financial system we often drive risk-taking be-
havior away from the spotlight and deep into the shadows, where it is
much harder to detect. The tequila crisis in Mexico in 1993 happened
because the Mexican banks were borrowing US dollars in New York to
lend to Mexican borrowers in pesos. Because the Mexican banks took
the currency risk, the Mexican authorities were justifiably concerned and
forbade this. The Mexican banks found it easy to get around the rules by
creating derivative transactions with the New York banks—Tesobonos.
Because the Mexican central bank, Banco de México, did not see these
transactions, it did not realize what was happening and could not step in
to prevent the crisis until it was too late. It is much better if risk-taking is
visible rather than hidden.
Finally, and even more insidiously, precisely because the financial au-
thority is trying to reduce financial risk, banks have an extra incentive
to take on more risk—an example of the Minsky effect. That perverse
outcome happens because of the way we measure risk. If all looks nice
and stable, the road is smooth—the great moderation before the 2008

crisis—and perceived risk is low. If everything is safe, what’s wrong with

a little bit more risk? The problem is we don’t usually notice all the ex-
tra risks until much later. It was decisions taken in the low-risk, great-
moderation environment at the start of the 2000s that created the con-
ditions for the crisis in 2008. Nobody had any idea until the crisis was
already under way.
The three factors—the encouragement of avoidance, the move to the
shadows, and the Minsky effect—came together in 2008, helped by fi-
nancial innovation: the creation of new and more complex financial
instruments. The late Paul Volcker, the éminence grise of US financial
policy making, alumnus of the London School of Economics, and the
slayer of the 1970s inflation when he was the chair of the Federal Reserve,
said in 2009, “I wish that somebody would give me some shred of neu-
tral evidence about the relationship between financial innovation recently
and the growth of the economy. . . . It was hard to think of a worthwhile
financial innovation since the ATM.”4
In 2008 it was all the various structured credit products like structured
investment vehicles, conduits, and collateral debt obligations (CDOs).
They allowed the buildup of very high levels of hidden risk, not least li-
quidity risk, that proved so damaging. The most interesting case of how
financial innovation can cause trouble is the IKB bank, a small, partly
state-owned, German bank. Before the 2008 crisis it was quite active in
the market for a particularly dangerous financial product called a conduit.
A conduit is a special purpose vehicle that invests in long-term bonds but
finances itself by borrowing short term. That is highly profitable most of
the time because short-term interest rates are lower than long-term inter-
est rates, and the difference between the two is profit. Profitable so long
as the yield curve is upward sloping—long-term interest rates are usually
higher than short-term interest rates—and when someone is willing to
lend short term. If either goes away, the strategy is no longer profit-
able. Naturally, the short-term lenders knew this and would normally
demand high compensation for financing a conduit. The conduits found
a neat way around that: a sponsor with a good credit rating that, for a fee,
guaranteed the short-term lending and promised to fund the conduit if
nobody else did. In good times, fantastic. The fee income rolls in, and
it looks like money for nothing. But there is hidden risk. When the crisis

started in 2007, the conduits could not roll over their short-term loans
and hence called on the sponsor, IKB, for help. Unfortunately, IKB did
not have the money needed. Facing bankruptcy, it was bailed out at the
cost of €9 billion to the German taxpayers.
IKB made a name for itself the following year by transferring money
to Lehmans a few hours after the latter defaulted. Apparently, IKB had
automatic rules in place for transferring money to Lehmans on Monday
mornings, and even if the entire world knew by Friday that Lehmans
was about to fail, and it had actually failed right after midnight Monday
morning, IKB didn’t think to cancel its transfer.
We have become a society of regulations, ruled by the risk Panopticon.

Jeremy Bentham would be proud. Many regulatory systems work well,
especially in the airline industry, where continuous improvements and a
culture of minimizing risk have led to a dramatic fall in fatalities even as
air traffic grows. Our view of how well regulations work is shaped by our
interactions with the best ones: airlines, traffic, food. Not surprisingly, de-
spite all the abuse and exploitation and crises we want to regulate finance
just as strictly. Unfortunately, it isn’t all that easy, and the reason it isn’t so
hinges on two concepts, risk and uncertainty, and the pivotal role played
by the riskometer.

5
The Myth of the Riskometer
Prediction is very difficult, especially if it’s about the future.

—Nils Bohr
In the Middle Ages maps sometimes came with the warning hic
sunt dracones, here be dragons (Figure 11). Where the mapmakers had
little information, they warned travelers of unexpected dangers, maybe
dragons. The modern riskometer should come with the same warning.
There are many ways to measure financial risk. The most accurate
is to study the deep structure of the financial system, identify all the
72

THE MYTH OF THE RISKOMETER 73
interlinkages and the hidden corners where risk is taken. That is difficult
and costly, and it is much more common to use a purely statistical ap-
proach, what I called a riskometer in a blog piece I wrote for VoxEU
.org titled “The Myth of the Riskometer.” It is a useful little thing, the
riskometer. I can plunge it deep into the bowels of Wall Street and it pops
out an accurate measurement of risk.
Riskometers are used all over. Anywhere from someone investing their
own money to risk managers controlling proprietary traders to a bank de-
termining the amount of capital it holds. Financial regulators concerned
with the stability of the entire financial system use one too. The riskome-
ters promise to distill the risk of entire financial institutions into one
number. It is really useful to have a single, unambiguous measurement
of risk. A number with all sorts of caveats is not nearly as helpful. The
decision makers, the people who run banks, and the regulatory agencies
are just like President Harry S. Truman, who is widely credited as having
demanded, “Give me a one-handed economist. All my economists say ‘on
the one hand . . . ,’ then ‘but on the other.’”
The prominence of riskometers is increasing by the day, and for a good
reason. They are cheap, quick, and objective—scientific, really. The alter-
natives are subjective, slow, and expensive. In the scientific world of risk,
where we have almost limitless data, sophisticated statistical methods,
and all the processing power one could want, how can the riskometer
not be the best way to measure risk? The problem is that riskometers can
capture only a caricatured view of risk. Any particular implementation
will focus on and often exaggerate some aspects of risk and ignore others.
That means riskometers are not nearly as accurate as most of us think,
most notably the senior decision makers. Those who are actually on the
ground, designing the riskometers and reporting risk to their superiors,
know better.
There is nothing new about riskometers, and we have been searching
for devices for measuring risk since the beginning of time. The earliest ex-
ample I know of is the riskometer the Chinese astronomer Zhang Heng
invented in AD 132 during the Han Dynasty. It is a metal urn containing
a bronze ball and eight dragon heads for each direction on the compass.
When an earthquake hit, the ball would roll onto the dragon head cor-
responding to the earthquake’s direction. The government was advised

74 THE MYTH OF THE RISKOMETER
both of an earthquake and where it happened, allowing the emperor to

send relief to the affected region.
The modern notion of risk and its applications to decision-making
dates back to Blaise Pascal’s work, the sixteenth-century French math-
ematician who both improved our understanding of risk and formally
linked it to how to manage future economic endeavors. Monsieur Pas-
cal might well be the first management consultant. The question then
is whether the caricature is sufficiently accurate. Will it lead us astray
by not measuring (or even hiding) the risk we care about and make us
waste our efforts on risk that is not all that important? To paraphrase
the statistician George Box: “All measurements of risk are wrong. Some
are useful.”
If I want to figure out the temperature outside, all I have to do is look at
the thermometer outside my kitchen window. Right now it tells me that
the temperature is 14.4° Celsius (57.9°F). If that is not precise enough, I
can spend more money and be told that the temperature is 14.392°C. If I
don’t trust my thermometer, I can pop out to the store and buy another
one. Or go to BBC’s weather service, which tells me that the temperature
in my postcode is 14.4°C. Thermometers are reliable and consistent, and
different ones give the same answer. On top of this, everyone sees tem-
perature the same way. Regardless of whether I am concerned about how
warm my house is or am controlling some industrial process, or if I am at
the North Pole or in the Sahara desert, the concept of temperature is the
same. 14.4°C always has the same meaning, no matter what I, or anybody
else, am up to.
What about risk? Can I use a riskometer in the same way as one might
use a thermometer? No. We need three separate layers of analysis for risk,
whereas only one is needed for temperature:
1. The recognition of what sort of risk is most important

2. A theory of how to quantify that risk
3. Statistical technologies for producing the actual risk measurements.
The latter two make up the concept of a riskometer, the choice of which
intimately depends on what the riskometer is needed for, the first layer.
To start with, what do I want? Suppose Paul, Ann, and Mary each in-
vest in Google stock. Their reasons for investing are different. Paul trades

on his own account, his reasons are speculative, and he aims to get out
in one week with a hefty profit. Ann is a fund manager, investing on be-
half of her bank, and her primary concern is beating her benchmark and
avoiding significant losses relative to her benchmark that would get her
fired. Mary is investing for the long term and worries about getting her
pension seventy years into the future when she is ninety-five years old and
relying on the financial industry for her material well-being.
Even though each of the three made precisely the same investment and
had access to the very same risk measurement technologies, their views on
risk are very different. Paul cares about day-to-day fluctuations over the
next week, Ann is worried about a substantial one-day loss sometime in
the next six months, and Mary needs the Google stock price to continue
to grow over the next seventy years and does not care about what hap-
pens to the stock in the meantime (and certainly not over the next week).
Their investment horizons are different, their objectives are different, and
therefore what risk means to them is different. Each needs a different
riskometer. Unlike temperature, where Celsius is the appropriate unit of
measurement regardless of what it is used for, for risk we need different
concepts depending on the end use.
So the next step for Paul, Ann, and Mary is to pick a riskometer. The
riskometer is a combination of two things: some concept of what risk is
and a statistical apparatus to produce a risk measurement. Start with the
concept. With temperature, we have three units of measure Celsius, Fahr-
enheit, and Kelvin. However, they all measure the same thing, tempera-
ture, and we know exactly how to go from one to the other. 100° Celsius
is 212° Fahrenheit and 373.15° Kelvin. It is not the same with risk, where
we have multiple concepts.
When it comes to an individual stock’s risk, I may be interested in
volatility, Value-at-Risk, or Expected Shortfall, just to mention the three
most popular. These are not just three measurements of the same thing
like Celsius, Fahrenheit, or Kelvin. It is like having three different opin-
ions of temperature with no apparent way to compare numbers produced
under one standard with another. The user has to pick the concept of risk
most appropriate to her, and if she uses a generic one—a one-size-fits-all
riskometer—the result will not be as good as if she picked what is the best
riskometer for her purpose.

I was recently asked to address fund managers from Latin America on

the idea of the riskometer and practical issues in risk management. I was
appalled when one of the delegates asked me a question on the use of
Value-at-Risk for pension funds. It is not often that I am speechless, but
that morning I was. As it turned out, the regulators in her country had
imposed a Value-at-Risk constraint on pension funds, in effect forcing
them to think about day-to-day risk rather than long-term solvency. Why
would they have done so? When I asked them, I was told that the reason
was probably that Value-at-Risk was seen as the best practice in risk man-
agement—it is fundamental in international banking regulations, the Ba-
sel Accords—and, without thinking it through correctly, the authorities
decided to apply it to pension funds.
The muddled pensions regulators are not alone. Most people, even
professionals who should know better, see financial risk measurements as
a single concept with no difference between day-to-day risk, the risk of
extreme losses, and long-term risk. This is nonsense. One cannot go from
one concept of financial risk to another without making what often are
entirely unrealistic assumptions.
After deciding on a concept of risk, the next step is statistical measure-
ment. When it comes to temperature, measurement is easy. Take some
mercury, put it in a tube, and since mercury expands with heat, the mer-
cury’s height in the tube tells us the temperature. It is the same with
prices. I can go to my Bloomberg terminal to check what the price of a
stock is. That is it. I know the price. Both prices and temperature are real-
time measurements. It is different with risk, as there is no single observ-
able phenomenon called risk. The reason is that what is measured cannot
be compared to any actual reality. In statistical language, risk is latent,
meaning that it cannot be directly observed. At best, one gets an imper-
fect measurement by observing how risk influences the world. If a stock
price fluctuates a lot, it might be highly risky, but then again it might not.
Once, when I was explaining the problem of measuring risk to my LSE
students, I did the following experiment. I told them to ignore what they
knew about the temperature in London at that moment and instead asked
them to look out of the window and guess what the temperature was,
based on how people were dressed. It was a typical October day, with a

temperature of 13° Celsius. I got numbers ranging from 8° to 24°. Those

on the lower end of the scale tended to come from tropical countries,
where people start putting on coats when the temperature is still in the
20s. Most of the rest came from cold countries, where people are lightly
dressed even when the temperature falls below 10°. The experiment shows
how biased we can be when it comes to measuring latent variables like risk
and how our background and experience shape the measurements.
Because risk is latent, the only way to measure it is to see what has
happened in the past. If I am interested in the risk of a stock, I have to
observe how the stock fluctuated in the past and infer a risk measure-
ment. Not all that straightforward. To begin with, how far to look back?
For some assets, we have a long history. The longest I know of is the
exchange rate between the Dutch guilder and the British pound, starting
in the early 1600s and extending to today if I allow the guilder to join the
euro in 1998. Some individual stocks have been traded for over a century.
I have observations on the most important stock market index in the
world, the Standard and Poor’s 500 index, going back to the 1770s. Yes,
the S&P 500 dates only back to 1957, but economic historians have re-
constructed its older values, some of which you can find on Bloomberg.
However, the US stock market was very different in the 1770s than it is
today, so I may want to start measuring risk more recently. But when?
Different starting points will give different risk measurements, and I have
no clear way of identifying which is more correct. There is no single cor-
rect answer to the question of how much data to use. We could ask the
experts, but they will probably say, “It depends,” and the more expert
they are, the more likely they are to say so.
If that was not enough, I also have to pick a statistical model. Here I
have a large number of choices, each with its strengths and weaknesses.
Some are relatively accurate but data-intensive, while others can provide
less accurate measurements with little data. Some need powerful com-
puters and highly trained experts. Others can be implemented in Excel
with minimum training. Some focus on extreme outcomes, while others
are more concerned with day-to-day movements. So how do we pick a
model? If you ask the experts, they will just suggest their favorite meth-
odology. Not very helpful.

When it comes to risk, the objectives, the concept of risk, and the sta-
tistical methodology should be considered simultaneously. This means
that different end users, all with the same investment and technical skills,
ought to measure risk differently. In the example above, what is risk to
Paul is irrelevant to Mary and vice versa. There are a lot of riskometers out
there. Because it is not very hard for someone who knows programming
and statistics to create yet another one, it is not surprising that academia
and consultancies are full of people churning out riskometers, all produc-
ing different measurements of risk for the same assets. It can be an easy
way of getting a PhD in statistics, physics, computer science, or econom-
ics. The same PhDs tend to get jobs in the financial industry, producing
riskometers for their government agencies and banks.
The European Central Bank’s Risk Dashboard

Far better an approximate answer to the right question, which is often
vague, than an exact answer to the wrong question, which can always be
made precise.
—John W. Tukey (1962)
There are few better places for seeing riskometers in action than
the European Central Bank’s risk dashboard. The dashboard is full of
interesting numbers (Figure 12). On the day after the Brexit vote in June
2016, it told us that systemic risk was 0.321853, on a scale of 0 to 1. That
was indeed much higher than the 0.185922 the week before, not to men-
tion the 0.058941 at the start of 2016, when we seemed to be especially
safe. The historically safest date was 27 September 2013, at 0.02106. For-
tunately, the Brexit systemic risk is not as bad as it was in December 2008,
when it hit 0.838846, worse than the number after Lehman failed in Sep-
tember 2008, when it was “only” 0.554620. These figures look exact. The
six significant digits tell us so: 0.321853 is much more accurate than a mere
0.3 and suggests the European Central Bank has pretty precise measure-
ment technology. But is that precision realistic? As Warren Buffett once
said, “We don’t like things you have to carry out to three decimal places.
If someone weighed somewhere between 300–350 pounds, I wouldn’t
need precision—I would know they were fat.”

Figure 12. ECB composite indicator of systemic stress (CISS) systemic

risk indicator. Credit: Lukas Bischoff/IllustrationX.
The 0.321853 is no more an accurate measurement than 0.3, and the

extra precision serves only to mislead. Despite the dashboard’s apparent
precision, there is a strange omission from it in that it does not tell us
anywhere what the numbers mean. Presumably the numbers are ordinal,
so 0.321853 is worse than 0.02106 and better than 0.838846, but by how
much? Is it a lot worse or almost the same? In other words, it omits any
analysis of statistical significance.
When I took my first course in statistics at university, the professor
hammered into us the importance of significance testing. Create confi-
dence intervals. If the standard deviation of the numbers on the systemic
risk dashboard is 0.1, the change in systemic risk from 0.185922 to 0.321853
on the day of the Brexit vote is not statistically significant. If the standard
deviation is 0.3, the Brexit systemic risk is not statistically different from
the December 2008 number of 0.838846. If it is 0.5, the historically high-
est and lowest numbers are not statistically different from each other. If
we require our undergraduate students to produce confidence bounds,
is it too much to ask the European Central Bank to do the same? How
should we, or anyone, make use of the dashboard otherwise? I could for-
give all of that if the dashboard gave a warning, indicating that risk was
building up and a crisis might be on the horizon. It does no such thing
(see Figure 12).

On 22 June 2007 it told us systemic risk was very low at 0.042938.

That is the month when the crisis was already under way, when the
quant funds faced difficulty, and when Northern Rock was starting to
run into difficulties selling its securitized mortgages. Indeed, measured
systemic risk was quite low in the five years before 2007. The indica-
tor sent the signal that everything was fine, and it was okay to take on
risk at precisely the same time all the bad decisions that led to the crisis
were being made.
Meanwhile, the highest number was recorded in December 2008, after
the worst was over: the peak of the crisis was late September and early
October. Imagine you observe a dam bursting. When is the risk highest?
Before it breaks or after? Before, of course, because the dam can’t burst
twice. It is the same with the highest risk measurement in December
2008, when it tells us the risk is at its all-time high at the time the crisis
was already subsiding. The measurement is misleading.
The European Central Bank could save itself a lot of trouble by taking
out a subscription to the Financial Times or perhaps the ECB’s home
newspaper, Handelsblatt. Because both the press and the dashboard react
to market prices, the newspapers’ front pages will tell us a crisis is under
way at the same time the systemic risk dashboard does. It would be much
cheaper, and the European Central Bank could reassign all its systematic
risk analytics team to something productive.
The First Riskometer

The riskometer was born in October 1994, when the investment
bank JP Morgan released Value-at-Risk to the world, along with a statisti-
cal technology to measure it. Two things, a concept and method, make
up a riskometer, and JP Morgan was the first to make one. The reason JP
Morgan created Value-at-Risk in the first place was that its chairman at
the time, Dennis Weatherstone, asked the bank’s quants (finance lingo for
statisticians working with financial data—experts in quantitative meth-
ods) to come up with a single number representing how much money the
bank might lose over the next twenty-four hours. The number was to be
ready in time for JP Morgan’s treasury meeting in New York at 4:15 pm
every day and hence has the informal name “the 4:15 report.”

Value-at-Risk promises to improve on the traditional risk measure,

volatility, which answers the question, How much fluctuation in stock
returns can we expect? The quants created Value-at-Risk based on the
query, What is the minimum loss on a day when we suffer large losses?
To illustrate how this works in practice, suppose the price of Apple stock
is $200, and I own fifty shares, so my portfolio has a value of $10,000. A
99 percent daily Value-at-Risk might be $200, meaning that one day out
of a hundred I expect to lose $200 or more, and on the other ninety-nine
days I will either make money or suffer losses smaller than $200. There is
nothing magical about the 99 percent, but it is the most common prob-
ability and the one mandated by financial regulations.
The idea behind Value-at-Risk dates back to the 1920s, but JP Morgan
made two notable contributions. It connected Value-at-Risk to a statisti-
cal apparatus in order to estimate it and made it freely available. It took
something that used to be proprietary and in-house, really known only
to a handful of people inside banks, and gave it to the world. From that
point on an army of researchers not only enthusiastically embraced the
Value-at-Risk riskometer but also created many new ones. Why would a
bank do such a thing? They are not exactly in the business of altruistic
creation of knowledge. The reason is regulations, according to Jacques
Longerstaey, one of the creators of Value-at-Risk: “All the banks would
go to the regulators and ask to be allowed to use internal models to lower
their capital requirements,” he said.1 So, JP Morgan gave us Value-at-Risk
so it could lower its capital and hence increase its profits.
Value-at-Risk became wildly successful and was almost immediately ad-
opted by the financial regulators when it was incorporated into Basel I in
1996, two years after its debut. It is one of those concepts that attract a lot
of criticisms, if not outright hate. Pablo Triana wrote a book called The
Number That Killed Us about all the evils of Value-at-Risk. Value-at-Risk
has few serious defenders (even though some are staunch defenders), but
the received wisdom is that it is very hard to do without it because the
alternatives are worse. As is so often the case, the usefulness of Value-at-
Risk depends on where it is to be used; in some cases it is useful, in others
useless and sometimes outright dangerous. Longerstaey came to regret
his creation, telling Dennis Weatherstone that “[he] created a monster by
asking for that one number.”2

The Dragon Bites—Fat Tails

There are two main challenges for any designer of riskometers:
fat tails and fluctuating risk. When Hurricane Harvey flooded Houston,
Texas, at the end of summer 2017 the meteorologists told us that it was a
once-in-five-hundred-year flood. One might think this was an extreme and
infrequent outcome, except that it was the third once-in-five-hundred-
year flood to hit Houston in a row. The meteorologists underestimated
the flood risk in Houston so badly because they used rainfall observations
over a century to project what might happen in half a millennium. In the
absence of data, all they have is an educated guess.
The most difficult problem in measuring risk is how to deal with very
large and infrequent negative outcomes. Because these significant losses
don’t happen very often, it is quite difficult to exploit historical observa-
tions to predict what sort of losses we might expect in the future: the
fat-tail risk problem. In the words of the former chairman of the Federal
Reserve Alan Greenspan: “As you well know, the biggest problems we
now have with the whole evolution of risk is the fat-tail problem, which is
really creating very large conceptual difficulties. Because as we all know,
the assumption of normality enables us to drop off the huge amount of
complexity in our equations. . . . Because once you start putting in non-
normality assumptions, which is unfortunately what characterizes the real
world, then these issues become extremely difficult.”3
At this juncture it becomes useful to define the term tail risk. The
most common way we describe statistical outcomes is by the bell-shaped
curve, often called the normal or the Gaussian distribution after Johann
Carl Friedrich Gauss, who came up with it in the late eighteenth century.
The bell curve’s tails sit on the far left and right side of the curve, away
from the middle part. They represent the likelihood of outcomes that are
either very large and positive—the right tail—or very negative—the left
tail (Figure 13). We say the tails are fat if the probability of a very large
outcome is higher than we would expect if the data followed the bell-
shaped normal distribution. This means that if I plot the normal distribu-
tion curve and a fat distribution curve together, the fat will lie above the
normal on the left and right sides, as shown in the figure, but be lower in
the middle. The fat tails are the unknown areas with dragons.

Figure 13. Fat tails and large losses. Credit: Lukas Bischoff/IllustrationX.
I have daily observations on the S&P 500 index dating back to 1928.
What is the most obvious way to translate these daily observations into
some measurement of risk? If you ask someone who has just taken a uni-
versity course in statistics, they might say “easy.” Calculate the standard
deviation, assume the data follows the bell-shaped normal distribution,
and voilà, we can calculate the likelihood of any possible outcome.
The standard deviation of stock market returns has another name: vola-
tility. Using my statistical package of choice, R, I find that the daily vola-
tility of the S&P 500 is 1.1 percent. Knowing the volatility is really cool.
I can calculate once-in-five-hundred-year losses, just like the meteorolo-
gists did for Houston, Texas. Suppose I invest $1,000 in the S&P 500.
Based on this analysis, I would expect the typical worst day every year to
give me losses of $26, while the worst day in my 88 year sample should
be a loss of $39, and over a millennium $45. Not so shabby! Except, of
course, it is not like that. If I take my eighty-eight-year history of daily re-
turns on the S&P 500, I find that the average annual worst loss is $47, not
$26. These differences become bigger with time. The worst day over the
past eighty-eight years, 19 October 1987, would give me losses of $229,
not the $39 from the normal. Using volatility and the normal to measure
risk would have lulled me into complacency—the dragons would have
eaten me. This is summarized in Figure 14, which shows both the type of
losses I would expect if I used the normal distribution versus what actu-
ally happened.

Figure 14. The largest losses. Credit: Lukas Bischoff/IllustrationX.
Volatility can be an accurate measure of risk if and only if returns are

normally distributed. Since they are not, volatility is an insufficient and
misleading measure of risk.4 Often it does not really matter all that much,
as the simplicity of volatility is all we need in most cases, situations in
which we care about day-to-day movements and not the dragons, as with
Paul in my example above. However, if we care about the dragons, be
aware: they bite only in times of stress. The next time you hear someone
use volatilities or standard deviations as measures of risk, especially ex-
treme risk, remember the dragons.
So if volatility and the normal distribution are not sufficient, what can
we do? There are strategies, but none is easy. Dealing properly with fat
tails requires well-trained researchers, a lot of data, fast computers, and
considerable time. Meanwhile, it is more expensive to hedge against tail
risk, and it is harder to explain. So, it is about time and resources and
communication and expectations. It is difficult enough to talk about
volatility in an accurate way to nonspecialists, yet, as this text perhaps
indicates, most discussion of fat tails tends to go over the heads of deci-
sion makers. That said, all is not lost: there are two other avenues one can
take. The first is based on understanding the deep nature of where risk
lies, how the people who work in the financial system interact, and how
they create risk.
If that is too complicated, we can go down the purely statistical route.
Here, we can do no better than follow the Dutch lesson because they

are the world leaders in predicting an especially deadly type of risk: flood
risk. For good reason, 26 percent of the land area of the Netherlands and
21 percent of its population are below sea level, and every once in a while
the sea rises so high that it floods all the areas below the sea level. The last
time this happened was in 1953, when 1,863 people drowned. As a conse-
quence, the Dutch government decided to build a tall seawall called the
Delta Works. I went to see the Delta Works in the summer of 2021. It is
an impressive structure, and if you find yourself in the Netherlands, I rec-
ommend paying it and the memorial museum to the 1953 disaster a visit.
The problem was that while it is possible to build a sufficiently high wall
to protect against all flooding, it is really expensive. Therefore, the gov-
ernment of the time decided that it would be acceptable for the Nether-
lands to flood once every ten thousand years. The Dutch statisticians and
engineers were tasked with determining the Delta Works’ height to meet
that once-every-ten-thousand-year requirement. No surprise the Dutch
have become world leaders in tail risk. The mathematical technique they
came up with is formally known as extreme value theory (EVT), also
called the power law after the underlying mathematical equation, or even
Pareto analysis after the Italian economist Vilfredo Pareto, who formu-
lated it in the nineteenth century.
Over the years I have spent a lot of time at Erasmus University in the
Netherlands working with my good friend Casper de Vries, the first per-
son to apply EVT to finance.5 Suppose I use the methods we developed
to estimate the losses on the S&P 500 index. As the S&P 500 risk figure
above shows, we get much closer to what has been observed in history,
and the once-in-a-millennium prediction is undoubtedly much more re-
alistic than the normal distribution predicts.
Even with all that EVT expertise, all the Dutch experts can capture is
an imperfect measure of risk. When they decided on the Delta Works’
height, they used historical observations on sea height, and, being Dutch,
they had records stretching back almost a millennium. Unfortunately,
the world had changed since then—even the 1953 numbers are no longer
accurate, given climate change. It is the same problem that faced the me-
teorologist predicting Houston’s rainfall. The statistical analysis can be
correct only if the world does not change very much.

Fluctuating Risk and the World of ARCH

ARCH and its ilk are fair-weather riskometers.
If we study a long history of financial returns, it becomes clear

that volatility is sometimes very low and, at other times, much higher. In
my eighty-seven-year sample of daily S&P 500 returns, annual volatility
is the lowest in 1964, at 4.8 percent, and the 1960s is the decade with the
lowest volatility. The 2006 volatility was also relatively low at 9.5 percent,
only to quadruple in 2008. We have to go back to the Great Depression
to observe similar magnitudes. History shows that we alternate between
periods of low and high volatility. The specialists call it volatility cluster-
ing (see an example in Figure 15).
Until 1982 the way one obtained volatility was by calculating the stan-
dard deviation of financial returns. Such analysis ignores volatility clus-
ters, so what was needed was a method that captures these changes in vol-
atility, and that was provided by Robert Engle while visiting the London
School of Economics. In 1982 he published a paper with the inauspicious
title “Autoregressive Conditional Heteroskedasticity with Estimates of
the Variance of the United Kingdom Inflation.” The first three words
Figure 15. Volatility clusters. Credit: Lukas Bischoff/IllustrationX.

of the title are abbreviated ARCH. The ARCH model was a revolution-
ary step in the estimation of risk, and in 2003 Engle got a well-deserved
Nobel Prize for developing it. His key insight was to assign weights to
historical observations. Suppose we have three days of data and give today
a 50 percent weight, yesterday 30 percent, and the day before that 20 per-
cent in the volatility calculation. By doing so, we see that the ARCH
model kills two birds with one stone. It both captures the volatility clus-
ters, and its mathematical formulation allows for returns to be fat-tailed.
At least that is the promise.
How well does the ARCH family of models work? Reasonably well.
It certainly fits historical data better than if one simply uses the standard
deviation of returns as volatility. For many applications that is all that is
needed, a reasonably good idea of what type of volatility to expect to-
morrow. For fat tails or systemic risk, we have to look elsewhere. Because
ARCH and its ilk are fair-weather riskometers.
Misusing History
A riskometer should not aim to explain the past well. Instead, its purpose is
to look for currently unknown future mistakes.
The final element in creating a riskometer is data. Data that de-

scribes the infinitely complex financial system. In a perfect world we would
have detailed observations on how market participants see the world: all
their information, models, and analysis; which asset classes are popular
and receiving money and which are deserted and sold; which bankers are
in ascendance, and who is on the way out; the regulatory environment
and how it is expected to evolve. Data of this type would give us a rea-
sonable understanding of what is happening today and allow us to make
solid guesses about likely future scenarios. Well, we don’t have these data.
Mostly we have the outcomes, the prices, and quantities that distill all
the information out there into risk measurements. Inevitably, a lot gets
lost in the process. The question is what. It depends on what the risk-
ometer is meant to capture. All riskometers use historical data to predict
what sort of risk we can expect in the future. How informative are these
historical outcomes? They capture how the market reacts to news but also

how it expects news to evolve. They are a good snapshot of what is hap-
pening at any given time. What such data misses are deep vulnerabilities.
Take the European Central Bank’s systemic risk dashboard yet again. It
told us that systemic risk was, on average, 0.06677640 in 2004, the saf-
est year in its history. What is missed is revealing. We now know, with
the benefit of hindsight, that a crisis was around the corner. The critical
hidden vulnerabilities were all the structured credit products composed
of subprime mortgages and other high-risk assets, all the excessive risk
nobody knew about, the hidden liquidity risk that bites only in times of
stress. The relevant information was out there somewhere, but informa-
tion is not the same as knowledge. If nobody connects the dots, informa-
tion is irrelevant. Every financial institution saw only its own exposures,
not everyone else’s, and nobody thought to add up the numbers. The
information was out there somewhere, but trying to make sense of all the
data coming out of the financial system in real time is like drinking from a
fire hose. We have to pick and choose. And in 2007 almost nobody picked
subprime and structured credit. It is easy, with the benefit of hindsight,
to ask, “How could you miss that? It was all in front of you.” But in
real time, not all that easy. Ultimately, most riskometers use just trading
prices and volumes, disregarding all the other relevant bits of informa-
tion. It is not because someone is deliberately trying to obfuscate. It is
just all we can do technically.
The primary challenge in creating riskometers is that risk comes from
the future, but we only know the past. That creates particular problems,
not least what is known in statistics as data snooping. The problem arises
because before using a riskometer to make important decisions, we would
like to know how good it is. We would ideally put it in practice, wait a few
years, and then pass judgment. We are too impatient for that, and what is
usually done instead is to figure out how well a riskometer measures his-
tory using a procedure called backtesting.
Suppose I take my long sample of S&P 500 prices and use all the obser-
vations from the beginning of the sample until the end of the year 1999
to forecast risk with a riskometer on the first day of the new millennium.
That gives me one forecast observation, which I can compare to the ac-
tual market outcome that day. I then move up one day and use all the
information up to and including the first day of the year 2000 to forecast

risk on the second day. By repeating this every day until today I will get a
few thousand observations and then use that information to evaluate the
quality of my riskometer. That is backtesting, something routinely done
in practice.
However, while it sounds like a fantastic way to evaluate the quality of
a riskometer, there is a problem. I know what happened between the year
2000 and now—risk was very low in the mid-2000s, we had a big crisis
in 2008, and all the other things that happened. The temptation is then
to try to explain as much as possible of what happened in the past—espe-
cially finding a way to predict the crisis of 2008.
When I was in graduate school, I took a course in Soviet economics,
and the professor told us a Russian joke: the best Soviet historians are
experts in forecasting history because government policies were justified
with reference to history. Thus history had to be “correct” if the histo-
rians were to survive. If I try enough riskometers, I will explain history
well, and I will predict the crisis in 2008 years in advance. A lot of re-
search has done precisely that. All of that work is useless because of what
statisticians call spurious correlation, in which variables seem related to
each other even though in reality they have no such causal relationship.
There are many examples of spurious correlations; the correlation be-
tween divorce rates in the American state of Maine and the consumption
of margarine is 99.3 percent.6 When I told this to someone, he responded
by saying, “It is a good thing I stopped eating margarine,” thereby fall-
ing for two fallacies: spurious correlations and assuming that this general
observation applies to the individual.
What explains the 99.3 percent correlation between divorce rates in
Maine and margarine consumption is that if I calculate the correlation be-
tween every possible economic time series available—and there are hun-
dreds of thousands of them—I will find some that are highly correlated
purely by random chance. Some of these might even appear to be more
sensible than the spurious margarine and divorce correlation, leading the
unwary, or unethical, analyst to the wrong conclusion. As H. L. Mencken
put it: “For every complex problem there is an answer that is clear, simple,
and wrong.” When it comes to riskometers, if I try one out and one only
I will get the correct confidence intervals for my risk measurements. If,
however, I arrive at the same riskometer as a result of trying out a large

number of explanatory variables and model specifications, these confi-

dence intervals will no longer be accurate. Any riskometer that performs
poorly in forecasting history will be changed, leaving us with those that
explain history well, like the Soviet historians. The problem is that such
riskometers will tend to do poorly in forecasting the future. Not much
more accurate than 99.3 percent is about the future relationship between
margarine and divorce in Maine.
Why Risk Is Hard to Measure

All models are wrong, but some are useful.
—George Box
A few years ago, when I was reading the Basel III proposals for
the global postcrisis financial regulations, I asked myself: How well does
what is being proposed work? How accurate are the riskometers that im-
portant regulations are founded on? To my surprise, I found almost no
public research on this, and as far as I can tell, nowhere in the thousands
of pages of Basel III is the accuracy of the mandated risk measurements
discussed. There is plenty of work on the various technical aspects of risk-
ometers, along the lines of “Does Method A work better than Method B
in some particular situation?” Indeed useful, but it does not answer the
more fundamental question of how accurate the riskometers we depend
on to keep us safe from finance are.
First, I did a blog piece on this question and then joined forces with
Chen Zhou, a coauthor of mine now working in the Netherlands.7 Our
paper is called “Why Risk Is So Hard to Measure,” and what it does is
to investigate how the most common riskometers work in practice. We
started by taking daily prices of all publicly traded stocks in the United
States from the early 1960s, measuring both the risk and the accuracy
of those risk estimates by the 99 percent confidence bound, that is, the
range of values where the risk measures will lie 99 percent of the time. We
then did a Monte Carlo experiment, simulating random outcomes from
our computer world based on our assumptions of how the world works,
measuring the risk in those outcomes. The benefit of such an approach is
that because we created the world, we knew exactly what the risk was and
therefore how accurate the riskometer’s estimations were.

What we found is that the number of observations used is crucial to

estimating risk accurately. Not unexpected, but what was surprising is
how many observations one needs. To get a reasonably accurate measure-
ment of risk, we require many thousands of daily observations. Consider-
ing that there are about 2,500 days when the financial markets are open
every decade, we need many decades of data. When we use four years of
data, the Value-at-Risk estimates are between $74 and $151, where the
true value is $100. Now four years of data are a lot; we usually see much
less used in practice. Three hundred trading days, or fifteen months, is
much more common, and here the Value-at-Risk measurements range
from $61 to $246.
What this means in practice is that two banks holding the same as-
sets could come to very different conclusions about risk. One might get
$61 and the other $246. The bank with $61 would think all is fine and
sharply increase its risk. The other bank faces $246 and would be forced
to scale down its risk. Very different outcomes when the underlying risk is
precisely the same because, after all, the two banks hold the same assets.
The only difference is that they use different riskometers. The bank with
the high numbers might want to switch to the low-number riskometer
and even do so in consultation with the regulators. The result would be
a race to the bottom, all banks chasing the same riskometer. Still, there
is no guarantee that the riskometer with the lowest number today will
do the same tomorrow; it might even provide the highest. Gaming the
riskometer in this may not be all that useful in the long run. The reason
is model risk.
When the Covid-19 crisis started in February 2020, it didn’t take long
for research explaining the virus’s propagation to emerge. Many models
were created, and while Covid-19 is a virus, the modelers weren’t limited
to epidemiologists, as computer scientists, physicists, doctors, statisti-
cians, economists, and financiers all got in on the action. Their models
were initially quite contradictory, and it seemed like the policy makers
often picked the models that best explained what they wanted to do in
the first place. Over time, as more and more information emerged, the
models started to converge.
The technical name for the degree of inaccuracy in the Covid-19 models
is model risk, the risk of making wrong decisions when relying on models.

Model risk is inversely related to the complexity of the problem at hand.

In the Covid-19 crisis the models were desperately needed at the outset,
as nobody knew how to counteract the virus. A lot of the early modelers
came under unfair criticism for the low quality of their initial predictions.
Yes, they turned out to be wrong in the end, but it is often better to have
poor predictions than none. And by the time the models become reason-
ably accurate they are no longer needed. It is always like that. Model risk
is lowest when the thing being modeled closely resembles something al-
ready well understood, like with airplanes. It is easy to simulate the move-
ment of aircraft in the computer because all the applicable laws of physics
are well known. Only the details differ. With a new disease like Covid-19
model risk is inevitably very high.
The same applies to the financial models that underpin the riskometers.
We are quite good at measuring day-to-day risk in highly liquid stocks,
when nothing much is going on in the markets, having a lot of data to
train the models and quick feedback if the models fail. Model risk is then
low. It is much harder to measure tail risk, and methods for doing so have
a lot of model risk. I looked at model risk in risk models for the stocks
of all American financial companies from 1970 to 2012 in a recent paper
titled “Model Risk of Risk Models.” We measured model risk by taking
the six most commonly used statistical methods for estimating risk, de-
fining model risk by how much they disagree. What we found was quite
striking. Model risk is lowest when the financial markets are quiet, as
in the years before the crisis in 2007, the mid-1990s, and the late 1970s.
Model risk is highest in times of crises, as in 2008. In other words, risk-
ometers have the annoying tendency of being reliable when all is good
and failing miserably when needed the most. Just like European Central
Bank’s systemic risk riskometer, almost every riskometer in use in 2006
told us risk was at a historical low. The riskometers all agreed, and model
risk was low. Unfortunately, they all agreed on the wrong thing.
Riskometers are used to make crucial decisions: regulating banks, con-

trolling pension fund investments, managing insurance company reserves,
and the day-to-day investment decisions of every financial institution.
They are fundamental to the robo-advisors now increasingly common in
investment management. The financial system as we know it today could

not function without riskometers. Those who understand the riskometer,

its strengths and weaknesses, know when it is a useful tool: perhaps in the
short-run comparison of portfolios where the underlying financial market
conditions are stable.
Unfortunately, too many people treat the numbers produced by a risk-
ometer as being akin to sacred truth and don’t question them nearly
enough. Riskometers are too often used inappropriately to make deci-
sions on tail risk in long-run investments, precisely the types of risk most
likely to negatively impact banks, pension funds, insurance companies,
individual savers, investors, and indeed the entire financial system. These
are the types of risk that riskometers are least useful for—when we get
bitten by the dragon.

6
Ideas Matter:
Risk and Uncertainty
Stability is destabilizing.
—Hyman Minsky
Uncertainty is not risk, an important distinction made by three

early twentieth-century economists: the American Frank Knight, the
Briton John Maynard Keynes, and the Austrian Friedrich August von
Hayek. It is one of the most important methodological breakthroughs
in the field of economics. The three economists expressed their disquiet
94

RISK AND UNCERTAINTY 95
with the established economic dogma at the time—nineteenth-cen-

tury classical economics—which maintained that the economy obeyed
Newtonian-like laws of nature. Such laws could be described with precise
mathematical equations spanning the entire range of possible outcomes,
discoverable by a simple measurement. An example is the classical theory
of prices, which says that prices are determined solely by the level of out-
put, technology, and wages.
A crucial but implicit assumption of such nineteenth-century thought is
ergodicity, a mathematical concept of stability. Ergodicity, a key assump-
tion in the natural sciences, implies that by doing repeated experiments or
observing the same phenomena over time we learn nature’s laws. All we
need are enough observations. The three thinkers rejected the assump-
tion of ergodicity for the economy as too simplistic and plainly wrong. In
doing so, they were a part of the wholesale revision of nineteenth-century
values and assumptions about society brought about by the World War I.
It started with Professor Frank Knight of the University of Chicago
arguing that the concept of risk was fundamentally different from uncer-
tainty. Risk is something one can quantify, describe with mathematical
and statistical models, and hence translate into numbers. If the economic
world is ergodic, risk is the correct way to describe it. So long as some-
thing is risky, one can make confident statements: there is a 1 percent
probability that England will win the next football World Cup, there is
a 25 percent chance Greece will need another bailout in the next three
years, there is a 48 percent chance we will make money by investing in
Amazon over the next six months. The only way we can make such ac-
curate declarations is if the statistical and mathematical models compre-
hensively describe the world, and every observer of the economy sees risk
in the same way. There is no room for subjective judgment. If one cannot
quantify the probabilities, we can say only that Greece may need a bail-
out. The Greek bailouts lie in the realm of uncertainty, not risk.
Uncertainty captures outcomes that cannot be adequately described
mathematically. When it comes to making decisions, uncertainty implies
that it is not sufficient to simply observe the world around us and process
those observations with statistical techniques. Something is missing. We
need to recognize that each person is different and sees and understands
the economy differently, via their own information and objectives and
abilities. If we do not, we will simply end up making decisions based on

96 RISK AND UNCERTAINTY
how we think the world is or ought to be, not how it is. Uncertainty, not
risk, drives economic activity. Investments and profits and losses happen
because we do not anticipate perfectly how events occur. Since every per-
son has different expectations about the future, their expectations drive
the uncertainty that drives the economy.
The same year Knight published his work on risk and uncertainty, John
Maynard Keynes took a more nuanced view, which he then further re-
fined in his General Theory of Interest, Employment and Money (1936): “By
‘uncertain’ knowledge . . . I do not mean merely to distinguish what is
known for certain from what is only probable. The game of roulette is
not subject, in this sense, to uncertainty. . . . The sense in which I am us-
ing the term is that in which the prospect of a European war is uncertain,
or the price of copper and the rate of interest twenty years hence, or the
obsolescence of a new invention. . . . About these matters there is no
scientific basis on which to form any calculable probability whatever. We
simply do not know!”1
His starting point was that humans are not naturally rational beings able
to anticipate and forecast everything perfectly, as the nineteenth-century
classical economists would have it. Instead, people routinely follow animal
spirits, instincts that guide human behavior. Keynes came to be rather
dismissive of statistical analysis, rejecting the notion that decisions can be
made on the basis of the frequency of past events—ergodicity. Instead,
he focused on the “degrees of beliefs” that humans can have, given their
knowledge at a given time, about the occurrence of future events.
My favorite expression of that sentiment comes from an economist
customarily thought of as being very far from Keynes: Ludwig von Mises,
one of the leaders of the Austrian school. Mises criticized economet-
rics—the statistical analysis of the economy—in 1962: “As a method of
economic analysis econometrics is a childish play with figures that do not
contribute anything to the elucidation of the problems of economic real-
ity.”2 It is delightfully ironic that many of Keynes’s followers went on to
do exactly what Mises and Keynes argued against, creating statistical and
mathematical models of the economy, treating uncertainty as risk in the
same way the nineteenth-century classical economists did.
The Nobel Prize winner George A. Akerlof, in a paper called “What
They Were Thinking Then: The Consequences for Macroeconomics dur-
ing the Past 60 Years,” makes the point that Keynes’s disciples deliberately

ignored his warnings about risk and financial crises so they could derive
simple mathematical models of how the macroeconomy works—models
which soon were proven to be not only wrong but also dangerous.
In 1945 Friedrich von Hayek, while a professor at the London School of
Economics, published an article titled “Use of Knowledge in Society” in
which he expressed views similar to those of Akerlof, agreeing with both
Knight and Keynes that quantifying risk was impossible. Hayek, however,
came to the point from a different direction. Instead of people being
motivated by Keynes’s animal spirits, he argued that it was technically
impossible to describe the world with precise mathematical statements:
“If we possess all the relevant information, if we can start out from a given
system of preferences, and if we command complete knowledge of avail-
able means, the problem which remains is purely one of logic. . . . This,
however, is emphatically not the economic problem which society faces.”3
Hayek was writing during World War II, when the economic policy de-
bate was between believers in central planning and Soviet-model scientific
socialism and those preferring a market-based economy. Most thinkers of
that era advocated central planning, seeing the capitalist system as crisis
prone and inefficient, where the Great Depression was caused by the fail-
ure of capitalism, while the Soviet Union escaped the Depression because
it was centrally planned. Almost every country in the world at the time
was actively considering central planning.
Hayek disagreed. To him, knowledge is dispersed among the various
individuals who make up society, information that is impossible to ag-
gregate into perfect knowledge. The farmer knows much more about
his fields than anyone else, he knows how best to plant his crops, and he
has a direct economic incentive to be as knowledgeable about his land
as possible. No central authority can acquire such information; all they
can hope for is some high-level summary information. This is why col-
lective farming does not work. The farmer makes better decisions than
the ministry of agriculture, and central planning of the economy cannot
be successful. In Hayek’s view, as long as markets are free of government
interference, market prices solve the problem of uncertainty, distilling es-
sential information into one number: the price.
Even though Keynes and Hayek are often seen as having very differ-
ent economic philosophies—a distinction much amplified by their disci-
ples—their views on uncertainty are quite similar. Keynes focused on how

uncertainty affects decision-making, while Hayek was more interested in

understanding how one can minimize and overcome uncertainty. These
views are highly complementary. Not surprisingly. They were both work-
ing in England at the same time, remained friends, and corresponded
extensively.
Where Keynes and Hayek diverge is in their policy conclusions. Keynes
argued that because of the markets’ animal spirits, they tend to get into
an undesirable state, justifying corrective government intervention. If we
don’t know how to quantify the risk of future investment decisions, they
“can only be taken as a result of animal spirits.” If the animal spirits are
too pessimistic, the government needs to increase confidence and reduce
uncertainty, thereby increasing investment. Hayek came to the opposite
conclusion, maintaining that while individuals and organizations might
know little about the future, the government knows even less. A company
knows the business it is in and hence has a better idea of future uncertain-
ties than the government. Because the government knows less about the
economy, its decisions are worse on average if it intervenes.
It therefore seems likely that if Keynes were writing today, he would
be in support of the objectives of the macroprudential regulators, the
crafting of rules to ensure a robust and safe financial system. After all,
his magnum opus, The General Theory, came out in 1936, right after the
Great Depression, which is often seen as having been caused by the free
market’s failure. Yet I suspect Keynes would not like how the regulations
are conceived today, at least to the extent they are based on risk measure-
ments; he would see them in the same way he saw nineteenth-century
classical economics. I do not think Hayek would like either the objectives
or the implementation of today’s financial regulations. He might argue
that because the regulators do not and cannot know all the risks in the
financial system, all they accomplish is increasing costs as well as inef-
ficiency and higher risk of crises. The central banks are making the same
mistakes as the central planners of yesteryear.
Hayek and Keynes, perhaps the two most influential economists of the
twentieth century, are among the most misunderstood thinkers in eco-
nomics. They have legions of groupies who delight in misrepresenting
their views. A friend of mine, Hannes Gissurarson, recalls in his new book
a dinner he had with Hayek in 1980 during which Hayek said to him, “We

should not become Hayekians, as the Keynesians were much worse than
Keynes and the Marxists much worse than Marx.”
The views of Knight, Keynes, and Hayek on risk and uncertainty were
mostly ignored after World War II. The disciples of Keynes, the most
influential of the three, disregarded this aspect of his work, preferring
instead to draw on the classical nineteenth-century views on risk in build-
ing the Keynesian models of the era. When Keynes himself stressed the
importance of uncertainty, the disciples paid little attention. Of course,
they did not see it that way—most economists of the era saw themselves
as the intellectual heirs to Keynes.
There are many reasons for this rejection of uncertainty in favor of
risk. Much better data collection after the war, new statistical techniques,
and the availability of computers to do calculations all led to a sense of
can-do. It was inevitable that we neglected uncertainty in our constant
attempts at controlling our environment. Risk and mathematical descrip-
tions of the world are essential for that, and uncertainty just gets in the
way. It tells people what they do not want to hear: it is not as easy to
measure and control. Invariably, the risk view wins out.
The person who formalized that best was Wassily Leontief, an econo-
mist educated at the University of Leningrad who became a Harvard pro-
fessor and eventually a recipient of the Nobel Prize. He saw the objective
of economics as the collection of facts and figures followed by mathemat-
ical models describing the relationships. This culminated in an approach
he called the input–output model, which reduces the entire economy
into a set of equations. The output of one sector becomes either an input
into other sectors or final consumption. Leontief’s model became very
influential in the mid-twentieth century and was one of the foundations
of central planning as practiced by the Soviet Union and other commu-
nist countries to this day.
The input–output model had a significant impact on the development
of a method called linear programming at the statistical control group
of the US Army in World War II. It was successfully put into practice in
organizing the Berlin Airlift. The army flew the maximum amount of
cargo into Berlin, given constraints like the number of available runways
in Berlin and the prevailing weather. A small set of equations can describe
the problem, and what matters is risk, not uncertainty.

One member of the army’s linear programming team was Robert Mc-
Namara, who became the secretary of defense in the 1960s and was famous
for using Leontief’s philosophy in conducting the Vietnam War.4 His
management philosophy was based on measuring everything that could
be measured, most famously body bags, and then using those measure-
ments to control outcomes. What could not be measured did not matter.
The problem with this approach was ably demonstrated by the sociologist
Daniel Yankelovitch, who called it the McNamara fallacy: “The first step
is to measure whatever can be easily measured. This is OK as far as it goes.
The second step is to disregard that which cannot be easily measured or
give it an arbitrary quantitative value. This is artificial and misleading.
The third step is to presume that what cannot be measured easily really
is not important. This is blindness. The fourth step is to say that what
cannot be easily measured does not exist. This is suicide.”5 In all fairness,
McNamara was trying to put structure on a very complex problem, using
quantitative methods as tools in all the Johnson White House’s political
battles. He was not the first to use statistics that way and not the last. It
has been quite common in the Covid-19 crisis.
One of McNamara’s successors, Donald Rumsfeld, expressed a much
better understanding of risk and uncertainty in 2002. While widely ridi-
culed for it at the time, it has come to be seen as a brilliant statement of
decision-making in a period of uncertainty: “Reports that say that some-
thing has not happened are always interesting to me because as we know,
there are known knowns; there are things we know we know. We also
know there are known unknowns; that is to say, we know some things we
do not know. But there are also unknown unknowns—the ones we do
not know we do not know. And if one looks throughout the history of
our country and other free countries, it is the latter category that tends
to be the difficult ones.”6
The problem with Leontief’s input–output model is the same as that
of nineteenth-century classical economics, McNamara’s warfighting phi-
losophy, modern financial regulations, and risk measurements. They all
create a simple, caricature view of a world that is infinitely complex. The
models assume all relevant factors can be summed up into simple math-
ematical equations and do not leave any room for uncertainty or com-
plexity or technological progress. Perhaps most important, they depend
on accurate measurements.

When I studied quantitative methods in graduate school, my professor,

James Henderson, told us the story of when he used to work at the IMF
on Latin America. The fund had encouraged the Latin American coun-
tries to create input–output models of their economies. This worked
quite well until Henderson noticed that, according to the input–output
model, the Latin American economies were not only very similar to each
other but also rather resembled the US economy. This was surprising
because the Latin American countries were much poorer than the United
States and their industries quite different. When Henderson looked into
this, he was told by the Latin countries that their statistical collection
was not as detailed as that of the United States, and when a number was
missing they would just use the US number.
The quality of measurements is only as good as the person making
them. A common problem, as noted by Josiah Stamp, one of the leading
statisticians in the United Kingdom in the early twentieth century: “The
government are very keen on amassing statistics. They collect them, add
them, raise them to the nth power, take the cube root and prepare won-
derful diagrams. But you must never forget that every one of these figures
comes in the first instance from the village watchman, who just puts down
what he damn pleases.”7
Subjective Probabilities
One person who continued the work of Keynes and Hayek on risk
and uncertainty was George Lennox Sharman Shackle. Shackle started
writing his PhD at the London School of Economics in the 1930s, with
Hayek as his advisor. However, after Keynes published his General Theory,
Shackle dropped Hayek and did his thesis on Keynes’s ideas. Hayek did
not hold it against him, and they remained friends. Shackle is one of
the great underrated economists. I don’t recall seeing him mentioned
anywhere when I was in graduate school, and it was not until my LSE
colleague Charles Goodhart brought him to my attention that I started
reading his work. Shackle argued that it was not possible to calculate the
probability distribution of all outcomes and thereby make rational eco-
nomic decisions. Economic data is not ergodic.8
Unlike Knight and Keynes, however, and their sharp distinction be-
tween risk and uncertainty, Shackle argued that something in between

existed, a degree of knowledge about an outcome’s probabilities. He

called this subjective probability, the confidence in the future occurrence
of a particular event. Subjective probability incorporates the probability
of someone’s judgment being incorrect, implying that an economic vari-
able, perhaps prices, can diverge from what should be expected. This
foresaw what has become known as psychological and behavioral eco-
nomics. Many observers have echoed Shackle’s views on probability, one
of whom, Warren Buffett, voiced his own notion of reasoned probability:
“The riskiness of an investment is not measured by beta (a Wall Street
term encompassing volatility and often used in measuring risk) but rather
by the probability—the reasoned probability—of that investment caus-
ing its owner a loss of purchasing power over his contemplated holding
period. Assets can fluctuate greatly in price and not be risky as long as
they are reasonably certain to deliver increased purchasing power over
their holding period. And as we will see, a nonfluctuating asset can be
laden with risk.”9
Stability Is Destabilizing
An important element is missing in the work of the four think-
ers discussed above, and that is how uncertainty arises and affects the
likelihood of outcomes. Hyman Minsky provided that link. Focusing on
financial crises, he argued that perceptions of risk affect peoples’ risk-
taking behavior and hence the likelihood of financial crises far into the fu-
ture. Minsky developed his theories in the context of the mid-twentieth-
century economic theories of his time, such as Leontief’s input–output
models. Minsky rejected these as being too simplistic because they ig-
nored how investment decisions are made and how firms are financed.
His theory of investment and crises was based on distinguishing between
three types of financing. The first, which he called hedge financing, is
the safest. Firms borrow little, and when they do they repay the loans
directly out of cash flow. The second type is speculative financing, riskier
since firms rely on cash flow to repay interest but roll the principle over.
The most dangerous, Ponzi financing, is where the cash flow is not suf-
ficient to repay either principal or interest, so firms are betting that the
underlying asset will appreciate enough to repay their liabilities. The cur-

rent trend in Britain for buy-to-let real estate is a prime example of Ponzi
financing: people buy properties to rent them out, financing mortgage
payments out of rent and increases in the price of the property.
Hedge financing is the most stable, but the other two are much more
tempting to investors. Moreover, when the economy is growing there
seems to be little reason to give up profits and play it safe. Indeed, it is dif-
ficult to do so. Suppose an economy starts safe, using only hedge financ-
ing. This installs confidence and motivates people to take more risk and
use speculative financing. In the beginning, it all looks good; economic
growth increases, making Ponzi financing increasingly attractive. But over
time we run out of feasible investments, and ultimately it all ends abruptly
and a crisis ensues. Financial crises happen because we think the good
times will last forever, and there is no reason not to make use of Ponzi
financing. Investors want to take on more risk, often helped by a lax reg-
ulatory environment and government encouragement. Ultimately, this
culminates in an unsustainable speculative bubble and a crisis.10
The conditions for a crisis are, therefore, ripest when we think risk is
low. This means that one of the best predictors of a financial crisis being
around the corner is when the pundits start talking about the current pros-
perity as lasting forever, such as the permanent era of stability in the 1920s
and the great moderation in the two decades before 2007. Often justified
by arguing that the laws of economics do not apply because “our country
has just such fantastic economic policy” or “we have become so clever that
we have learned how to prevent crises, investing optimally and behaving
rationally, ensuring permanent stability and high growth.” Not so fast.
When we start seeing cultural reasons for prosperity, it is time to run.
Not surprisingly, Minsky did not get much recognition in his lifetime.
He was not doing mainstream economics and was rejected by the Keynes-
ian school he came from, both because of his emphasis on uncertainty
and because of his criticism of deterministic economic models. Yet he
was never wholly forgotten, and after the global crisis in 2008 he be-
came quite the celebrity. People in the know often call crises a Minsky
moment because his theory explains them so well. “If there is excessive
optimism in the boom period, it will lead to an accumulation of conflicts
[in the economy], which may end up with a so-called Minsky moment,”
said Zhou Xiaochuan, then governor of the People’s Bank of China, in

October 2017. Meanwhile Minsky’s critics and those who shunned him
are long forgotten.
Minsky’s instability hypothesis implies that an observation of low risk
should create the conditions for a future crisis. This proposition was ex-
pressed in 2014 by the then chairwoman of the Federal Reserve Janet
Yellen: “Volatility in markets is at low levels. . . . [T]o the extent that low
levels of volatility may induce risk taking behavior . . . is a concern to me
and to the Committee.”11
The relationship between low volatility and crises has remained con-
jectural, not verified empirically. That got me curious, and I joined forces
with a couple of coauthors, Marcela Valenzuela and Ilknur Zer, to see if
such a relationship existed, writing a paper on the subject titled “Learn-
ing from History: Volatility and Financial Crises.” There are two reasons
this had not been done earlier. The first is that we need a very long history
of volatilities and crises, and the necessary data was not readily available.
To that end we collected data on monthly stock market observations,
spanning 60 countries and 211 years.
Still not sufficient. If we test a statistical model of the relationship be-
tween volatility and crises, we find it does not exist. That, however, does
not mean there is no relationship; it is merely more complex. What mat-
ters is volatility being different from what people have come to expect. To
test that hypothesis, we first need to estimate expected volatility and then
use deviations from that in a statistical model. Low volatility is, then,
when volatility is below expectations and, conversely, high volatility is
where volatility exceeds expectations. And low volatility turns out to be
statistically significant.
We find a clear chain of causality. Unexpectedly low volatility sends
the all-clear signal. We therefore have no qualms about taking on more
risk. To do that, we borrow money to make risky investments. In the
short run, all is fine, but over time, some or even many of the loans turn
out to be dodgier than expected. Perhaps they were used to invest in real
estate, creating a house price bubble that eventually bursts. Not surpris-
ingly, loan defaults mount, banks get into difficulty, and a crisis ensues
(Figure 17). Unexpectedly low volatility predicts both rapid credit growth
(increasing leverage in the banking system) and the incidence of crises up
to ten years into the future. Perhaps surprisingly this causal relationship

Figure 17. Volatility and crises. Credit: Lukas Bischoff/IllustrationX.
holds only for unexpectedly low volatility. High volatility has no predic-
tive power for crises. It is only associated with them. In other words, high
volatility happens at the same time as crises and cannot be used as a crisis
predictor, only as a crisis indicator.
Goodhart’s Law
The last step in the evolution of how we see risk and uncertainty
is the analysis of what happens when governments try to regulate eco-
nomic activity by targeting measurements of the economy. This link was
made by my LSE colleague and, I am privileged to say, coauthor Charles
Goodhart. Very few are as adept at distilling complicated government
policies into the bare essentials, identifying what works, and clearly dem-
onstrating why other policies are destined for failure.
He is best known for Goodhart’s law: “Any observed statistical regu-
larity will tend to collapse once pressure is placed upon it for control
purposes.”12 His law has a natural implication for regulations. Once the
government tries to regulate some activities, they immediately become
unreliable as indicators of economic trends. The context of this statement
relates to the use of monetary policy to achieve the optimal trade-off
between inflation and unemployment. This all started when Bill Phil-
lips, a professor of economics at the LSE, wrote a paper implying that
there might be a long-term, stable, and negative relationship between
unemployment and inflation. The lower the unemployment, the higher
the inflation. Various economists then suggested that central banks and

the fiscal authorities—because fiscal policy was thought to be the more

powerful at that time—could use their policy instruments to choose the
optimal trade-off between inflation and unemployment.
Putting his training as an engineer to work, Phillips developed the
Monetary National Income Analogue Computer (MONIAC) in 1949, us-
ing hydraulics to model the British economy’s workings. The MONIAC
was a large contraption, about 2 meters high, 1.2 meters wide, and 1 me-
ter deep, composed of several transparent plastic tanks linked by pipes.
Each tank captured some aspect of the UK economy, and the flow of
money was represented by colored water flowing through the pipes. Eco-
nomic principles determined the flow of water between the tanks. The
MONIAC operator could set various economic parameters, such as tax
rates and investment rates, by turning the valves controlling the flow be-
tween the tanks. Fundamental to the MONIAC was the negative associa-
tion between inflation and unemployment, a relationship that the central
banks could exploit. If unemployment was too high, all the central banks
had to do was print money.
Although the mechanical MONIAC was too unwieldy for practical
use, the underlying mathematical equations were soon translated into the
then recently available digital computers. Unsurprisingly, the idea behind
the Phillips curve was quite attractive to policy makers. The central bank
can control the level of economic activity simply by tweaking inflation.
The resulting activist monetary policy became quite popular in many
countries.
I experienced Phillips’s notion of monetary policy firsthand in my first
job as an economist. A summer intern at the Central Bank of Iceland in
1987, I was asked to work on a model of optimal currency devaluations.
It was based on taking the five most important sectors of the economy
and finding out by first approximation how they would be affected if the
Icelandic krona was devalued. A 1980s version of Phillips’s MONIAC and
based on macroeconomic models that were already twenty-five years out
of date and proven to be wrong. I did manage to upset my superiors, not
by telling them the model was wrong—which I did—but by program-
ming it with a menu-driven spreadsheet so that they could simply plug
in a number and immediately see pretty charts and tables for how the
economy would react. They got upset because the very simplicity of the

spreadsheet exposed the naïveté of the central bank’s economic model.

You see, the previous version of the model was programmed fifteen years
earlier in an obscure programming language called Fortran that nobody
understood, allowing the senior staff in the central bank to believe the
analysis was rigorous.
The problem with Phillips’s model and the other macro models of the
era is that it is not possible to exploit the negative correlation between un-
employment and inflation for policy purposes. The Nobel laureate Robert
Lucas expressed that most clearly in his “Econometric Policy Evaluation:
A Critique,” in what has come to be known as the Lucas critique: “Given
that the structure of an econometric model consists of optimal decision
rules of economic agents, and that optimal decision rules vary systemati-
cally with changes in the structure of series relevant to the decision maker,
it follows that any change in policy will systematically alter the structure
of econometric models.”13
The observed relationship between monetary growth, inflation, and
real economic activity holds only as long as the central bank is not try-
ing to use monetary policy to achieve and reduce unemployment. Once
it tries to do so, the relationships immediately break down. The reason is
that the causality runs in one direction only. Inflation is high because the
economy is growing rapidly, but not the other way around. If the cen-
tral bank tries to exploit the association, it might succeed once because
people mistake the new money for actual prosperity. But they are not
going to be repeatedly fooled by the fact that the central bank is printing
money. Instead, the freshly printed money just ends up increasing infla-
tion because people understand that the central bank is pulling the wool
over their eyes, and they will simply demand higher wages to pay more
for the same goods but will not work any harder, so the economy does
not grow. The relationship between unemployment and inflation breaks
down precisely when the central bank tries to exploit it.
While Goodhart originally phrased his law in the context of monetary
policy, he has since extended his interpretation to financial regulations.
He argued in Risk, Uncertainty and Financial Stability that, following
Shackle’s approach, measurements of risk may be useful for the control of
day-to-day outcomes in financial markets while being mostly useless for
anyone concerned with the risk of large, infrequent adverse outcomes—

systemic risk or tail risk—because such analysis needs to be forward look-

ing. By definition, no risk measurement based on historical data is ever
able to adequately capture the probability of future outcomes and will
especially miss out on large and infrequent events; events that include an
element of surprise—the unexpected and unexpectable.
Suppose a regulator observes that banks with high-risk readings have
historically been more prone to fail than those with low risk. In response,
she persuades her agency to impose a rule whereby any bank with a high-
risk reading will be subject to extra scrutiny and required to hold ad-
ditional levels of capital. Will doing so make the financial system more
stable? Not necessarily. First, the relationship between the high-risk read-
ings and bank failures could be misleading. More important, the bank
is likely to target the measurements and try to make its readings look as
low as possible while not changing its risk-taking activities. And they will
not find this all that difficult to do. Banks find it easy to manipulate risk
measurements in a way that is fully compliant with the letter of financial
regulations, and there may be nothing the authorities can do about it
even if they find out. Therefore, if the financial authorities use some risk
measurements to control the financial system, they may just end up dis-
covering that the measurements cease to be reliable: a direct application
of Goodhart’s law.
The idea that the economy obeys laws akin to the laws of physics is
seductive. All we have to worry about is risk, and we can ignore uncer-
tainty. We then can control an unruly economy, allocate resources in the
best possible way, manage risk, and prevent calamities. If only it were
so. The economy is different from the physical world. It is based on the
behavior of human beings, many of whom are intent on doing exactly
what they want to do—rules or no rules. Recognizing that fact is the
genius of the six thinkers discussed above. They all understood it is only
in the physical world that we can describe the relationship between out-
comes and probabilities precisely—where all that matters is risk and un-
certainty is irrelevant. In physics, math captures all; not so in the society
of humans that form the economy.

7
Endogenous Risk
A data-driven risk management process cannot capture endogenous risk,

only exogenous risk.
The pedestrian Millennium Bridge, opened by Queen Eliza-

beth II on 10 June 2000, was the first new bridge to span the River
Thames in London for a century. Thousands of people crowded onto
the bridge on the day of its opening, supposedly not a problem, as the
bridge was designed to cope easily with such large numbers. However,
109

110 ENDOGENOUS RISK
Figure 19. Albert Bridge in London.
within moments of the bridge’s opening it began to wobble violently,

to the great embarrassment of its designers, Arup Engineering and the
architect Norman Foster. In the process it earned the nickname the Wob-
bly Bridge. The swaying came as a surprise, as no such outcome had been
predicted by Arup’s extensive computer modeling and human testing.
It is well known that soldiers marching across bridges can cause them to
collapse—the reason they are asked to break step when crossing bridges.
Marching soldiers generate harmonized frequencies that can create feed-
back between the internal frequency of a bridge and soldiers’ steps, lead-
ing to a collapse. If you look closely, you can often see signs on bridges
telling soldiers to break step when marching (Figure 19).
The problem caused by soldiers marching across bridges is well known,
and Arup did consider that in their modeling. They even had a precise
number: it would take 167 marching soldiers for the Millennium Bridge
to wobble. But those crowding on the bridge on opening day were not
soldiers, but people from all walks of life who, for the most part, did not
know each other. The chance of them spontaneously marching was con-
sidered next to impossible. But the designers missed something impor-

ENDOGENOUS RISK 111
tant. Every bridge is designed to move with the elements, and the Mil-
lennium Bridge was supposed to sway gently in response to the Thames
breeze. Soon after it opened it was hit by a gust of wind and moved side-
ways, as expected. The pedestrians’ natural reaction was to adjust their
stance to regain balance—lean against the movement. Herein lies the
problem. They pushed the bridge back, making it sway even more. As an
ever-increasing number of pedestrians tried not to fall, the bridge swayed
more and more. What happened was that when at least 167 pedestrians
crowded onto the bridge in windy conditions, a feedback loop emerged
(Figure 20). It was always present, lurking in the background, but needed
particular conditions to emerge.
It is the same in the financial system. The distinguishing feature of
all serious financial crises is that they gather momentum from the en-
dogenous responses of the market participants themselves, like a tropical
storm over a warm sea gains energy as it develops. As financial condi-
tions worsen, the market’s willingness to bear risk disappears because the
market participants stop behaving independently and start acting as one.
They should not do so, as there is little profit and a lot of risk in just fol-
lowing the crowd, but they can be forced to by circumstances.
So, how did Arup miss the potential for the Millennium Bridge wob-
bling? For the same reason, most financial regulations fail to prevent sys-
temic risk. Arup modeled the impact of individual members of the crowd
Figure 20. Millennium Bridge feedback loop. Credit: Lukas Bischoff/

IllustrationX.

112 ENDOGENOUS RISK
in isolation. But to properly understand the risk, it is necessary to study

every aspect jointly: the weather, the bridge’s mechanics, pedestrians’ be-
havior, and the danger arising from individuals acting as one. All of these
factors have to be studied simultaneously or, as we economists say it, in
equilibrium. The Arup engineers were aware of the danger arising from
soldiers marching across the Millennium Bridge. They just never imag-
ined a couple of hundred civilians who didn’t know each other would end
up doing precisely that. Apparently, these feedback effects were known
in some civil engineers’ circles, but the natural instinct of companies to
keep information proprietary meant it was not made public. It took Arup
eighteen months to figure out what had happened, and it eventually came
up with two separate solutions. They could either put a dampener on the
bridge’s movements or create noise—some mechanism that creates ran-
dom movements to cancel out the synchronized swaying. Arup opted for
the former, and it is working well. Still, the random noise solution would
have been more interesting, at least from a financial system point of view.
There is a moral lesson in all of this. When something goes wrong, we
always want to find someone to blame, someone to punish. On the Mil-
lennium Bridge, the natural self-preservation instinct of the people on
the bridge caused the wobble—the precautionary principle. Nobody did
anything wrong. They were just trying not to fall. Paradoxically, doing
the right thing can bring about the worst possible outcome, and we often
see that in crises. Market participants’ self-preservation instinct causes
liquidity to dry up in times of strain, taking us over the precipice into a
full-blown crisis. Who was to blame for the Millennium Bridge’s wobble?
The engineers and the architects. Certainly not the pedestrians. Who is to
blame for a financial crisis? It is not as clear-cut, but the financial authori-
ties who set the rules of the game do share the blame. How to find the
fundamental causes of instability? Studying beauty contests might help.
Beauty contests were popular in British newspapers in the 1930s. They
published pages of photographs of women, encouraging readers to vote
for whomever they thought was the prettiest. But there was a twist: a
lottery ticket was given to those who voted with the majority. This in-
spired John Maynard Keynes when he was writing in his General Theory
about how market prices are formed: “It is not a case of choosing those
[faces] which, to the best of one’s judgment, are really the prettiest, nor

ENDOGENOUS RISK 113
even those which average opinion genuinely thinks the prettiest. We have
reached the third degree where we devote our intelligences to anticipat-
ing what average opinion expects the average opinion to be. And there
are some, I believe, who practice the fourth, fifth and higher degrees.”1
The readers did not choose their favorite based on whom they thought
the prettiest. Rather, they voted strategically to maximize the chance of
voting with the majority and so get a lottery ticket. In the same way,
speculators don’t choose stocks on the basis of the fundamentals of the
company, they try to out-think other speculators.
In 2002 Hyun Song Shin, then a professor at the London School of
Economics (and now the economic advisor and head of research at the
Bank for International Settlements, the central banks’ central bank), and
I proposed a new direction for the literature on risk and uncertainty. We
focused on the origin of risk and how the behavior of people leads to risky
outcomes. In our view risk can be either endogenous or exogenous.
A dictionary definition of the term “endogenous” refers to outcomes
having an internal cause or origin. How an infectious disease spreads
through a population is endogenous to the nature of that same popula-
tion. If we always keep a safe distance between ourselves and our fellow
countrymen, we will not get infected, but if we choose to live cheek by
jowl with other people, our chance of infection is high. Your chance of
getting a cold is endogenous to your behavior and those around you:
one reason why taking the New York subway can be hazardous to one’s
health; and why social distancing was so important in the Covid-19 crisis.
The opposite of endogenous is exogenous, whereby outcomes have an
external cause or origin. When an asteroid hit the Gulf of Mexico sixty-
five million years ago, wiping out the dinosaurs, that was an exogenous
shock. There was certainly nothing the dinosaurs did to cause their de-
mise. The risk of an asteroid hitting Wall Street is exogenous.
Suppose I wake up this morning and see on the BBC website that there
is a 50 percent chance of rain. If I then decide to carry an umbrella when
leaving my house, my doing so has no bearing on the probability of rain.
The risk is exogenous. Suppose, instead, I wake up this morning and see
on the BBC website that there is some negative economic news about the
United Kingdom, and in response I decide to buy a put option on the
pound sterling—I profit if the pound weakens. My actions make it more

114 ENDOGENOUS RISK
likely the pound will fall. Not by a lot, mind you, a tiny, tiny amount. But
tiny is not zero—there is an endogenous effect.
I would not be doing anything wrong. On the contrary, I am behaving
prudently, hedging risk, like the pedestrians on the Millennium Bridge,
who were trying not to fall into the water. Just like a single pedestrian
on the Millennium Bridge did not make it wobble, me alone buying this
put option will not make the pound sterling crash. Another ingredient
is needed, some mechanism coordinating the actions of many people so
when we act in as one, our combined impact is strong. The chance of that
happening is endogenous risk. All that is needed to turn some shock—
the financial market version of the gust of wind hitting the Millennium
Bridge—into a crisis is for a sufficient number of people to think like me
(all wanting to prudently protect themselves) for the currency to crash.
It is the self-preservation instinct of human beings that so often is the
catalyst for crises.
No crisis is purely endogenous or exogenous: they always are a combi-
nation of an initial exogenous shock followed by an endogenous response.
The same initial exogenous shock can one day whimper out into nothing,
and the next blow up into a global crisis. When Covid-19 infected the first
person in Wuhan, all the subsequent events could have been averted if
that person had behaved differently. But nobody knew at the time. The
reason the exogenous shocks blow up into a crisis is that they prey on hid-
den vulnerabilities no one knows about until it is all too late.
The Dark Side of Well-Meaning Rules

The twin roles of market prices drive endogenous risk. The first
is familiar, an idea fundamental to Friedrich Hayek’s theories, that mar-
ket prices embed all relevant information in the economy into one vari-
able: the price. In a simple investment example often used in introductory
undergraduate finance courses, the price of a stock is determined by the
present discounted value of future dividends, where the discount factor
embeds the uncertainty about the future.
Less familiar is the notion that prices are also an imperative to action,
forcing us to behave in a particular way, perhaps even against our self-
interest. This happens because most market participants are subject to

ENDOGENOUS RISK 115
a myriad of regulations, codes of conduct, and restrictions imposed by

their trading partners and other stakeholders. These include accounting
rules, legal obligations, disclosures, bank capital, risk constraints, and
mark-to-market guidelines.
Suppose a bank holds some asset that falls in price by an unusually large
amount (Figure 21). The measured risk of that asset will then increase,
and risk weights will go up, so the bank’s risk-based capital ratio might
fall dangerously close to the minimum set in regulations. To protect it-
self, the bank may have no recourse but to sell that asset. In other words,
a change in prices causes riskometers to show higher risk, which in turn
triggers forced selling, an example of how prices can be an imperative to
action. That is exactly what happened in the Gamestop market turmoil in
January 2021 when speculators fought each other, some shorting, others
going long. Small speculators trading on the Robinhood platform were
eventually unable to trade because Robinhood itself was forced to limit
trading because it clears its trades via the Depository Trust and Clear-
ing Corporation (DTCC), and the heightened price volatility prompted
DTCC to demand $3 billion in collateral from Robinhood.
A bank may not react to market outcomes in the way we expect it to
because the constraints it is under dictate a certain type of behavior. If the
bank is large its trading activities will have a significant price impact. One
could say that the presence of constraints undermines the integrity of the
Figure 21. Prudent banks and endogenous feedback. Credit: Lukas

Bischoff/IllustrationX.

116 ENDOGENOUS RISK
prices, taking them away from their fundamental values. In extreme cases,
prices can become so distorted that they lead to undesirable extreme out-
comes, like bubbles and crashes. That said, these constraints don’t bite
all of the time, not even most of the time. It is only in times of stress
that they significantly affect market prices. That observation goes a long
way toward explaining why we decide to use riskometers and why they
perform so poorly. The riskometers describe the world quite well when
everything is quiet but do not capture behavior changes in times of stress:
why they are fair-weather instruments.
What does this mean for regulations? That we should do away with all
the rules and constraints? Far from it. The rules are by and large very ben-
eficial: helping to keep the financial system orderly, protecting investors,
and preventing abuse. However, they do have a dark side. Well-meaning
rules can act as a catalyst for making market participants act in unison,
just like the Millennium Bridge’s design made all the pedestrians march
like soldiers. A good example is buying stocks on margin.
The first time stock markets became accessible to the general public was
in 1920s America, as buying stocks was the preserve of the wealthy and
connected before that. After World War I anyone in America could buy
stocks. Not only that, but they also didn’t have to invest much money.
One could bring $100, borrow $900, and buy $1,000 worth of stock—be
leveraged nine times. This buying on the margin became very popular.
Money poured into the stock market, and since the United States was the
only country where stock markets were fully open to the general public,
money flowed to Wall Street from all over the world. The result was the
mother of all stock market bubbles, one that came to a sticky end in Sep-
tember 1929, triggering the Great Depression.
The dark side of the well-meaning margin rules shows its face when
the bubble is bursting. Suppose I buy ten shares of IBM at $100 each,
$1,000 in all. I put up 10 percent, or $100, of my own money and borrow
$900. The entity financing this transaction will want some protection
against the stock falling in price, insurance called a margin. Suppose the
margin is equal to the initial $100. If IBM’s price falls by 5 percent to
$95, the investment is now worth only $950. I still owe $900, while my
net value ($1,000–$950) has fallen to $50. However, the entity financing
the transaction insists on 10 percent of the original amount, $100, so I

ENDOGENOUS RISK 117
have to make up the $50 shortfall. This is called a margin call, an immedi-
ate demand for $50. I have two choices: sell enough of the stock to pay
back the borrowed funds or find $50 elsewhere. Many investors will have
no choice—they can’t find the $50 in time and have to sell. How likely
is a day like that? I looked at the history of daily IBM stock prices and
got 22,696 observations. Out of those, there are 117 days when the price
of IBM fell by 5 percent or more, so the likelihood of a 5 percent price
drop is 0.52 percent, or about once every 9½ months. If the price falls by
10 percent or more, we are wiped out; such days have happened 13 times,
or almost one day out of every seven years.
Investors may have no choice but to sell the stock to meet the margin.
Yet if a large number of investors are in the same situation, a significant
volume of sells hits the markets simultaneously. Prices fall even more,
creating yet more margin calls, more investors have no choice but to sell,
and prices fall more. An endogenous risk vicious feedback loop emerges
from the dark (Figure 22).
Here, the margins work like the wobbly Millennium Bridge. Pedes-
trians were trying not to fall, leaning against the swaying of the bridge,
perversely causing it to sway even more. In the case of margins, it is the
automatic protection for the lenders that is at the root of the damage.
One of the enduring images of the crash of 1929, even if now debunked,
is desperate investors jumping out of the sky-high windows of their Wall
Street offices. The putative reason was margins. Investors got margin
calls, their capital was wiped out, and they were set to be declared bank-
rupt by the end of business that day.
Figure 22. Margin call feedback loop. Credit: Lukas Bischoff/IllustrationX.

118 ENDOGENOUS RISK
Some rules, like margins, constrain market participants only in particu-

lar situations. Others work all the time, having a rather surprising im-
pact on prices by making investors continually react in unexpected ways
to market developments. The way prices and quantities are supposed to
relate to each other is clearly illustrated in an example economics profes-
sors are fond of. When the price of oranges increases, we buy fewer (and
farmers want to sell more), and when the price falls, we buy more (and
the farmers want to sell fewer)—the law of supply and demand. There
is no exception. We will buy some goods almost without considering
cost—goods economists call highly inelastic, like tobacco and heroin. We
buy them almost regardless of the price, but we don’t buy more when the
price goes up.
Are there any goods that we buy more of when the price goes up? Yes,
a strange asset called a Giffen good, named after the nineteenth-century
Scottish economist Robert Giffen. It is a hypothetical good, useful to
illustrate weird behavior in markets, something that exists in theory but
not in practice. Every other good exhibits the intuitive property that we
buy less when the price goes up. If you are thinking of counterexamples
like high fashion, yes, demand can go up when prices increase. But that
is because the good’s exclusivity is also valued—if I want to be the only
person on my street with a posh handbag, I am willing to pay more for it.
However, the constraints I have been discussing can make investors buy
more when prices increase and sell when prices fall: the law of supply and
demand apparently gets reversed. Sometimes these constraints can even
be self-imposed, for instance, when investors use automatic trading rules.
My favorite example of how constraints can cause unfortunate outcomes
is the biggest stock market crash in history, commonly known as Black
Monday. On 19 October 1987 stock markets around the world fell by
23 percent. The underlying cause is the best example of endogenous risk
I have ever seen.
It all started in 1976, when two professors at the University of Califor-
nia, Berkeley, Hayne Leland and Mark Rubinstein, came up with a clever
idea to protect investors from stock market crashes: portfolio insurance.
If an investor is interested in buying insurance against a large price drop,
all she has to do is buy a put option, which gives her the right (but not
the obligation) to sell a stock at a preagreed (strike) price at some point in

ENDOGENOUS RISK 119
the future. Suppose the price of Apple is $200 today. If we buy a five-year
put option on Apple with a strike price of $180, we get the right to sell an
Apple stock for $180 whenever we want for the next five years.
What if no option is available? Or the option is too expensive? Leland’s
and Rubinstein’s portfolio insurance solved that with the magic of finan-
cial engineering: what is formally known as dynamic replication, whereby
one can create a financial instrument that looks and behaves like an op-
tion. So, if it looks like a duck, swims like a duck, and quacks like a duck,
it must be a duck, right? Not quite. There is a key difference. In portfolio
insurance, one has to buy or sell the asset being insured every day to
replicate the option correctly. If the price increases, we have to buy, and
if the price falls, we have to sell. In other words, a buy dear–sell cheap
strategy—we apparently violate the laws of supply and demand.
All that was needed for portfolio insurance to cause a crisis was for a
sufficient number of people to use it as a trading strategy. And that is what
happened in September 1987. The US government postmortem of the
1987 crash estimated that around $100 billion was placed in formal port-
folio insurance programs, representing around 3 percent of the precrash
market capitalization. From Wednesday, 14 October to Friday, 16 October
1987, the market declined by around 10 percent. The sales dictated by
those who used portfolio insurance strategies amounted to $12 billion,
but actual sales were only $4 billion. This meant a substantial amount of
pent-up selling pressure accumulated over the weekend, causing the S&P
500 index to fall by 23 percent on Monday, 19 October.
The stock market crash of 1987 is a classic example of the destabilizing
feedback effect on market dynamics of concerted selling pressure from
mechanical trading rules, like the sell-on-loss considered here. Again, just
as in the Millennium Bridge example, the underlying destabilizing be-
havior is entirely invisible so long as trading activity remains below some
critical but unknown threshold. It is only when this threshold is exceeded
that the endogenous risk becomes apparent, causing a market crash.
A common view of crises maintains that they arrive from the outside,
like the above asteroid that is about to hit Wall Street. That is not true.
The main driver of crises is endogenous risk, underpinned by the system’s
hidden mechanisms, just like sell-on-loss trading rules. Unfortunately,
there is too much of a tendency to focus on the triggers of crises, not the

120 ENDOGENOUS RISK
mechanism. The problem is that there is a tiny number of mechanisms

but an almost infinite number of triggers. They are visible but not all that
important. Some are even human.
Andreas Georgiou used to be Greece’s chief statistician but then com-
mitted the crime of accurately reporting the size of the Greek budget defi-
cit in 2009 without allowing the board of the statistics agency to first vote
on the size of it. One might think a politically appointed board should
not decide on statistical measurements. But it got worse for Andreas. He
was pursued in the courts for this crime and convicted in June 2018 by the
Greek Supreme Civil and Criminal Court and sentenced to two years on
probation. Why did the Greek authorities get upset? Because he reported
the true number of 15.4 percent and not the finance ministry’s estimate
of 13.6 percent, thereby triggering the Greek crisis. Andreas may have
been the trigger, but he was certainly not to blame, and the crisis would
have happened anyway, even if he had fudged the numbers. The crisis
mechanism was already in place. The Greek crisis was not the first time
the trigger was confused with the underlying cause: see the 1914 crisis
following the assassination of Archduke Ferdinand and, in 1763, the end
of the Seven Years’ War. In both cases a crisis was inevitable: it was not a
question of if, but when.
We should not focus on the trigger but on the underlying vulnerability,
the mechanisms that cause crises. When getting too concerned about the
triggers, we risk what my skiing instructor once called paralysis by analy-
sis. If one thinks about skiing down a one-kilometer, 45° slope, one might
conclude it is best to take a lift down. Paralysis by analysis. It is the same
with triggers because there are so many of them; if you start searching
for and protecting yourself against all potential trigger events, that will
end up being all you do. Focusing on the triggers can lead policy mak-
ers seriously astray, But that is often what they end up doing. The press
reports on the triggers, the public demands that someone be blamed, and
the likes of Andreas Georgiou get prosecuted. Meanwhile, those respon-
sible, like the Greek politicians, get off scot-free. Our attention should
be focused on the mechanisms, and the financial authorities and hedge
fund managers should know that. Sadly, they are often no better than the
Greek prosecutors.

ENDOGENOUS RISK 121
Bubbles 101: Perceived and Actual Risk

Crises take (almost) everybody by surprise. It could not be any other
way because if many people anticipate a crisis, they will get out of the
market—either preventing the crisis or causing it to happen immediately.
However, even if we are surprised, crises don’t happen randomly. They
are the culmination of the accumulation of risky behavior over years and
decades. Few people have said it better than the former head of the Bank
for International Settlements, Andrew Crockett: “The received wisdom
is that risk increases in recessions and falls in booms. In contrast, it may
be more helpful to think of risk as increasing during upswings, as financial
imbalances build up, and materializing in recessions.”2
In a paper I wrote with Hyun Song Shin and my LSE colleague Jean-
Pierre Zigrand, we applied endogenous risk to the measurability of risk,
classifying risk into two categories, perceived risk, what the riskometer
measures, and actual risk, the underlying hidden risk that is undetected
until it suddenly materializes in a crisis. Riskometers, by definition, can
measure only perceived risk because they are based on using fluctuations
in market prices to infer the riskiness of the underlying assets. If the fluc-
tuations are low, the risk measurements will follow. If the market is confi-
dent, prices are rising, fluctuations are low, and perceived risk is low. The
use of riskometers based on perceived risk is one reason bubbles happen.
Market prices get increasingly disconnected from the underlying reality,
and the risk of a significant correction (actual risk) steadily increases. Sup-
pose some investors buy an asset. Consequently, the price of it increases
slightly, which in turn encourages more people to buy, making prices rise
a little more. The result is a series of small price increases, which, when
translated to price fluctuations, imply volatility is falling.
If we then use a common device to measure risk-adjusted returns, the
Sharpe Ratio, proposed by the Nobel laureate William F. Sharpe, the asset
looks better and better.3 Prices increase steadily, perceived risk falls—a
fantastic investment, steady profits at low risk. “Money for nothing,” as
the 1980s Dire Straits song would have it. A bubble born of the positive
feedback between rising prices and falling perceived risk. Prices continue
to rise faster and faster until some trigger event causes them to crash.
Because everybody wants to sell before the prices drop, they all rush to

122 ENDOGENOUS RISK
Figure 23. Money for nothing. Actual and perceived risk. Credit: Lukas
the exits, and the prices crash immediately. When things are good, we are
optimistic and buy, which endogenously increases prices—the bubble
feeds on itself. This eventually goes into reverse, and negative news preys
on falling prices, with the markets spiraling downward much faster than
they went up. Yet again, prices go up the escalator and down the lift (or
elevator if in America). Only after the prices have crashed does perceived
risk increase. By then it is too late.
So, what happens to actual risk through all of that? It increases along
with the bubble. After all, it captures the fundamental risk, the risk of a
market crash, hence falling when the bubble deflates (Figure 23). I often
wish someone could have convinced the financial regulators that the risk
after 2008 was much lower than the risk before, and they should have
been encouraging risk-taking and not derisking the system.
Long-Short-Term Management and the Perfect Storm

In the summer of 1998 the financial markets were turbulent, not
least due to the hedge fund Long-Term Capital Management (LTCM).4
LTCM was the most celebrated hedge fund of its time. It was founded in
1994 with illustrious partners like Myron Scholes and Robert C. Merton
(of Black-Scholes fame), who shared the 1997 Nobel Prize. LTCM started
with over $1 billion in capital and was very successful despite its exception-
ally high fees. A typical hedge fund charges “2 and 20,” 2 percent of an

ENDOGENOUS RISK 123
investor’s assets as fees plus 20 percent of profits earned. LTCM charged

2 and 25, and its investors were required to commit their funds for at least
three years. In the first two years investors enjoyed 43 percent and 41 per-
cent returns, respectively, after fees, and $10 million invested in 1994 was
worth $40 million four years later. In September 1997 LTCM’s net capital
was $6.7 billion, and the value of its investments was $126.4 billion, the
difference being borrowed money. It returned $2.7 billion to investors in
December 1997 to focus on investing its partners’ own money.
LTCM’s success created the impression that it would make extraor-
dinary profits from applying technical expertise unavailable to anyone
else—just magic. Scholes summed up the strategy thus: “LTCM would
make money by being a vacuum sucking up nickels that no one else could
see.” LTCM drove tough bargains on financing and got very low rates
and special deals because its prime brokers did not want to be left out
of the LTCM business. They were prepared to cut corners both on the
rates charged and on the amount of collateral demanded. The mainstay
of LTCM’s trading strategies was convergence or relative value trades,
which seek to exploit differences in the price of similar assets. The un-
derlying principle is that two fundamentally identical assets should have
identical prices, as otherwise there is a profit-making opportunity: arbi-
trage. LTCM’s risk management system indicated in 1998 it would take
a ten-standard-deviation event (otherwise known as a 10-sigma event)
for it to lose all of its capital in a single year. This means the risk system
expected LTCM to go bust once every 3.3 × 1020 years.5 By comparison,
the age of the universe is only 1.3 × 1010 years. No wonder the partners re-
turned funds to the investors to focus on their own money. They just did
not expect to fail—their riskometer told them so. Meanwhile, LTCM’s
trading environment was increasingly choppy, as its very success spawned
copycats. More and more players with similar trading strategies piled into
the market, and profit-making opportunities started disappearing. In re-
sponse, LTCM ventured into new areas, most notably the volatility on
the S&P 500, measured by the Chicago Board Options Exchange’s Vola-
tility Index (VIX). The long-run average VIX until the middle of 1998
was 16.8, and the VIX exceeded that level throughout 1998. Volatility is
mean reverting, and with volatility way above average, money was to be
made by betting the VIX would fall.

124 ENDOGENOUS RISK
On the day I am writing this, the VIX is 35. Suppose you see the same
value when you read this and get the brilliant idea to log onto your online
broker and buy a short VIX fund—betting it will fall to the long-run av-
erage. Think twice. While it is almost certain the VIX will fall, the cost of
maintaining the short VIX fund might exceed 35 percent a year, so unless
the VIX falls relatively soon you are in for losses.
That can happen quite quickly. Just ask the unfortunate investors in
Nomura’s Next Notes S&P500 VIX Short-Term Futures Inverse Daily Ex-
cess Return Index ETN, which lost 96 percent the very same day Nomura
launched the fund on 5 February 2018. What happened is that the VIX
had fallen from 28 at the start of 2016 to 13.54 by the end of January
2018, and, if one follows the basic principles of momentum investing, the
VIX was destined to fall more. Nomura launched its short VIX fund on
Monday morning Japanese time with the previous week’s VIX value and
quickly sold ¥32 billion worth to investors. Unfortunately, heightened
uncertainty hit the US market the same day, so when the US stock market
opened the VIX quickly rose to 37.32—boom, Nomura’s luckless inves-
tors lost ¥32 billion in the blink of an eye.
Back to LTCM. It was eyeing the steady increase in VIX throughout
1998 and decided to get in on the action. However, because the profit
margin on betting that the VIX would fall was too small for LTCM it
opted to borrow to increase leverage. LTCM became so prominent in
the volatility market that it earned the nickname “The central bank of
volatility.”
Except, as LTCM piled into the market, the VIX continued to rise.
LTCM suffered losses of −6.42 percent in May 1998 and −10.14 percent
in June. Then Russia defaulted in August, triggering a market panic, and
the VIX continued to rise, reaching 45 in August. By early September
LTCM’s equity had tumbled to $600 million. As the debts were constant
and the positions worth less, leverage rose sharply, reaching 125. LTCM
was in serious difficulty, getting margin calls it did not have the cash to
cover and hence was forced to liquidate its positions. That made the VIX
rise even more. A vicious feedback loop was set in motion: higher VIX
causing margin calls, leading to liquidation, making the VIX rise further,
repeat.

ENDOGENOUS RISK 125
Realizing the disastrous consequences of letting LTCM fail, the New

York branch of the Federal Reserve organized a $3.6 billion bailout from
its major creditors to avoid a wider market collapse. The participating
banks got a 90 percent share in LTCM in return. After the bailout, the
panic abated, and the positions formerly held by LTCM were eventually
liquidated at a small profit to the bailers. In light of the 2008 crisis a de-
cade later, it is quite amusing that the LTCM bailout was controversial;
the government did not put up any of its funds—only LTCM’s counter-
parties did—but even that was seen as beyond the pale.
One LTCM partner said, “What happened to LTCM was a perfect
storm—a 100-year flood.”6 But does that stand up to scrutiny? Endog-
enous risk analysis suggests otherwise. The unprecedented price move-
ments in the summer of 1998 were not the results of terrible luck, just
as the Millennium Bridge did not wobble because of bad luck. Instead,
it was only a matter of time before a small exogenous shock would hit
LTCM, causing its failure. Once the bubble started bursting, the internal
dynamics of the feedback loop took hold with a vengeance. Under the
right conditions, the crisis was a near certainty because of the endog-
enous feedback loops between margin calls and distress.
LTCM bet on the mean reversion of an asset that fluctuates around its
long-run mean. Profits were made, but only by those who bailed LTCM
out. Why did LTCM not profit? The explanation is provided by an obser-
vation often misattributed to Keynes, who supposedly said when asked
by an investor in 1931 about market conditions: “The market can stay ir-
rational longer than you can stay solvent.” The very high VIX levels were
caused by the uncertainty created by the very existence of LTCM. A neces-
sary condition for the VIX to return to its long-run mean was for LTCM
to fail. A perfect storm did not hit the fund. Its failure was a near certainty.
Endogenous Risk and the Financial System

Endogenous risk doesn’t usually leave much of a footprint on
the financial markets. While it lurks in the dark corners of the system,
most of the time all we see is exogenous risk. When markets appear to
be efficient, with prices that follow random walks, as in Burton Malkiel’s

126 ENDOGENOUS RISK
book A Random Walk Down Wall Street. But the good times don’t last.
Endogenous risk will show its face when some trigger pushes us to act in
concert with each other, like the pedestrians on the Millennium Bridge.
Price movements will be amplified—bubbles and crashes.
The spirals of coordinated selling that are unleashed by endogenous
risk are normally held back by the inherent stabilizing forces in the mar-
kets: the arbitrageurs, the hedge funds, the sovereign wealth funds, the
Warren Buffetts, the Soroses. They step up to the plate in crises to buy
cheap assets, putting a floor under prices. An excellent expression of this
is from Baron Rothschild in the eighteenth century, who is reported to
have said, “Buy when there’s blood in the streets, even if the blood is your
own.” The self-interest of the speculators benefits society, as in Adam
Smith’s classic statement, “It is not from the benevolence of the butcher,
the brewer, or the baker that we expect our dinner, but from their regard
to their own interest.”
What lets endogenous risk off its leash is decisions and policies that
serve to harmonize market participants’ behavior. Rules that prevent
them from erecting a floor under the markets by buying all the assets that
are so undervalued because of the crisis. The riskometer puts the contrast
between endogenous and exogenous risk in the sharpest relief. Almost
all methods of measuring risk are based on the assumption that risk is
exogenous because that is the easiest way to deal with risk. All one has to
do is to collect some daily historical financial data—market prices, credit
default swap spreads, interest rates, trading volumes—and feed them into
a riskometer. Nothing wrong with that if all we care about is exogenous
risk: short-term fluctuations.
If, however, we care about tail risk, banks failing, and crises, we have no
choice but to find some way to measure endogenous risk. That is not easy.
Large losses and crises happen because of risk everybody missed, so iden-
tifying endogenous risk before it is too late is like searching for a needle
in a haystack, except that we don’t even know what the needle looks like.
We know only that the needle is there when we stick our hand into the
haystack and it gets pricked. After a crisis, everybody knows what went
wrong, and we prevent a repeat: closing the barn door after the horse has
bolted. Meanwhile, the forces of the next crisis start gathering strength
somewhere where nobody is looking.

ENDOGENOUS RISK 127
The 2008 crisis is an excellent example of how we missed all the warn-
ing signs. While I suspect many people thought financial products based
on the American housing market were dodgy and opted not to buy them,
a few profited from it. The book (and movie) The Big Short by Michael
Lewis tells the story of a few plucky players who did exactly that. Mean-
while, no government authority had an inkling of the danger lurking
right under their noses. The lesson from 2008 is that the buildup of en-
dogenous risk happens almost wholly out of sight. Eventually, the vulner-
abilities hit the hidden trigger—risk got a little bit too high on a certain
day when the markets had little tolerance for it—and it all blew up.
The challenge for investors and the financial authorities is that while
endogenous risk is ever-present, it cannot easily be measured. As endog-
enous and exogenous risk move in opposite directions, the riskometers
get it wrong in all states of nature, reporting too little risk before a crisis
and too high after it, just like the European Central Bank’s systemic risk
dashboard.
Outcomes in the financial system are determined by the interaction of

the institutions and individuals that make up the financial system. That’s
why my colleagues and I proposed the term “endogenous risk” to cap-
ture the resulting risk. However, even though endogenous risk gives a
useful description of the world, it is difficult to work with in practice. It
cannot readily be quantified or incorporated into riskometers or put onto
risk dashboards. The general practice is to assume risk is exogenous, that
it arrives from outside the financial system, impacting the system but not
being affected by it. Like an asteroid hitting Wall Street (see Figure 18).
As a practical matter, it may be perfectly okay to assume most risk is
exogenous. It just depends on our objectives. Anyone concerned with
day-to-day risks or risks of frequent small losses would not be very far off
by focusing on exogenous risk. It is only those who care about tail risk—
the infrequent large outcomes that cause significant losses, the failure of
banks, and financial crises—who should focus on endogenous risk. What
is worrying is how easily the riskometers based on exogenous risk can be
manipulated.

8
If You Can’t Take the Risk, Change Riskometers
Beauty is skin deep.
We often run into street artists who, for a small fee, will draw a
caricature of us. One that exaggerates facial features for comical effect,
perhaps emphasizing the nose and diminishing the chin. One example is
the caricature of me made by Ricardo Galvão, who did many of the draw-
ings in this book (Figure 24). Risk measurements are just like those street
caricatures. The riskometer is the creation of its designer, who makes
all sorts of decisions, balancing ease of implementation with accuracy,
128

CHANGE RISKOMETERS 129
emphasizing what he cares about, and dismissing the rest. Riskometer de-
sign is highly subjective, and two designers faced with the same problem
will create riskometers that measure risk quite differently.
The reason for all of the subjectivity is that risk is not directly mea-
sured, like prices or temperature. It can be inferred only by how prices
move, and to do that we need a model. And since by its very nature every
model is subjective, the risk measurements are a product of the underly-
ing assumptions. And that means the accuracy of risk measurements is
much lower than is generally presumed. Certainly much less reliable than
the thermostat that keeps the risk manager’s office at a steady 22°C. More
important, all that subjectivity makes riskometers easy to abuse. Perhaps
with blatant dishonesty, as when someone deliberately tweaks the models,
so they tell us risk is $1 million when it is really $5 million. Plenty of that
going on, but I think intellectual laziness is more common. Using risk-
ometers to pretend risk is measured and managed so we can demonstrate
we are diligent and compliant. We follow all the regulations while doing
something else. And because it is so hard to verify that the riskometers are
accurate there is little the regulators, the compliance officers, and all the
other guardians of good practices can do.
A friend of mine, Rupert Goodwin, used to make a living selling risk
systems. One day he went to a bank that had just been audited by the lo-
cal financial authority. The regulator came in and asked the risk manager
if he used a riskometer. When the risk manager said yes, the regulator
ticked off a box and left. The most obvious way a riskometer goes wrong
is when it is simply a tick-the-box exercise:
1. Is a risk model used? ✓

2. Is the Value-at-Risk number below $1 million? ✓
I suspect laziness is the usual explanation for such tick-the-box risk

management, but deliberate misunderstanding of what riskometers can
and cannot do is much more interesting. The powers that be have opted
for deliberate ambiguity. Senior bankers and regulators generally have
a limited understanding of statistics, and even when they do it’s likely
out of date and won’t cover the detailed mechanics of the modern risk-
ometer. They have to make decisions and are much more comfortable
not knowing—the deliberate ambiguity. Like sausages. If you are going

130 CHANGE RISKOMETERS
to eat them, you are better off not knowing how they are made. Some
two decades ago I made exactly that mistake when I read a book called
Fast Food Nation by Eric Schlosser, which went into the gory details of
how fast food is made. Not many books have changed the way I live my
life, but this one did, and I have studiously avoided fast food ever since.
I would have been happier if I hadn’t read Fast Food Nation—ignorance
can be bliss. I suspect most senior bankers and regulators take the same
view, happier not knowing how the riskometers that govern their world
work. The easiest way to do that is to choose a measurement of risk that
somehow maps all the portfolios’ complexities and even entire financial
institutions’ risks into one number—Value-at-Risk. So long as the Value-
at-Risk number is within an acceptable range, all is fine.
The creator of the all-important Value-at-Risk riskometer, the JP Mor-
gan bank, came to get badly hit by its creation three decades after bring-
ing it into the world. That is when a member of the bank’s London staff,
nicknamed the London Whale, caused a $5.8 billion loss because he was
apparently trying to profit from the bank’s Value-at-Risk. It started with
Bruno Iksil, a senior trader in a division called the chief investment of-
fice (CIO), whose function was to hedge the bank’s credit risk. That
year the Value-at-Risk for the CIO division exceeded $95 million, prob-
lematic since the total target Value-at-Risk for the entire bank was only
$125 million.1
There are two ways to deal with this problem. Either reduce the risk of
the portfolio or change the riskometer. JP Morgan seems to have chosen
the latter. How do we know this? The person in charge of the Value-at-
Risk model at the CIO, Patrick Hagan, sent an email to his colleagues
with the subject “Optimizing regulatory capital,” using his private Yahoo
account. The email was subsequently made public by the Senate com-
mittee investigating the loss.2 And that exposes the crucial difference
between thermometers and riskometers. A favorite saying of President
Truman was, “If you can’t take the heat, get out of the kitchen.” When it
comes to risk: If you can’t take the risk, change riskometers.
As it turned out, JP Morgan’s new riskometer missed out on some of
the critical risks facing the CIO, risks that soon would lead to the $5.8 bil-
lion loss. In the words of JP Morgan’s quarterly securities filing: “This
portfolio has proven to be riskier, more volatile, and less effective as an

economic hedge than the firm previously believed.” Of course, if one de-
liberately picks a riskometer that signals the lowest risk, the portfolio will
be riskier than we think it is.
Manipulating Riskometers 101

A few years ago I addressed a small group of very senior financial
regulators, telling them that when I lecture to my LSE students on risk
forecasting, I like to give them the following anecdote: “Suppose my job
is to forecast the risk of some portfolio, measured by Value-at-Risk. One
day my boss calls me into her office: ‘Jón, the Value-at-Risk is too high,
and you have to make it lower. However, our traders are doing a fantas-
tic job, and I want to keep the nice returns. Now go and play with your
computers and make it happen. And by the way, don’t get caught make
sure you comply with all regulations.’”
And then I show the students how to make it happen. The senior deci-
sion makers were aghast: both that I would have the audacity to show the
students how to do such a thing and that it was easy to do. The answer
to why is that these are simple tricks a lot of people know. If the students
become risk managers, they should learn those schemes so as to know
what to keep an eye out for.
The easiest way to manipulate the risk measurements is to pick a risk-
ometer that delivers a lower number. On my website, extremerisk.org,
I forecast risk for several assets every day with the most commonly used
techniques. Let’s focus on the S&P 500 index, where we find today that
our Value-at-Risk is $22 million, supposing a $1 billion portfolio, getting
the Value-at-Risk from a common technique called historical simulation.
Historical simulation is just one of many techniques one can use to cal-
culate Value-at-Risk, and one of the most commonly used. Suppose I use
the six most common calculation methods (Figure 25).
By switching from historical simulation to another method, expo-
nentially weighted moving average (EWMA) (incidentally, created by
JP Morgan in the same report in which it proposed Value-at-Risk), the
Value-at-Risk falls from $21.8 million to $8.5 million—accomplished with-
out doing anything else. There are two problems with this trick. To begin
with, there is no guarantee EWMA will continue to result in the lowest

Figure 25. Value-at-Risk on a $1 billion portfolio of S&P 500 on 2 January

2018. Credit: Lukas Bischoff/IllustrationX.
Value-at-Risk. Indeed, on some days it will be the highest. Even worse, it

does not meet our criteria for not being detected. The bank’s compliance
people and the regulators will take notice and likely take a dim view of
such switching of riskometers, especially if done often.
A better way to manipulate riskometers is to cherry-pick the assets one
puts into the portfolio. Choose stocks that robustly provide the juicy
returns we want but do not contribute very much to the measured risk.
Technically, we add another constraint into the portfolio optimization
problem: maximize returns subject to two constraints, low reported risk
and high true risk. This is possible because the riskometer is a caricature,
and what we are doing is searching for assets that exploit the weaknesses
in our riskometer: the tiny chin and big nose. It is easy enough to do by
anyone with moderately good quantitative skills and is practically un-
detectable by both the compliance department and the regulators. When
I tried to do this for a sample portfolio, it took me only a few minutes
to reduce the Value-at-Risk by over 70 percent—without changing the
expected returns much.
The final way to manipulate is to use options. I show a particularly
egregious example in my book Financial Risk Forecasting. This approach
is based on judiciously buying and selling options around the actual and
desired Value-at-Risk numbers to minimize measured risk and not affect

much else.3 Doing exactly that is a blatant abuse of the bank’s risk man-
agement system and presumably would be picked up by the risk con-
trollers. However, if one goes about it more subtly, it will likely never
be detected.
There are many similar ways one can manipulate riskometers. Some
are easy to detect, others are known only to the person taking the risk.
If you are skeptical and think I am just an academic making up extreme
examples that would never see the light of day in the real world, think
again. The reason the UBS bank failed in 2008 was precisely because of
what I’m describing here.
Many banks failed in the 2008 crisis, but the most interesting for the
topic at hand is the Swiss bank UBS. We are quite fortunate that the
Swiss Federal Banking Commission had UBS produce a postmortem
titled Shareholder Report on UBS’s Write-Downs. It is a fascinating docu-
ment, clearly and clinically highlighting all that went wrong: a superb
example of a desire to manipulate the risk management process and, ul-
timately, self-delusion. The main culprit was $19 billion in losses on col-
lateralized debt obligations (CDOs) composed of US subprime mort-
gages in 2007.4
Given that it used Value-at-Risk to measure the risk, UBS apparently
did not realize the CDOs were risky. Value-at-Risk is the worst risk-mea-
surement methodology for CDOs because by construction it cannot pick
up the risk in an asset with a steady income and the occasional substantial
losses.5 The bank could have done much better. It had comprehensive de-
tails on each mortgage it bought, and nothing prevented it from analyz-
ing these data, which would have told it that something was fishy. Indeed,
some of UBS’s competitors did avoid large losses on subprime mortgages
by carefully analyzing the data they had on them. The UBS risk managers
opted for a riskometer that was tailor-designed not to capture subprime
mortgage risk. This fed into the calculations of the bank’s overall riskiness
and was dutifully reported to senior management, the Swiss authorities,
and UBS’s auditor, Ernst & Young. None was concerned. UBS lost sight
of the fact that it was deceiving itself when it thought it was fooling the
regulators.
Another exploit of riskometers is to take advantage of the risk manage-
ment techniques that are meant to protect us. Use the regulations to hide

risk. Under the Basel regulations, banks are supposed to measure their
trading risk and report the risk numbers to the authorities. The technique
for doing so is one beloved by consultants and designers of dashboards:
traffic lights. The traffic-light rule says that a bank is allowed to have three
days a year where its losses exceed the Value-at-Risk measurement. The
banks are given some leeway, and so long as they exceed the Value-at-
Risk only four times a year they stay in the green zone. If they exceed it
between five and nine times they are in the amber (yellow) zone and if
they exceed more than nine times, the red zone.
A bank in the green zone must hold capital at least equal to three times
the Value-at-Risk; if in the amber zone, three to four times Value-at-Risk;
if it hits the red zone, four times the Value-at-Risk plus the likelihood of
being subjected to intrusive scrutiny. Obviously, the higher the Value-at-
Risk, the more the bank has to invest in low-yielding capital and less in
high-return, high-risk assets. It’s no surprise that the bank wants to mini-
mize its measured risk, hence the amount of costly capital it has to hold.
Recall the example of the London Whale.
Imagine the following: Suppose a bank knows its true Value-at-Risk,
but the regulator knows only what the bank tells it. Further, suppose the
bank fully intends to comply with the letter of the regulations, the traf-
fic lights, but still wants to take on more risk. What will happen? Chen
Zhou, a coauthor, and I looked into this question in a paper titled “Why
Risk Is So Hard to Measure.” As it turns out, it is quite easy to under-
report risk while remaining fully compliant with the traffic lights rule.
Particularly interesting to us was how banks react to being controlled by
the traffic rule. They will change the composition of assets, shifting from
assets that fluctuate a lot in price (but not excessively) to assets whose
prices are usually relatively stable, but on occasion are subject to sub-
stantial losses. In the language of finance, the banks come to prefer assets
with lower volatility and higher tail risk. Such assets have the advantage of
making the bank look good, but at the expense of making it more likely
the bank will fail. Like UBS’s CDOs.
Finding such assets is easy. As we had CRSP, the entire database of
daily stock returns in the United States, on hand, we simply searched for
stocks that simultaneously matched some profitability criteria and stan-
dard risk management standards while staying in the green zone. The
results confirmed the theoretical prediction. Because of how the regula-

tions worked, the best way to bypass them was to increase the very risk we
don’t want the banks to take. In all fairness, the example above applies to
Basel II, which now has been surpassed by Basel III, which also replaces
Value-at-Risk with Expected Shortfall, which is not subject to this par-
ticular exploit.
Capital Structure Arbitrage

One of the best ways to see how riskometers get abused in prac-
tice is in what is called capital structure arbitrage, the manipulation of
capital rules to make the capital ratio look as high as possible while keep-
ing it very low in practice: all without violating the letter of the law or
drawing the attention of the authorities. What makes capital structure
arbitrage possible is the very complexity of bank capital. As there is con-
siderable scope for mistakes and even outright manipulation of the capital
calculation, it is no surprise banks are quite adept at such machinations.
Any bank wanting to be seen as having high capital while actually hold-
ing little capital can use clever financial engineering tricks to make bank
capital appear to be almost anything the bank wants. There are consulting
firms entirely dedicated to helping them do precisely that.
Such capital structure arbitrage helped many banks get into difficulty
in the 2008 crisis. I looked at this in detail in my last book, Global Finan-
cial Systems: Stability and Risk, when I saw that banks that appeared to
be highly capitalized were going bust. That should not have happened
because if they were making risky loans, the risk weights and hence capital
should have reflected that risk. As it turned out, the capital was illusionary
because of capital structure optimization.
The capital ratio comprises two parts, capital in the numerator and risk-
weighted assets in the denominator. Capital structure arbitrage involves
maximizing the numerator and minimizing the denominator to make the
ratio as high as possible. The denominator captures the interaction of the
value of assets, often loans, and their risk weights. Since most loans are
illiquid and have no market price, the only way to find their value is to use
a valuation model. It is even harder to calculate the risk weights because
we have to use a riskometer. Since both parts of the denominator calcula-
tion are based on models, the banks have ample scope to pick models that
make the denominator as small as possible. Similarly, the numerator is

Figure 26. Capital ratios for European banks before the 2008 crisis.
Credit: Lukas Bischoff/IllustrationX.
easy to manipulate because capital is composed of both equity and capital

instruments that turned out to be not very equity-like, not the least hy-
brid instruments.
I can illustrate this by looking at two ways of calculating a capital ratio:
tier 1 capital divided by risk-weighted assets and equity divided by total
assets (leverage ratio). The former allows plenty of scope for manipula-
tion, but the latter does not. The leverage ratio is always smaller because
of the risk weights (Figure 26 shows the two ratios over time). The risk-
weighted capital ratio is trending upward, sending the signal that the
banks were becoming more capitalized and safer as time passed. The le-
verage ratio trends downward, indicating that the banks are becoming
less capitalized and riskier.
The Inevitable Crisis in Quantland

Wednesday is the type of day people will remember in quant-land for a
very long time. Events that models only predicted would happen once in
10,000 years happened every day for three days.
—Wall Street Journal, June 2007
The June 2007 crisis in Quantland heralded the opening shots

of the global crisis that reached its peak fourteen months later. The term

“Quantland” refers to the activities of a particular type of investment

fund that replaces decisions by human beings with computer algorithms.
These funds are often called quant funds, short for funds that use quan-
titative strategies.
The classical way to manage a fund is to do what one of the world’s
most successful investors, Warren Buffett, does with his Berkshire Hath-
away fund: use fundamental analysis to analyze companies’ relative
strength. Quant funds dispense with that, instead collecting vast amounts
of statistical information: prices of every conceivable asset, the state of the
economy, and whatever else they can get their hands on. This data is fed
into algorithms that identify how all this data affected prices in the past,
and use the result to forecast prices. The algorithm is then let loose on
the markets, automatically buying and selling assets of all types—stocks,
bonds, derivatives, and so on—much faster than any human being can.
Humans take only an indirect role by monitoring the algorithms’ deci-
sions to make sure they don’t go too crazy.
The people working for quant funds differ from those working for
more traditional fund managers. Not accountants or economists or fi-
nance experts, they are more likely to have studied technical subjects like
computer science, physics, and statistics, often holding doctoral degrees
from the world’s best universities. All the quant funds think they have
the best information and algorithms and computers and human capital,
stealing a march on their quant competitors, not to mention the much
slower traditional funds.
By the summer of 2007 the quant funds had been very successful,
with the average fund outperforming its benchmarks by 2 percent. Some
did much better. Goldman Sachs’s flagship Global Alpha fund made
48 percent from April 2005 to March 2006, not quite as impressive as
the 140 percent it made in its first year, 1996. The quant funds got their
comeuppance in June 2007 when prices began to move in ways that were
the opposite of those predicted by the computers. The algorithms were
programmed to get out of losing investments, so when prices fell, they
automatically sold. And because they sold, prices fell, and because prices
fell, they sold more. They were all selling simultaneously, so prices moved
in lockstep, as they usually do in crises. Figure 27 shows a typical endog-
enous risk feedback loop.

Figure 27. Quant crisis, 2017. Credit: Lukas Bischoff/IllustrationX.
All the riskometers and clever pricing models failed to anticipate such
an eventuality. Losses were substantial. Global Alpha lost 7.7 percent in
July 2007 and a further 22.7 percent in August. It was not alone, as most
other quant funds suffered similar losses. By comparison, the US stock
markets fell 3.3 percent in July and increased 1.3 percent in August. Global
Alpha was eventually closed in the autumn of 2011 after even more losses.
What happened was that the masters of the quant funds overestimated
their cleverness. Each thought it was the smartest, using unique and
state-of-the-art techniques to make money. They weren’t, as underneath
they were all doing the same thing. It wasn’t visible when nothing was
happening as in the tranquil years of the great moderation. All that was
needed was a trigger. Typical endogenous risk crash, just like the wobble
of the Millennium Bridge.
Crises happen when a hidden risk factor hits an unseen trigger, pre-
cisely what happened here. The hidden risk factor was how all the algo-
rithms were programmed to react in the same way to market turmoil.
Completely unknown until a sufficiently large price drop triggered their
self-preservation tendencies: David Viniar, the then Goldman chief finan-
cial officer, was tasked with explaining Alpha fund’s losses to the world.
“We were seeing things that were 25-standard deviation moves, several
days in a row.” Under the normal distribution, the only way such a state-
ment has any meaning, a 25-standard-deviation loss is expected to happen
one day out of 10137, which is 1 with 137 zeros after it. For comparison,
NASA tells us that the universe’s age is about 14 billion years, while the
Earth is a lot younger at 4.5 billion years.

Either the quant funds got really, really unlucky or they weren’t nearly
as clever as they thought they were. Through the lens of endogenous
risk, the 2007 Quantland crisis was inevitable. The combination of the
great moderation with aggressive market participants aiming to exploit
the tranquility. After all, what is our natural reaction when we perceive
risk as being very low? Certainly not to sit on our hands and be conser-
vative. No, we follow Minsky and take more risk. In the beginning with
no impact on the markets, but eventually more and more funds pile into
the action. Hugely profitable at the start. Returns are higher and higher
because everybody enters the markets, while measured risk remains low
because the prices are going up steadily but with little volatility.
The riskometers couldn’t fathom such an eventuality in 2007. Because
it wasn’t in the data, it wasn’t detected. It wasn’t in their DNA. Funda-
mentally, when the quants started to use the riskometers to make invest-
ment decisions they changed the financial system, undermining the qual-
ity of the risk measurements. The 2007 quant crisis was inevitable. Risk is
endogenous even if the riskometers assume otherwise.
Demand for Junk–Financial Alchemy:

CDOs and the 2008 Crisis
Before Dmitri Mendeleev published the periodic table in 1869,
the concept of the chemical element was not well understood. People
even believed in alchemy, the turning of cheap base metals into gold.
They just had to find the right recipe. Fabulous riches awaited. While
Mendeleev showed alchemy is not possible in the physical world, the
modern practice of alchemy in finance is still alive—turning cheap, un-
loved assets into something everybody desires.
Financial alchemy was at the heart of the global crisis in 2008, not
least the CDO. The CDO gets its name because it uses various types
of debt, such as mortgages, loans, or bonds, as collateral for a CDO. It
became increasingly apparent in the early 2007s that CDOs were a cause
for concern. That is when I was asked to do a new course in LSE called
Global Financial Systems and started to look into CDOs. I found practi-
cally no background reading on how they work and especially what the
risk in them is. The only material out there was technical discussions

about creating CDOs and marketing blurbs on why they were fantastic.
All about alchemy: the ease and benefit of turning high-risk junk assets
into gold. Nothing about risk.
I ended up calling around former students working in the financial in-
dustry who were kind enough to connect me to experts willing to talk me
through the mechanics of CDOs. What I learned was decidedly fright-
ening. If I want to get the current price of a stock, all I have to do is to
look up the price on Bloomberg. Not that simple for CDOs, as there is
no market price to be found there. Instead, I need a complicated model
just to get the price. And an even more complicated model is required
to get the risk. Plenty of places for things to go wrong. Not only that,
but the models used to get the prices and risk of the CDOs contained a
fatal flaw. The way a CDO works is that a bank buys risky debt. Subprime
mortgages, junk bonds, or anything, so long as it is risky. Suppose I
buy one hundred subprime mortgages. Over time some of the mortgage
borrowers may get into difficulty. They may get sick or lose their job or
whatever, and default. However, it is quite unlikely that all one hundred
default.
Now comes the alchemy. Every month I expect one hundred mortgage
payments to be made. I then promise Mary the first five payments, no
matter which mortgage they come from. Yiying then buys the right to
get the next twenty payments. The important thing is that if ninety-five
of the mortgages default, Mary gets all her payments but Yiying nothing.
Morgane buys the right to the next sixty-five payments, and Paul buys
the right to the payments from the last ten mortgages to be paid every
month. Mary’s investment is the safest and Paul’s the riskiest, so Mary
will pay the most for her right and Paul the least. These rights are called
tranches (French for “slices”).
Mary’s right to the first five payments is very safe and gets a AAA rat-
ing from the credit-rating agencies. These tranches are often called super
senior because it is so certain they will pay out. Yiying’s tranche is less safe
and gets only a AA rating, while Morgane’s are even lower at BBB. The
middle tranches are known as mezzanine, Italian for “intermediate floor.”
Paul is the last to be paid, and his tranche is called equity or, colloquially,
toxic waste. Suppose we assume that if one family defaults on its mort-
gage, that says nothing about the likelihood of any other family default-

ing. Usually an innocent assumption, as each family’s default is specific to

them, perhaps illness, disability, or death. Even though individual families
are likely to default, the likelihood that all default is very, very low.
Alchemy. We have taken high-risk assets and turned them into super-
safe assets. What’s missing? The chance of an economic slowdown. When
the economy is doing well, mortgage defaults are mostly independent of
each other. But in recessions factories close and people get laid off. Be-
cause all the employees lose their jobs, they no longer patronize the local
restaurants and supermarkets and hairdressers, so the economic misery
spreads. A sizable number of families in that community can end up de-
faulting on their mortgages.
When things are good, mortgage defaults are individual, in recessions
contagious. The problem is that the previous recession in the United
States was in 1992, and subprime mortgages only really took off in 1993.
The data used to train the riskometers didn’t contain any recessions, so
the riskometers didn’t know about the possibility of factories closing and
people being thrown out of their jobs. The common practice of sourc-
ing subprime mortgages from a single county in the United States made
the problem worse because people living in the same geographic area are
likely to depend on the same employers.
But why would one want to put subprime mortgages into the CDOs
in the first place? The reason is that some tranches are easier to sell than
others. The safest and the least safe—super senior and toxic waste—had
few buyers, while the mezzanine was the most popular, so it makes sense
for the CDO’s creator to make the mezzanine tranches as big as possible.
However, that means that the assets cannot be too safe. For instance, if
you put US government bonds into the CDO, the safest tranche’s size in-
creases. Consequently, the CDOs created a demand for junk. A friend of
mine worked in the treasury office of a bank in 2008, and one day he got
a phone call from a large international bank offering to lend him a rather
substantial amount of money. My friend was a bit taken aback by this and
asked the banker if he had seen Bloomberg that day and noted his bank’s
very high credit default swap (CDS). The banker said, “Yes, that’s why
I’m calling you. I need something risky to structure.” My friend took the
loan and told me the large international bank took a 96 percent loss when
the global crisis hit a few months later.

Not many buyers wanted the equity tranches, so some smart bank-
ers found a way to tranche them up in a product called CDO-squared.
Clever, except the model risk is amplified because now we have the model
risk of the initial CDSs and the model risk of the CDO squared. Madness.
We even got some CDO-cubed!
Because the super senior and equity tranches were hard to sell, the
banks often held them on their books. Naturally, because the super se-
nior were AAA rated, no problem. While the equity tranches were risky,
the models didn’t find them excessively risky, so the capital charges were
limited. That is what caused all the bank problems in the second part
of 2007. As the CDOs got downgraded, all of a sudden the banks were
holding much riskier assets and they had a much lower value than initially
assumed. The banks found themselves in serious difficulty and close to
violating their capital constraints. The technical reason why UBS failed.
If one of the criteria for high profits is maximizing the use of junk as-
sets, it shouldn’t be surprising that it did not end well. Richard Bitner
describes this eventuality in his book Confessions of a Subprime Lender:
An Insider’s Tale of Greed, Fraud, and Ignorance. Nobody cared about
quality. All they wanted was the highest possible number of the riskiest
mortgages imaginable.
Cat and Mouse Games

When I used to do executive education courses on risk forecast-
ing, most students worked in supervisory agencies, risk management, or
compliance and wanted to get up to date on the latest techniques. But
one day I got a trader. We got talking, and when I asked her why she was
taking the course, she said that her risk manager was using Value-at-Risk,
and she wanted to know how he arrived at the risk numbers so she could
do what she wanted without interference from the pesky risk controllers.
While we are used to thinking about the regulators arguing with banks
about how much risk they can take, a much more adversarial debate oc-
curs between traders and risk managers. The former want to invest in
what they like when they like, while the latter are tasked with ensuring the
traders take risk in line with what the company wants. Not too little and

not too much. There is a perpetual tension between the risk managers’
and traders’ objectives, what economists call a principal-agent problem.
Because both the traders and risk managers know all about the strengths
and weaknesses of riskometers, a cat and mouse game ensues. The traders
try to exploit these weaknesses, whereas the risk managers try to close the
loopholes. One trick used by the risk managers is Chinese walls, a term
dating back to the Great Depression when the US government wanted
to minimize the conflict of interest between a bank producing suppos-
edly objective research on a company while also handling its initial public
offerings. Banks were allowed to keep both functions in the same bank
but were required to put the Great Wall of China between them. The
risk manager’s ideal case is when the riskometer is entirely hidden from
the traders—the Great Wall of China stands between them. The night-
mare for any risk manager is exercising control over a trader who is an
expert in riskometers. Even worse is when the traders used to work in
risk management or compliance before being promoted to trading and
know all the tricks. Jérôme Kerviel, the Société Générale trader who cost
his employer $6.9 billion, started his career in compliance before being
promoted to trading.
If Mary is a proprietary trader, all the risk manager should tell her is
that her risk is too high or too low or just right. He should not tell her
that her Value-at-Risk is $2 million and should be only $1.5 million or
even that he is using Value-at-Risk, and he should certainly not tell her
that he got the $2 million from a GARCH model. The more the trader
knows about the risk measurements, the easier it becomes for her to ma-
nipulate the risk management process. This is precisely what the student
in my executive education course was trying to accomplish.
The practical problem is that while Chinese walls sound great in the-
ory, they don’t work well in practice. After all, the risk model cannot be
hidden because it must produce results. Meanwhile, as profits are threat-
ened the Chinese walls lead to accusations of unfairness, discretion, and
incompetence. The resulting political game will be won by risk manag-
ers only immediately after a crisis, when memories of past losses are fresh
and fears widespread. In normal times the profit-making trader will have
much more political power than the risk manager. Senior management

turns a blind eye to transactions when significant profits are made. Just
like Credit Suisse, which muzzled its risk management department and
for its trouble lost at least $4.7 billion on Archegos Capital and cost
its clients up to $3 billion from the collapse of Greensill Capital. All in
early 2021.
The Four Principles of Correct Riskometer Use

Risk comes from the future, but we know only the past.
Soon after I wrote “Risk and Crises,” a blog piece on the prob-
lems of riskometers, I got an interesting comment from a risk manager:
“As a risk manager I fully recognise the shortcomings of any model based
on or calibrated to the past. But I also need something practical, ob-
jective, and understandable to measure risk, set and enforce limits, and
encourage discussions about positions when it matters. It is very easy to
criticise from the sidelines—please offer an alternative the next time.”6
He is right. It is easy to fall into the trap of excessive nihilism, criticizing
without providing alternatives. The riskometer is useful, we have to man-
age risk, and if we don’t use the riskometer to do so, we might as well
stick a wet thumb in the air. So, in response I joined forces with a friend,
Robert Macrae, and wrote a couple of blog pieces titled “Appropriate
Use of Risk Models.” Our fundamental point is that risk is defined only
in terms of outcomes we wish to avoid, and that how best to estimate risk
critically depends on what we want. In chapter 5 above on the myth of
riskometer, I gave the example of Paul, Ann, and Mary, who each invest
in Google stock but for very different reasons. Even though each holds
the same portfolio and has access to the same technology, they measure
risk differently. Simply because Mary cares about risk seventy years in the
future, Ann in the next six months, and Paul the following week.
Robert and I came up with five principles of the appropriate use of
riskometers. The first is that risk comes from the future, but we know
only the past. All a riskometer can do is project the past into the fu-
ture, and many conditions have to hold for the projection to be accurate.
The actual history has to represent the future, with no nasty surprises in
store—ergodicity. Meanwhile, the person creating the riskometer has to

be respectful of history and avoid data snooping. If we don’t think history

is a reliable guide to the future, don’t use riskometers; do something else.
The second principle is that the point of a riskometer is not to predict
bad outcomes. All it is supposed to do is tell us something bad may hap-
pen. So, even if everything goes wrong or nothing goes wrong, we do not
have sufficient grounds to criticize or praise the riskometer. A principle
that is violated frequently, for example, in postmortems on big losses.
The benefit of hindsight is that after a bad outcome everyone is an expert,
and criticizing is easy.
The third principle is that how one should calculate risks depends on
what one fears. Avoid at all costs one-size-fits-all approaches when it comes
to risk, like Value-at-Risk. We are all different and have different needs, so
other users need different riskometers. Unfortunately, everything pushes
toward the one-size-fits-all riskometer. When the same riskometer is used
by everybody, individual failure is covered by collective failure.
The fourth principle is, don’t probability shift. It is easy to estimate
a model for short-term risk (like 95 percent daily Value-at-Risk) and use
that to get an estimate of a once-in-a-millennium event, as I explained
earlier. While easy enough to do, for the numbers to have any meaning
we need at least two strong assumptions—ergodicity and the correct
parametric form—both of which are violated in practice.
The final principle is that financial risk is endogenous, created by the
interaction of the human beings who make up the financial system. En-
dogenous risk limits what any riskometer can do in that risk forecasting
involves predicting the behavior of people. As a general rule, a riskometer
performs best when focused on short-time horizons and nonextreme
outcomes, in cases where a lot of people are trading an asset. The risk-
ometer fails when asked about illiquid, long-term, and extreme risk be-
cause that involves predicting human behavior, too much to ask of a com-
puter algorithm.
Riskometers ought to come with warnings, like medication does: If

used correctly, the drug will help you, but there is a risk of side effects
and even death, so be careful. Just like the chance that medication might
kill you is not a good reason not to take medicine, the riskometers’ side
effects don’t mean we shouldn’t use them. Not measuring risk just means

sticking the proverbial wet thumb in the air. While not perfect, risk-
ometers indicate risk and, if used correctly, can be quite useful. And if
abused, it can similarly cause a lot of damage. Just like medication. Good
risk managers tell me they use many riskometers simultaneously. By get-
ting multiple measurements of the same risky position and knowing how
each of the riskometers performs, they can combine subjective judgment
with the objective output of the riskometers in their decisions.
The potential for manipulation is real, as the examples of JP Morgan’s
Whale and UBS show, not to mention all the rogue traders. Manipulation
is hard to detect until everything goes awry, no matter how well inten-
tioned a bank is. Even if the senior management fully intends to obey the
spirit and the letter of the rules, internal incentives such as bonuses and
promotions will lead traders to take advantage of the risk management
system: to abuse the riskometer. What is especially worrying is how regu-
lations can encourage banks to take the worst type of risk—tail risk—and
so perversely make it more likely the bank will fail. The reason is simple:
the tendency to use riskometers not as a risk-control device but to maxi-
mize profits: the London Whale problem.
Does anyone care? Despite all the colossal failures in 2008, riskometers
are still in widespread use, and banks and regulators are increasingly us-
ing them. The simple truth is that the modern financial system would not
function without riskometers. If we ask someone to manage our money,
we need to monitor and control the risk they take. In practice, that means
using riskometers. Just remember not to ask for too much. To those
whose job it is to care about the entire financial system’s stability, the
riskometers’ failings are especially pertinent.

9
The Goldilocks Challenge
What do you see? Light. Darkness. The balance is so much bigger.

—Star Wars: The Last Jedi (2017)
When I lecture on bank bailouts in my LSE masters course I ask

my students to imagine they are the finance minister of their country.
Suppose, as the finance minister, you receive the CEO of your country’s
biggest bank one morning, telling you that her bank will go bankrupt
by five o’clock that day unless you give her a $50 billion bailout. And if
her bank fails, it will take your economy down with it, causing a systemic
crisis. You have two clear choices. Call her bluff and refuse, risking a fi-
nancial meltdown, or give some of the wealthiest people in your country
$50 billion. Not an easy decision. At that moment, it would not matter
what your political leanings are, right or left wing. A libertarian who
believes the government should not regulate or bail out private entities;
147

148 THE GOLDILOCKS CHALLENGE
or a socialist hating the right-wing financial elite who would like nothing
more than to tax and regulate them out of existence. The only question
to consider: Is the banker telling you the truth? If you think she is, you
have no choice but to give in.
The reason is that no government can ignore a systemic crisis. The con-
sequences are so severe that we would do almost anything to prevent one.
Perhaps the worst outcome is that the empirical evidence suggests that
governments lose power after a major crisis. You would be out of a job,
and that would not do. When faced with the $50 billion request demand,
the more information you have about your financial system and the bank
in question, the better. That will not only help you figure out whether
you are being played, but also aids you in finding a better, cheaper solu-
tion. If you had had access to good information earlier, you might even
have been able to prevent the bank from failing in the first place.
The nineteenth-century US federal government did not impose much
in terms of regulations on banks, leaving that to the states. The United
States did not even have a central bank. When a crisis came along, the
private sector would sort it out. Not surprisingly, the United States was
very crisis prone, and there were increasing demands to establish a central
bank and start regulating finance. Still, politics was not in favor. It took
the severe financial crisis of 1907 to finally overcome the political opposi-
tion, and that was because of how the private sector dealt with the crisis.
In 1907 the most important banker in the country was John Pierpont
Morgan, the founder of the eponymous bank, which was later split by the
Glass-Steagall Act in 1935 into JP Morgan and Morgan Stanley. He acted
as the de facto central bank, providing liquidity himself and leaning on
other bankers to do the same, in the process enriching himself and his
buddies while punishing his enemies. The way he behaved in the 1907
crisis was seen as being beyond the pale, and the political leaders realized
that a central bank could not be avoided. It could not be called a central
bank for political reasons, but the Federal Reserve System, or the Fed,
was set up in 1913.
Even then, the Fed did not get many powers, nor did it want them.
During the Great Depression the Fed did everything it could to do noth-
ing. The definitive 1963 study by Milton Friedman and Anna Schwartz
showed that the Fed’s failure to increase liquidity in the Depression was

THE GOLDILOCKS CHALLENGE 149
the main reason it became so severe. Their conclusions determined why

we had so much liquidity created during and after the 2008 crisis: the
central bankers were dead set on avoiding their Great Depression pre-
decessors’ mistakes. I once talked to a central bank governor who told
me that what motivated him in the 2008 crisis was the portrait of his
Depression-era predecessor in the main conference room. He would look
at it and say to himself, “What would you have done? I will do the op-
posite.” Not until President Roosevelt took office did the US government
give the Fed both the mandate and the obligation to respond to crises.
Being forced to intervene in times of crisis without adequate plan-
ning is the worst-case scenario, as the Argentinians learned to their cost
in 1994. A few years earlier their government had proclaimed it would
never bail out any bank, only to reverse course embarrassingly following
the tequila crisis in Mexico in 1994. It is far better to be prepared for the
eventual crisis—it will happen.
It is not enough to ensure the central bank understands the system and
is ready to step in if a sufficiently bad event happens. Even that decision
has secondary consequences requiring further regulations, and so forth.
Once they have taken one step, they need to go the whole way. It is like
getting married. I cannot just pick the things I like about my partner and
leave the rest—she is a package deal. It is the same with financial regula-
tions. Still, we should not expect too much. The authorities can never be
fully prepared, and they will make mistakes. Big, costly, stupid ones. It
is easy to criticize with the benefit of hindsight, but regulating finance is
a tough job. It is difficult, perhaps outright impossible, for any financial
authority to stop the excessive risk-taking that leads to a crisis, as bubbles
make everybody feel warm and cuddly. Since the public sees only the ben-
efits, the politicians follow. So do the bankers, even if they should know
the danger the bubble poses.
The technocrats should be in the know. However, if they warn the pub-
lic, they risk being denounced, losing their income, or even being pros-
ecuted, as Greece’s former chief statistician, Andreas Georgiou, learned.
In my native Iceland, when the pending crisis became clear for all to see
in the spring of 2008, government ministers still denounced those warn-
ing it was all going to end in tears, with some even calling on the banks
to increase risk. Ultimately, it is tough to follow the advice of William

McChesney Martin Jr., the former Fed chairman, who said the Fed’s
most important job was “to take away the punch bowl just as the party
gets going.” The financial authorities face a difficult Goldilocks challenge:
not regulate too little, as the United States did in the nineteenth century;
and not too much, like Cuba or North Korea today. It has to be right.
The 1866 Lesson

No country had any financial regulations to speak of at the start
of the nineteenth century. When they thought of banks at all, the govern-
ments cared only about forcing them to buy government bonds. The only
financial crime was theft, sometimes punishable by death, which applied
only to poor people stealing to eat, not bankers robbing their business
partners, clients, the government, or the general public. There were no
rules on disclosure, and it was perfectly legal and morally acceptable to lie
in public statements. This way of life changed after the 1866 crisis, when
Overend & Gurney (O&G) went bankrupt. O&G was one of London’s
most respected financial institutions and the largest bank in the world fo-
cused on commercial lending. The day after it collapsed, on 30 September
1866, it was called the “greatest instrument of credit in the Kingdom.”1
The development of a commercial credit market was a major reason the
United Kingdom became the world’s first industrialized country. Bud-
ding British capitalists found it easier to raise money—the venture capi-
talism of the day—than their counterparts anywhere else in the world.
One of the riskiest and most profitable types of venture was shipping, and
the development of new ships and shipping lines was the high-tech indus-
try of the mid-nineteenth century. For most of its existence O&G didn’t
participate in such high-risk lending, but the partners eventually came to
desire an increase in their already immense wealth and decided to finance
shipping technology. But they bet on the wrong technology and soon
faced bankruptcy. To save themselves after O&G was effectively bank-
rupt the partners decided to incorporate, selling shares to the public and
telling potential shareholders the bank was completely safe, with no bad
loans. At the time, this was legal, and there was no obligation for any firm
to publish accounts, reveal its actual financial situation, or tell the truth.

When O&G failed, Walter Bagehot, then editor of The Economist, wrote
that the partners ran their business “in a manner so reckless and foolish
that one would think a child who had lent money in the City of London
would have lent it better.” The partners believed they would be bailed
out by the Bank of England, which sent three bankers to look at O&G’s
books. It did not take them long to realize O&G was broke. The Bank of
England faced a delicate decision: if O&G failed, there would be panic; if
it were saved, the other firms in the finance game would also expect to be
rescued. The Bank of England chose to let O&G fail. It is not clear why
it made that choice. Moral hazard was clearly a significant concern, but
other factors weighed on it as well. The Bank of England was a private
institution competing with the likes of O&G, and its future profits were
likely to be enhanced by the failure of an important competitor. Because
the Bank of England did not do any bailouts nor provide support of any
kind, even refusing to grant loans against government securities, panic
spread through the banking system. The market for otherwise safe assets
like government bonds—gilts—dried up.
The financial institutions of the nineteenth century were partnerships,
with the notable exception of the Bank of England. This meant that
O&G’s partners should have been liable for all losses, except they man-
aged to incorporate in the nick of time. The third senior partner had the
surname Barclay, and a few years later he started a bank with the epon-
ymous name, where the last Overend became the largest shareholder.
The partners of O&G eventually faced private prosecution because the
government did not feel they had done anything wrong and refused to
prosecute them. The only crime they could be charged with was theft. In
the private prosecution, the partners hired the government’s most senior
lawyer to defend them, so the same lawyer both represented the accused
and was the boss of the judges ruling on the case. The partners were
acquitted.
At the time, there were no established procedures for dealing with the
failure of a big bank. But as the O&G 1866 crisis turned out to be one of
the worst crises of the century, the government had no choice but to do
something. It made the Bank of England commission Walter Bagehot to
investigate how it should respond to future crises. He published a white

paper in 1873, establishing three principles enshrining the Bank of En-

gland’s role as a lender of last resort:
1. The central bank should lend freely

2. At a penal rate of interest
3. On good banking securities.
Bagehot’s advice proved to be good, and the Bank of England success-

fully dealt with the banking panics of 1878 and 1890 by following his rules,
while the panics of 1847, 1857, and 1866 led to acute financial crises. Bage-
hot’s analysis of the 1866 crisis and his recommendations for how the
government should resolve crises are the first modern analysis of finan-
cial stability and policy remedies. It has been hugely influential and has
shaped policy throughout the world ever since. The refusal of the Federal
Reserve to follow Bagehot’s rules was a significant contributor to the
Great Depression from 1929 to 1933. We see the influence of Bagehot’s
principles to this day, and he was frequently invoked in the 2008 crisis.
Bubbles and Our Amazing Capacity for Self-Deception

One of the trickiest Goldilocks challenges is bubbles. Are they
simply the consequence of speculation gone sour and should hence be
pricked? Or is all the rapid price growth real, to be appreciated and nour-
ished, not stamped out? Japan and China give nicely opposite answers to
this question.
The land the Imperial Palace in Tokyo sits on is not very large—when
last in Tokyo, I walked all the way around it in less than an hour. In
1989 that land was more valuable than the entire landmass of the state
of California. Japanese home buyers needed multigenerational loans to
buy homes. That was at the very tail end of an extraordinary economic
boom that started right after World War II and ended in 1992 with one
of the biggest financial crises ever seen. In the three decades before 1992,
the average annual economic growth in Japan was 5.1 percent. It has been
0.7 percent since. The stock market has fared even worse. The annual
return was 9.9 percent but only about 0.5 percent since.2 The Japanese
bubble is typical of financial bubbles. In the beginning Japan’s rapid eco-
nomic growth was based on sound foundations, high-quality human

capital, and the rebuilding of a war-devastated economy. But because the

growth was so rapid it looked like the Japanese businessmen and policy
makers could do no wrong, a perception that masked the underlying vul-
nerabilities. When Japan ran out of productive industrial investments, real
estate became the logical target, which explains the astronomical value of
the land under the Imperial Palace in 1989.
Real estate bubbles are often the last thing to happen before a crisis.
When Robert Aliber, a professor at the University of Chicago, one of
the best-known experts on financial crises, and the coauthor of the book
Manias, Panics, and Crashes: A History of Financial Crises, visited Iceland
in the spring of 2008, he caused a scandal when interviewed on televi-
sion. What Robert did was to count the number of building cranes in
Reykjavík, and use that number to predict a crisis. Everybody said he was
wrong. The Icelandic crisis, the worst in the global crisis that year, started
four months later.
The Japanese themselves didn’t see a bubble, finding all sorts of rea-
sons why things were different in Japan. One of the best crisis predictors
is people invoking their culture. Japan had a unique economic model,
rejecting outdated European and American short-term capitalism and
creating a unique, sustainable Japanese economic system. The Japanese
businessman had morphed into what the Japanese liked to call the eco-
nomic man, uniquely able, working in a superior system. Then reality
came knocking. The laws of economics are universal.
Bubbles are insidious. We start by taking a little risk. Things turn out
okay. We take more and more risk at a faster and faster rate until it all
collapses. Bubbles happen all the time. Internet stocks in the late 1990s,
subprime mortgages and structured credit products before the 2008 cri-
sis, the Roaring Twenties, the Nifty Fifty stocks in the early 1970s, real
estate in Spain in the 2000s. We can go far back in history. The Dutch
tulip mania in the 1600s is the first documented bubble but surely there
were many before. Is the financial market Covid-19 boom simply a bub-
ble destined to burst? Is China: The Bubble that Never Pops in Thomas
Orlik’s book or as George Magnus has it in his Why Xi’s China Is in
Jeopardy?
Who is right? Looking back after a bubble bursts, it is easy to act wise
and say, “We had this huge bubble. Why were the speculators so stupid,

and why did the authorities not do anything? All we had to do was to
restrict credit. Really simple stuff.” Yet bubbles still happen. The prob-
lem is that their very existence can be verified only after they burst—
we need the benefit of hindsight. We can never be certain before the
bubble bursts. The cost of getting it wrong, calling something a bubble
incorrectly—what statisticians call a type I error, the incorrect rejection
of a true hypothesis—can be very costly.
Suppose the Chinese authorities had decided in 1995 that “oops, China
is overexpanding, we are in a bubble and have to slow down.” Then
China would not have enjoyed the 470 percent economic growth over
the next couple of decades. One can just as easily become dazzled by all
that growth. China will stop growing one day, and then its market will be
seen as having been in a bubble. But crying wolf is a game one is destined
to lose. It could happen tomorrow or in twenty years.
Even supposedly well-identified bubbles, like the internet dot-com
bubble of the 1990s, might not necessarily be such a bad thing. Bubbles
can cause prices to explode, but they also release a lot of funds for risky
investments, some of which end up being the seed money for the lead-
ing companies of the future. The online giants of today, like Amazon
and Google, originated from the dot-com bubble. Intel and many of the
semiconductor companies came out of the Nifty Fifty bubble in the early
1970s, while many companies like Coca-Cola had their IPOs during the
Roaring Twenties bubble. We don’t even know if we should call these
episodes bubbles. The Nobel Prize–winner Fama French, the godfather
of the theory of efficient markets and the person just about least likely
to believe in bubbles, argued recently, “For bubbles, I want a systematic
way of identifying them. It’s a simple proposition. You have to be able to
predict that there is some end to it. All the tests people have done trying
to do that don’t work. Statistically, people have not come up with ways
of identifying bubbles.”
It is difficult for the financial authorities to meet the Goldilocks chal-
lenge posed by bubbles. The political economy is always in favor of allow-
ing the bubble to grow. And the authority aiming to prick them faces a
dual risk. If they prick the bubble, they will always be accused of destroy-
ing growth. If they allow it to happen and things then blow up, they will
be accused of incompetence or worse.

How Iceland Failed the Goldilocks Challenge

One of the first questions facing a designer of regulations is
whether to use rule-based or principle-based regulations. The former fo-
cuses on preventing undesirable behavior, much like a speed limit that
prohibits us from going too fast. The latter says it is better to focus on
the outcomes, that we should drive in a way that does not lead to a crash,
taking into consideration things like the weather, the state of the road we
are on, traffic, and the quality of the car we are driving. Neither approach
is satisfactory, and we need some combination of both. If we go too far
in either direction we are likely to end up in trouble. The Icelandic crisis
in 2008 exemplifies this conundrum.
After the Icelandic banking system collapsed in October 2008, the head
of the banking regulators was asked in a television interview what the
financial supervisor’s job was. He replied, “To ensure the banks don’t
break the law.” To a lawyer, that is the right answer, but to me, an econo-
mist, the journalist asked the wrong question. It should be “What is the
purpose of regulations?” a question with a straightforward answer. The
objective of regulations is not to make sure banks don’t break the law.
Instead, it is to ensure they provide essential functions to the economy,
efficiently and at a reasonable cost while not behaving in a way that harms
society. The Icelandic banks violated all the regulations’ objectives while
not breaking any law—and got away with it.
The emphasis on not breaking the law might not be too problem-
atic if the law is sufficiently general, allowing the regulators to enforce
the law’s spirit. The problem in Iceland is that its legal system has an
unusually strong focus on literal interpretations of the law and is very
reluctant to take into account the purpose of the law. In most coun-
tries the regulator would determine the law’s spirit and issue instruc-
tions to achieve that. Hard to do in Iceland. This led to some curious
outcomes. A few years before the crisis, the banking supervisor told a
bank, call it bank A, that it needed to increase its equity—issue and
sell new shares—but the bank did not want to do so because it would
dilute existing shareholders. The most direct way around this was for
the bank to issue new shares and purchase them itself. But that is a fake
transaction and is not allowed.

The Icelandic banks found a neat way around this. Bank A sold the
newly issued equity to bank B, and bank B sold its newly issued equity to
A. While this might mean the banks are now exposed to each other, they
then made a contract for difference, compensating each other for losses
and reimbursing each other for profits. If the price of A fell, A would
make up B’s losses, and, conversely, if A’s stock price went up, B would
give the profits to A. To the Icelandic regulator and its strict tick-the-box
legal point of view, no problem. Of course, it was all fake and, while le-
gal, did not afford any protection. Even worse, it gave the appearance of
protection, encouraging market participants to engage with these banks
as if they were safe. After the crisis, I mentioned this to a few acquain-
tances in supervisory agencies in other countries. They were aghast, all
saying that this sort of transaction would not have been allowed in their
country. Even if not explicitly forbidden, it would be seen as a violation
of the regulations’ spirit.
But if the regulators were too relaxed before the crisis, there is a danger
of going too far in the other direction after it has happened. The Ice-
landic regulator again shows this better than most. Ever since the 2008
crisis its main preoccupation has been to impose rules that would have
prevented the crisis, closing the barn door after the horse escapes. Ice-
land has some of the world’s highest bank capital charges and extensive
restrictions on what the financial system can get up to. Consequently,
financial services in Iceland are the most expensive in Europe, and it is
really hard to make the sort of risky investments so necessary to underpin
future growth. Instead, Iceland bet its fortune on tourism, an industry
that does not need a lot of capital, only to see it come crashing down with
the Covid-19 crisis in 2020.
The Icelandic regulators are hardly unique in failing the Goldilocks
challenge, even if they fell for it particularly hard. It has happened in all
jurisdictions and exemplifies a particular form of procyclicality. During
upswings, regulations become increasingly lax, amplifying the boom; af-
ter a crisis, they become excessively strict, magnifying the downturn. By
2020 the cry for fewer regulations was getting increasingly loud, and,
with Covid-19 everywhere, financial stability played second fiddle to eco-
nomic prosperity. Many countries already have relaxed capital standards,
and most signs point to even more relaxation. Unfortunately, the last cri-

sis episode always has an undue influence on how we think about future
crises. The policy makers are always at risk of falling for the successful
general’s syndrome, just like the French army in the 1930s.
The Maginot Line: Preparing for the Past

When the French military after World War I was preparing for
another German invasion, they thought the best way was to construct
a great defensive barrier along the French–German border. They called
it the Maginot Line. There was only one problem: while the French
prepared for another World War I with trenches and heavy artillery, the
Germans came up with Blitzkrieg. They invaded via the Ardennes For-
est—which the French assumed was impenetrable because they had not
stayed on top of new technology—bypassing the Maginot Line entirely
and eventually capturing it from behind. The French fell for the successful
general’s syndrome—preparing for the last war while the enemy prepares
for the next.
It is difficult for financial authorities to avoid the successful general’s
syndrome. The last crisis dominates the headlines and is studied intensely.
Inevitably, prevention of the past dominates the regulations of the future.
The problem is that while there will be a crisis, it is almost certain that the
next crisis will happen in a different place, somewhere no one is looking,
like Covid-19.
That is precisely what happened in the years and decades before 2008,
when old-fashioned financial crises were thought to be a thing of the past
and what really mattered to the central banks was inflation. It is under-
standable why central bankers would think that way. High inflation was
endemic in the 1970s and early 1980s, and there seemed to be no way of
conquering it. Everyone was obsessed with inflation. As far as I know, the
only time a US president has talked about the third derivative—the rate
of change of the rate of change of the rate of change—in a public speech
was Richard Nixon in 1972: “The rate of increase of inflation is going
down.” Stagflation—a combination of stagnation and inflation—was
the word of the 1970s. With all the effort focused on inflation, the central
bankers eventually defeated it, and by 2007 they were patting themselves
on the back for having solved their main problem. They had reduced

monetary policy to an engineering problem: small tweaks in interest rates

would allow them to hit their target inflation rates. As it turned out, they
hadn’t mastered inflation all that well either: they did not hit their 2 per-
cent inflation target for over a decade afterwards.
Most central banks thought it essential that their mission not be pol-
luted by other dirty objectives such as financial stability. I remember being
in central bank conferences in the mid-2000s in which senior central bank
officials debated whether financial stability belonged within their remit and
most argued no. After all, the main problem of their forefathers, financial
stability, had been solved. Crises happened only in developing countries,
not in the United States or Western Europe. Echoing the French army’s
backward-looking preparations in the 1930s for the next war, most central
banks scaled down their financial stability divisions in the 1990s and early
2000s. The Bank of England was among the most aggressive, laying off
or pushing out those not doing macroeconomics, the reason why it was
so badly prepared in 2007. In doing so, the central bankers forgot the
lessons of the previous crisis. Macroeconomics had evolved in the second
half of the twentieth century as if the financial sector didn’t exist. When
I was in graduate school, my macroeconomics courses completely ig-
nored the financial system. The father of macroeconomics, John Maynard
Keynes, must have rolled over in his grave because his magnum opus, The
General Theory of Interest, Employment and Money, strongly emphasized
the financial system’s importance in the macroeconomy. When 2008 came
along, the central bankers and macroeconomists rushed to the library to
dust off the history books and the economic texts of half a century earlier.
They were no better than the French generals in the 1930s in falling for
the successful general’s syndrome.
It is so easy to fall for the successful general’s syndrome because most
information we have about financial stability relates to what happened in
the past. A bureaucratic entity charged with creating transparent rules
will find it much easier to make rules preventing recurrence than an-
ticipating what might happen. Central banks are rigid, bureaucratic, and
formal, well suited for precise processes that don’t often change, not so
much for out-of-the-box thinking. One might then try to solve this state
of mind by using principle-based regulations. A bank is supposed to be-
have in a way that meets the broad objectives of financial regulations, not

take on too much risk, finance the all-important small- and medium-sized
enterprises, and at the same time not go bust. It is one of those things
that sound good in theory but are not so easy in practice. It is hard to fig-
ure out what the objectives would be and practically impossible to verify
whether a particular bank is complying.
Principle-based regulations require innovation, and if the authorities
try to be too innovative in their regulatory designs, they run the risk of
fostering spectacular regulatory failures. This does not mean that cre-
ativity in financial regulation, like principle-based regulations and light
touch and sandboxes or whatever the buzzword du jour is, is a bad idea.
The alternative can be just as bad and even worse, as the tick-the-box-
obsessed Icelandic regulator shows. So we need some balance between
the tick-the-box approach and abstract, principle-based regulations if we
are to meet the Goldilocks challenge.
After the 2008 crisis, the word—in the UK and in places that take its
lead—is conduct. We even have a Financial Conduct Authority in the
United Kingdom, the idea being that financial institutions should be-
have properly and avoid doing anything untoward. That means focusing
on individual behavior, with a special emphasis on compliance. After the
financial 2008 crisis, the LIBOR, foreign exchange, and terrorist financ-
ing scandals, financial institutions have beefed up their compliance func-
tions. The number of risk and compliance staff in HSBC in 2014 was
10 percent of its entire workforce. Every bank intensively monitors its
staff. Computers and senior managers read messages and emails, tele-
phone calls are recorded, and the banks are spending enormous efforts to
ensure that employees behave properly. Many senior managers must read
daily emails and messages from their underlings, selected by an artificial
intelligence engine. In his 2016 book Why Aren’t They Shouting? Kevin
Rodgers mentions the anecdote of a Russian quant working for him in
his Deutsche Bank office, saying, “It’s great. I was too young to be spied
on under communism, but now I have the chance to experience it under
capitalism. I guess you could say that’s progress.” Even so, banks can’t
prevent some employees from going rogue, and abuse is inevitable. When
the next scandal comes along, are we willing to ratchet up monitoring
of bank employees even further? Will the cost of doing so in both finan-
cial and civil liberties terms be acceptable? Even then, will that prevent

future scandals? Francesch Castello, the banker beheaded in Barcelona

in the 1300s, might suggest it does not. The Goldilocks challenge will
not be solved by finding the right balance between principles and rules.
Because it is not enough to focus on the behavior and the incentives of
the regulated. We should also look at the motives of the financial regula-
tors. Are they captured or excessively risk averse like the Chinese air traffic
controllers?
Over the years, as a happy consequence of its rapid economic growth,
the airspace above China has become extremely crowded. One reason is
that the military controls a substantial part of it, but the other (of bigger
interest to me here) is Chinese air traffic controllers’ incentives. Their
job is stressful, having to maintain constant high concentration, juggling
a large number of aircraft, and managing the limited airspace efficiently.
The problem is that accidents are inevitable, and the only way to pre-
vent them is not to have any air traffic—just like the only way not to have
financial crises is not to have a financial system. Well, that is not feasible,
but the instinct of any air traffic controller who wants to minimize the
likelihood of accidents is to maximize the spacing between aircraft so they
will not collide. If taken to the extreme, air traffic will become very slow.
To solve that particular Goldilocks challenge, American and European air
traffic controllers are incentivized to manage traffic efficiently, tasked with
simultaneously maximizing the volume of air traffic while maintaining ac-
ceptable higher levels of safety. Air traffic controllers who make mistakes
and report them promptly benefit from leniency programs. The aviation
authorities have created a highly effective risk culture. It is different in
China. Their air traffic controllers are severely punished when they make
mistakes but get no benefit when air traffic flows smoothly. Thus they like
to keep the space between aircraft between ten and sixteen kilometers,
not the five in America.
Bank regulation continually pushes toward the Chinese air traffic
model. When the banking system is functioning well and with no scan-
dals, the supervisors don’t get much credit. Even worse, we harp on the
cost of regulations. Then, when something goes wrong we blame the
regulator. The head of the regulatory agency will be roasted in the press
and summoned to give testimony before the senate on live television. The

type of question she is going to get is, “You had all the information and
all the powers, why did you let the bankers exploit the public?”
The danger is that the supervisors’ incentives end up being focused
on the prevention of failure at all costs: the supervisor becomes too risk
averse. In other words, the supervisor’s incentive problem is the inverse
of that of the banker, and it is essential to have some mechanisms in
place to prevent excessive supervisory risk aversion. One way to do so is
through a cost-benefit analysis of regulations, easy to conceive of, hard
to do in practice. The opposite also frequently happens, what is known
as regulatory capture, whereby a supervisory agency no longer services
society but instead works in the interests of the regulated. The best recent
example is the American Federal Aviation Administration (FAA) and the
troubles of the Boeing 737 Max aircraft. The FAA seems to have out-
sourced most monitoring of the 737 Max to . . . Boeing. When foreign
regulators started to prohibit the 737 Max from flying, the FAA resisted. A
classic case of regulatory capture. Regulatory capture is all too common.
In Britain a recent case is the horsemeat scandal, in which food processors
got away with selling horsemeat as beef. The Food Standards Agency not
only failed to discover the substitution despite frequent inspections, but
also did not want to punish the guilty nor change practices afterward.
The banks are superb lobbyists, aiming to create banking regulations
favoring themselves while discouraging entry into the banking system.
And for good measure protect banks’ profits and ensure the odd bailout.
If the supervisors don’t play ball, the banks just go directly to the politi-
cians in charge of the supervisors. If that wasn’t enough, governments
have many other reasons for regulating besides keeping the financial sys-
tem running efficiently. Most, if not all, like to have “national cham-
pions”—the reason we have the systemically important financial insti-
tutions (SIFIs)—and that political desire gives the banks considerable
power over the regulators. The government may have altruistic ulterior
motives, such as requiring banks to provide unprofitable banking services
to disadvantaged sectors of society. Charles Calomiris and Stephen Haber
argue in their book Fragile by Design: The Political Origins of Banking
Crises and Scarce Credit that the political objective of providing housing
for low-income families in the United States led to an unholy alliance

between left-wing advocates of the poor and the bankers, preventing ef-
fective regulations and thus creating fertile ground for the 2008 crisis.
They further argue that some countries suffer an excessive number of cri-
ses because bankers and politicians join forces to create a financial system
favoring well-connected special interests at the expense of society.
Why Bail Out Banks and Not Carmakers?

When the finance minister in the archetypal example gets the re-
quest for $50 billion (see above), the banks are demanding a bailout.
The main problem for the government is that giving in results in moral
hazard. Even the remotest possibility of a bailout creates a moral hazard.
Moral hazard happens when we benefit from taking risk while someone
else bears the cost of that risk. Insurance is a typical example. When one
is insured, the insurance company pays out if something terrible happens.
Moral hazard is where being insured induces us to misbehave. Perhaps
a guy’s laptop is old and gets conveniently lost when he is on holiday,
the type of moral hazard insurance companies face continuously. But
they know this, and their fees reflect the risk. That is the key: the risk is
priced in.
So why even contemplate bailing out the banks in the first place? When
the global crisis hit in 2008, I remember listening to BBC’s main current
affairs program, Today, on the topic of carmakers in difficulty. The spokes-
person for the carmakers’ association made a strong case that carmakers
were strategically important for Britain and had to be bailed out if they
failed. We see many such demands in the 2020 Covid-19 crisis. We have all
these businesses that are innocent victims of the virus demanding bailouts
and quite often getting them. One of the worst legacies of the Covid-19
crisis will be the lingering expectation of state support for private indus-
try when things go wrong.
I don’t think we should bail out carmakers or any other private firm,
except perhaps banks. The reason is that there is a crucial difference be-
tween a bank and a carmaker. The failure of even the largest carmaker
is quite unlikely to be a vital public concern. The shareholders, credi-
tors, and possibly the employees suffer, but competing carmakers benefit.
There is no significant damage to the economy, and the failure may be

good for the economy if it leads to more innovation. The role of the
government should be limited to managing the bankruptcy process. That
doesn’t stop it from interfering, and the government routinely bails out
corporations, like the United States’ bailing out of General Motors in
2009 at the cost of $11 billion. While there might be good political rea-
sons for bailing out carmakers, there is little economic sense in doing so.
Failure is an essential part of the capitalist economy. Inefficient companies
fail, and other, better ones take their place—Joseph Schumpeter’s cre-
ative destruction.
While the economy can quite easily cope with even the largest corpora-
tions’ failure, it is not the same with banks. Modern society cannot func-
tion without continuous banking services. We depend on debit cards to
pay for lunch every day, and companies need to pay employees and sup-
pliers on Fridays. It all goes via banks, so any hiccups in the services pro-
vided by banks are quite disruptive to society at large. While it is relatively
straightforward to transfer control of failing corporations to the creditors,
it is not the same with banks because their business is moving money. In a
bankruptcy, ownership of money and obligations need to be clear. It can
take a long time to establish claims; Lehman Brothers is still in litigation
a decade and a half after its failure.
It gets worse. Bank failures are contagious. The only reason a bank stays
in business is because its clients believe in its solvency and permanency.
If the clients lose that faith, they will all rush to withdraw their money or
terminate their business relations—a bank run. The social cost of a bank
failure vastly outweighs the money lost in a bankruptcy because any dis-
ruption in the provision of banking services directly hits everybody. The
result is what we economists call externality. The private cost or benefit is
dominated by the cost or benefit to society at large.
Bank bailouts also create moral hazard. If the government steps in
when banks fail, the banks are likely to take on more risk than they other-
wise would. Unfortunately, the moral hazard from bank bailouts is dif-
ferent from that in insurance because it is usually not possible to charge
those receiving bailouts for the privilege. After all, the bank is already
failing, and we have to give it money: there is no point in kicking it when
it is down. So why not charge them an insurance premium that pro-
tects the government in case it has to give a bailout? Easy in theory, hard

in practice. The reason is that the insurance premium should reflect the
bank’s riskiness, but if it was so risky why didn’t we prevent it in the
first place? Besides that, it isn’t that easy to measure the risk in the first
place. If we then charge a fixed insurance fee, we are merely punishing
the prudent.
The hardest decision to get right in the Goldilocks challenge is bailouts.
Bailing out banks means rewarding the very entities that got into trouble
in the first place, but not bailing them out can cause a much more costly
crisis. All is not lost, and there are ways to do bailouts right. A bailout is a
transfer from one part of the population to another—someone pays and
another benefits. How to determine the winners and losers is a political
question, why the government, not a bureaucrat, has to decide to do a
bailout. Once a bailout is decided on, who should do it? The central bank
or the ministry of finance? The central bank might be ideal because it can
print money on demand and so cannot run out of money. However, there
is no such thing as a free lunch. The consequence of monetizing bailouts
is inflation, and the cost is borne by people on fixed incomes, like retir-
ees. That might seem somewhat academic now that the central banks can
apparently print all the money they want without consequences. But the
central question remains: Even if the central banks could run the print-
ing presses at full speed 24/7 without worry, who should benefit? Do we
build a hospital or school, lower taxes, or do a bailout with the freshly
printed money?
To get a handle on the question of bailouts, I came up with a classifica-
tion scheme for bailouts in my book Global Financial Systems. To start
with, some bailouts should be done by the ministry of finance and others
by the central bank. The central bank should do the bailout in liquidity
crises, where everybody is clamoring to convert liquid assets into cash—
as in 1907, 1914, and 1866. In a liquidity crisis, the market for even the saf-
est of assets disappears, as when investors went on strike in the autumn of
2007. A solvent bank with tangible assets might fail simply because it can-
not sell them at a reasonable price to meet immediate demands for cash.
Suppose the crisis was not only about liquidity but also solvency, as in
2008. The banks made bad decisions, some of their loans are worthless,
and they are facing bankruptcy. The banks’ difficulty is caused by their
misbehavior, so they are not merely the victim of the crisis. A liquidity

injection is not sufficient. The government needs to directly support the

banks to shield the economy from the worst. That is best done by the
ministry of finance, not the central bank, because it involves using tax-
payers’ money to bail out a private entity, needing political support to be
legitimate.
So what is the best way for the ministry of finance to bail out a bank if
another bank can’t take it over? It has five choices:
1. Guarantee the bank’s debt

2. Take over (buy) the bank’s bad debt
3. Lend money to the bank
4. Buy the bank’s preference shares
5. Force the bank to recapitalize by issuing new shares sold to the
government at a nominal cost or just nationalize the bank if it is
in a particularly bad way.
For taxpayers, option 5 is the best, followed by 4, all the way to the worst,
1, since the taxpayer gets the upside if things go well. The banks, of course,
see it differently. They prefer the taxpayer to absorb all the losses and will
lobby for debt guarantees and, barring that, the government assuming
bad debt. The last thing they want is equity dilution or preference shares.
So which outcome prevails? It depends on a bank’s power. If it can force
option 1 it will do so, whether via lobbying, scaremongering, or bribery.
Perhaps the same family that owns the bank runs the government? It’s
been known to happen.
In the 2008 crisis the European banks were generally successful in forc-
ing the government’s worst options, 1 and 2. Why? Scaremongering and
the authorities’ inexperience. The banks knew what was happening, and
the government was not prepared, so the banks maneuvered the authori-
ties into their preferred option. Ireland made the worst choice, guaran-
teeing bank debt and earning a sovereign default for its trouble. Europe
since then has tried to find better outcomes for the taxpayer. The Spanish
government brilliantly managed to implement option 5 in the summer
of 2017 when it resolved Banco Popular and sold it to Santander for one
euro. It is not easy to be so ruthlessly efficient, as is shown by the Euro-
pean country with the highest number of bank failures in recent years,
Italy. Their politics gets in the way, as in the resolution of the world’s

oldest bank, Banca Monte dei Paschi, in 2017. It had been failing slowly,
and anybody following Italian banks knew it was only a matter of time
until it failed. Unfortunately for the Italian taxpayers, the government
vacillated while the costs mounted, eventually deciding to bail it out at a
cost of €5.4 billion.
The Monte dei Paschi case and many other bank failures in Italy and
elsewhere illustrate how easy it is to take advantage of the taxpayer. The
reason is shown in The Wealth Effect: How the Great Expectations of the
Middle Class Have Changed the Politics of Banking Crises by my LSE
political science colleague Jeff Chwieroth and his Australian coauthor,
Andrew Walter. They argue that bailouts have become a middle-class
good. Rescuing banks benefits the middle classes, which then lobby in
favor of bailouts, which is why the Italians continue to spend money they
can ill afford to bail out their banks.
It is easy for the financial authorities to adopt a principled position
when nothing is happening, proclaiming they will protect the taxpayer
and minimize moral hazard. It is much harder to stick to it when push
comes to shove, as the Argentinians learned in 1993 and the Italians to-
day. The political pressure is enormous. Perhaps the investors are pension
funds or grandmothers as in Italy, and by having them take a hit the eco-
nomic and political consequences will be severe. The temptation is always
to give in: the taxpayers can surely afford it.
Goldilocks enters the Bear family’s house, tasting the porridge of Papa
bear, Mama bear, and Baby bear. She finds Papa bear’s porridge too hot,
Mama bear’s too cold, and baby bear’s porridge just the right tempera-
ture. The Goldilocks challenge is to find the appropriate balance between
too hot and too cold. The most important and the most challenging part
of the regulators’ job is the Goldilocks challenge. Regulate too much,
and the economy doesn’t grow; regulate too little, and we get damaging
crises. We need the right regulation intensity.
It is not easy. Of all the human activities we try to regulate, the finan-
cial system is by far the hardest to control. The incentives of the regula-
tors and bankers do not align well with society’s interests. The mission’s
complexity means it can be difficult to avoid regulators who fall into one

of two extremes, either excessive risk aversion or getting captured by the

regulated. The regulators are under constant pressure from lobbyists,
pundits, and politicians, all demanding action, disregarding the need for
an integrated approach to regulations. Meanwhile, those who are regu-
lated react unpredictably, in some cases driving risk-taking into the shad-
ows or perhaps not taking the right type of risk.

10
The Risk Theater
Over the past decade, G20 financial reforms have fixed the fault lines that
caused the global financial crisis.
—Mark Carney (2017)
In the movie Minority Report, a special police unit called Pre-

Crime predicts murders before they occur and thus arrests killers before
they kill. Premeditated murders do not happen, leaving only crimes of
passion, unplanned and hence not preventable. By 2008 almost every-
body—policy makers, politicians, bankers, journalists, pundits, and aca-
demics—thought the problem of financial crises, at least in developed
countries, had been solved. We now know better, and the financial au-
thorities have a firm mandate from the highest levels of the political
leadership: “Do something about finance. We will give you the resources
and powers and political support to make it happen.” In response, we got
168

THE RISK THEATER 169
a new breed of PreCrime—macropru—financial regulators put in charge

of macroprudential policy, tasked with preventing premeditated miscon-
duct before it all goes wrong.
A vast agenda of regulation has been created. Leading the way is the
G20, the forum for the governments and central banks of the twenty
largest economies. The G20 delegates the job of dealing with finance
to the likes of the Financial Stability Board and the Basel committee.
Almost every country has set up its own financial stability board and put
its central bank in charge of macropru. It is still early days, but a clear
picture is emerging of the next generation financial policy. We have new
authorities and older ones with a bigger role. They have more data, more
resources, and more power. The financial sector retains a formidable lob-
bying power but has only partially succeeded in resisting the push for
regulations. So, how well does it work? Some parts are quite successful,
others a waste of resources, and others are outright damaging. What we
have gotten is a lot of risk theater. I take the term “risk theater” from the
notion of the security theater, coined by Bruce Schneier, referring to the
practice of implementing measures supposed to improve security while in
practice doing little to achieve it. A highly visible demonstration of secu-
rity: checkpoints, guards, cameras, the banning of products. Restricting
people’s behavior, privacy, and liberty. But when scarce resources priori-
tize visible security, actual security is deemphasized.
Turning Banks into Volvos Will Not Help

The safest carmaker in the world is Volvo. It was the inventor of
the seat belt, which it made freely available to the world in the 1950s. The
Volvo XC90 was crowned the safest car in the United Kingdom: no driver
or passenger deaths occurred since its launch in 2002. Volvo’s objective is
zero fatalities. Ironically, a woman named Elaine Herzberg was killed by a
self-driving Volvo XC90 in 2018, but it wasn’t the Volvo’s fault. Uber, the
company testing the self-driving XC90, had disabled the car’s automatic
self-breaking. The XC90 knew it was supposed to break, but Uber’s soft-
ware didn’t allow it to.
Suppose the financial authorities manage to make every financial in-
stitution prudently run—as safe as Volvos. Nobody makes crazy invest-
ments. Everybody follows the rules and measures risk correctly—all are

170 THE RISK THEATER
prudent. Surely, then, no bank will fail, and financial crises will not hap-
pen, just as there would be very few car crashes if everybody followed the
speed limit, nobody texted while driving, nobody drove drunk, every-
body obeyed all the traffic laws, and we all had Volvos.
Can regulations prevent financial crises? The evidence from the years
after World War II, the Bretton Woods era of 1944 to 1972, superficially
suggests yes. The financial system was then very heavily regulated, and al-
though banks failed, only two banking crises are recorded anywhere in the
world over those twenty-eight years. But the financial system was almost
entirely national, with little cross-border banking and very expensive finan-
cial services. Perhaps suitable for its era, but not the twenty-first century.
The world changed. After the collapse of Bretton Woods in 1972 and the
emergence of the Washington Consensus, banking became global. Regu-
lations followed. But the new regulations were different from the Bretton
Woods ones. The focus was no longer on tight controls; instead, we aimed
to control risk. Making sure the banks were prudently run—Volvos.
It was all fine for a while, but then in 2008 banks started to fail for
completely unexpected reasons. That should not have happened since
Volvos are safe. The problem was that the prevailing regulatory thinking
focused on each bank’s behavior in isolation. What the authorities missed
was the importance of the system. A financial system composed entirely
of prudent banks is inherently unstable. We now know there were plenty
of hitherto unknown risks, especially liquidity risk, so fundamental to the
2008 crisis. The system is not merely the sum of the individual banks.
But financial regulations still mostly focus on each financial institution in
isolation, using the riskometer as their main tool.
An interesting case study of what this means in practice happened on
Thursday morning, 15 January 2015. That day Switzerland’s central bank,
the Swiss National Bank (SNB) did something that was either quite ex-
traordinary or inevitable, depending on how one thinks about risk. The
Swiss decided to allow their currency, the franc, to float freely. For some
years before that, the SNB had pegged the franc to the euro at 1.2 francs.
Immediately after the franc was set free, it strengthened by 16 percent.
The SNB decision really should not have been that surprising. Those
who followed the Swiss economy and local politics—especially those who
knew who owns the Swiss National Bank—expected the exchange rate

peg to break. It was not a matter of if but when the franc would be set
free. The roots of the announcement grew when Switzerland’s economy
began outperforming its eurozone neighbors. Before the global crisis
started in 2007 the euro was a success, a stable currency investors wanted.
The crisis put paid to those thoughts: investors holding liquid funds and
seeking safety began looking for new destinations. But there are not that
many stable, well-run countries in the world. Switzerland is one of the
few. Unsurprisingly, money poured into Switzerland, and, as in any mar-
ket, when there are more buyers than sellers prices go up.
The first time the franc and euro traded in January 1999 the exchange
rate was 1.6 francs to the euro, and it was almost in the same place when
the global crisis started in June 2007. From that day on, the euro steadily
fell, reaching a low of 1.05 francs in August 2011. Though nice for Swiss
consumers, the fall was not all that good for Swiss exporters, and the SNB
did what so many other central banks like to do in that situation: fix the
exchange rate, in their case at 1.2.
It is easy, at least in principle, to prevent a currency from rising. Just
print money and use it to buy foreign currency on the open market. The
problem is that when Switzerland prints money to buy euros, it is infla-
tionary since the money supply is expanding. To prevent inflation, central
banks like to counteract the inflationary forces by sterilization, soaking
up the freshly printed money by selling bonds. That works because if
the central bank buys one thousand euros’ worth of foreign currency by
printing twelve hundred francs and then sells a bond worth twelve hun-
dred francs, the amount of money in circulation does not change. So far,
so good. But the central bank eventually runs out of bonds to sell and
can no longer sterilize. Meanwhile, it is piling up paper losses. Not only
does it forgo interest payments on the bonds it sells, but also if the cur-
rency eventually floats, the central bank will incur a substantial loss on its
foreign currency holdings.
Central banks are rarely structured as corporations, but the SNB is:
60 percent of its shares are owned by public institutions, like regional
governments—the cantons—while the remaining 40 percent trade on
the open market. The cantons depend on dividends from SNB, so they
are naturally unhappy about the central bank running a big loss-making
position. So, even though the SNB supposedly operates independently of

its shareholders, that policy is not true in practice. The SNB has to meet
two objectives simultaneously: make money for its shareholders while
also helping the economy. A currency peg makes those two objectives
mutually exclusive. On the last day before the announcement, the SNB
held 471 billion francs’ worth of foreign currency. The size of the Swiss
economy, its GDP, was 565 billion francs. The SNB lost 78 billion francs
on 16 January. If it had waited even longer to abandon the peg, the even-
tual losses would have been higher. What does any of this have to do with
Basel or the risk theater? Suppose we had gone to bed on 14 January 2015
and used the main Basel-approved riskometers to forecast the likelihood
of a 16 percent appreciation of the Swiss franc (Table 1).1
The first two riskometers, EWMA and GARCH, found the likelihood
of that happening so low that it could not be calculated. For the third,
MA, the likelihood was smaller than once in every universe. For the fourth
riskometer, t-GARCH, it was once every 14 million years, and for the
fifth, EVT, once every 109 years. The riskometers were quite inconsistent.
But then, did these five riskometers pick up on what had happened and
forecast risk better after the appreciation? For one, t-GARCH, the risk of
another 16 percent appreciation increased 104 times, while for another
(EVT) it went up only by 14 percent. Which of these two is more likely
to be correct? Neither. After the currency had appreciated by 16 percent,
it was implausible it would see another appreciation of the same magni-
tude. It is like a dam bursting: we don’t expect it to burst again for some
time. The risk went down, not up, as the riskometers had it. It is not
like I specially picked lousy riskometers; on the contrary, these five are
state-of-the-art, recommended by the authorities, and used by industry.
Table 1. Riskometers (VaR) and the likelihood of the Swiss currency appreciation
Riskometer Frequency in years Context
EWMA Never Too low to be calculated by a computer

GARCH Never Too low to be calculated by a computer
MA 2.7 × 10217 Age of the universe is 1.4 × 1010 years
t-GARCH 1.4 × 107 Age of the earth is 4.5 × 109 years
EVT 109 A bit less than one crisis in a typical lifetime
Source: Author’s calculations

Two, t-GARCH and EVT, are especially lauded for their ability to pick
up tail risk.
When I was drafting my blog on the riskometers and the Swiss an-
nouncement, I sent a copy to Robert Macrae, a friend who ran a hedge
fund, and asked him to comment on it. He responded by saying that my
findings were not all that interesting because to a currency trader us-
ing such riskometers to determine risk and nothing else was naive. Any
market participant would use a hybrid approach, combining a riskometer
with a detailed study of the Swiss economy. A comment I have heard many
times: It is not fair to criticize riskometers merely because they failed to
predict the likelihood of the Swiss currency appreciation. However, Rob-
ert ran his own fund, and the authorities had no say in how he managed
risk, so he could pick any riskometer he wanted. A regulated bank does
not have that luxury. The authorities push banks to use a riskometer and
not just not any riskometer: banks can choose only among a few, and as
time passes they are increasingly told precisely how to measure risk.
I have presented the Swiss currency appreciation case many times to
audiences in both the private and public sectors. Those in the private
sector tend to react by saying something like, “So what? We know this
already.” In other words, they know that the risk measurements used to
regulate banks are very unreliable. I find it especially amusing to hear
that from the very risk managers whose job it is to report on Basel to the
authorities. They have to use the officially sanctioned riskometers, even
though they have no faith in them.
The reaction from regulators is more interesting. They often respond
by saying that while the example holds true, it is irrelevant. The riskome-
ters in Basel III are designed for microprudential regulations, the man-
agement of day-to-day risk, not for extreme outcomes, so the Swiss cur-
rency appreciation is not a fair test of them. The regulators are expressing
a nuance that is generally lost on the rest of the world. After all, there is
more to microprudential regulations than day-to-day volatility; measur-
ing the likelihood of banks failing is key to their mission, and their pre-
ferred riskometers don’t capture such risk.
The use of riskometers in regulations can have curious consequences.
The BBC Panorama program came out with a provocative episode in
2014 titled “Did the Bank Wreck My Business?” about how two banks

bailed out by the government, RBS and Lloyds, were unfairly destroying
small companies. What the BBC missed was why the banks behaved this
way. They were merely doing what they were supposed to do under the
post-2008-crisis Basel regulations—derisk. The risk theater is to blame
for the destruction of the small companies, not the capricious bankers, as
Panorama argued.
The reason is because the global crisis in 2008 showed that the global
financial regulations in place at the time, Basel II, were not up to the task.
It was just too easy for banks to manipulate their capital—exploit capi-
tal structure arbitrage. The Basel committee wasted little time in com-
ing up with its successor, Basel III, mostly in place by now. It happened
with lightning speed, taking only a decade—yes, ten years is super fast
by international regulation standards. That haste meant the committee
did not have time to make any fundamental reforms. Basel III is only
an incremental improvement on its predecessor, chipping away some of
the most glaring deficiencies while leaving the underlying philosophy in
place—Basel II on steroids, if you will. There is much talk about “Basel
IV” addressing all the difficult issues, but I suspect regulatory fatigue may
get in the way.2 Basel III remains microprudential for most parts, leaving
the system’s stability to the newly established macroprudential regulators.
The primary focus of Basel III, like that of its predecessors, is capital.
The 2008 crisis showed that the way capital was calculated was problem-
atic because many banks with supposedly high levels of capital failed.
Basel III’s main thrust is in increasing both the amount and the quality
of banks’ minimum capital: the reason the banks discussed in the BBC
Panorama program had supposedly been wrecking businesses. Because
the banks have to hold much more capital for the riskiest types of loans,
the inevitable consequence is that the availability of loans to small- and
medium-sized enterprises sharply decreased while the interest rates on
those loans increased significantly.
The regulators are proud of their creation. Banks now hold more
capital, have more stable funding and robust systems for managing risk,
promising a more resilient banking system, one that stimulates economic
growth. All these reforms left the banks in much better shape for dealing
with the Covid-19 crisis in 2020. The banks complain, but they complain
surprisingly little and are much happier with Basel III than is often as-

sumed. Compared with the very loud protests made by insurance compa-
nies and asset managers over being designated as systemically important,
the banks appear to be outright acquiescent. In and by itself, worrying.
There is much to like in Basel III. Still, I have reservations. To begin
with, Basel III still assumes the stability of the financial system is ensured
if each bank is a prudent Volvo. Furthermore, Basel III not only fails to
address procyclicality adequately, but also makes the problem worse. Al-
right, the standard response is “Basel III has a new type of capital buffer
that is adjusted countercyclically according to the financial cycle.” Not
quite, as the adjustment buffers are small, and a temporary relaxation
may not encourage them to lend more. The Covid-19 lowering of the
capital ratios did not stimulate lending. More fundamentally, Basel III
implements its predecessor more intensively but does not ask the critical
question, What do we need from financial regulations? Yes, it does ad-
dress things like financial stability and the provision of services. Still, it
does not position itself properly within the context of what we want from
the financial system and its regulations. So, yet again, risk theater. Perhaps
worst of all, it mostly ignores the SIFIs.
The SIFIs are those systemically important financial institutions whose
failure will cause a systemic crisis. Before 2008 nobody worried about the
SIFIs; on the contrary, the prevailing wisdom maintained they were the
safest of institutions. Globally diversified financial institutions that could
offset losses in one area with profits in all the others. Well, that turned out
to be wrong. The largest banks did fail. Worse still, they failed with cata-
strophic consequences, like the world’s biggest bank at the time, RBS-
ABN AMRO, not to mention Lehman Brothers.
One might ask why we allow financial institutions that are so dan-
gerous. The answer is simple: politics. Even if the financial authorities
preferred no SIFIs, their political masters overrule them. In other words,
SIFIs exist only because politicians want them to exist. Why do politi-
cians like SIFIs? The straightforward reason is that they are quite handy.
A single financial institution that can provide financial services wherever
your country’s companies operate is beneficial. If a German company
based in Düsseldorf has businesses in America, Brazil, and Korea, it helps
to have a single bank that can service it in all these places. The other
reason SIFIs exist is prestige. If your bank has your country’s name on

it—think Deutsche Bank—of course you want your bank to be big and
powerful. Besides, having global banks in your jurisdiction allows you to
project power. Austria gains power and prestige by having the regional
SIFIs in its jurisdiction. The reason there are so many SIFIs in insecure
midsize European countries (Figure 30).
So, how dangerous are the SIFIs? To answer that, I need to go back to
bank capital. The numbers reveal that the SIFI problem is much worse for
smaller countries (see Figure 30). The total assets of the largest American
bank, JP Morgan, are only 13 percent of the GDP of the United States,
while the total assets of the Swiss UBS bank exceed its GDP at 136 per-
cent. One can use many ways to look at how vulnerable these SIFI banks
are. The most obvious is the leverage ratio, the ratio of capital to total
bank assets. The lowest ratio is Deutsche Bank’s at 4 percent, while JP
Morgan’s exceeds 6 percent, and the Chinese ICBC’s is almost 8 percent.
No wonder Deutsche is always in the news, and not in a good way. It is
the SIFI bank the pundits think will fail next.
The Basel III financial regulations say that a bank has to have a leverage
ratio of at least 3 percent. So, by subtracting three from the leverage ratio
Figure 30. SIFI size and GDP. Credit: Copyright © Ricardo Galvão.

Figure 31. Distance to default. Credit: Lukas Bischoff/IllustrationX.
and multiplying it into total assets, we can see how big a loss the bank
would have to suffer in order to be forced into bankruptcy—the distance
from default (Figure 31). It would take a loss of only $14 billion for the
Royal Bank of Canada to be shut down and $15 billion for the Dutch
ING bank. Meanwhile, it would take $89 billion for JP Morgan and over
$200 billion for the Chinese ICBC bank.
The failure of any bank whose assets are a significant portion of GDP
will be tough to resolve. Of these banks, the United States could easily
absorb even the largest losses arising from JP Morgan’s failure, and China
would not find it too hard to handle ICBC’s default. However, it would
be challenging for other countries, especially Switzerland, the Nether-
lands, and Spain, where the largest bank’s assets exceed GDP. It is much
better to try to prevent failure by containing the size of the largest banks.
PreCrime Macropru to the Rescue

We are all macroprudentialists now.
—Claudio Borio
While the Basel Accords are microprudential, focused on each

bank’s behavior individually, systemic risk is left to the macroprudential

(macropru) authorities: the PreCrime police of the finance world. While

the label macropru is new, it has always been one of the four central planks
of the governments’ financial policy, now just sporting a new name.
I suspect that Walter Bagehot and his nineteenth- and early twentieth-
century counterparts would not have been the least surprised by the crisis
in 2008. It was not so different from the crises they knew, like those in
1763, 1866, 1907, and 1914. What would have shocked them was how com-
placent and unprepared their twenty-first-century successors had become.
Just as many policy makers in developing countries, like those affected by
the Asian crisis of 1998 and the various Latin American crises, watched
events unfold in 2008 with a wry weariness, if not schadenfreude. After
being lectured by the developed countries for so many years, the shoe
was now on the other foot. Still, to their credit the financial authorities
quickly dusted off the history books in 2008, responding appropriately
by applying past lessons. What is especially impressive is that the policy
mistakes that led the financial turmoil in 1929 to become the Great De-
pression were deliberately avoided. The necessary liquidity was applied,
trade was not restricted, and the banks kept alive. But not without a cost.
Never again. The cost of responding to the 2008 crisis was so high that
something had to be done to make the financial system more resilient. In
response, we got the doctrine of macropru. The central banks set up fi-
nancial stability divisions. An army of people, a vast amount of resources,
and considerable powers have been given to the PreCrime macropru reg-
ulators. It quickly became a truism that macropru was the way forward, as
Claudio Borio’s quote above indicates.
So what is the objective of macropru? There is no single definition.
That said, distilling the various notions out there, I find two. Prevent
excessive risk from accumulating in the financial system, and if and when
a crisis happens, contain it in the most efficient manner possible. While
these objectives are sensible, if not complete, they are high level and need
to be translated into specific policies. And that is precisely where there
is no consensus: not between countries, not inside countries, not even
within particular policy organizations. I find it amusing that a central
bank can provide different and conflicting views on macropru without
seemingly being concerned with the inconsistency or, worse still, not even
realizing it. It is difficult to create effective macropru, and we are still find-

ing our way. I often hear words to the effect that macropru at present is
how monetary policy was in 1950—give it fifty years and it will become
as sophisticated. One can do a lot of damage in fifty years, and those who
know their monetary history will recall that the stagflation of the 1970s
was in no small part due to the poor monetary policy of the era. The fail-
ure to hit inflation targets since 2008 shows that monetary policy is still
far from perfect.
There are many directions the macropru authorities can take. Most
are passive, focusing on fixed rules that hold through the financial cycle.
Many such macropru implementations have been very successful in pre-
venting crises—my favorite is the ban on buying stocks on margin in
the United States in 1934. (Should I be worried about having a favorite
macropru rule?) The alternative is active macropru, leaning against the
wind in a discretionary manner. If risk is building up, tighten capital and
liquidity standards. Relax when risk and growth are low, as we did follow-
ing the Covid-19 crisis. If the market is too risk averse, follow Keynes and
encourage it to take more risk. Such discretionary macropru policies are
designed to be countercyclical, dampening out the financial cycle. Active
macropru demands much more of the financial authorities than passive
macropru. They need estimates of systemic risk and its impact on the real
economy, from the early signs of a buildup of stress all the way to the post-
crisis economic and financial resolution. They need tools to implement
effective policy remedies in response to changes in risk. And the authori-
ties need legitimacy, a reputation for impartiality, and political support.
To start with, systemic risk needs to be measured, not an easy job.
There are many indicators of systemic risk out there, such as the Euro-
pean Central Bank’s CISS. Most are prone to what statisticians call type I
and type II errors: falsely finding something that is not there or failing to
detect something that is. Not a problem with an easy solution. Systemic
crises are infrequent, less than once every forty-three years on average,
giving the statisticians little to work with. To complicate matters, the
financial system’s structure will be very different from one crisis to the
next. Having identified that systemic risk is increasing, the authorities
have to respond with one of the tools at their disposal. There are many
tools in the macropru tool kit. Some are meant to limit procyclicality—
lean against the wind—some shield the real economy, yet others are used

to respond to imminent threats. Some tools are surgical and specific, like
loan-to-value ratios aiming to limit the amount of money people can bor-
row when buying a house. Others are blunt—sledgehammers—like bank
capital ratios which affect all bank activities. The blunt instruments may
kill the patient, while the surgical ones may not work.
Frustrating the job of the macropru designers is the continuing evolu-
tion of the financial system. The past informs the tools, but the threats
come from the future, like driving by looking into the rearview mirror.
Meanwhile, the impacts and side effects of the tools are poorly under-
stood. The most visible macropru problem, and politically the most im-
portant, is real estate, a common cause of financial crises. The macropru
policy makers are always on the lookout for real estate bubbles. Simple
things, bubbles. We borrow from banks to buy homes, and in response
prices go up and the economy blossoms, encouraging more people to
borrow to buy homes. Everyone feels happy in the short run, but over
time fault lines emerge and a crash becomes increasingly likely. Both
the bubble itself and the eventual crash create problems. Rising housing
prices directly affect inequality. Homeowners get richer, and the rest are
left out, with political consequences. Governments can be forced to im-
plement policies that further stimulate housing prices—like the various
policies helping first-time buyers and high-risk borrowers. The political
desire to help poor households acquire property in the United States was
the main reason for the emergence of the subprime mortgage market, a
leading cause of the crisis in 2008. Real estate is one of the most common
causes of banking crises. Elementary. Has to be solved, right?
Unsurprisingly, the macropru authorities have identified real estate as a
significant priority. So what are they to do? One of the main tools in use
today is the loan-to-value ratio, whereby one can borrow only a certain
percentage (say, 80 percent) of the value of a house. However, while real
estate is undoubtedly a macropru concern, the remedies deal only with the
symptoms, not the causes. House prices are directly affected by economic
growth and various government policies, like zoning laws, help-to-buy,
tax deductible mortgage interest, ultralow interest rates, and subsidized
mortgages for high-risk borrowers. Macropru has no impact on any of
those, and all the macropru authority can do is to mop up after the other
policy domains. Meanwhile, just using the macropru tools will expose the

authority to considerable public hostility. Worse still, if the said authority

is the central bank, being responsible for macropru may undermine the
central bank’s ability to effectively execute monetary policy. Real estate
shows how difficult macropru is. Other parts of the government are re-
sponsible for causing the problem, and all the macropru authorities can
do is clean up. However, the worst challenge for them is politics.
Covid-19, and how the authorities responded to it, also demonstrates
the procyclicality of macropru. The authorities decided in March and
April 2020 to relax capital constraints on banks in the hope that they
would lend to Covid-stricken small- and medium-sized enterprises. But
banks by and large did not respond the way the authorities wanted them
to. They didn’t want to lend, both because they saw these companies as
bad credit but also out of precaution, not knowing how the Covid eco-
nomic crisis would end. But these capital relaxations remain in place. And
if the Covid economic crisis is V-shaped or K-shaped, the relaxation of
capital may finally be used by banks to stimulate the economy at a time
it is already overstimulated. Yet another example of procyclical macropru.
Politics Gets in the Way

While the hope is that the credibility of monetary policy rubs off on
macroprudential policy, it is more likely that the fuzziness of financial
stability and the interplay of political pressures will instead undermine
monetary policy.
—Jeff Chwieroth and Jón Daníelsson
The most powerful bureaucrat in the world is the chair of the

Federal Reserve System in the United States, currently Jerome Powell.
He has more power than General Joseph Dunford, the head of the Joint
Chiefs of Staff, even though the latter has nuclear weapons in his tool kit.
The reason is that General Dunford reports to the president, while Chair-
man Powell reports to nobody. Paul Tucker, the former deputy governor
of the Bank of England, in his book Unelected Power, questioned why a
democratic society gives a bureaucrat like Powell such remarkable powers.
The reason is monetary policy, whereby the wisdom of independent
central banks is backed up by decades of research, including the work of
multiple Nobel Prize winners. Politicians cannot be trusted with interest

rates—for that, we need an independent professional body. In the bad

old days, when politicians ran monetary policy, they used interest rates to
enhance their electoral prospects. Reduce interest rates a few months be-
fore elections to stimulate the economy. Some countries still do this, like
Turkey. Politicians don’t want to delegate their power to the central bank:
witness President Trump’s criticism of the Fed and President Erdogan’s
repeated firings of Turkey’s central bank governors.
It is easy to hand monetary policy to an independent central bank be-
cause the policy is well defined. There is one unambiguous measurement
(inflation), the objective is clear (perhaps 2 percent inflation), and there
are two tools (the price and quantity of money). If the central bank fails
in its job, it is clear for all to see. It is acceptable to have an independent
central bank if its actions and outcomes are clearly monitored.
Some countries and central banks took independence and monetary
policy to an extreme. The importance of the mission implied purity, so
financial regulations and financial stability belonged elsewhere. The key
reason the United Kingdom created the Financial Services Authority
in 1997 and why many countries followed suit. Now the pendulum has
swung the other way, and we want to house financial stability in the cen-
tral bank. For two reasons. First, the hope is that the reputation and the
power of the central banks as guardians of money will rub off on financial
stability. Second, the central bank is the only institution that can create
money on demand and therefore has to be at the center of fighting finan-
cial crises.
Will it work as hoped? No. The central bank in charge of financial sta-
bility faces a complex, ill-defined policy domain for which there is no clear
consensus on either the problem or the objective. The indicators at their
disposal are imprecise and conflicting. The surgical tools are ineffective,
and the powerful tools blunt. Worse, and even more so than monetary
policy, macroprudential policies result in clearly identifiable winners and
losers, subjecting the policy authorities to intensive lobbying and political
pressure. The ultimate decisions of anything that affects economic policy
in a direct and personal way should fall under political oversight. It would
be undemocratic if it did not.
Nowhere has the conflict between politics and independent regula-
tions become more evident than in macropru’s real estate policy. In 2013

Norwegian Parliament members threatened to withdraw the regulatory

agency’s ability to determine the loan-to-value ratio when the regulator
planned to lower it. Similar debates have happened in many countries.
The reason is apparent. If we cannot borrow to buy houses, we do not
enjoy the benefit of wealth creation and security that comes with own-
ing a property. The policy creates clearly identifiable losers. This is why
less democratic countries like Malaysia, Singapore, Thailand, and Hong
Kong find it easier to implement macroprudential policies. One example
is Hong Kong’s 60 percent loan-to-value ratio limits for first-time buyers
and a maximum debt-service ratio of 40 percent for borrowers acquir-
ing a second property. Singapore is another excellent example of what
is possible when democracy does not get too much in the way, as it has
implemented multiple rounds of real estate interventions and can do so
without too many political considerations. It would be unthinkable for
European countries and especially the United States to deploy such in-
trusive macropru tools. Democratic opposition and industry lobbying
would get in the way.
Indeed, the biggest problem for the macropru policy makers is that
they have to ignore political risk, as I discussed in a blog piece cowritten
with Robert Macrae, “The Fatal Flaw in Macropru: It Ignores Political
Risk.” Very few major stress events are caused purely by excessive risk-
taking, the target of macropru. Most have politics as a primary driver. War
or the transition between political systems, as in Russia in 1919, Germany
in 1923, Japan in 1945, and China in 1949. Houses for the poor, as in the
United States in 2008. Covid-19 in 2020. Today, politics is the sole cause
of the severe financial and economic crises in Venezuela and Zimbabwe.
Take Brexit. During the referendum campaign the Bank of England
warned of serious economic consequences from Brexit, putting itself on
the losing side of an acrimonious political debate. As a result, the bank
came under repeated attacks from the new political leadership—the bank
has had to affirm its independence and request support from the new
government in a way we are not used to seeing. The new governor of
the bank now loudly proclaims his support for the government’s Brexit
policies. The post-Brexit politicization of the Bank of England highlights
the dangers facing central banks when they include politics in their con-
siderations. Politics drive systemic risk. How deeply can or should a civil

servant working in the European Central Bank venture into populism?

While the election of President Trump heralded a dramatic change in US
economic priorities, it was hard to see how the Fed could react.
In practice, and despite whatever may be said about their indepen-
dence, the financial authorities are authorized by, controlled by, and gain
their legitimacy from the political leadership. Unsurprisingly, the man-
date from political leadership stipulates that the central banks look at
financial and economic risk, not the risk emanating from the politicians
themselves. Making it risky for the financial authorities to incorporate
political risk as a determinant of systemic risk, despite its importance.
As a result, political risk is mostly missing from the macroprudential
debate despite having always been the primary cause of systemic risk. It
is not only institutionally challenging for the financial authorities to pub-
licly anticipate crises with predominantly political causes, but also makes
it difficult for them to contain such crises once they happen. That is why I
think those who hope the quality of monetary policy will help macropru
are misguided. Independent monetary policy is possible only because of
the clarity of the mission. The messiness and politicization of macropru
are much more likely to undermine the central banks’ legitimacy and
their reputation for impartiality, making it much harder to mobilize sup-
port for difficult policies.
Of Hammers and Nails

A few years ago I was invited to a conference at a European insur-
ance regulator to discuss the impact of risk measurements on the efficacy
of insurance regulations. The reason for the invitation is not that I’m an
expert in insurance regulations, which I am not, but rather because a cou-
ple of years earlier I had coauthored three blog pieces on the European
Solvency 2 regulations, focused on the systemic risk arising from valuing
the reserves of insurance companies at market values.3 The subtext to the
event was that the Financial Stability Board (FSB), a policy body work-
ing at the behest of the G20 group of countries, was worried about the
systemic consequences of insurance companies and asset managers. The
reason it was so concerned was the pivotal role of the insurance company
AIG in the global crisis in 2008, when AIG was the world’s most active

seller of CDSs via its London-based and France-regulated bank, the risk
of which was missed by its New York State regulators.
The financial authorities now focus on the systemic threats emanating
from insurance companies and asset managers, institutions that control
trillions of dollars of assets with relatively little oversight and, as AIG
showed, can make business decisions that threaten global financial stabil-
ity. Both industries are fragmented, with nothing akin to globally coor-
dinated banking regulations, and, until 2008, nobody thought either in-
dustry was systemic.4 While insurance companies and asset managers may
pose systemic threats, the nature of the threat is poorly understood, and
until very recently there had been no studies of how these two industries
interact with the rest of the financial system and the threats that poses to
global stability. We have studied the systemic fragility of banks for over a
century. There is no comparable research on insurance companies or asset
managers. When all you have is a hammer, everything looks like a nail,
and when the FSB started looking at the systemic risk in the nonbanking
parts of the financial system, all they had was analysis of banking stability.
The type of work I have discussed extensively in this book: bank runs, fire
sales, liquidity dry-ups, and, most important, capital.
That was the reason I was invited to the conference at the insurance
regulator. The FSB opted to analyze the fragility of insurance compa-
nies and asset managers by looking through the lens of banking fragility,
which meant capital. But their risks are quite different. What kills insur-
ance companies is writing insurance contracts too cheaply so they can’t
meet their eventual obligations. There is nothing systemic about that
because most insurance payouts are uncorrelated with systemic financial
risk. The AIG crisis happened because it decided to become a bank, and
the relevant authorities did not pay attention. Asset managers use neither
much leverage nor derivatives. They are systemic consequences of asset
managers, but they are in exchange traded funds with illiquid assets like
small company bonds. Capital has nothing to do with that.
So, when applying bank fragility analysis to asset managers and insur-
ance companies, the financial authorities just end up saddling them with
unnecessary burdens which the clients—us—pay. We might get the per-
ception of safety but not actual safety: risk theater. The more funda-
mental danger is that if we increasingly harmonize regulations around a

Basel-type philosophy, including not only banks but also other activities
in the financial system, like insurance, asset management, and nonbank
banking, we get monoculture.
I recently gave a presentation at a central bank conference and made a
throwaway remark that macropru could be procyclical. That is, it could
perversely amplify the financial cycle instead of dampening it. To the au-
dience, that was heresy, and some senior staff members got cross with
me. After all, the fundamental promise of macropru is that it is counter-
cyclical, dampening out the natural cycles in the financial system. So,
could macropru be procyclical? Well, yes. I eventually cowrote a blog on
this, “Why Macropru Can End Up Being Procyclical,” arguing that dis-
cretionary macropru—leaning against the wind—has considerable scope
for amplifying the financial cycle. Suppose the macropru authorities were
successful in smoothing out the financial cycle. Would market participants
respond to this gratefully and say, “What a great job the central bank is
doing?” No, the market would see the resulting low risk as an invitation
to take more risk: the Minsky effect. We have seen many examples in the
past, like the Greenspan put.
Another reason macropru may be procyclical is the difficulty in measur-
ing risk—riskometers aren’t exactly reliable. Figure 32 shows a hypotheti-
cal time path of risk over one year. The target risk is three. In the first
month the risk is too high at five, but nobody realizes it yet. A couple of
months later the riskometers pick up on risk having been high, alerting
the authorities, who start planning their reaction and, a few months later,
decide on what to do. Eventually, in month twelve, the policy response
is implemented. Meanwhile, risk has been steadily falling and is already
below the target by the middle of the year. By the time the policy inter-
vention takes place in month twelve, risk is too low. Instead of bursting a
bubble, it exacerbates the derisking already taking place, and risk crashes
to one, way below the target level of three. A problem caused by reacting
with a time lag to indicators of risk that are themselves measured with a
time lag: the policy response can come too late and be procyclical.
By the end of the year the appropriate policy response would have been
to increase risk to stimulate economic activity, not to decrease risk. Such
a procyclical policy response is common. A recent example was Japan in

Figure 32. Procyclical macropru. Credit: Lukas Bischoff/IllustrationX.
2007, when the authorities issued guidance restricting bank lending to

real estate developers just when foreign lenders were also withdrawing
from this market, thereby leading to a severe credit crunch. And there is
a real danger that all the Covid-19 stimulus may simply end up inflating
the postvirus bubble. Political pressures and lobbying further amplify the
policy procyclicality. We sometimes hear complaints like “The banks are
failing because they have already overextended credit” or “Helping Wall
Street to increase lending leads to even bigger moral hazard” or “Macro-
pru is discredited because it was supposed to have prevented this credit
event in the first place—why should it do better this time?” All of these
objections call for a procyclical policy response.
I don’t think the potential for procyclical macropru would have sur-
prised Friedrich Hayek. His argument against central planning was that
economic information is created locally—by the farmer, the mechanic,
and all the companies making up the economy. As decisions are taken on
higher and higher levels, the decision makers necessarily have to aggre-
gate information. Unfortunately, when we take an almost infinite amount
of local knowledge and summarize it in a few key macropru indicators,
quite a lot is lost along the way. It is much better for the farmer than the
ministry of agriculture to decide what to grow and when. The macropru

authorities wield enormous power, just like the planning ministries of the
Soviet Union, but may not be any better at getting the result they desire.
Mark Carney’s confident proclamation (see chapter epigraph above)

that the problem of crises has been fixed sounds promising but hard to
verify in practice. Financial crises are not frequent. The typical OECD
country can expect such crises less than once every forty-three years.
Mark Carney will be long retired before we can validate his statement.
I am not as optimistic. There are many good things in the new financial
regulations, such as the dual requirement for the leverage- and the risk-
weighted capital ratios and the emphasis on the type of capital that is
hardest to manipulate. Capital was too low before the 2008 crisis, and
the banks depended too much on short-term funding. However, some
aspects of the new financial regulations are not useful and even down-
right dangerous, like the continuing focus on individual prudence and
dependence on riskometers; antidiversity and the neglect of the system
are the worst. Central bank research has given us so much evidence on the
importance of diversity and the system for financial stability. It is unfor-
tunate that so little of that work is reflected in actual policy. The overall
thrust of financial policy is moving in the wrong direction, with too much
risk theater. I would much rather the focus of financial regulations be on
diversity.

11
The Uniformity, Efficiency, and
Stability Trilemma
Common beliefs and action are the enemy of financial stability. The best
way to get stability is diversity, both in regulations and financial institutions.
The most popular banana in the nineteenth century was the Gros
Michel. It tasted good, came in big bunches, and had a bruise-resistant
peel. Ideal for export. But the Gros Michel banana had a vulnerabil-
ity. Each banana was a clone of every other, so they were all genetically
identical. It came as no surprise when the fungus Fusarium oxysporum,
189

190 UNIFORMITY, EFFICIENCY, AND STABILITY TRILEMMA
nicknamed the Panama disease because of the extensive banana planta-

tions in Panama, came calling at the end of the nineteenth century. It
managed to wipe out almost every Gros Michel. Evolution did not pro-
vide resistance because the Gros Michels were clones. If one banana was
vulnerable, so was each and every other one. Genetic diversity protects.
When the Spanish flu hit in 1918 tens of millions of people died, but our
genetic variety protected us. If humans had all been clones, we could have
become extinct.
Bananas show monoculture at its best and worst. Monoculture reduces
uncertainty so growers can optimize their plantations to achieve maxi-
mum efficiency. The trade-off is stability, in this case the vulnerability
to disease. When choosing among stability, efficiency, and uniformity,
we can pick only two (Figure 33). What does any of this have to do with
finance? After all, the financial system is not full of clones, and banks are
not bananas, even though I do get the impression that monkeys run some
of them. The forces of creative destruction are as powerful as ever, with
plenty of new ideas and new businesses pushing diversity, not uniformity.
Unfortunately, strong forces oppose diversity. Decision makers tend to
opt for efficiency and uniformity, rejecting stability. Of course, that is
not how they would put it, and I am sure many will be offended by my
words here. All say they want stability, adding that they prefer diversity,
not uniformity. Those are words. Actions say otherwise.
The consequence of choosing uniformity and efficiency is lower volatil-
ity and fatter tails, the successful smoothing out of day-to-day price fluc-
tuations at the expense of large future crashes—tail risk (Figure 34). The
light gray line is the natural fluctuations of some hypothetical returns.
The fat black one is where we successfully squeeze out short-term risk but
then, by 2028, end up with a crisis.
The private institutions in the financial system are becoming more
similar every day. The reasons are subtle. Not many call for less diver-
sity, at least not explicitly; the antidiversity preference instead tends to
show up indirectly. As when calling for best practices. It is hard to oppose
best practices. After all, why would anyone ever want to have second-best
practice? But there can only be one best practice, so if all pick the best
practice, they all have the same practice: monoculture. Fairness and level
playing fields are also drivers of uniformity, as they demand the same

UNIFORMITY, EFFICIENCY, AND STABILITY TRILEMMA 191
Figure 34. Lowering volatility and fattening the tails. Credit: Lukas
rules for everybody. And if the rules are about specific conduct, like all
the micro risks the microprudential regulators concern themselves with,
we get harmonization of action. The financial industry also pushes for
monoculture. Almost every financial institution has the same objective:
maximizing profits subject to constraints. That takes them closer to other
institutions’ optimal asset mix, leading to more crowded trades and
procyclicality.
While there are not many solid measurements of diversity, one metric
points in that direction: the number of banks is steadily falling. Half a
century ago we had a wide variety of financial institutions. Every coun-
try had its own financial regulations, and they were all different. Each
country had many types of banks, regulated and operated differently, all
driving diversity and stability. The most diverse of all was the United
States. It used to leave banking regulations to the states, and, of course,
they were all different. Banks in Arkansas operated under rules that were
different from those of banks in New York, resulting in herd immunity.
Per the Glass–Steagall Act banks were not allowed to operate across state
borders. Then the rules were relaxed, banks could operate across state
lines, they were increasingly federally regulated, and state regulations be-
came harmonized. One consequence of these changes is that the number
of banks has fallen sharply. The federal reserve bank in St. Louis has col-
lected data on the number of banks in the United States from 1984, when

there were 14,400 commercial banks. By 2020 the number had fallen to
4,404, a reduction of 3.3 percent a year.1
Sam Peltzman, an economics professor at the University of Chicago,
came up with a radical idea in 1975: risk compensation. When regula-
tions are enacted to improve safety, the unintended consequence is more
risk. Take American football. At the beginning of the twentieth century
football players started to demand helmets to protect themselves against
injuries. In the beginning, helmets reduced injuries, but over time, as
the helmets became better and better, the way of playing changed. Play-
ers started using their heads as ramrods, running headfirst into the op-
position. Because helmets improved, the shocks to the head became
more violent, and injuries increased—the players compensated for being
protected by taking on more risk. The vicious feedback between better
helmets, more aggressive playing, and injuries showed how the laudable
objective of protecting players by making them use helmets could per-
versely harm them. In the sister sport of rugby, heads are unprotected,
but because the players don’t use them as ramrods, head injuries are much
rarer than in American football.
Risk compensation is particularly pernicious in the financial system,
driven by common beliefs, that is, everyone observing the world in the
same way. Blame the riskometers. If everybody uses the same one, we all
end up seeing the world in the same way. Financial institutions obviously
don’t all use the same riskometer, as they are a part of the secret sauce that
makes some better than others. There is, however, a limit to how different
the riskometers can be, especially in banks. Industry trends directly influ-
ence even the largest banks. Their modelers studied in the same universi-
ties and read the same academic papers. They go to the same conferences
and move freely between banks. Not to mention regulations, which also
make risk management practices more homogeneous. All pushing banks
toward similar ways of modeling risk, so they increasingly see the world
the same way.
Beliefs are only half the picture. The other half is common action,
also driven by riskometers. Half a century ago, before the widespread
use of statistics and risk management and regulations and best practices,
every financial institution was quite free to do what it wanted. They were
primarily partnerships and allowed to make the dumbest decisions. Not

today. We have a myriad of controls from regulators, media, shareholders,

accountants, stakeholders, and politicians, all wanting to constrain what
financial institutions can get up to. All pushing toward reacting the same
way to shocks.
Common beliefs and actions are the enemy of stability. Instability arises
from a lack of diversity, when we move as a herd. Like the Millennium
Bridge wobbling when all the pedestrians behaved as one (see chapter 7).
When the wisdom lunacy of the crowds tells us risk is low, more risk
makes perfect sense: Minsky, yet again. Everybody tells us that—the risk
takers, the risk controllers, the regulators, the journalists, the academics,
and the pundits. We then all pile into the same risky assets. Then a shock
comes along. It could be any of the ones I discussed earlier, like Archduke
Ferdinand’s assassination in 1914 or the realization in 2008 that CDOs
full of subprime mortgages were not as clever as assumed, or a virus like
Covid-19. It all goes pear-shaped. The same people now tell us risk is too
high, that we must derisk. Harmonization of belief and action is pro-
cyclical, amplifying the boom-and-bust cycle up and down. It doesn’t
take much to set these processes of coordinated buying and selling in
motion. Figure 35 shows an Economist cartoon that is more realistic than
we would want it to be.
The Kaizen Revolution in Finance

The largest antihunger charity in the United States is the Food
Bank for New York City, which feeds about 1.5 million people every year.
The Food Bank is financed by donations, most of which take the form of
money or food. But when the charity approached Toyota, it got Kaizen,
Japanese for “continuous improvement,” the key to Toyota’s success.2
It worked wonders for the Food Bank. In one soup kitchen in Harlem,
Toyota’s engineers reduced the wait for dinner from ninety minutes to
eighteen. Volunteers packing supplies for the victims of Hurricane Sandy
used Kaizen to shorten the time it took to pack a box from three minutes
to eleven seconds.
Finance went through its own Kaizen-type revolution, in three phases.
The first commenced in 1950, when Harry Markowitz defended his doc-
toral thesis in economics at the University of Chicago, later published

Figure 35. The Economist on procyclicality. Credit: Kevin KAL Kallaugher,

Baltimore Sun, Kaltoons.com.
under the title “Portfolio Selection.” His critical insight was that inves-
tors should focus on two variables: expected returns and the variance of
asset returns, and it has been known as the mean–variance model ever
since. Markowitz’s thesis was controversial from the very beginning,
and he almost did not graduate. Milton Friedman felt that the mean–
variance model contained too much statistics and not enough economics
and therefore did not deserve a PhD in economics. Even so, it earned
Markowitz the Nobel Prize in Economics in 1990, as his work was by then
the cornerstone of financial economics.
Markowitz formalized an ancient idea: the more risk we take, the higher
the return we demand. His insight was to reduce a complicated decision-

making problem to a simple choice between only two variables and, in

the process, sidestepping four complex issues. The first issue is that in
the real world, everybody has different objectives—or utility functions,
in the language of economics. Earlier I discussed how Paul, Ann, and
Mary all have the same portfolio and technology but need different risk
measurements. Well, Markowitz assumed they are all identical, and hence
all prefer the same riskometer. The second problem he avoided is the
distinction between risk and uncertainty by simply ignoring uncertainty.
He assumed we know variance of asset returns, that variance is risk, and
there is no uncertainty. Third, he assumed ergodicity, that the statistical
laws governing the financial markets are constant over time. In practice,
that means both the expected returns and variances never change, so once
you know them you have what you need to make investment decisions
until the end of time. Finally, he surmised that every market participant
is tiny (atomistic) and has no impact on the evolution of prices—all risk
is exogenous. His implicit assumption of atomistic market participants
is crucial because otherwise trading will change prices, undermining his
other assumptions.
Even though the technology behind investment decisions has come
a long way since Markowitz, I suspect quite a lot of fund managers still
use his technique, mostly unmodified. And even when his model is aug-
mented, most improvements are still based on the same four assump-
tions: identical investors, where uncertainty doesn’t matter, along with
ergodicity and exogenous risk.
Markowitz’s mean–variance framework was theoretical—supposing
we know the vital statistics. They are not known in the real world and
have to be estimated, the bit of the puzzle supplied by the second phase
of the Kaizen finance revolution. Many researchers and institutions con-
tributed, but two deserve special mention: Robert Engle’s ARCH model
in 1982 and JP Morgan’s Value-at-Risk method in 1994. Yet it wasn’t until
1996 that the second Kaizen phase got the official stamp of approval,
when the Basel committee updated Basel I with the Amendment to the
Capital Accord to Incorporate Market Risks, heralding the birth of sci-
entific financial risk management. The regulators thought until then that
risk belonged in buckets. Money lent to the government in the risk-free
bucket. Most loans to corporations are low risk, while those to hot dog

stands are high risk. The 1996 amendment upended all of this, introduc-
ing the internal ratings-based (IRB) approach. IRB meant that the most
sophisticated banks got to use a riskometer to identify the risk of each
of its activities, like determining bank capital, what JP Morgan wanted
when it proposed Value-at-Risk two years earlier. The timing of the 1996
amendment is not coincidental.
Practically every financial institution followed the lead. Academics
started churning out papers on risk management, universities created
courses in risk management and risk forecasting, and risk consultants
pushed into every financial institution. A wave of optimism hit the finan-
cial risk management community. It looked as if financial risk had been
successfully reduced to an engineering problem, Kaizen-style—financial
risk management became scientific. Just like structural engineers design
safe bridges, financial engineers create safe banks and financial systems.
I am as guilty as anybody. I paid for my house in London by teaching
executive education courses on implementing Value-at-Risk techniques.
And have written a book called Financial Risk Forecasting. One of my LSE
masters courses is called Quantitative Methods for Finance and Risk Anal-
ysis, all about implementing the techniques in my risk-forecasting book.
While the first phase of financial engineering was about theory and
the second statistics, a significant problem was left unaddressed. Risk-
ometers are complicated beasts. They want to be fed with a lot of com-
plex financial data, and although there is a lot of data in finance, it is not
easy to work with. Databases are inconsistent, full of errors, and much of
the data is incredibly complicated, with arcane, inconsistent conventions.
There can be multiple standards for measuring the same thing. To this
day there is no single surefire way to identify a particular financial institu-
tion in this sea of data. Suppose two financial institutions trade with each
other. Even if both report the trade to the authorities, it may be impos-
sible for the authority to match the trades because of the lack of uniform
trade identifiers.
We need a central place where financial institutions can access all the
data they need, cleaned and synchronized. They can feed in their posi-
tions, run ready-made models on the data, or, even better, have artificial
intelligence manage it all. The financial system version of Amazon Web
Services (AWS). I am a huge fan of AWS. Before I started using it I was

always looking for computational resources, building expensive machines

in my office, and trying to get accounts on high-performance computers.
Then I discovered AWS, and it was a revelation. Instead of doing difficult
things in-house, having to be an expert in hardware and operating sys-
tems and high-performance software, I could outsource it all. The com-
plexity and costs went down and efficiency went up. I use AWS every day.
The financial system version of the Amazon cloud is Aladdin and Risk-
Metrics, products of BlackRock and MSCI, respectively. There are other
vendors, but as cloud computing is a business with increasing returns to
scale, those two are dominant, especially Aladdin. But I have a soft spot
for RiskMetrics. It is the name given by JP Morgan to the model they rec-
ommended for estimating Value-at-Risk when they proposed it in 1994,
the EWMA model. JP Morgan then realized that the name RiskMetrics
was too valuable to be given to a mere statistical model and assigned it to
the risk management division, which they eventually spun off. To this day,
you run into people referring to the EWMA model as RiskMetrics. While
I understand JP Morgan’s sensible commercial decision to repurpose the
name, I wish I could still use it instead of the unpronounceable EWMA.
Aladdin’s and RiskMetric’s promise is that users won’t have to buy and
maintain expensive, inconsistent databases and hire an army of systems
engineers and programmers and financial engineers to run the models.
They can outsource it to the cloud. Let artificial intelligence take care of
all the gory details.
So What Does the Kaizen Revolution Do to Risk?

The three phases of financial engineering—that is, the theory of
Markowitz, the empirics of Robert Engle and JP Morgan, and the risk-
ometers to the cloud—make it much easier to manage financial risk.
All good, right? Not so fast. All that technology has not overcome the
more fundamental problems about the reliability and inconsistency of
the riskometers.
Demonstrating how riskometers disagree is easy. The Swiss currency
appreciation discussed in chapter 10 is one example. Suppose I take a
common concept of risk, one that is embedded in Basel III—Expected
Shortfall—and apply it to returns on the S&P 500 index and Amazon

Table 2. Expected shortfall on a $1,000 portfolio, 1 September 2020
Asset HS EWMA GARCH t-GARCH EVT
S&P 500 $64.7 $20.2 $18.4 $34.2 $67.8

AMZN $66.2 $53.0 $43.8 $107.0 $70.9
Source: Author’s calculations
stock prices. I then estimate the Expected Shortfall with six of the most
common techniques and show the results in Table 2. The highest risk
reading is more than three times that of the lowest. Similar results would
obtain for other assets and times. For a lot more examples you can go to
my website, extremerisk.org, where I estimate risk every day.
Does it matter that the riskometers are so inconsistent? In theory it
shouldn’t. A good risk manager knows riskometers are imprecise. They
know the strengths and weaknesses of each, treating them as a portfolio
of methods, picking the best one for the problem at hand. For example,
among the six in Table 2, EWMA is really simple, and nothing can go
wrong when it’s implemented. It reacts quickly to changing information.
However, just like an overeager teenager, it can respond way too strongly.
HS is the opposite, calm and stable: like a seasoned bureaucrat, it never
gets ruffled. Which do you prefer? It depends. HS’s conservativeness is an
asset for day-to-day operations, but when a shock hits, you need a quick
reaction, and that is when EWMA can be invaluable. Then, there is the
dark horse of EVT, short for extreme value theory. Like a wise man who
has nothing to say about day-to-day occurrences but tells us what we
need to know about the extremes when everything goes wrong.
All the riskometers have their good sides, and a good risk manager will
be guided by her intuition and experience. Seeing EWMA is $20.2 and
GARCH $18.4, while HS is $64.7, tells her that short-term risk not only
has fallen quite a bit recently but also is still relatively low by historical
standards. EVT says long-term risk is not affected very much by Covid-19.
The risk manager knows each number tells a different part of the story,
and by using all of them she gets a much more complete picture.
The problem is that such subjective judgment may not be acceptable.
Where the professional risk manager draws important lessons from the
diversity of measurements, the uninitiated see a problem. It is the same

asset. How can risk be so different? You know, they are all measuring
the same true process—risk—and if they disagree, one must be accu-
rate and the rest wrong. And that is precisely the conclusion the Basel
committee and the European Banking Authority reached when they saw
the inconsistency in risk measurements. The solution then is to pick one
riskometer—of course, the best of breed—and mandate its use. So does
that matter? Yes, if every bank has to use the same riskometer, they always
see the risk in the same way—their beliefs get harmonized. The regula-
tors would respond by saying that the banks can choose their own risk-
ometers; yes, to some extent, but the wiggle room is really small.
There is an interesting consequence of mandating all banks to use the
same riskometer. Under current financial regulations, banks have to mea-
sure risk from proprietary trading by 97.5 percent Expected Shortfall. In
plain English that means risk is the amount of money they expect to lose
on the worst day in every two months. So, if a bank holds $1 million
worth of Amazon stock and uses the EWMA riskometer (see Table 2),
then the risk on the day I did the measurement was $53,000, that is, the
bank was expected to lose $53,000 one day every two months.
The law of the land is 97.5 percent Expected Shortfall. Does it matter?
Yes, for two reasons. The first is the question of whether the 97.5 percent
Expected Shortfall is the correct risk to measure. After all, it captures
the worst day every two months and says little about any other type of
event. Earlier, I gave the example of Mary, Ann, and Paul, who all invest
in Google but all have different objectives for their investment. Paul cares
about short-term fluctuations, Ann is concerned about big losses in the
next half-year, and Mary worries about the pension she will start draw-
ing in half a century. The 97.5 percent Expected Shortfall risk measure
provides no valuable information for Mary and her pension problem and
little for Paul and Ann. It would be unfortunate if Mary lived in one of
the countries mandating the use of Expected Shortfall for pension funds.
The more interesting question is what the 97.5 percent Expected Shortfall
does to the market.
Imagine the total amount of risk is a balloon and that the balloon
covers all possible outcomes. We first squeeze the balloon in one place
(Figure 36). What happens? The obvious. The place where we squeeze
narrows, and the risk pops out everywhere else. In the second figure,
three hands squeeze in different places, so risk is much more uniformly

Figure 36. Squeezing risk. Credit: Lukas Bischoff/IllustrationX.
managed and never balloons out. Because we target the same part of the
risk universe—the 97.5 percent Expected Shortfall—that type of risk falls.
But like a balloon, it ends up getting squeezed out elsewhere. And one
part that will get squeezed is the extreme left tail, the big, nasty events
that blow up banks and cause banking crises. Perversely, by mandating
the 97.5 percent Expected Shortfall, the financial authorities increase the
most dangerous types of risk. The law of unintended consequences. Yet
another example of lowering volatility and fattening the tails.
When the Basel committee and the European Banking Authority ac-
knowledged that riskometers give very inconsistent measurements, the
interesting thing is how they saw the problem and what they proposed.
They could have said, “Okay, this is a known problem, and we want to in-
corporate best practices in risk management into the regulation process.”
That means running multiple riskometers on the same assets and using
the strengths of each in meeting the objectives of the regulations. How
good risk managers work in practice. No. That is not what the authori-
ties said. They expressed concern and concluded that the inconsistency

meant that some banks were using low-quality riskometers. It was es-
sential to require banks to use only the best. In other words, we need
to find the best riskometer and force all banks to use that and only that
riskometer. Why would the authorities have come to that conclusion? It
is not that they don’t know better. Having talked to a lot of regulators,
I have concluded the answer has to do with philosophy. If regulation is
risk-based, we must measure risk accurately, meaning models should give
the same assessment of the risk of a particular exposure. Otherwise, risk-
based regulations, like the Basel Accords, are fundamentally unsound. As
the regulators see it, their mission depends on the one true model, even
if few would admit to that.
The risk philosophy has further interesting consequences because of
bank lobbying. If we make regulations risk sensitive—that is, set risk tar-
gets for banks and specify the methodological approaches as we do in the
Basel Accords—then the output of the riskometers has a material impact
on banks. Suppose we have two banks, call them A and B, with an identical
portfolio, perhaps $1 million in Amazon. If A uses the t-GARCH riskome-
ter and B GARCH, then A has a risk of $107,000 and B only $43,800.
Because minimum capital is three times the risk, just imagine how loudly
A will scream, doing everything in its power to switch riskometers.
It all ends with a race to the bottom, whereby every bank wants to use
the riskometer that gives the lowest risk readings. The only way to pre-
vent that is for the authorities to decide on which riskometers are accept-
able. We then get an iterative process between banks wanting to measure
risk as low as possible and the regulators reducing the banks’ freedom
to measure risk as they see fit. The resulting ratcheting between the risk
philosophy and bank interests can end only one way: the regulators end
up dictating how banks should measure risk. Convergence to the “one
true model.”
I think a lot of people would react to this by saying, “So what!” We
are adopting best practices, the best riskometer, which means we will
measure risk in the best possible way. That might be correct if we look
at each bank in isolation and if they are small. However, for larger banks
and especially for the financial system in its entirety, there are at least
three serious consequences: procyclicality, transfer of responsibilities, and
ossification.

Start with procyclicality. I’ve spilled a lot of ink on it in this book and
so will be brief. If everybody ends up using similar or identical riskome-
ters, we harmonize beliefs and action. Everybody reacts in the same way
to shocks, all buying and selling the same assets, amplifying the cycles:
procyclicality and systemic risk.
The second problem is more insidious. The more we transfer responsi-
bility to the financial authorities, the more obligation they have for ensur-
ing things go well. Suppose the authority is in charge of how risk is mea-
sured; then, when the next crisis comes along, the critics will say, “Why
did the authorities pick such a stupid way to measure risk?” The banks
will say, “We did nothing wrong. We followed the guidelines coming
from the authority. Give us a bailout.” So, because the regulators end up
being the risk modeler for the banking system, they are also responsible
for its well-being, and hence to blame for large losses.
The final problem is that even if the authority manages to pick the best
riskometer of all possible worlds—the one riskometer that will not fail—
it will still end up ossifying. The technical issue is both the pace and the
process of designing regulations. Financial regulations, especially global
rules, change slowly. Basel I took effect in 1992. The Basel II’s design
process started in the mid-1990s, and it was partially implemented only
in 2007. As for the postcrisis regulations, Basel III has been in discussion
since 2008. One and a half decades pass between each iteration of the
global banking regulations. Throughout the process the regulators are
bombarded by lobbying from industry and governments. They have to
be seen as fair and transparent. The rules need to be as technically un-
demanding as possible.
Even if the designers of financial regulations manage to pick the very
best riskometer today, it will be put into practice years in the future and
used for decades thereafter. Over that time the riskometer will become in-
creasingly out of date—ossified—capturing irrelevant risks and neglecting
important ones, giving banks ample scope for manipulation. Supervisory-
mandated riskometers are much more likely to stagnate than riskometers
developed internally in a competitive environment, Kaizen style.
When I read the Basel committee’s and the European Banking Author-
ity’s reports on risk, I wrote a blog piece on the implications, “Towards a
More Procyclical Financial System,” arguing that it was inevitable for the

authorities to coalesce around a single riskometer. A few weeks later I did

a presentation at a central bank event, and a few of the people criticized
me, saying words to the effect of “We are not stupid. Everybody knows
harmonizing riskometers is procyclical. The reports you cited are just ex-
pressing that for political reasons; we don’t really mean it.” Okay, having
participated in the drafting of such reports, I know the writing process is
intensely political. A lot of people and entities want to shape the results.
So every line is the result of fraught negotiations. Maybe neither the
Basel committee nor the European Banking Authority meant to imply
that they prefer riskometer harmonization. Maybe they tried so hard to
paper over disagreements that a misunderstanding slipped out. But I have
more evidence. Harmonizing riskometers is precisely what the authorities
do in practice.
Finance and Scientific Socialism

One of the things I find so fascinating in finance is that while so
many of the people who work in the financial system proclaim a belief
in free markets—indeed, many see themselves as libertarians—they go
about their business in a very socialistic way, with no apparent irony. Bail-
outs are the obvious example: privatize profits and socialize losses. Just
about every bank in the United States got a bailout in 2008.
More relevant here is that the way we measure and manage risk has
its roots in an old idea called scientific socialism, a concept proposed by
Friedrich Engels in his book Socialism: Utopian and Scientific. We don’t
hear much about scientific socialism anymore, but it used to be quite fa-
miliar. The idea is that we can avoid all the worst aspects of capitalism by
scientifically creating a better socialist society. Scientific socialism gives us
testable theories based on empirical observations. When reality does not
happen as predicted by theory, ad hoc assumptions are added to make
theory fit the facts.
The idea that theory can be made to fit the facts is why Karl Popper, in
his book The Open Society and Its Enemies, described scientific socialism
as a pseudoscience. Theory can be incompatible with all observations and
still be scientific, but the opposite is not true. A theory that is compat-
ible with all possible outcomes precisely because it has been modified to

accommodate a particular observation is not scientific. Popper argued

that science is founded on testable theories based on empirical observa-
tions; theories are accepted, changed, or refuted on the basis of actual
outcomes. Since one cannot test the claims of scientific socialism, and
hence not prove it or disprove it, it is pseudoscientific.
What does that have to do with the management of financial risk?
Quite a lot, actually. No measurement of risk can be properly validated
because risk is latent. Risk is not directly observable, so there is nothing
concrete a risk forecast can be related to. We have to compare it to some
other model’s output, one that has its own set of untestable assumptions
and can create an infinite number of statements about what is essential,
most of which are mutually inconsistent. In Karl Popper’s language, since
we can’t falsify a risk forecast, it is not scientific.
The riskometer is pseudoscientific in the same way scientific socialism
is, but it doesn’t present itself as such. It is the objective, scientific way
to measure risk. By using the riskometer, we supposedly move away from
subjectivity, incompetence, and even corruption. But we don’t. The edi-
fice on which we build our control of the financial system, the riskometer,
is no more scientific than scientific socialism. And all the free market sup-
porters working in finance have adopted one of the most fundamental
tenets of Marxism in their day-to-day work.
The definition of the term “trilemma” is that when faced with three
choices we can only pick two. When it comes to managing a risky financial
system, we have three options: uniformity, efficiency, or stability. Which
do we pick? So far, all the incumbent powers in the financial system, the
private institutions, and the regulators favor uniformity and efficiency.
The commercial institutions because it reduces competition and makes
money for them, and the regulators because it makes their job easier.
There are two lessons one can take from this: either we need better tech-
nology or we do something different.

12
All about BoB:
Robots and the Future of Risk
I’m sorry, Dave, I’m afraid I can’t do that.

—2001: A Space Odyssey
I was recently at a conference on financial regulation where one

of the speakers gave a presentation on how artificial intelligence, AI,
would change the world for the better. I got into a debate with him.
What follows is a fictionalized account of the conversation, taking into
account other, similar discussions:
speaker, excited technology enthusiast.

AI is changing the world. Computers double in speed every eighteen
205

206 ROBOTS AND THE FUTURE OF RISK
months. We now have self-driving cars and robo-advisors and all

sorts of fantastic technology that is rapidly improving society. AI
will revolutionize the financial system and give us fantastic financial
regulations.
m e , o l d f o g e y / c l u e l e s s s k e p t i c . What about
all the recent AI failures, like Facebook’s and YouTube’s inability
to eliminate scams, hate speech, antivaccination propaganda, and
advertisements for fake products?
s p e a k e r . Give it time. Ten years ago nobody thought cars
could drive themselves. Remember Moore’s law.
m e . So is there anything you can’t see computers doing better
than humans?
s p e a k e r . There are many things AI cannot do today, but
technology is advancing rapidly, and AI can do new things every
day. There is nothing AI will not be able to do, and it will all be
wonderful. Just keep your mind open.
m e . But the financial system is very complex, and it is surely
beyond the ability of any computer algorithm to keep it under
control.
s p e a k e r . No. That is precisely why AI will be so valuable. Put
it in charge, and it will master all the complexity human beings
can’t. Financial crises will be a thing of the past, and the financial
system will be superefficient. Just wait and see.
I walked away intrigued but skeptical. The next thing I did was to sit
down and investigate whether my interlocutor was right. I ended up writ-
ing a series of articles on the subject.
We might think the financial system is the ideal application for AI.
After all, it generates almost infinite amounts of data, plenty for AI to
train on. Every minute decision is recorded, trades are stamped to the
microsecond. Emails, messages, and phone calls are recorded. But data
does not equal information, and making sense of all this data flow is like
drinking from a fire hose. Even worse, the information about the next
crisis event might not even be in this sea of data. I’ve concluded that the
speaker was mostly wrong.

ROBOTS AND THE FUTURE OF RISK 207
What Is Artificial Intelligence?

The computer says no.
—Carol Beer
A lot has been written about AI, and there is no need to repeat
that here. I highly recommend Stuart Russel’s book Human Compatible.
But I need to establish where I’m coming from, so bear with me. The
idea behind AI is that a computer learns about the world so it can make
decisions by itself. It can be something as simple as playing a game, as
complex as driving cars, and it can even regulate the financial system.
Describing AI is not straightforward; not even the experts agree. Let’s
start with machine learning, a computer algorithm that uses available data
to learn about the world that created that data. The algorithm studies all
the patterns and complicated causal relationships. What is magical is that
it can do so without human intervention: unsupervised learning. That is
different from the way we usually do science, where we start with some
idea of how the world might work—a theory—and see if data is compat-
ible with that theory.
A supermarket might use machine learning to figure out where best to
place Coca-Cola cans to maximize sales. The data scientist takes all the
historical observations on Coca-Cola sales, the weather, and demograph-
ics. She runs the data through her machine-learning algorithm and then
tells the supermarket where best to place the cans on the shelf. That is a
lot of data. For a chain like Walmart it can easily be quadrillions of ob-
servations. Precisely what is called big data. The critical thing is that the
machine-learning algorithm doing this doesn’t need to know anything
about Coca-Cola or supermarkets—it just gets data and finds the patterns.
There is no such thing as a free lunch, and there is a trade-off. Machine
learning needs a lot of data, much more than most other statistical appli-
cations. It needs big data. The reason is that it knows nothing about the
world and so has to learn everything from the data. Human beings know
the world and bring prior information—cultural, economic, historical,
and so on—to bear on a problem and thus need a lot less data. They
know theory, allowing traditional statistics to work with small datasets.
While machine learning is all about extracting information from a data
set, the objective of AI is to make decisions based on that data. AI is used

to make a lot of decisions today. In the 2000s British comedy show Little
Britain, a recurring sketch featured a bank loan officer named Carol Beer,
who responded to every customer query by typing into her computer and
then answering, “The computer says no,” even to the most reasonable of
requests.
The term AI is a bit of a misnomer, though. The AI of today isn’t in-
telligent in the sense a human being is intelligent. It merely knows a lot
about what-if rules. If the traffic light flashes red, stop. If it flashes green,
look at the traffic and then go. AI replicates the human brain, sort of. The
average human brain has eighty-six billion neurons, all interconnected by
synapses to form neural networks. A computer with a large enough num-
ber of artificial neurons wired in the same way could become intelligent—
in theory. We haven’t reached that stage yet. We are still competing with
the average insect. Take cockroaches. They are among the smartest of in-
sects and can learn and adapt to their environment, not to mention being
the only animal that supposedly will survive a nuclear war. They are not
the most social of animals but do exhibit complex social behavior. AI has
not caught up with cockroaches. In the words of the theoretical physicist
Michio Kaku, “At the present time, our most advanced robots . . . have
the collective intelligence and wisdom of a cockroach; a mentally chal-
lenged cockroach; a lobotomized, mentally challenged cockroach.”1
Moore’s law, named after Gordon Moore, the cofounder of Intel,
observed that the number of transistors in a microchip doubles every
eighteen months. Will that help? That was exactly what the technology
enthusiast I debated with argued. When I did my PhD, I ran my com-
puter code on a $27 million Cray Y-MP supercomputer. Time on the
Cray was carefully rationed, and the computing jobs of an economics
PhD student did not get first priority. I learned that the queue of jobs
tended to finish early over the weekend, so by waking up at four on Sun-
day mornings and going to the office I could have the Cray all to myself.
I did this way too often. When I was in Paris recently, I saw a Cray Y-MP
in the Science Museum. The iPhone in my pocket is many times faster.
The speed of computation has increased exponentially since before I was
born, and it shows little sign of slowing down.
Will Moore’s law help AI catch up with human intelligence? The short
answer is no. Moore’s law is about the speed of computations, and the

problem with AI catching up with humans is not about speed. It is about

concepts. The quality of today’s AI algorithms simply cannot take us very
far, regardless of how fast computers are. Yet breakthroughs can happen
overnight, and someone may already have thought of a completely new
algorithm, one that allows AI to surpass humans. Or it may never happen.
The pundits call that eventuality the technological singularity, and
some think it is not too far away. In the words of Ray Kurzweil, Google’s
director of engineering, “2029 is the consistent date I have predicted for
when an AI will pass a valid Turing test and therefore achieve human lev-
els of intelligence. I have set the date 2045 for the ‘Singularity’ which is
when we will multiply our effective intelligence a billion fold by merging
with the intelligence we have created.”2
However, we don’t need the singularity for AI to be useful; it doesn’t
need human intelligence and not even that of a cockroach or an ant to be
useful. The more confined a problem is, the better AI performs, which is
why it excels at playing games with a known set of rules and clear objec-
tives. A celebrated recent AI is Google’s AlphaGo Zero, which, being
instructed only with the rules of the game of Go, was able to learn over
the span of three days how to conquer its predecessor AlphaGo, which
had earlier defeated the Go world champion.
Does being good at playing complicated games mean that one is good
at other things? A lot of people think so. The Soviet Union venerated its
chess players, believing that if one could master the complexities of chess
strategy, one should also be able to manage a centrally planned economy
or fight a war. Lenin was a serious chess player, and Stalin even more so.
However, AI that is good at chess is much more likely to be an idiot sa-
vant, like the main character in the movie Rain Man. Being good at chess
says nothing about the ability to do other things, such as regulating the
financial system or driving a car. Games like chess have a predefined num-
ber of actions a player can take, and all the information about the state
of the game is visible to the players. In the language of game theorists,
information is complete, and the action space is finite.
This is also why AI can, sort of, drive cars. There is a lot of information
to learn from, the rules are well understood, and they don’t change much.
Even then, AI can drive only on highways, and then only if the weather is
not too bad. It is far from being able to drive in cities, not to mention in

deep snow, where all the signs are hidden. Perhaps the worst challenge for
a car-driving AI is human beings. We are unpredictable and can behave
in a way that upsets self-driving cars. In response, some AI designers have
proposed reprogramming humans.3
The more we move away from the confined space of board games and
driving on the highway, the worse AI performs. It is not good at play-
ing games in which information is incomplete and the action space ill-
defined. I suspect it would not do well in my favorite game, Diplomacy. I
once played Diplomacy over the internet with a few friends, one of whom
is a politician. He beat us hands down, showing deviousness and strategy
that none of the other players had. I find it hard to believe AI will beat
him any time soon. It is especially challenging for AI if the rules evolve
during play, as happens in most human endeavors.
I have my own personal test for robots and artificial intelligence. Today,
I can pay someone who has never been to my house $100 to do my laun-
dry. He comes to my university office, and I give him money, keys, and
my home address. He finds his way to my house, gets in, finds my laundry
baskets and washing machine, figures out how to operate it and where
to find the detergent. He manages to do the laundry and put it into my
closet and cupboards, and then he slides my keys through the mail slot
on my door when done. All without any explanation or instruction. The
technology involved is older than I am. When I find AI that can do that,
I will be impressed.
The Bank of England Bot: BoB

A few years ago the chief economist of the Bank of England,
Andy Haldane, envisioned a new way to control the financial system, one
in which the regulator will track “the global flow of funds in close to real
time (from a Star Trek chair, using a bank of monitors), in much the same
way as happens with global weather systems and global internet traffic. Its
centerpiece would be a global map of financial flows.”4 Let’s give him a
name, BoB, short for the Bank of England bot (see Figure 37). BoB is the
future AI who oversees the British financial system. BoB will take all the
data and human behavior and use that to figure out all the contingencies
and identify fragilities, inefficiencies, and systemic risk. Suppose further

that other major financial centers also develop their version of BoB, like
Fran and Edith, and all AI are friendly and cooperating with each other.
The financial institutions will also have their AI: Gus, Mary, and Betty.
Is this a pie-in-the-sky futuristic vision doomed to failure, like the flying
cars of the 1970s? No. While BoB and friends don’t exist yet, the technol-
ogy to create them is already here. Well, most of it. We just lack the will.
Roboregulators already exist in microprudential regulatory agencies.
The latest buzzword is RegTech, short for regulation technology. Its
chief cheerleader, the UK’s Financial Conduct Authority (FCA), defines
RegTech as the “adoption of new technologies to facilitate the delivery
of regulatory requirements.” I was involved with RegTech research over
the past few years because of a joint research program conducted by the
FCA, my research center, and other interested parties.
The starting point is the rulebook, all the rules and regulations which,
if printed out, create a stack of paper two meters high. The FCA has
translated the rulebook into an AI engine so it can check it for incon-
sistencies and give faster and better advice. Financial institutions corre-
sponding with the FCA bot find that it answers much better than its
human colleagues. AI is also revolutionizing risk management in banks.
The first step in creating a risk management AI is to develop and manage
riskometers, an easy task for AI. It quickly learns all the approved models,
the data is readily accessible, and it is easy for it to create riskometers. A
lot of financial institutions have AI engines today that do precisely that.
The AI then needs to learn about all a bank’s investments, the individuals
who made them, and, voilà, we have a functioning risk management AI.
The necessary information is already inside banks’ information technol-
ogy infrastructure, and there are no insurmountable technological hur-
dles along the way. And if there are, just use Aladdin or RiskMetrics. The
cost savings will be enormous. The bank can replace most risk modelers,
risk managers, and compliance officers with AI. The technology is already
here. All that remains to be done is to inform the AI of a bank’s high-level
objectives. The machine can then automatically manage risk, recommend
who gets fired or gets a bonus, and advise on how to invest.
Risk management and microprudential supervision are the ideal uses
for AI—they enforce compliance with clearly defined rules and processes
based on vast amounts of structured data. They have access to closely

monitored human behavior, are guided by precise, high-level objectives,

and produce directly observed outcomes. Just as in games, information is
mostly complete and the action space finite. There is still some way to go,
but it is eminently conceivable for AI to take over many current functions
given the trajectory of technological advancement. The main hindrance is
not technology but legal, political, and social considerations.
The macroprudential problem is different because all the precondi-
tions for AI to perform well are absent. Data is scarce. The events being
controlled are mostly unique and very infrequent. After all, the typical
OECD country suffers a systemic crisis only one year out of every forty-
three, giving AI precious little to train on. Even worse, when the authori-
ties and the private institutions learn from past crises, they tend not to
repeat mistakes, instead creating new ones, frustrating AI. I looked into
the main issues facing AI doing macroprudential policy in a recent paper
with two of my collaborators, Robert Macrae and Andreas Uthemann,
titled “Artificial Intelligence and Systemic Risk” (the remainder of this
chapter is based on that work). There are four areas we identify as being
of particular concern: procyclicality, unknown unknowns, need for trust,
and optimization against the system.
Banking is inherently procyclical. Banks lend freely in good times, am-
plifying booms, then in turn they contract-lend when things turn sour,
causing a credit crunch. Modern financial regulations and risk manage-
ment practices further exacerbate procyclicality because of their depen-
dence on riskometers and risk dashboards. Riskometers work by looking
at recent history, and if that recent history seems stable the riskometers
tell us risk is low. The problem is that price data tends to be more calm
in upturns than in downturns, and any backward-looking, data-driven
process, including the machine learning that feeds AI, will identify risk as
being low in quiet times and high after a crisis—the measurement pro-
cess itself is procyclical. The way we use the output of riskometers is also
procyclical. They feed risk-control systems and regulations so that the
portfolio managers and loan officers and the traders who make decisions
on risky investments also amplify the financial cycle. Of course, there is
nothing inherently AI about that. But my contention is that AI makes
the problem worse because of the particular way it interacts with risk
measurement and management.

The AI engines at financial institutions and regulators all have access to

the same data and machine-learning techniques, all have significant com-
putational resources. The various AI engines will inevitably converge to
the same riskometer much more rapidly than human risk modelers, thus
harmonizing knowledge. AI will also harmonize action, how the financial
institutions react to new information. It has a much better understand-
ing of best practices than any human being does. It knows what should
be done and what prevented. The AI will push banks to manage risk in
the same way.
All of this is procyclical because of how it interacts with the banks’ objec-
tive functions, maximizing profits subject to constraints. The more similar
information is, the closer the various banks’ solutions become to each
other. That means crowded trades so the financial cycle gets amplified.
Ultimately, AI is procyclical because it favors monoculture. Because it has
more information than human risk managers, the solutions it recommends
become increasingly similar across the industry. Even then, the AI would
not necessarily be all that procyclical if it were not for all the external con-
straints, most important, regulations. The feedback between regulations
and AI is the most dangerous driver of monoculture and systemic risk.
If we ask AI to manage financial stability, where will it look for danger?
With the benefit of hindsight there are plenty of warning signals for the
2008 crisis, and many have criticized the human regulators for missing
all of them. Could AI have done any better? Unlikely. Suppose there are
no observations of the consequences of subprime mortgages being put
into structured credit products with hidden liquidity guarantees, crossing
multiple jurisdictions, institutional categories, and countries. Then there
is nothing for AI to learn from. It is conceivable that an appropriately in-
structed AI engine might, in 2008, have noticed the connections between
house prices, mortgage defaults, and default correlations, the factors that
determine the prices of CDOs. It could also have realized the fragility of
the structured credit products when liquidity evaporates. Even if AI iden-
tified each individual step, the likelihood that it would put all the pieces
together is quite remote, necessary for the chain of vulnerabilities to be
discovered. We are asking a lot, not only of AI but also of the banks and
the national financial authorities who would have to allow such intrusive
data-driven international supervision.

The ability to successfully scan the financial system for systemic risk
hinges on where the vulnerability lies. Everyday factors, well founded in
economic theory, drive financial crises. Yet the underlying details are usu-
ally unique to each event. After each crisis, regulators and financial insti-
tutions learn, adapt processes, and tend not to repeat the same mistakes.
I suspect that BoB will focus on the least important types of risk, the
exogenous risk that is readily measured while missing the more dangerous
endogenous risk. It will automate and reinforce the adoption of mistaken
assumptions that are already a central element of current crises. In doing
so, it will make the resulting complacency even stronger. In other words,
BoB will be what the Soviet central planner Hayek warned us against in
his article “The Use of Knowledge in Society.” The problem is one of
how information aggregates, and BoB will have all the individual pieces
but will not know how to connect the dots. BoB measures all the minute
details of how a financial institution operates so he can understand all the
risks. He then aggregates the results to quantify not only the risk of an
individual institution but also systemic risk. Quadrillions of bits of tiny
risks that end up as simple summary measures. The risk-weighted as-
sets of a bank and the European Central Bank’s systemic-risk dashboard.
How well does that work? About as well as it did for Gosplan, the State
Planning Committee of the USSR: almost all the relevant information is
lost. Okay, you may retort. The current human-centered setup is no bet-
ter. Nothing in this tells us AI makes it worse, and it could certainly do
better because it will be much better at solving the aggregation problem.
Perhaps, if it were not for trust.
I have presented the work discussed here many times, and the most
frequent pushback I get is on the question of trust: “Jón, I may buy
your conclusions, but it doesn’t matter because we will never put AI in
charge of anything important.” I disagree because trust has a sneaky way
of creeping up on us. Twenty years ago almost nobody trusted internet
banking. Then, seeing that it works, almost everybody today uses online
banking. Very few people would have trusted computers to fly aircraft or
drive cars a quarter of a century ago, and now we mostly do. We are happy
for AI to control surgical robots. AI is proving its worth in critical day-to-
day applications, and that creates trust. The more we see AI outperform-
ing human beings, the more we prefer AI decision makers to humans.

The guardians of financial stability, the central banks, are already us-
ing AI. I recently gave a talk on AI at a central bank, where the audience
assured me AI made no decisions. Well, that day I could not easily enter
the central bank because the AI controlling the security system didn’t like
me. When I told that to the audience, they responded by saying, “We
meant AI controls nothing important.” They got it wrong. The central
banks already employ AI in a variety of places, doing a good job today,
building trust. And there are plenty of cost savings to be had. The staff
might not like it, but the board sees the benefits of replacing an army of
PhD-level economists with computers.
And herein lies the problem. In the 1980s AI called EURISKO used a
cute trick to defeat all of its human competitors in a naval wargame: it
sank its slowest ships to achieve better maneuverability than its human
competitors. An example of what is known as reward hacking, something
human beings are, of course, expert in.5 EURISKO’s creator, Douglas
Lenat, notes that “what EURISKO found were not fundamental rules
for fleet and ship design; rather, it uncovered anomalies, fortuitous inter-
actions among rules, unrealistic loopholes that hadn’t been foreseen by
the designers of the TCS simulation system.” Each of EURISKO’s three
successive victories resulted in rules changes intended to prevent a repeat.
The only thing that proved effective in the end was to disinvite Lenat and
his AI.
And that is the problem with AI. How do we know it will do the
right thing? Human admirals don’t have to be told they can’t sink their
own ships. They just know. AI has to be told. But the world is complex,
and it is impossible to create a rule covering every eventuality. BoB will
eventually run into cases in which he makes critical decisions in a way no
human would. And that is where humans have the advantage over AI. Of
course, human decision makers mess up more often than BoB. But there
is a crucial difference. The former also come with a lifetime of experience
and knowledge of relevant fields, like philosophy, history, and ethics, al-
lowing them to react to unforeseen circumstances and make decisions
subject to political and ethical standards without it being necessary to
spell them out.
Before putting humans in charge, we can ask them how they would
make decisions in hypothetical scenarios and, crucially, ask them to justify

their decisions. They can be held to account and be required to testify to

Senate committees. If they mess up, they can be fired, punished, incarcer-
ated, lose their reputation and even their lives. You can’t do any of that
with AI. Nobody knows how it reasons or decides, and it cannot explain
itself. You can hold the AI to account, but it will not care. But AI has
some advantages. The human might be incompetent, callous, or corrupt,
unlike AI, which, paradoxically, can behave more humanely than humans.
Judges sentence people more harshly when they come up in front of them
an hour before lunch as opposed to an hour after. AI has no such biases.
How important, then, is the question of trust? It depends on what
we ask of AI. Trust is not all that important for microprudential regula-
tions because the existing rulebook already specifies the objectives and
rules. There is little danger of AI making seriously wrong decisions. If it
does, the short reporting timescales, the large number of repeated ob-
servations, and the relative unimportance of decisions mean we realize
the problem quickly, allowing us to react in time. It is different with
macroprudential regulations, macropru. The problem is unbounded, the
system is infinitely complex, the events under control are very infrequent,
and almost all are unique. There is little to train the machine on, and the
wrong decisions will be catastrophic. If we let a machine with fixed ob-
jectives interact with an infinitely complex environment, the outcome is
unexpected behavior. AI will run into cases in which it makes critical deci-
sions in a way no human would—the financial version of sinking its own
ships. The chance of a catastrophic outcome increases over time when the
machine is forced to reason about unforeseen contingencies. The conse-
quences could be disastrous, perhaps a Minsky moment.
Optimization against the System

Any observed statistical regularity will tend to collapse once pressure is
placed upon it for control purposes.
—Goodhart’s Law
Every time we try to control people’s behavior, they evade. It’s

in our human nature. Why the Panopticon exists. We have to continually
guard against those who seek to undermine the controls. For the lack

of a better phrase, I will call these people hostile agents, those intent on
exploiting the system for private gain by optimizing against the system.
People do this all the time, and there is nothing uniquely AI about it. But
AI has inherent weaknesses that make optimization against the system
particularly dangerous.
The starting point is that the very fact we exercise control changes
the system—exactly why the epigraph above is so prescient. The world
looks different with and without the control. In most applications, opti-
mization against the system is not very important. Human drivers exploit
self-driving cars for their own advantage. They know the self-driving car
follows the rules and drives conservatively and predictably. It is easy for
the human to gain an advantage when it comes to merging traffic, at four-
way stops, and in any situation in which drivers are competing. However,
such optimization against the system is quite limited. The cost of failure
is local and small, and, most important, the drivers do not influence the
rules of the game.
And that is the crucial difference between most AI applications and the
regulation of the financial system. While human drivers taking advantage
of a Tesla on the freeway cannot build new roads to disadvantage the
Tesla or change the traffic rules or even move signs around, their counter-
parts in the financial system can do all of that. Because the rules and the
structure of the financial system are mutable, the financial system is a par-
ticularly fertile ground for optimization against the system. There is a lot
of money to be made, the adversaries are smart and well resourced, the
system is infinitely complex, and there are plenty of places to misbehave.
Ultimately, the financial system’s controllers are forever doomed to be on
the losing end of a cat and mouse game.
The complexity of finance offers many ways for economic agents to
bypass regulations, perhaps by creating new financial instruments that
mimic controlled instruments but are regulated differently, the driver of
much financial innovation. A good example is high-frequency trading.
The SEC in the United States has repeatedly changed the rules, but the
high-frequency trading firms are still quite able to exploit regular inves-
tors, as noted in Michael Lewis’s book Flash Boys.
Most hostile agents don’t deliberately break the law. Most, but not all.
Rogue traders have always been around. Take Nick Leeson and Kweku

Adoboli. Leeson was trading futures contracts for the Barings bank in
Singapore and exploited the “error account,” numbered 88888—mean-
ing super lucky in Chinese. It was supposed to be used to correct mistakes
in trading, but Leeson used it to hide his trading losses, which amounted
to $1.4 billion. Adoboli, trading on behalf of UBS, was making unau-
thorized trades, entering false information into the bank’s computers to
hide his actions. His eventual losses amounted to $2 billion. Leeson and
Adoboli were just individuals who illegally manipulated the control sys-
tems, and, though materially important to their employer, their impact
on society was negligible.
A more insidious example of optimizing against the system is when an
entire bank or a group of financial institutions join forces in destabilizing
the system. There may be nothing illegal about what they are doing. It is
quite possible they don’t even know that they are destabilizing, and those
in charge may be blissfully unaware of the consequences of what is going
on. A good example is all the dangerous financial instruments created be-
fore the crisis in 2008. Nobody had an overview of all the CDOs and con-
duits and all the other nefarious instruments that proved so damaging.
They were all optimizing against the system in their own little parts of it,
where everything looked okay. It was damaging only in the aggregate. We
don’t need a single big entity. The consequences are just as serious if op-
timization against the system involves many small, hostile agents. Their
profit may be maximized if they coordinate their hostile actions. Acting
as a wolf pack, sharing information, outwitting the controllers.
Even worse is a hostile agent intent on causing damage, a terrorist or a
rogue nation. It is even harder to detect such agents because they don’t
play by the standard rules. The rogue trader is motivated by profit and a
desire not to be caught, limiting what he gets up to. If someone doesn’t
care about making a profit, it is much easier to cause damage. Can AI
solve these problems? After all, BoB can patrol the financial system much
more extensively than any human being can and use machine learning to
find all the hidden connections. He knows everything that has happened
and is a much better enforcer than humans. The microprudential AI will
have the advantage over hostile agents. The stakes are small. There is
ample information about repeated events, plenty of data for the AI to
train on and learn all the hostiles’ nefarious tricks. And that is why AI

will be so valuable for both microprudential regulations and internal risk

management and control, why we will be able to outsource to AI the vast
majority of jobs in these functions.
Yet again, it is different for macroprudential regulations, where the
hostile agents, intent on profiting or even causing damage, have the ad-
vantage over BoB. The macroprudential problem is a much bigger game
than the microprudential. There will be hostile agents optimizing against
the macro authorities, and while the AI might catch most, even almost
all, it takes only one slipping through the cracks to cause a systemic crisis.
Since the financial system is infinitely complex, there is only a small area
for the controllers to patrol, so the hostiles just have to find another one
that is not monitored. Often the boundaries between jurisdictions.
The Standard Defenses Won’t Work

When someone hacks financial regulations, the regulator cannot respond
within hours or days. It might have to wait decades.
When I have discussed optimization against the system with AI

experts, they all recognize the problem but also add that defenses do ex-
ist. A self-driving car faced with hostile humans intent on taking advan-
tage can respond by mimicking the human drivers’ aggressiveness and ir-
rationality. The AI selectively obeys the rules, even making its own rules.
Except such a defense strategy does not work well in financial regula-
tions. The reason is the need for transparency, fairness, simple rules, slow
decision-making, and silos. Financial regulations must be transparent and
enforced fairly. Randomized responses have to be programmed into the
macroprudential AI, but it would be unacceptable for the AI to decide by
itself which rules to enforce, not to mention being allowed to make new
rules. Meanwhile, even if permissible, randomized responses will require
the AI designers to specify the statistical distribution for how it will re-
spond. Given time, that distribution can be reverse engineered.
The need for fairness and transparency also implies that the rules have
to be simple and easily implemented. When I read financial regulations, it
is remarkable how basic the models are, much simpler than those taught
in university courses. But that very simplicity frustrates the standard

defenses. It is hard to randomize responses when the rules are simple and
crystal clear. And the rules don’t change very often. The Basel rules are
updated every couple of decades or so. Even at the local level, the relevant
laws have to be passed by Parliament and then implemented by the super-
visory agency. A process that is slow, transparent, with plenty of lobbying.
When someone hacks financial regulations, the regulator cannot respond
within hours or days. It might have to wait decades.
Then we have the silos. When the Icelandic banks failed in 2008, I
talked to a senior European regulator about why the problems and misbe-
havior were not detected. He said, “Simple. The banks did not misbehave
in any single jurisdiction. It was only in aggregate that they were causing
serious damage.” While the European authorities have now plugged that
particular loophole, the problem remains. Even individual countries have
multiple regulators, controlling ill-defined and ever-changing domains.
When the German bank Wirecard failed due to fraud in 2020, the rel-
evant regulator, Bafin, disclaimed any responsibility because it said it had
decided it did not need to regulate Wirecard. Apparently nobody had
responsibility for one of the largest German banks. The regulators vigor-
ously patrol the boundaries of their domains. It’s even worse internation-
ally, where regulators are jealous of their data and powers, refusing to
share and cooperate across borders. All helping the hostile agents to oper-
ate across jurisdictions without hindrance and frustrating AI. Ultimately,
the AI engine’s innate rationality, coupled with demands for transparency
and fair play, puts it at a disadvantage over human regulators.
BoB might still have a fighting chance against the hostile agents if the
financial system’s structure remained static. If the system never changes,
BoB learns more about the system every day and one day might be able
do a perfect job. The problem is that the financial system isn’t static. Not
only is it infinitely complex for all practical purposes, it is endogenously
infinitely complex, meaning the complexity is continually evolving in re-
sponse to what everybody in the system is up to. The worst outcomes
happen when seemingly unconnected parts of the financial system reveal
previously hidden connections. The vulnerabilities spread and amplify
through opaque channels, the dark areas nobody thought to be worried
about. Competition makes the system adversarial, and any rules aiming
to contain risk-taking become obstacles to be overcome.

The hostiles take full advantage. They are like cockroaches, scurrying
out of the lit areas the authorities monitor into the shadows, and in an
infinitely complex system find plenty of dark areas to feed on. The hos-
tiles, like everybody working in the financial system, have an incentive to
increase its complexity in a way that is very hard to detect. There are many
ways to do so, perhaps by creating new types of financial instruments
that have the potential to amplify risk across apparently distinct parts of
the system.
All of these factors frustrate BoB’s mission, but the ultimate problem
is that BoB isn’t smart enough to ensure financial stability and never will
be. His computational problem is harder than that of his adversaries. BoB
has to patrol the entire financial system, searching for the hidden corners
inhabited by the hostile agents where instability thrives. The hostiles have
to find only one vulnerability, one dark corner where they can feed. And
that is a much simpler problem. The hostiles always have a computational
advantage over BoB. Moore’s law or technological developments do not
help. The more we use AI in the financial system, the more the advantage
tilts toward the hostile agents.
The Need for a Kill Switch

OK, computer, I want full manual control now.
—Douglas Adams, Hitchhiker’s Guide to the Galaxy
Suppose BoB is put in charge of financial stability, where he talks

to his counterparts in the private sector, perhaps Gus, Mary, and Betty.
Those three are grandmasters in portfolio construction and risk manage-
ment, trained in the entire corpus of historical asset price movements.
They know everything that has ever happened and have a sensible struc-
ture permitting extensive extrapolation.
Suppose a new type of instability emerges. Being the most aggressive of
the bunch, Gus smells a profit opportunity and goes on the attack, real-
izing that by attacking Mary and Betty he will make lots of money. And
so he does. It may even dawn on him that attacking BoB is profitable.
What prevents such an eventuality? Either Gus has to be told to desist or
the profit-maximizing masters of Gus will have to foresee that eventuality.

But it is impossible to predict everything. The alternative is to have a kill

switch. Will it be possible to turn BoB and the other AI off? The longer
we leave them in charge, the harder it will be. BoB’s knowledge of the
financial system and his internal representation of data will be unintelligi-
ble to humans, and turning him off will disrupt the system in unforeseen
ways. Even in a crisis, as when BoB is clearly making the wrong decisions,
the side effects of turning him off might be worse.
And then there is a question of whether BoB will actually allow us to.
That is precisely what Dave was trying to do in the quote from 2001: A
Space Odyssey. The AI engine has a purpose programmed into it, and be-
ing turned off prevents the AI from meeting its objectives, and therefore
it may not allow us to do so.
Artificial intelligence will become increasingly helpful to the financial

authorities—financial logic makes it certain, the cost savings are just so
great. But not overnight. BoB will first have to earn our trust, but he will.
That is fine for some purposes. BoB will do a fantastic job with micro-
prudential regulations, allowing the authorities to dispense with expen-
sive, slow, inconsistent humans. Replacing them with AI that performs
cheaper and better on all levels. Not a problem because the micropruden-
tial problems are contained and involve repeated outcomes and precisely
identified regulations, all playing to BoB’s strength.
It is different with macropru. The events under control are very infre-
quent, and each is unique. The cost of failure is high. To perform well
BoB will have to exercise control across borders and silos, randomize
his responses, and create rules in a nontransparent manner. He has to
understand causality and unforeseen cases and reason on a global rather
than local basis. Meanwhile, he has to identify threats that have not yet
culminated in bad outcomes. These are either unacceptable, beyond cur-
rent capabilities, or both.
The cost of AI making the wrong decision can be catastrophic. Once
we realize we can’t live without AI, it will be too late to do anything
about it. We might want to have a kill switch to turn BoB off if he starts
misbehaving, but we will not get one. If AI will not save us, how best to
regulate?

13
The Path Not to Take
The past informs the tools, but the threats come from the future.
It is devilishly difficult to bend the financial system to our will. We

want the impossible: fat investment returns and high economic growth
without too much risk. Not easy. The financial system is among the
most elaborate of human constructs, so how does one make something
infinitely complex do what we want? There are many ideas out there.
Perhaps the best we can do is the setup we have now, what I call the
modern philosophy of financial regulations. Some think technology will
save us, maybe cryptocurrencies or central bank digital currencies. Oth-
ers favor political solutions, like the libertarians, who think it best if the
223

224 THE PATH NOT TO TAKE
government stays away from finance, or the socialists, who want tough
regulations to bend it to the will of society. And then there is the way for-
ward I will propose in chapter 14, putting the inherent stabilizing forces
of the financial system to good use.
What Is the Point of the Financial System

and Financial Regulations?
The financial system is necessary, bringing financial intermedia-
tion: the channeling of funds from one person to another across time and
space. It reallocates resources, diversifies risk, allows us to build pensions
for old age, and helps companies make multidecade investments. Finance
is also dangerous. Financial institutions exploit their clients. Banks fail
and cause crises. The bankers lack empathy and are arrogant while hav-
ing no qualms about demanding bailouts when they mess up—privatize
profits and socialize losses. Finance is not popular except with some of
those who make a living from it. It is needed, and hence tolerated, but
not accepted, and society’s response is to enjoy the benefits of the finan-
cial system while also regulating it heavily. Not an easy task, and there
certainly is no consensus as to how to do it.
So what do we want to achieve from the financial system? Plenty of
answers out there. But supposing one dispenses with all the politics and
special interests and asks a typical person, I think we would get a simple
answer: high economic growth without too many costly recessions and
crises. In other words, maximize long-term economic growth and mini-
mize the cost of crises and recessions:
cumulative long-run economic growth
Maximize ————————————————––
cumulative cost of crises and recessions
It is hard to translate such a general objective into actionable policy, as

so many special interests push for what works best for them. The bankers
hope to eliminate competition and be given free rein otherwise. The reg-
ulators want to demonstrate they have done their job. The anticapitalist
activists aim to eliminate all speculation and even close down privately
owned banks. The politicians desire quick growth before the next elec-
tion. The pundits crave attention. So how to go about meeting the ob-

THE PATH NOT TO TAKE 225
jective and overcoming all the objections from the special interests? Start
with learning from history, the last financial crisis, 2008, with a detour
to Voltaire. In his book Candide, ou l’Optimisme (1759), Voltaire tells
the story of a young man, Candide, who lives a sheltered life in a sup-
posed paradise on earth. His mentor is Professor Pangloss, who likes to
proclaim “All is for the best in the best of all possible worlds.” We get
from him the word “Panglossian,” an excessively optimistic view of the
world.
The global crisis in 2008 should not have happened. The years before
2008 were the great moderation, evoking the 1920’s permanent era of
stability. The financial engineers had tamed risk. Yes, there were losses
here and there, even large ones like Enron or WorldCom, and developing
countries did have their crises, like Korea in 1998. But even then it was
the accountants and lawyers and politicians who failed, not the financial
engineers. Crises were a thing of the past, at least in the developed world.
Even in the developing world crises happened only when the good advice
of the IMF was rejected. By 2007 we were so safe that the IMF itself was
on the verge of being drastically downsized because there were no crises
and nothing for it to do.
It was the best of all possible worlds. The rules and institutional struc-
ture, founded on the rigor of scientific risk management, protected us.
The riskometers told us we had never been as safe as in 2007. For good
reasons. We have used statistical analysis to manage risk ever since Blaise
Pascal in the sixteenth century. The three waves of Kaizen took financial
engineering to the pinnacle of its success. The financial industry collects
vast amounts of data, which it processes with sophisticated models on
superfast computers, all overseen by highly educated financial engineers,
graduates of the world’s best universities. How can that not make the
financial system do what we want? Elementary. While the financial en-
gineers focus on all the micro risks—the grains of sand being examined
above (Figure 38)—new and dangerous forms of risk emerge where no-
body is looking—the endogenous risk monster that had been hiding all
along. Why did we miss it? Because the financial system is infinitely com-
plex, the financial engineers can only patrol a tiny part of it, while the
monster hides where they are not looking. The Panglossian great mod-
eration, meanwhile, encouraged even more risk-taking, à la Minsky. And

we did not know because the riskometers measure only exogenous risk,
not endogenous risk.
Then—boom!—2008 happened. Northern Rock, Bear Stearns, AIG,
and finally Lehman. We had a full-blown global crisis on our hands. The
exogenous risk indicators shot up. The authorities reacted quickly, ini-
tially with improvised responses and eventually the doctrine of macropru.
Are we protected? I don’t think so. Regulations are backward looking,
aiming to prevent the mistakes of the past. The rules that would have pre-
vented a repeat of 2008 ossify, while the world continually changes. The
market participants become increasingly good at evading the rules—new
risks get taken where nobody is looking, and the cycle repeats (Figure 39).
When you start hearing the world described in terms of a permanent era
of stability or the great moderation or growth and safety being ensured
by financial engineers or your country’s prosperity being due to cultural
Figure 39. The forty-three-year cycle of financial crises. Credit: Lukas


superiority, or the central banks proclaiming all is fine, it is time to run.

When Professor Pangloss discussed the best of all possible worlds, he was
only protecting Candide from seeing it.
Cognitive Failures
The CIA had a lot of information on Al Qaeda in early 2001 and
much forewarning of the 9/11 attack. In his book Rebel Ideas: The Power
of Diverse Thinking, Matthew Syed clearly illustrates the cognitive failures
that led the CIA to dismiss the threat. The CIA staff was predominantly
upper-middle-class, white, Ivy-League-educated men, and when all the
intelligence on Al Qaeda was filtered through their cultural lens an attack
was inconceivable. Like the CIA in 2001, so many fund managers, bank-
ers, and regulators have enormous power and resources. What frustrates
their mission is four cognitive failures that blind them to the threats and
opportunities. They make facts fit their biases, and when they don’t, find
other facts.
The first cognitive failure is a fallacy of composition, whereby we infer
that something must be true if all, or even some, parts of it are true: “Hy-
drogen (H) is not wet. Oxygen (O) is not wet. Therefore, water (H2O)
is not wet.” The fallacy of composition in financial regulations is that if
all the individual micro risks are kept under control, the entire financial
system is safe, implying that the financial system is the simple sum of all
the individual activities within it, so all the scientist has to do is study the
grains of sand on the beach (see Figure 38).
Suppose each and every bank is prudent—they are all Volvos, as I called
them earlier. They all do what they are supposed to do, with none taking
crazy risks. Still, they have to make risky investments like mortgages and
loans to small- and medium-sized enterprises; otherwise, they would not
be a bank. Some of those risky investments have a market price, but the
value of the majority can be determined only by a model, while risk in all
is measured by riskometers. The fallacy of composition means that even if
all the banks are prudent, the system is not safe.
Blame it on shock absorption, or rather the inability of prudent banks
to absorb shocks. Suppose some external shock arrives—Brexit or a tsu-
nami or Ukraine or Trump or an earthquake or China or Covid-19, or

just anything—so that the price of some assets held by the Volvo banks
falls. The consequence is that exogenous risk shoots up, inevitable by de-
sign since the riskometers are based on short-term historical fluctuations.
What is our Volvo supposed to do? Dispose of its riskiest assets and hoard
liquidity, of course, because that is prudent. But such selling will just
make the prices fall more, which makes exogenous risk increase further.
Cue more selling and a vicious feedback between falling prices, evapora-
tion of liquidity, and higher risk. If there is no buyer (because all possible
buyers are prudent), it can end only in tears (crisis). Nobody does any-
thing wrong. It is like passing a hot potato from one person to another:
they all pass it on in order to not get burned by it (Figure 40). Financial
stability is not achieved by making all the financial institutions prudent.
The second cognitive failure is invariance, the view that the financial
system does not change when observed and controlled. After all, the
world of physics is invariant, so why not the financial system? There is,
though, a crucial difference: endogenous risk. Anybody who interacts
with the financial system changes it, whether as asset managers, salespeo-
ple, regulators, journalists, investment bankers, university professors, or
finance ministers. Some significantly, others not so much. But while their
impact can be tiny, it is never zero. That may not matter much, and most
of us can safely behave as if risk were exogenous. But for the guardians of
financial stability and those worried about tail risk, like pension funds, the
Figure 40. The hot potato feedback. Credit: Lukas Bischoff/IllustrationX.

distinction is crucial. Accepting invariance means underestimating the

forces of large losses and instability.
The third cognitive failure is believing riskometers measure true risk.
I likened that in my earlier discussion of the myth of the riskometer to
seeing the riskometer as a device akin to the thermometer. That we can
somehow plunge a riskometer deep into the financial system and get an ac-
curate reading of risk. The problem is that the riskometers capture merely
a small part of the vast domain that risk is. They promise to summarize
all that risk into one number, and invariably a lot is lost in the process.
When employed to capture a bank’s risk or the risk of an entire financial
system, the measurements are essentially content-free. The riskometer is
comfortable to those in charge. They get a feeling that because the risk
dashboard only shines green, all is good. The riskometer misdirects them
to look only at the numbers that tell them what they want to hear.
The final cognitive failure is taking financial stability as an objective in
itself. It is not. Stability is only an instrument. The objective is sustainable
economic growth with not too many costly crises. The financial system is
a tool, and financial stability is important only insofar as it takes us toward
our objective. Policy makers who focus excessively on stability and see
it as the be-all and end-all of their mission are doing a disservice to the
society they are supposed to serve.
The four cognitive failures blind decision makers to what really mat-
ters. And that has many real world consequences. To begin with, when
we, with the best of intentions, aim to reduce or even annihilate risk,
we may just end up displacing it. It is possible to eliminate financial risk,
as Cuba and North Korea have shown: don’t have a financial system.
But that is an all-or-nothing approach, and we don’t like that, preferring
growth, which means risk. And because the financial system is so complex
there are plenty of places for taking said risk. Risk is like a balloon. By
reducing one type of risk, we may just displace it elsewhere, often into
the more shadowy parts of the system, where the controllers have less
oversight. Out of sight, out of mind.
The cognitive failures also make us neglect the network that links ev-
erybody in the financial system together. Okay, the regulators and bank-
ers will tell you that everybody is connected to everybody else, that the
network not only exists but also is of utmost importance. However, while

Figure 41. The silos. Credit: Copyright © Ricardo Galvão.
it is easy to acknowledge the network’s importance, it is much harder to

incorporate it into standard practices. I once heard a deputy governor of
a central bank complain about network analysis, saying that all such work
ever shows is that everybody is connected to everybody else, which he
already knew.
The reason the network gets ignored is all the silos. Everybody work-
ing in the financial system lives inside a silo (Figure 41). The financial
engineers are experts in their personal subareas, foreign exchange, loans
to southern European small- and medium-sized enterprises, American
subprime mortgages, treasury management, and all the myriad business
areas of the modern financial institution. Even the largest banks operate
solely within particular parts of the system. And the financial authorities
all live in silos. They are tasked with regulating insurance in a US state
or market conduct in Germany. There is no authority in charge of the
systemic risk of a single country, and certainly not the whole world. Be-
cause nobody has the mandate to control the network, nobody does. All
we have is a patchwork of government agencies regulating a small part of
it—their own silo—and they don’t like to collaborate with other silos. If
you think this can’t be true, ask the regulators about sharing data across
jurisdictions, and they will all tell you how much resistance data-sharing
meets. While ignoring the network helps the decision makers to sleep
well, the problem is that the network both dampens and amplifies shocks.

Some of the most powerful and destructive forces in the financial system
not only lie on the boundaries of the silos but also actively exploit them.
The most damaging consequence of the cognitive failures is short
termism—the dissonance of the short and long run. Almost every eco-
nomic outcome we care about is long term. Pensions, the environment,
house prices, education, you name it, all are about what happens years
and decades hence. The short run isn’t all that important. Day-to-day
fluctuations in stock prices or real estate values or loan portfolios or inter-
est rates don’t matter much to most of us. So, does the way we measure
and manage financial risk reflect the importance of the long run? By and
large, no. We proclaim we care about the long term but actually just end
up managing short-term risk.
The reason is simple. It is really hard to measure long-term risk because,
after all, extreme infrequent events are, by definition, very scarce. The
problem of measuring long-term risk is entwined with that of standard
risk management practices. Consider a sovereign wealth fund that cares
about very long-term risk, decades into the future, where such a time
perspective is written into its laws and mandates. However, the fund is
monitored quarterly, and if it performs poorly over a few quarters, ques-
tions are raised, bonuses may not be granted, the head of the fund could
be summoned to appear in front of a parliamentary committee, and some
people may be fired. That makes the managers of the sovereign wealth
fund care about quarterly, not the decennial, performance regardless of
what the legal mandate says.
And the implications are unfortunate. If the short-term risk dashboards
are reassuring, as they were in 2006, we may easily take on undesirable
levels of risk, oblivious to the dangers. The impact on financial markets
will be lower volatility and fatter tails—day-to-day fluctuations become
smaller, while the chance of catastrophic long-term outcomes increases.
There are real world consequences to all of that. Economic growth
suffers. Growth has been slowing in the developed world over the past
few decades, a phenomenon called secular decline (Figure 42).1 “Hold
on, Jón,” I suspect many of you will think when reading this. “There are
many causes of secular decline, it is poorly understood, and to blame the
cognitive failures is a bit far-fetched.” Fair enough, they are not the only
cause, but do make it worse. For the economy to grow, we need risk:

Figure 42. Secular decline in economic growth. Credit: Lukas Bischoff/

IllustrationX.
entrepreneurs who have easy access to capital at each level of a company’s

growth, from the very first incubation state all the way up to the IPOs and
beyond. Because loans to start-ups have high short-term risk, they attract
high capital charges, making it doubly costly for banks and other financial
institutions to finance the new companies that will drive future growth.
Meanwhile, the large, established corporations have high credit ratings,
good relations with banks, and easy access to financing. But they are
not growing very much. The very sectors that reliably deliver economic
growth are starved of funding. Earlier I gave an example of such depri-
vation when discussing a BBC Panorama program on how banks were
hurting the all-important small- and medium-sized enterprises, SMEs,
so vital to growth. While the program makers blamed predatory bank
behavior, the actual impact was caused by the Basel III capital regulations
that make it much more expensive to lend to the SMEs.
The Covid-19 Lesson in How the Financial Authorities

See Their Mission
What are policymakers to do? First and foremost, reduce uncertainty.
Do so by removing tail risks, and the perception of tail risks.
—Olivier Blanchard (2009)
The Covid-19 crisis in 2020 and the global crisis in 2008 are a
lens for looking at how the financial authorities think the financial system
should be controlled. While there is no single term or phrase that cap-
tures how they see their mission, I would like to propose “the modern

philosophy of financial regulations.” As far as I can tell, no policy author-

ity has expressed any view on what the purpose of regulations is beyond
the most vacuous. The single exception is the Bank of England, whose
senior policy makers manage fairly consistent views. Perhaps the financial
authorities don’t want to tell us what the modern philosophy is. After all,
they are masters of constructive ambiguity, trying very carefully to say as
little as possible in a lot of words. The official reason is, of course, that
constructive ambiguity is a necessary evil because of moral hazard: coor-
dinating the markets while keeping them guessing. I think the real reason
is more prosaic. The lack of clarity helps them to get the job done and pa-
per over differences of opinion. I have read many official documents, and
the deliberate lack of transparency shows up everywhere, even in places
where it shouldn’t. A few years ago I read the Basel III market risk regula-
tions, trying to understand how the main component of market risk was
meant to be calculated. It is a simple calculation and should be easy to
explain. The document was quite unclear despite spending many words
on the explanations, and I could not figure out how to do the calculation.
I would love for everybody to ask their central bankers and regulators how
they see financial regulations benefiting their country. Please let me know
if you get an answer. In the absence of clarity, I will use deduction. Having
read enough speeches and policy reports, participated in plenty of confer-
ences, and had many talks with regulators, I have arrived at the following
definition of how the financial authorities see the modern philosophy.
Definition: The modern philosophy of financial regulations as

indirectly expressed by the policy authorities:
1. Calm markets in distress with whatever it takes, in order to reduce

uncertainty
2. Afterward, identify those parts of the financial system that were at
the heart of the distress, and figure out what happened—lessons
learned. Aim to impose corrective regulations so as not to get a
repeat
3. By ensuring every financial institution behaves prudently, meaning
each individual risk is kept under control, the system is safe.
When I asked some friends working for the regulatory agencies what
they thought of my definition, they all agreed with the first two parts.

The third is more controversial and not how many people working for
the macroprudential authorities see it. (The microprudential regulators
would agree with my third part.) But my interlocutors agreed that was
only words and personal opinions. Actions were consistent with all three.
One of them added that the annual Risk Monitoring Exercise epitomized
the problem.
Covid-19 nicely demonstrates the modern philosophy in practice. The
initial Covid-19 shock was purely exogenous, so the virus was akin to
Archduke Ferdinand’s assassination, which set in motion the systemic
crisis of 1914. What happened after the virus hit was the typical endoge-
nous response. The epidemiological, social, and economic outcomes were
all the result of the interaction of the human beings who make all the
decisions—endogenous risk.
Covid did not cause a financial crisis. At best, we suffered turbulence.
Why? There are two schools of thought. The authorities maintain it is
thanks to both the post-2008 regulations leaving the financial system
highly resilient and also the swift policy responses in March and April
2020 (system bailout). In the words of the head of the Basel commit-
tee, Pablo Hernández de Cos, in 2021, “The global banking system has
remained broadly resilient. . . . The initial Basel III reforms, alongside an
unprecedented range of public support measures, are the main explana-
tions for this outcome.” Since the initial shock was purely exogenous, I
suspect the financial system would have had little trouble absorbing the
virus shock even without the bailout and Basel III. The FSB, with input
from all the leading financial authorities, published a report in November
2020 on what went wrong, how they responded, and lessons learned. The
document is called the Holistic Review of the March Market Turmoil and
sheds light on how the policy authorities see their mission:
The policy response was speedy, sizeable and sweeping. . . . Absent

central bank intervention, it is highly likely that the stress in the
financial system would have worsened significantly.
Some parts of the system, particularly banks and financial market
infrastructures, were able to absorb rather than amplify the macro-
economic shock, supported by the post crisis reforms.
The March turmoil has underscored the need to strengthen resil-
ience in the NBFI [nonbank financial intermediation] sector.2

In other words:
1. There was excessive market turmoil

2. Our job is to reduce uncertainty, and we did so successfully
3. The part of the system we regulated after the last crisis in 2008 did
well (a great job: we turned the banks into Volvos)
4. Unfortunately, we had to provide liquidity to the NBFIs to calm the
markets
5. So we will regulate the NBFIs to prevent a recurrence.
Left unsaid, but implied, is that the template for the regulations of the
NBFIs is bank regulations. So did it all work as well as the FSB suggests in
its Holistic Review? My collaborators and I have been looking at that very
question. Two days after the worst day in the equity markets in March
2020, we published a paper, “The Coronavirus Crisis Is No 2008,” on
how the Covid-19 crisis differed from the 2008 crisis. We concluded that
as the Covid-19 turmoil was not the same as that in 2008, a new policy
response was called for. In particular, liquidity injections would not be
very effective this time around.
Our subsequent research has continued to support these conclusions in
a paper titled “The Calming of Short-Term Market Fears and Its Long-
Term Consequences: The Central Banks’ Dilemma.” We took advantage
of a unique data set on option markets that allowed us to identify how
the financial markets’ fear of large losses changed owing to all the central
bank interventions.3 As the data is rich in time and space, spanning a large
number of stocks and countries and maturities from one week all the way
up to thirty years into the future, we have a comprehensive picture of how
the markets reacted to the Covid interventions.
The primary objective of the policy interventions in March and April
2020 was the immediate calming of market fear, but only in the short run.
Long-term fear would ideally not fall since that signals moral hazard. But
that is precisely what happened. The impact on long-term market fear,
even ten years into the future, was larger than at the short term. The les-
son the financial markets took from the interventions was that the central
banks stand ready to do what it takes. The market will be less vigilant and
take on more risk—moral hazard.
I can demonstrate that in more detail by taking one example out of our
paper: the Fed relaxed bank capital requirements under which the world’s

largest bank, JP Morgan, works. Fear in JPM’s stock price fell significantly
in response, both in the long and the short run. As the lowering of bank
capital requirements is explicitly designed to make banks take more risk,
one would have expected long-term fear to increase. However, the mar-
kets took it the other way, long-term fear fell—moral hazard increased.
That result, and others in our paper, lead me to conclude that the mod-
ern philosophy is destabilizing. Why? History provides guidance. The
economic profession went off the rails in the 1950s and 1960s, after John
Maynard Keynes died and before the pathbreaking work of Robert Lu-
cas. The prevailing view in those lost years was that one could control
the economy with a static framework, what I have termed Excelonomics.
Achieve policy objectives—high growth, low unemployment, low infla-
tion—by tweaking the parameters of a static economic model. The finan-
cial authorities and the governments loved it, as this way of doing eco-
nomics made them feel powerful, even omnipotent. Except it didn’t work.
The Lucas critique showed us why. Expectations matter, and economic
agents react to policies in a way that undermines the policy objective. All
we got was the word of the 1970s “stagflation”—inflation and stagnation.
Expectations also matter when it comes to macroprudential policy.
Banks were strongly affected by the 2008 crisis, so the policy authorities
sharply increased the intensity of regulations and the required levels of
bank capital. When Covid came, it all seemed to work splendidly since
the banks were hardly affected by the virus. But the risk now had spilled
over to the shadow banking system. You see, the economic agents that
comprise the financial system do not take regulations lying down. They
react to them, changing the financial system in the process. Once the
regulations take effect, they apply to a system that no longer exists: the
Lucas critique.
The authorities now ask the right question: “What can we do to pre-
vent a repeat?” but come to the wrong answer: “More regulations and
more control.” Fair enough, that might be the political outcome you
desire, but there are consequences. The first is that diversity suffers since
the lesson learned is that the nonbank sector needs to be brought under
official control that is similar to that of the banking sector. Financial in-
stitutions will become more like each other and the financial system more
procyclical. Systemic risk will increase.

Furthermore, the financial authorities gain even more power. But that
power comes at a cost. To begin with, the more power a state agency
has, the more democratic oversight it needs. We can’t have unelected
bureaucrats making decisions of fundamental importance to society when
they have no direct democratic legitimacy. But even worse, it makes
the financial authorities even more responsible for stability. They have
all the information and the power, and hence get blamed when things
go pear-shaped. And then it is much harder to resist bailing out private
institutions.
Then we have the very high cost of the new regulations, and who pays
the cost? The banks’ clients—us. Not necessarily a big problem in the
United States, where only a third of financial intermediation comes via
the banks. But in Spain, the home country of the head of the Basel com-
mittee (Governor Pablo Hernández de Cos), 96 percent of company fi-
nancing is provided by the banks. Spain is not exactly doing well econom-
ically and might think that the worst policy would be to further increase
the cost and reduce the availability of company loans.
And finally the political consequences. Bailouts are dangerous things,
and nobody likes them except the recipients. They drive populism and
moral hazard and undermine the credibility of the state. If the financial
authorities can do no better than design a setup requiring a bailout every
decade, neither the authorities nor the governments that empower them
appear to be competent or honest. Fortunately for the authorities, all the
Covid-19 bailouts passed unseen, lost in all the virus hoopla, unlike what
occurred in 2008.
The ultimate consequence of the modern philosophy of financial regu-
lations is to make us question the private financial system. Why not bring
it under direct state control?
The Magical Solutions

While the modern philosophy of financial regulations is the way
we control the financial system today, there is no shortage of alternative
proposals for how to fix the financial system. Start with cryptocurrencies.
While the crypto advocates are a broad church, a common view is that
the established institutions of the state and the private sector are corrupt

and cannot be trusted. Repeated crises with their bailouts and quantita-
tive easing mean it would be much better to replace the central banks
with algorithms—mathematics can be trusted. In steps bitcoin. It is a
complicated subject, meriting much more space than I have here, and
I have written extensively on it elsewhere.4 The specific form of money,
whether the fiat money we use today or cryptocurrencies, is not all that
important for the objectives of financial policy; what matters is the power
the financial authorities have and how they exercise it. So, cryptocurren-
cies are not the solution.
What, then, about their cousin, central-bank digital currencies, or
CBDCs. The idea is that the central banks create a new digital form of
the fiat money every country uses today. Perhaps as a token on a block-
chain. There was a lot of enthusiasm about CBDCs a few years ago. They
promised to solve so many of the problems we have with the financial
system, allowing targeted bailouts, the fine-tuning of the money supply,
and provision of ample information about what financial institutions are
up to, especially all the liquidity flows. The central bank governor can
then manage everything with their AI—BoB. Then the disadvantages be-
came clear. In their purest form, CBDCs mean the central bank controls
all the money in the economy because it controls the blockchain. Every
transaction is visible to it, so the central bank not only closely monitors
what citizens are up to but also makes all the loans. We don’t want that,
so today’s CBDC proposals aim merely to improve the payments system.
Here, the authorities are haunted by PayPal, which came out of nowhere
two decades ago, and by the time they woke up it was too late to do any-
thing about PayPal. So, the primary motivation for CBDCs today is to
forestall alternative payment systems, especially those under the control
of foreign companies. Certainly a worthwhile goal, but does nothing to
solve the problems I am discussing here.
If technology is not the solution, then what about politics? On the
libertarian wing of the political spectrum, the root of the problem is the
state, especially its regulations, bailouts, and currency mismanagement.
By forswearing regulations and bailouts, we will see crises, but not nearly
as many as now, since the reason we need bailouts is that the govern-
ment promises them. That is true, but the laissez-faire position makes
sense only in theory, not in practice, because in a democratic society we

demand the governments help us and politicians respond. It is simply not

credible to expect the politicians not to regulate and bail out the financial
system when needed. We need a solution that recognizes that reality.
And that brings me to socialism. If the state has to bail the financial
system out every time things go wrong, in 2008 and 2020 and 20??, why
not just have the state run the banks? Many commentators call for heavy
regulations, perhaps pointing to the golden era of Bretton Woods, when
the financial system was nailed down and there were no financial crises.
Impose very high capital requirements along with strict limits on activi-
ties and risk. Even just nationalize the whole system and let the govern-
ment run it. Certainly doable, but only at a cost. Finance is, after all, a
service. The safer the service becomes, the more costly it is. The more the
state gets involved with the provision of finance, the more political, bu-
reaucratic, and corrupt it becomes. Who pays for all that? We do, in lower
economic growth. I don’t think the socialist solution is viable either.
We need a balanced financial policy that protects us from the financial

system’s excesses and ensures that we are given good services and in-
novation at a reasonable cost. Society wants the financial authorities to
find a balance between safety and growth and for money managers to
focus on long-term performance and risk. That is not what we get now.
There is too much focus on short-term risk, not enough on long-term
risk, and certainly not enough on economic growth. We rely too much
on risk measurements that provide a veneer of scientific management of
risk. There are too many risk dashboards and excessive control of day-
to-day activities, and not enough thought given to long-term risk and
performance. The financial engineers who optimize the investment port-
folios and design modern regulations and risk management systems have
created this superpowerful microscope to look at all the individual risks.
They are like the scientist who is an expert at studying the grains of sand
on a beach but misses the incoming tsunami (see Figure 38).

14
What to Do?
Diversity stabilizes.
What to do about the financial system? Plenty of pundits tell us

they know how to make it deliver for us, with no crises, great investment
returns, and robust economic growth. Good luck with that. We may wish
to have our cake and eat it too, but there is always a trade-off between
safety and risk. All the easy solutions target the most visible exogenous
risk, ignoring the hidden dark forces behind tail risk and crises. All is not
lost, as there are sensible ways to get what we want. I have discussed the
issues in detail throughout this book and below distill them into five prin-
ciples for getting the best out of the financial system.
240

WHAT TO DO? 241
Focus on Endogenous Risk

The first principle is to recognize that the real threat comes from
endogenous risk, the dark forces that hide until it is too late. We too often
ignore endogenous risk and focus our efforts on the visible exogenous
risk that leads us down the garden path (Figure 43). Endogenous risk
captures the fundamental forces of instability and losses. Excessive lever-
age. The belief that liquidity is infinite in an interconnected financial sys-
tem that depends for its very existence on that infinite liquidity. Prudent
desires to protect oneself. The promise of government bailouts. Focus on
the visible while ignoring the hidden. Good control inside the silos and
willful disregard of the world outside. The hope that good times will last
forever. Low risk that encourages risk-taking, Minsky style.
Every crisis has some of these fundamentals at its core, like De
Neufville’s in 1763, John Pierpont Morgan’s in 1907, and Covid in 2020.
The problem for anybody concerned with risk is that these fundamental
causes of losses and instability are difficult to grapple with. It is hard, to
the point of impossible, to measure such endogenous risk. Nobody knew
about the dangers from subprime mortgages put into structured credit
products in the years before 2008 until it was too late to do anything
about them. The temptation then is to control the system by what we can
measure, not by what matters.
Endogenous risk lies dormant for years and decades until it is brought
to life by some exogenous shock. The Covid-19 virus triggered three fun-
damentals: liquidity, the precautionary principle, and bailout promises.
The system had operated as if liquidity were infinite, but then the vi-
rus made the liquidity providers cautious—poof, a liquidity crisis. The
same precautionary principle that made the Millennium Bridge wobble
in 2000 and investors to go on strike in 2007. Add to that the promise of
bailouts; the risk takers knew the financial authorities stood ready with
an open checkbook, so they took much more risk than they would have
otherwise. The Covid financial turmoil was inevitable. If the virus had not
triggered it, something else would have.
While only a few well-understood fundamental causes of crises exist,
the number of triggers is infinite. Assassination of an archduke in 1914,
virus in 1918 and 2020, sour bets on shipping technology as in 1866, a

242 WHAT TO DO?
small price drop in 1987 and 2007, end of war in 1763, etc., etc. The trig-
gers are as varied as they are numerous. The key difference between the
triggers and the fundamentals is visibility—the triggers are simple and
for all to see, while the fundamentals are obscure. And that very visibility
leads us down the wrong path, a trigger that causes a crisis today may
whimper out into nothing tomorrow. It would be much better to ignore
the triggers and focus on the fundamental driver of crises and bad perfor-
mance—endogenous risk.
Beware of the Riskometer and False Resilience

The second principle is to be wary of false resilience. It is easy
to set up some framework that tells us what we want to hear. All is fine
because that is what the risk dashboard tells us. Meanwhile the endog-
enous risk monster is laughing at us (Figure 44). If the risk dashboard
tells us risk is low, we want to take more risk. But that is false resilience.
The crisis in 2008 did not happen because all the bankers went risk mad
that year. No. It was all the crazy risk they took in the go-go years of the
early 2000s, and when 2008 came along there was nothing to do except
Figure 44. Monster under the bed. Credit: Copyright © Ricardo Galvão.

WHAT TO DO? 243
mitigate the worst. We were already in a crisis. And then, once something
bad happens, we like to learn the lessons. Figure out what went wrong so
it can never happen again—closing the barn door after the horse escapes.
That also brings false resilience. The forces of instability congregate in the
dark areas where no one is looking, so by its very definition, the danger
will emerge somewhere else next time.
The main driver of false resilience is that the risk we measure, all the
micro risks, tends not to be the type of risk we most care about. The rea-
son is the riskometer, the magical device that pops out measurements of
financial risk when plunged deep into the bowels of the financial system.
Financial regulations, risk control, and portfolio management depend on
the riskometer, more and more every day. Why? Because it is seen as
scientific and objective, helping decision makers collapse a complicated
problem into a small set of precise numbers on a risk dashboard.
And that is where things go wrong. The riskometer is not nearly as
scientific and objective as its proponents think. They have been waylaid
by precise scientific instruments, like the thermometer, which allows us
to measure temperature as accurately as we want, in real time. There is
only a single unambiguous notion of what temperature is, and it is easy
to use the thermometer to control temperature with real-time feedback
techniques. When the temperature is too high, turn down the thermo-
stats; that’s why the problem of keeping the risk manager’s office at a
steady 72°F or 22°C is easy. We can’t implement such feedback mecha-
nisms with most financial risk, even if plenty have tried to. Why? To begin
with, there is no uniform view of what is important and hence what risk
to target. Is it day-to-day volatility? Tail risk, the hefty price-drops that
cause sudden big losses, bankruptcies, and crises? Or the slow drip-drip-
drip movements of prices downward, with no significant fluctuations and
no discernible tail risk, but prices that only go south? Is it the chance of a
pension not delivering on that comfortable retirement fifty years hence?
The likelihood our country will suffer a systemic crisis next year?
Each concern calls for a different concept of risk—what is risk depends
on what we care about. It is unfortunate that the easiest risk to measure,
and hence the one most widely used, is short-term, day-to-day events—
volatility or its close cousins Value-at-Risk and Expected Shortfall. These
have little or nothing to say about tail risk or crises or the solvency of

244 WHAT TO DO?
your pension fund. Astonishingly, the very financial regulations and risk
management practices that are meant to keep banks safe, protect our
pensions, and prevent crises are so often based on nothing more than
day-to-day price fluctuations.
Even after picking a concept of risk, we are left with the problem of
measurement. There are dozens of competing techniques out there that
deliver widely different measurements of the same risk, with no clear way
to discriminate among them. All purporting to be state-of-the-art, and
each with its own groupies. And even then we have measured only the
risk in a single asset, perhaps a stock, a loan, or a derivative. The next
step is even harder, the aggregation of risk across time and space. How
to go from all the micro risks to the portfolio, department, bank, and
the system, today and years and decades into the future. The more we
aggregate risk, the less accurate the result is. There is a subtle point at
work here. While it is obviously true that systemic risk is the aggregate of
all the individual micro risks, that conceptual notion does not mean we
know how to do the calculations. It is a common problem in science. You
can know everything about a human being’s physiology and biology and
know nothing about them as a person. We can’t easily aggregate risk be-
cause of the complex interactions between all the individual risks. In real
life, outside of the universe of the risk modeler, the strongest connections
between risk factors manifest themselves only in times of extreme stress.
They are simply not seen otherwise.
Why? Liquidity is the most obvious reason. Liquidity is ample most of
the time, even seemingly infinite. But it is by and large not measurable
and has the annoying tendency of evaporating when most needed, in
times of stress. Becoming the common crisis factor that affects all assets
and liabilities, exposing all the hidden connections we never knew existed,
until its too late. If we measure risk in normal times, we underestimate
each asset’s risk and especially how it relates to other assets, because the
very factor that makes them strongly related—liquidity—is not visible.
So while it is easy to come up with a number for the aggregate risk of a
bank or a country or even the entire world, the calculations’ accuracy is
very low. I think many readers will disagree since it is standard practice in
finance to do precisely that. Yes, it is easy to come up with a number for
aggregate risk. It is not so straightforward to do it accurately. Those who

WHAT TO DO? 245
propose doing so might be confused by physical systems. Math captures

all in physics and not in finance. I saw the problems with such aggregate
risk measures firsthand a few years ago when I joined forces with three co-
authors, Kevin James, Marcela Valenzuela, and Ilknur Zer, in studying the
accuracy of measures of systemic risk. We wrote two papers, “Model Risk
of Risk Models” and “Can We Prove a Bank Guilty of Creating Systemic
Risk? A Minority Report,” and, by looking at all the leading indicators of
systemic risk, concluded that they were very inaccurate, and could not be
recommended for use.
Like so many other tools, the riskometer is helpful when used appro-
priately, like controlling risk on the trading floor. It is dangerous when
in the hands of someone who either does not know about its limitations
or, what I think is much more common, willfully ignores all the warning
signs because it helps them do their job. They might argue “We have to
control the amount of risk. The riskometer gives us usable measurements.
It is not perfect but allows us to exercise some control.” But by doing so
only reinforces the illusion of control.
Ultimately, the riskometer drives false resilience, the view that because
the risk dashboard flashes green, we have risk under control. The conse-
quences of false resilience are unfortunate. Investment management that
focuses excessively on short-term risk, like daily fluctuations or quarterly
performance. Investors get deceived. Perhaps when their portfolio man-
agers tell them the portfolio risk is $15,000 or there is a 25 percent chance
of a loss over the coming year. Misleading, since the accuracy of such
assertions depends on the reliability of the risk calculations and the rel-
evance of that risk to the investor. Furthermore, because the asset manag-
ers calculate the risk measurements, they have the incentive to choose a
riskometer that presents their promises in the best possible light. Which
often means one based on short-term exogenous risk.
While money managers may care only about the short run, most inves-
tors don’t. Pension funds, sovereign wealth funds, family offices, insurance
reserves are all concerned with losses spanning years or even decades. The
problem is that these investors rely on outsiders to manage their invest-
ments, which means performance monitoring. And that monitoring is
more often than not done with risk dashboards populated by the output
of riskometers that emphasize the short term. Investors may then look

246 WHAT TO DO?
for false diversification. Perhaps buying private equity funds, which do

not exhibit much short-term volatility even though they are cointegrated
with equity markets in the long term.
The macroprudential regulators are similarly waylaid by false resilience.
By the very definition of their existence, they are meant to concern them-
selves only with long-run risk, act countercyclically, slow the system down
in good times, and stimulate it when hit by shocks. There is much more
to macropru than the measurement of risk, but if your job is to think
about bad outcomes years and decades into the future, you need some
way to measure long-term resilience. And the go-to tool for that is the
riskometer. Combine recent fluctuations in the financial markets with ac-
counting information, where the relevant variables are chosen by how
well they forecast past stress events. Hope that informs the future. The
regulators might think they are doing a splendid job when they ensure all
the micro risks are low—the fallacy of composition. And when they mea-
sure aggregate financial-system risk—like ECB’s CISS—measurement
error is very high. The systemic risk dashboards may flash green when the
actual systemic risk is high, giving oxygen to the forces of instability, or
just as likely do the opposite, making the authorities clamp down when
systemic risk is low, the economy suffering, so a stimulus is called for.
Beware of false resilience. True resilience focuses on the fundamental
drivers of losses and instability, not on measures of risk or fixing what
went wrong last time.
Remember the Objectives

The third principle is to always keep your objectives in mind.
Maximum return with acceptable risk for investors and economic growth
with reasonable financial stability for the regulators. Consider the en-
tirety of the problem together: think globally, not locally.
It is easier for investors to do so, as profit maximization is at the front
of everybody’s thoughts. The concern here is the dissonance of the short
and long term, since it is all too common to focus excessively on the short
term, by using risk measurements that capture short-term fluctuations,
while ignoring the tail risks that can be so damaging to long-term in-
vestment returns. The objectives of investors can end up being sacrificed

WHAT TO DO? 247
for scientific risk management techniques that only mitigate short-term

threats to investment performance, statistical techniques that promise ac-
curacy, sold by spurious precision—expected return on this portfolio is
25.234 percent over the next two years while risk is 12.228 percent. Most
risk calculations use a riskometer based on some model the vendor has
created, one that is tailor-designed to signal what most benefits the ven-
dor, not the client. Those investors who are concerned with long-run
risk, which means most of them, should dispense with performance con-
trols that push to the short term and insist that money managers also care
about long-run risk and the risk that matters to them.
It is much harder for the regulators to keep the objectives in mind,
because they don’t even agree on what they are. As I argued earlier, the
objective of financial policy should be to maximize long-term economic
growth and minimize the cost of crisis:
cumulative long-run economic growth

Maximize ————————————————––
cumulative cost of crises and recessions
The policy authorities should clearly express what their objectives are,
why they are regulating, and what they aim to accomplish. High-level
statements are welcome, like those made by Chairman Powell, emphasiz-
ing low, steady inflation and high employment, but such clarity is mostly
absent in the actual documents that outline policy actions.
And straying firmly into controversy, unlike so many other commenta-
tors, I think a key purpose of central banks and other financial authori-
ties is to help the financial system operate at higher risk levels than it
otherwise would do. As a general rule, more risk means more economic
growth, and by helping the system to operate at higher levels of risk than
it otherwise could safely do, we all benefit.
Blow Up the Silos

The fourth principle is that the authorities should focus their ef-
forts on both maximizing growth and minimizing the cost of crises. We
don’t do that today because the policy makers live in silos. They care deeply
about their own corner of the world but are instructed to ignore the rest of
the universe. A few years ago I gave a presentation at a central bank and had

248 WHAT TO DO?
the privilege of sitting next to the governor over dinner. We got to discuss-
ing the topic of silos, and I asked him about how central bank policy since
the crisis in 2008 affects inequality. He told me that on a personal level he
was deeply concerned about inequality, but from his central bank’s point
of view it was irrelevant because inequality was not in its legal mandate.
Policy makers optimize locally, not globally. Researchers of financial
stability are usually no better, even though they should be unencumbered
by the silos. I have heard a lot of conference presenters claim that “the
financial system is dangerous, I have identified the most important risks,
and this is how you measure and control them. If you follow my sugges-
tion, we meet our objective.” Derisking the very part of the system they
have spent years studying.
The silos can lead to strange outcomes. The Bank of England decided
in June 2017 to tighten capital constraints on commercial banks because
it thought they were taking on too much risk. That month it also opted
to keep interest rates low to encourage banks to make more risky loans to
small- and medium-sized enterprises in order to stimulate the economy.
These policy decisions are obviously contradictory. While I have no idea
how the bank came to this point, I can think of only two reasons: Either
it wanted to please two separate political interests and was cynical enough
to recognize that nobody would spot the contradiction. Or the decisions
were made by two distinct parts of the bank—the financial stability and
monetary policy divisions—and they didn’t coordinate with each other.
Overcoming the silo mentality requires that financial policy be done
in a more holistic way, and the only authority able to make that happen
is the government. It can mandate the necessary interagency and inter-
silo cooperation, force the various policy authorities to optimize globally
and not locally. I can hear the objections: “Government agencies should
focus on a single objective; if they have multiple mandates, one will lose
out.” “It will politicize the central banks.” “It will lead to muddled and
confusing policy-making.” “We will get inflation.” Every agency will find
a host of reasons for why this is both a horrible idea and practically im-
possible. Balderdash. It is certainly doable, and one country, Singapore,
leads the way.1 The Monetary Authority of Singapore collaborates closely
with both the Ministry of Finance on fiscal policies and the Ministry of
National Development (land supply policies). Singapore has been quite

WHAT TO DO? 249
successful in meeting my objective, high long-run growth and a stable

financial system. There is no inherent reason such global optimization
cannot be done elsewhere. All it takes is will.
Diversity to the Rescue

The enemy of stability and good investment performance is uniformity.
The final principle for getting the best out of the financial sys-
tem is to embrace diversity, the most potent force of financial stability
and good investment performance. Suppose some shock hits the markets,
perhaps a virus like Covid-19 or the fight between the Reddit investors
and the short-selling hedge funds in January 2021. If I react to a shock
by buying and my buddy Ann by selling, our reactions cancel each other
out—together, we create countercyclical random noise. If, instead, we
both buy or sell, we procyclically amplify the price movements. We are
procyclical when we see and react to the world in the same way, and
countercyclical when we don’t. The reason why Baron Rothschild was
so prescient when he wrote two and half centuries ago that the best time
to buy property was in the middle of a Civil War—buying in crises stabi-
lizes. We need the Rothschilds, the Soroses and Buffetts, all the sovereign
wealth funds, someone who sees buying opportunities during turmoil.
For such individuals and entities to exist, they have to be unencumbered
and free to invest the way they see best.
The enemy of financial stability and good, stable, long-term invest-
ment returns is uniformity. The more similar financial institutions are, the
higher systemic risk becomes, because they will amplify the same shocks
and inflate the same bubbles. The Millennium Bridge wobbled because
the pedestrians on the bridge acted like a troop of soldiers, not like civil-
ians. And that is what always happens in times of turmoil. Market par-
ticipants become much more uniform in outlook and action than they
usually would be.
Not many disagree. The relationship between diversity and systemic
risk is well understood. But that is theory. Practice is different, and the
incentives of the banker and the regulator push toward uniformity. There
are good and bad reasons for this. The most apparent is best practices.

250 WHAT TO DO?
Nothing wrong with that. Nobody likes to use the second best. But there
is a problem when it comes to risk. Because best practices mean using the
one state-of-the-art riskometer and risk management technique, all will
see and react to risk in the same way—uniformity, not diversity.
Increasing returns to scale also erode diversity. Banking favors the large
because the fixed costs of financial services are so huge. The bigger the
banks become, the cheaper it is to service all the complex needs of their
large clients. Competitive forces drive mergers. While sometimes bemoan-
ing the falling number of banks, the financial authorities encourage it in
practice, happy to use mergers when resolving crises and failing banks.
Financial regulations further favor uniformity. There are two costs
a bank has to pay when complying with regulations, variable and fixed
costs—understanding how the regulatory apparatus works, knowing the
legal environment, and the like. While well meaning and generally useful,
there is a dark side. Because the fixed cost is substantial, the bigger the
bank, the cheaper it is, per unit of size, to comply—increasing returns to
scale (Figure 45).
The problem of antidiversity regulations is especially pernicious in Eu-
rope because the financial authorities there have to deal with transna-
tional regulations of national banks and the politics of managing unruly
countries on a European level. That means they need to be seen as pro-
viding a level playing field—why Basel III applies to all banks instead of
Figure 45. The fixed and variable cost of regulations. Credit: Lukas

WHAT TO DO? 251
only to the largest, which would be much more appropriate and what the
rest of the world wanted. The result is uniform rules that treat all banks,
large and small, in the same way. Since we need complex rules for the
largest, the cost of complying favors the large.
Bank-based financial systems also favor uniformity. In the United States
only about one-third of lending to companies is done by banks; the rest
happens via the bond markets and various other nonbank entities that
intermediate funds from savers to companies. In the rest of the world
about 90 percent of credit comes from banks, over 80 percent in the
United Kingdom, 92 percent in Germany, and 96 percent in Spain. And
even then the typical bank in the United States is much smaller relative to
the size of the economy than banks in other countries. There are about
4,400 banks in the United States and only 200 in Japan, even though the
Japanese economy is about half the size of the U.S. economy. The large
number of banks in the United States plus their small share of overall
credit make the US way of funding companies much more resilient, and
much easier for the country to whip its banks into shape after a crisis, as
it did following 2008. Bank-based financial systems provide less financing
for innovative and risky companies, the cost of finance is high, and it is
very hard to regulate the banks without imposing significant economic
costs. The stillborn European Capital Markets Union was meant to help,
but the power of the incumbent interests was just too great.
The financial authorities face a thorny way forward. They have to juggle
a lot of issues and are subject to ferocious lobbying. But they should em-
brace diversity. Actively encourage new financial institutions, especially
those whose business models are different from everybody else’s. Regu-
late financial institutions for consumer protection, but don’t control all
the micro risks. Unfortunately, the financial authorities are not keen on
diversity. Their rules actively get in the way of start-ups and new busi-
ness models. The start-ups have to comply with regulations suited for
the largest banks and, before getting licenses to practice, having to set up
the myriad of functions that make up the modern financial institution;
like the board, capital, management, IT system, and compliance. A slow,
cumbersome, and very expensive process. And at the end of the day there
is no guarantee they will get a license. The reason the regulatory pro-
cess is so antidiversity and anti-start-up is that the regulators are worried

252 WHAT TO DO?
about making a mistake; they are like the Chinese air traffic controllers
I discussed earlier. Focused on risk to themselves, not on the benefits to
society. That risk aversion means the licensing process favors uniformity
and the incumbents.
What is lacking is risk culture. The financial authorities could do well
by learning from their counterparts in other fields, like aviation. The air-
line industry is regulated with a view to simultaneously maximize the
benefit to society and keep risk under control, and we see the outcome.
The cost of flying is steadily falling while safety gets better every year. The
central banks and regulators need such risk culture.
The financial authorities should be made to explain how the job they
are doing benefits the rest of us. Telling us what the objective of what
they are doing is, outlining how they are achieving that goal, and how it
all fits in with what the other agencies are doing.
Diversify the regulators. If we put a single regulator in charge of every-
thing—the super regulator so common today—we end up with a gov-
ernment agency that prefers uniformity, one that shares the goals of the
incumbent interests and loathes what is different. We need competition
between regulators, so we get agencies that both regulate and defend
their part of the industry, protecting heterogeneity along the way.
While there are no silver bullets when it comes to the financial system,
some ways of controlling it are better than others. The worst thing is to
fight risk, the preferred option of the financial authorities. The financial
system is like the Hydra, and though the authorities can cut off as many
heads as they want, they will always grow back. It is much better not to
try the impossible, instead just taking advantage of the inherent forces of
stability and the five principles for how to deal with the system.
The first is to recognize that the real danger is endogenous risk, not the
type of risk we usually measure. There are few fundamental reasons why
the financial system does not do what we want it to do, and that is where
our attention should be. It is all too easy to focus on the triggers of crises
and bad investment performance, as they are there for all to see while the
fundamental causes are hidden.
The second is to be aware of riskometers and false resilience. It is easy
to convince oneself all is okay, that we are fully hedged against the worst
tail events and comfortably safe since the financial authorities have sys-

WHAT TO DO? 253
temic risk under control. It is another matter altogether to make it hap-

pen. Riskometers misdirect us away from the real threats, especially when
we ask them to do the impossible, creating aggregated risk measures for a
portfolio, bank, or the entire financial system. True resilience focuses on
the fundamental drivers of losses and instability, not on measures of risk
or fixing what went wrong last time.
The third principle is to remember the objectives. There is a reason
why we regulate finance, it should not be regulation for regulation’s sake,
and the authorities should clearly show how what they are doing helps
meeting their ultimate objectives, solid economic growth and financial
stability. Fund managers should use risk management methods that re-
flect the needs of their clients, typically long-term tail risk, and not just
measure and manage short-term risk.
The fourth principle is to blow up the silos, think globally, not locally.
The financial system is a network connecting all, but most who work
within it are instructed to ignore the network. Their job is invariably con-
fined to a silo, and while they may care deeply about their silo, they don’t
pay much attention to the world outside them. We pay for that because
the forces of bad investment performance and crises take advantage of the
silos, operating on the boundaries where no one has oversight.
Finally, the best way to bend the financial system to one’s will is to
embrace diversity. We need a new risk culture like the one that works
so well for the airline industry, so that the regulators are incentivized to
allow new types of financial institutions to set up shop. The more differ-
ent from each other the institutions of the financial system become and
the more heterogeneous the regulations are, the more stable and diversi-
fied the system will be, and hence the better investment portfolios will
perform. The only way to overcome the forces of uniformity is for the
government to force the regulators to embrace diversity.
What gets in the way of us getting what we want out of the financial
system is all the incumbent interests, the private firms that want to protect
their franchises and the risk-averse regulators who want to demonstrate
they are doing their job. The special interests are antidiversity, and that is
where the government can step in. It can mandate the financial authori-
ties to do what is necessary. All it takes is political will, and most of us
will benefit.

Notes
1. Riding the Tiger

1. Box, “Science and Statistics,” 791–99.
2. Goodhart, “Public Lecture at the Reserve Bank of Australia.”
2. Systemic Risk
1. Schnabel and Shin, “Liquidity and Contagion: The Crisis of 1763,” 929–68, and
Quinn and Roberds, “Responding to a Shadow Banking Crisis: The Lessons of 1763,”
1149–76.
2. International Monetary Fund, Bank for International Settlements, and Financial
Stability Board, Guidance to Assess the Systemic Importance of Financial Institutions,
Markets and Instruments: Initial Considerations, 2.
3. From Hans Christian Andersen’s nineteenth-century tale “The Emperor’s New
Clothes.”
4. Carville, television interview.
5. Black-Scholes refers to the statistical techniques for pricing options developed by
Fischer Black and Myron Scholes in 1973, earning them the Nobel Prize in 1997.
6. Black, “Hedging, Speculation, and Systemic Risk,” 6–8.
7. As quoted in Hoyt, The Cyclopedia of Practical Quotations.
8. Holder, Senate Judiciary Committee testimony.
3. Groundhog Day
1. Capra, It’s a Wonderful Life.
2. The primary source of historical banking crises is Reinhart and Rogoff, This Time
It’s Different. I combine that with the IMF crisis database, Laeven and Valencia, “Sys-
temic Banking Crises Revisited.”
3. Prince, interview in the Financial Times.
255

256 NOTES TO PAGES 35–105
4. Martin, “Address before the New York Group of the Investment Bankers Asso-
ciation of America.”
5. As quoted in Day, A Wonderful Life: S&L HELL: The People and the Politics be-
hind the $1 Trillion Savings and Loan Scandal.
4. The Risk Panopticon

1. Who watches the watchers?
2. Chapter 13 in www.globalfinancialsystems.org/slides.
3. Goodhart, “Risk, Uncertainty and Financial Stability.”
4. New York Times, “Paul A. Volcker, Fed Chairman Who Waged War on Inflation,
Is Dead at 92.”
5. The Myth of the Riskometer

1. As he told Dunbar in “Value-at-Risk Inventor Longerstaey on the Perils of Over-
simplification.”
2. Ibid.
3. Greenspan, “Maintaining Financial Stability in a Global Economy.”
4. Researchers have known about the fat tail problems for over half a century, at
least since the work of the future Nobel Prize–winner Eugene Fama and his PhD
advisor, Benoit B. Mandelbrot.
5. Jansen and de Vries, “On the Frequency of Large Stock Returns: Putting Booms
and Busts into Perspective.”
6. http://tylervigen.com/spurious-correlations.
7. Daníelsson, “The New Market-Risk Regulations.”
6. Ideas Matter
1. Keynes, The General Theory of Interest, Employment and Money.
2. Mises, The Ultimate Foundation of Economic Science.
3. Hayek, “The Use of Knowledge in Society.”
4. See Rosenzweig, “Robert S. McNamara and the Evolution of Modern
Management.”
5. Yankelovitch, Corporate Priorities: A Continuing Study of the New Demands on
Business.
6. Rumsfeld, U.S. Department of Defense (DoD) news briefing.
7. Josiah Stamp is recounting a story from Harold Cox, who quotes an anonymous
English judge.
8. Goodhart, “Risk, Uncertainty and Financial Stability”; Shackle, Keynesian
Kaleidics.
9. Buffett, “Berkshire Hathaway 2011 Letter to Shareholders.”
10. Minsky, “The Financial Instability Hypothesis: An Interpretation of Keynes and
an Alternative to ‘Standard’ Theory.”
11. Yellen, press conference.
12. Goodhart, “Public Lecture at the Reserve Bank of Australia.”

NOTES TO PAGES 107–184 257
13. Lucas, “Econometric Policy Evaluation: A Critique,” in The Phillips Curve and
Labor Market.
7. Endogenous Risk
1. Keynes, The General Theory of Interest, Employment and Money.
2. Crockett, Marrying the Micro- and Macro-Prudential Dimensions of Financial
Stability.
3. The average return in excess of the risk-free rate divided by volatility.
4. For a lucid description of events, see Lowenstein, When Genius Failed: The Rise
and Fall of Long-Term Capital Management.
5. Ibid.
6. Ibid.
8. If You Can’t Take the Risk, Change Riskometers

1. Fitzpatrick, “J.P. Morgan to SEC: That Model Change Doesn’t Count as Change”;
Henry and LaCapra, “JP Morgan and Other Banks Tinker with Risk Models.”
2. Flitter, “Emails Show JP Morgan Tried to Flout Basel Rules—U.S. Senate.”
3. Suppose a bank buys a put option, that is, the right but not the obligation to sell
an asset in the future at a price called the strike price, X1. The bank then writes a put
option for a lower strike price called X2. The effect of the former is to provide risk
against the price of the asset falling below X1, while the second exposes the bank to
large losses if the price of the asset drops below X2.
4. CDO stands for collateralized debt obligations, a structured credit product used
to house subprime mortgages.
5. If the likelihood of losses is below the Value-at-Risk probability, typically one day
out of a hundred, then the risk is recorded as zero.
6. Comment on Daníelsson, “Risk and Crises.”
9. The Goldilocks Challenge

1. Described in Elliot, Overend & Gurney, A Financial Scandal in Victorian
London.
2. Real total return, including dividends and adjusting for inflation.
10. The Risk Theater

1. Details on the specifics can be found on my website at www.modelsandrisk.org/
Swiss.
2. The Basel committee came out with some adjustments to Basel III in December
2017. While sometimes dubbed Basel IV, they do not have (or deserve) that formal
moniker.
3. Daníelsson and Keating, “Valuing Insurers’ Liabilities during Crises: What EU
Policymakers Should Not Do”; Daníelsson et al., “A Prudential Regulatory Issue
at the Heart of Solvency II”; Daníelsson et al., “Countercyclical Regulation in Sol-
vency II: Merits and Flaws.”

258 NOTES TO PAGES 185–248
4. I discussed the systemic importance of asset managers in a blog piece, Daníelsson

and Zigrand, “Are Asset Managers Systemically Important?”
11. The Uniformity, Efficiency, and Stability Trilemma

1. https://fred.stlouisfed.org/series/USNUM.
2. El-Naggar, “In Lieu of Money, Toyota Donates Efficiency to New York Charity.”
12. All about BoB

1. Kaku, The Future of Quantum Computing.
2. Kurzweil, Communication to Futurism.
3. Kahn, “To Get Ready for Robot Driving, Some Want to Reprogram Pedestrians.”
4. Haldane, “Maxwell Fry Annual Global Finance Lecture: Managing Global Fi-
nance as a System.”
5. EURISKO is certainly not the only such example. For a list of similar examples,
see Krakovna, “Specification Gaming Examples in AI.”
13. The Path Not to Take

1. Moving average economic growth for high-income countries, those in which
2015 GNI per capita was $12,476 or more. Data source: World Bank, World Develop-
ment Indicators.
2. Financial Stability Board, Holistic Review of the March Market Turmoil.
3. Kindly provided by the data vendor IHS Markit.
4. See modelsandrisk.org/cryptocurrencies.
14. What to Do?

1. Monetary Authority of Singapore, “Macroprudential Policies: A Singapore Case
Study,” 321–27.

Bibliography
Adams, Douglas. The Hitchhiker’s Guide to the Galaxy. London: Pan Books, 1978.
Admati, Anat, and Martin Hellwig. The Bankers’ New Clothes: What’s Wrong with
Banking and What to Do about It. Princeton: Princeton University Press, 2014.
Akerlof, George A. “What They Were Thinking Then: The Consequences for Macro-
economics during the Past 60 Years.” Journal of Economic Perspectives 33, no. 4
(2019): 171–86.
Aliber, Robert Z., and Charles P. Kindleberger. Manias, Panics, and Crashes: A His-
tory of Financial Crises. New York: Palgrave Macmillan, 2015.
Bagehot, Walter. Lombard Street: A Description of the Money Market. London: H. S.
King, 1873.
Bank for International Settlements. Report on the Regulatory Consistency of Risk-
Weighted Assets for Market Risk. Basel: Bank for International Settlements, 2013.
Basel Committee on Banking Supervision. Amendment to the Capital Accord to Incor-
porate Market Risks. Basel: Basel Committee on Banking Supervision, 1996.
———. Fundamental Review of the Trading Book: A Revised Market Risk Framework.
Basel: Basel Committee on Banking Supervision, 2013.
BBC. “Did the Bank Wreck My Business?” Panorama, 2014.
Bernstein, Peter L. Against the Gods: The Remarkable Story of Risk. New York: John
Wiley, 1996.
Bevilacqua, Mattia, Lukas Brandl-Cheng, Jón Daníelsson, Lerby Ergun, Andreas
Uthemann, and Jean-Pierre Zigrand. “The Calming of Short-Term Market Fears
and Its Long-Term Consequences: The Central Banks’ Dilemma.” SSRN Elec-
tronic Journal, 2021.
Bitner, Richard. Confessions of a Subprime Lender: An Insider’s Tale of Greed, Fraud,
and Ignorance. New York: John Wiley, 2008.
Black, Fischer. “Hedging, Speculation, and Systemic Risk.” Journal of Derivatives 2
(1995): 6–8.
259

260 BIBLIOGRAPHY
———, and Myron Scholes. “The Valuation of Option Contracts and a Test of Mar-
ket Efficiency.” Journal of Political Economy 27 (1973): 399–418.
Blanchard, Olivier. “(Nearly) Nothing to Fear but Fear Itself.” The Economist, 2009.
Borio, Claudio. “The Macroprudential Approach to Regulation and Supervision.”
VoxEU.org, 2009.
Box, George. “Science and Statistics.” Journal of the American Statistical Association
(1976): 791–99.
Buffett, Warren. “Berkshire Hathaway 2011 Letter to Shareholders.” 2011.
———. “Why Stocks Beat Gold and Bonds.” Fortune, 2012.
Calomiris, Charles W., and Stephen H. Haber. Fragile by Design: The Political Origins
of Banking Crises and Scarce Credit. Princeton: Princeton University Press, 2014.
Capra, Frank, dir. It’s a Wonderful Life. 1946.
Carney, Mark. “Ten Years On: Fixing the Fault Lines of the Global Financial Crisis.”
Banque de France Financial Stability Review, no. 21 (2017).
Carville, James. Television interview. 1992.
Chwieroth, Jeffrey M., and Jón Daníelsson. “Political Challenges of the Macropru-
dential Agenda.” VoxEU.org, 2013.
Crockett, Andrew. Marrying the Micro- and Macro-Prudential Dimensions of Finan-
cial Stability. Basel: BIS, 2000.
Daníelsson, Jón. “The Emperor Has No Clothes: Limits to Risk Modelling.” Journal
of Banking and Finance 26 (2002): 1273–96.
———. “The Myth of the Riskometer.” VoxEU.org, 2009.
———. Financial Risk Forecasting. New York: John Wiley, 2011.
———. “Risk and Crises.” VoxEU.org, 2011.
———. Global Financial Systems: Stability and Risk. London: Pearson, 2013.
———. “The New Market-Risk Regulations.” VoxEU.org, 2013.
———.“Towards a More Procyclical Financial System.” VoxEU.org, 2013.
———. “What the Swiss FX Shock Says about Risk Models.” VoxEU.org, 2015.
Daníelsson, Jón, Paul Embrechts, Charles A. E. Goodhart, Con Keating, Felix Muen-
nich, Olivier Renault, and Hyun Song Shin. An Academic Response to Basel II.
London: LSE Financial Markets Group, 2001.
Daníelsson, Jón, Kevin James, Marcela Valenzuela, and Ilknur Zer. “Model Risk of
Risk Models.” Journal of Financial Stability 23 (2016).
———. “Can We Prove a Bank Guilty of Creating Systemic Risk? A Minority Re-
port.” Journal of Money Credit and Banking 48 (2017).
Daníelsson, Jón, Frank de Jong, Roger Laeven, Christian Laux, Enrico Perotti, and
Mario Wuthrich. “A Prudential Regulatory Issue at the Heart of Solvency II.”
VoxEU.org, 2011.
Daníelsson, Jón, and Con Keating. “Valuing Insurers’ Liabilities during Crises: What
EU Policymakers Should Not Do.” VoxEU.org, 2011.
Daníelsson, Jón, Roger Laeven, Enrico Perotti, Mario Wuthrich, Rym Ayadi, and
Antoon Pelsser. “Countercyclical Regulation in Solvency II: Merits and Flaws.”
VoxEU.org, 2012.

BIBLIOGRAPHY 261
Daníelsson, Jón, and Robert Macrae. “The Appropriate Use of Risk Models: Part I.”
VoxEU.org, 2011.
———. “The Appropriate Use of Risk Models: Part II.” VoxEU.org, 2011.
———. “The Fatal Flaw in Macropru: It Ignores Political Risk.” VoxEU.org, 2016.
———. “The Dissonance of the Short and Long Term.” VoxEU.org, 2019.
Daníelsson, Jón, Robert Macrae, Dimitri Tsomocos, and Jean-Pierre Zigrand. “Why
Macropru Can End Up Being Procyclical.” VoxEU.org, 2016.
Daníelsson, Jón, Robert Macrae, and Andreas Uthemann. “Artificial Intelligence and
Systemic Risk.” Journal of Banking and Finance (2021).
Daníelsson, Jón, Robert Macrae, Dimitri Vayanos, and Jean-Pierre Zigrand. “The
Coronavirus Crisis Is No 2008.” VoxEU.org, 2020.
Daníelsson, Jón, and Hyun Song Shin. “Endogenous Risk.” In Modern Risk Manage-
ment: A History. London: Risk Books, 2003.
Daníelsson, Jón, Hyun Shin, and Jean-Pierre Zigrand. “Endogenous Extreme Events
and the Dual Role of Prices.” Annual Reviews 4 (2012).
Daníelsson, Jón, Marcela Valenzuela, and Ilknur Zer. “Learning from History: Vola-
tility and Financial Crises.” Review of Financial Studies (2018).
Daníelsson, Jón, and Chen Zhou. “Why Risk Is So Hard to Measure.” Amsterdam:
De Nederlandsche Bank NV, 2016.
Daníelsson, Jón, and Jean-Pierre Zigrand. “Are Asset Managers Systemically Impor-
tant?” VoxEU.org, 2015.
Day, Kathleen. A Wonderful Life: S&L HELL: The People and the Politics behind the
$1 Trillion Savings and Loan Scandal. New York: W. W. Norton, 1993.
Diamond, Douglas W., and Philip H. Dybvig. “Bank Runs, Deposit Insurance, and
Liquidity.” Journal of Political Economy 91 (1983): 401–19.
Dunbar, Nicholas. “What JP Morgan’s Release of VaR Has in Common with Sex
and Computer Viruses.” 2012. http://www.nickdunbar.net/articles/what-jp
-morgans-release-of-var-has-in-common-with-sex-and-computer-viruses/.
———. “Value-at-Risk Inventor Longerstaey on the Perils of Oversimplification.”
Bloomberg Briefs, 2012.
Elliot, Geoffrey. Overend & Gurney, a Financial Scandal in Victorian London. Lon-
don: Methuen, 2006.
El-Naggar, Mona. “In Lieu of Money, Toyota Donates Efficiency to New York Char-
ity.” New York Times, 2013.
Engle, Robert. “Autoregressive Conditional Heteroskedasticity with Estimates of the
Variance of the United Kingdom Inflation.” Econometrica 50 (1982): 987–1007.
Engels, Friedrich. Socialism: Utopian and Scientific. London: Swan Sonnenschein, 1880.
European Banking Authority. “EBA Interim Report on the Consistency of Risk-
Weighted Assets in the Banking Book.” 2013.
Fama, Eugene. “Mandelbrot and the Stable Paretian Hypothesis.” Journal of Business
36, no. 4 (1963): 420–29.
———. “Are Markets Efficient?” Posted at https://review.chicagobooth.edu/
economics/2016/video/are-markets-efficient, 2016.

262 BIBLIOGRAPHY
Financial Stability Board. Holistic Review of the March Market Turmoil. Financial Sta-
bility Board, 2020.
Fitzpatrick, Dan. “J.P. Morgan to SEC: That Model Change Doesn’t Count as
‘Change.’” Wall Street Journal, 2013.
Flitter, Emily. “Emails Show JP Morgan Tried to Flout Basel Rules—U.S. Senate.”
Reuters, 2013.
Friedman, Milton, and Anna Jacobson Schwartz. A Monetary History of the United
States: 1867–1960. Princeton: Princeton University Press, 1963.
Gissurarson, Hannes. Twenty-Four Conservative-Liberal Thinkers, Part II. Brussels:
New Direction, 2021.
Goodhart, Charles A. E. “Public Lecture at the Reserve Bank of Australia.” 1974.
———. Risk, Uncertainty and Financial Stability. London: Financial Markets Group,
London School of Economics, 2008.
———. The Basel Committee on Banking Supervision: A History of the Early Years
1974–1997. Cambridge: Cambridge University Press, 2011.
Greenspan, Alan. Discussion at Symposium: Maintaining Financial Stability in a
Global Economy, at the Federal Reserve Bank of Kansas City (1997): 54.
Haldane, Andy. “Managing Global Finance as a System. Maxwell Fry Annual Global
Finance Lecture,” Birmingham University, 2014.
Hayek, Friedrich von. “The Use of Knowledge in Society.” American Economic Re-
view 35, no. 4 (1945): 510–30.
Henry, David, and Lauren Tara LaCapra. “JP Morgan and Other Banks Tinker with
Risk Models.” Reuters, 2013.
Hernández de Cos, Pablo. “Basel III Implementation in the European Union.”
BCBS Speech, 2021.
Holder, Eric. US Senate Judiciary Committee testimony. 2013.
Honohan, Patrick, and Daniela Klingebiel. “The Fiscal Cost Implications of an Accom-
modating Approach to Banking Crises.” Journal of Banking and Finance 26 (2003).
House of Commons library. “Financial Services: Contribution to the U.K. Econ-
omy.” Briefing Paper Number 6193, 2017.
Hoyt, Jehiel Keeler. The Cyclopedia of Practical Quotations. London: Funk & Wag-
nalls, 1907.
International Monetary Fund, Bank for International Settlements, and Financial Sta-
bility Board. Report to G20 Finance Ministers and Governors. Guidance to Assess
the Systemic Importance of Financial Institutions, Markets and Instruments: Initial
Considerations (2009): 2.
Jansen, Dennis, and Casper G. de Vries. “On the Frequency of Large Stock Returns:
Putting Booms and Busts into Perspective.” Restat 73 (1991): 18–24.
Johnson, Rian, dir. Star Wars: Episode VIII—The Last Jedi. 2017.
Kahn, Jeremy. “To Get Ready for Robot Driving, Some Want to Reprogram Pedes-
trians.” Bloomberg, 2018.
Kaku, Michio. The Future of Quantum Computing. At https://www.youtube.com/
watch?v=YgFVzOksm4o (2011).

BIBLIOGRAPHY 263
Keynes, John Maynard. A Treatise on Probability. London: Macmillan, 1921.

———. The General Theory of Interest, Employment and Money. London: Macmillan,
1936.
Kindleberger, Charles P. Manias, Panics, and Crashes: A History of Financial Crises.
3d ed. London: John Wiley, 1996.
King, Mervyn. “Inflation Report Press Conference.” Bank of England, 2007.
Knight, Frank Hyneman. Risk, Uncertainty and Profit. New York: Houghton Mifflin,
1921.
Kohn, Meir. “Early Deposit Banking.” Hanover, NH: Department of Economics,
Dartmouth College, 1999.
Krakovna, Victoria. “Specification Gaming Examples in AI.” https://vkrakovna
.wordpress.com/2018/04/02/
Kubrick, Stanley, dir. 2001: A Space Odyssey. 1968.
Kurzweil, Ray. Communication to Futurism. 2017.
Laeven, L., and F. Valencia. “Systemic Banking Crises Revisited.” IMF Working Pa-
per, 2018.
Lenat, Douglas B. “EURISKO: A Program That Learns New Heuristics and Domain
Concepts: The Nature of Heuristics III: Program Design and Results.” Artificial
Intelligence 21, nos. 1, 2 (1983): 61–98.
Lewis, Michael. The Big Short: Inside the Doomsday Machine. Detroit: Gale Cengage
Learning, 2010.
———. Flash Boys. New York: Penguin, 2015.
Lowenstein, Roger. When Genius Failed: The Rise and Fall of Long-Term Capital
Management. New York: Random House, 2000.
Lucas, Robert. “Econometric Policy Evaluation: A Critique.” In The Phillips Curve
and Labor Markets, ed. K. Brunner and A. Meltzer, 19–46. 1976.
Magnus, George. Red Flags: Why Xi’s China Is in Jeopardy. New Haven: Yale Univer-
sity Press, 2019.
Malkiel, Burton Gordon. A Random Walk down Wall Street. New York: W. W. Nor-
ton, 1973.
Mandelbrot, Benoit B. “The Variation of Certain Speculative Prices.” Journal of Busi-
ness 36 (1963): 392–417.
Markowitz, Harry. “Portfolio Selection.” Journal of Finance 7 (1952): 77–91.
Martin, William McChesney, Jr. “Address before the New York Group of the Invest-
ment Bankers Association of America.” 1955.
Marx, Karl. Das Kapital: Kritik Der Politischen Oekonomie. Verlag von Otto Meissner,
1867.
Merler, Silvia, and Jean Pisani-Ferry. “Who’s Afraid of Sovereign Bonds?” Bruegel
Policy Contribution, 2012.
Minsky, Hyman. “The Financial Instability Hypothesis: An Interpretation of Keynes
and an Alternative to ‘Standard’ Theory.” Nebraska Journal of Economics and Busi-
ness 16 (1977): 5–16.
———. Stabilizing an Unstable Economy. New Haven: Yale University Press, 1986.

264 BIBLIOGRAPHY
———. The Financial Instability Hypothesis. Annandale-on-Hudson, NY: Jerome

Levy Economics Institute, 1992.
Mises, Ludwig von. The Ultimate Foundation of Economic Science. Princeton: Van
Nostrand, 1962.
Monetary Authority of Singapore. “Macroprudential Policies: A Singapore Case
Study.” BIS Papers, no. 94 (2017): 321–27.
New York Times. “Paul A. Volcker, Fed Chairman Who Waged War on Inflation, Is
Dead at 92.” 2019.
OECD. G20/OECD INFE Report on Adult Financial Literacy in G20 Countries.
OECD, 2017.
Orlik, Thomas. China: The Bubble That Never Pops. Oxford: Oxford University Press,
2020.
Popper, Karl. The Open Society and Its Enemies. London: Routledge, 1945.
Prince, Charles (“Chuck”). Interview in the Financial Times, 2007.
Quinn, Stephen, and William Roberds. “Responding to a Shadow Banking Crisis:
The Lessons of 1763.” Journal of Money, Credit and Banking 47, no. 62 (2015):
1149–76.
Reinhart, Carmen, and Kenneth Rogoff. This Time Is Different: Eight Centuries of
Financial Folly. Princeton: Princeton University Press, 2009.
Rodgers, Kevin. Why Aren’t They Shouting? London: Penguin, 2016.
Rosenzweig, Phil. Robert S. McNamara and the Evolution of Modern Management.
Cambridge: Harvard Business Review, 2010.
Rossi, Hugo. “Mathematics Is an Edifice, Not a Toolbox.” Notices of the AMS 43,
no. 10 (1996).
Rumsfeld, Donald. U.S. Department of Defense (DoD) news briefing, 2002.
Russel, Stuart. Human Compatible. London: Allen Lane, 2019.
Santayana, George. “Reason in Common Sense.” In The Life of Reason, 1:284. New
York: Scribners, 1921.
Schlosser, Eric. Fast Food Nation: What the All-American Meal Is Doing to the World.
Boston: Houghton Mifflin, 2001.
Schnabel, Isabel, and Hyun Song Shin. “Liquidity and Contagion: The Crisis of
1763.” Journal of the European Economic Association 2, no. 6 (2004): 929–68.
Schneier, Bruce. “Beyond Security Theater.” New Internationalist, 2009.
Schumpeter, Joseph. Capitalism, Socialism and Democracy. London: Harper & Broth-
ers, 1942.
Shackle, George L. S. Keynesian Kaleidics. Edinburgh: Edinburgh University Press,
1974.
Shapiro, Fred R. The Yale Book of Quotations. New Haven: Yale University Press, 2006.
Silver, David, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang,
Arthur Guez, Thomas Hubert, et al. “Mastering the Game of Go without Human
Knowledge.” Nature (2017).
Skalweit, Stephan. Die Getreidehandelspolitik und Kriegsmagazinverwaltung Preußens
1756–1806. Berlin: Acta Borussia, 1931.

BIBLIOGRAPHY 265
Smith, Adam. An Inquiry into the Nature and Causes of the Wealth of Nations. Lon-
don: W. Strahan, T. Cadell, 1776.
Soto, Hernando de. The Mystery of Capital. New York: Basic Books, 2000.
Syed, Matthew. Rebel Ideas: The Power of Diverse Thinking. New York: Flatiron Books,
2019.
Triana, Pablo. The Number That Killed Us. London: John Wiley, 2011.
Tukey, John W. “The Future of Data Analysis.” Annals of Mathematical Statistics 33
(1962): 1–67.
UBS. Shareholder Report on UBS’s Write-Downs. UBS, 2008.
Viniar, David. “Goldman Pays the Price of Being Big.” Financial Times interview,
2007.
Whitehouse, Kaja. “One ‘Quant’ Sees Shakeout for the Ages—‘10,000 Years.’ ” Wall
Street Journal, 2007.
Yankelovich, David. Corporate Priorities: A Continuing Study of the New Demands on
Business. D. Yankelovich Inc., 1972.
Yellen, Janet. Press conference. Federal Reserve Board, 2014.

Index
Page numbers in italics refer to illustrations.
Adams, Douglas, 221 Arup Engineering, 110, 111–12

Admati, Anat, 59 Asian debt crisis (1998), 8, 37, 178
Adoboli, Kweku, 217–18 asset management, 185–86
Against the Gods (Bernstein), 51 asteroids, 4
AIG (American International Group), Austria, 38–39
184–85, 226
airlines, risk culture of, 252, 253 backtesting, 89
Akerlof, George A., 96–97 Bagehot, Walter, 151–52, 178
Aladdin (risk management system), 197, bailouts, 13, 44, 48, 147, 224, 238; during
211 Covid-19 pandemic, 47, 162, 234, 237,
alchemy, 139 241; in Greece, 43, 47, 95; for Long
Aliber, Robert, 153 Term Capital Management, 125; as
AlphaGo Zero, 209 middle-class benefit, 166; as moral
Al Qaeda, 227 hazard, 162, 163; popular resentment
Amazon, 154 of, 237; as socialistic practice, 203
Amazon Web Services (AWS), 196–97 Banca Monte dei Paschi, 166
Amsterdam, 6, 12, 35 Banco Ambrosiano, 62–63
Andropov, Yuri, 2 Banco de México, 69
arbitrage, 68, 123, 126, 135–36, 174 Banco Popular, 165
ARCH (autoregressive conditional bank capital, 58–60, 176, 180
heteroskedasticity), 86–87, 143, 195 Bank for Commerce and Credit Inter-
Archegos Capital, 144 national (BCCI), 63–64
Argentina, 149, 166 Bank for International Settlements, 8,
artificial intelligence (AI), 5, 159, 196, 65, 113
197, 205–22 bank–government doom loop, 45
267

268 INDEX
Bankhaus Herstatt, 61–62 Black, Fischer, 16

Banking Act (1933), 38 Black Death, 12
Bank of England, 13–14, 20, 39, 40–41, Black Monday (1987), 9, 118–19
45, 210, 233; contradictory policies of, BlackRock, 54, 197
248; crisis of 2008 bungled by, 158; Blanchard, Olivier, 232
establishment of, 55; in panic of 1866, Bohr, Nils, 72
151–52; politicizing of, 183; as regula- Borio, Claudio, 177, 178
tor, 63–64, 248 Box, George, 3, 74, 90
Bank of Ireland, 47 Bretton Woods system, 28, 60, 61, 170,
Bank of Italy, 62 239
Bank of Japan, 65 Brexit, 15, 78, 183
Bank of United States, 31–32 bubbles, 32, 103, 104, 116, 121–22,
bankruptcy, 17, 27, 31, 33, 163 152–54, 180
banks: fragility of, 29; government debt Buffett, Warren, 78, 102, 137
held by, 44; necessity of, 28, 55; Burrows, Jonathan, 54
number of, in U.S. vs. Japan, 251; as Bush, George H. W., 15
political target, 27; retail clients of,
39; runs on, 29–30; wholesale fund- Calomiris, Charles, 161–62
ing of, 39 Calvi, Roberto, 62
Barcelona, 56 Candide (Voltaire), 225
Barings, 218 Das Capital (Marx), 58
Basel Capital Accords, 90, 195, 201; cost capital conservation buffer, 60
of complying with, 250–51; delays capital ratio, 59, 65, 135
in implementing, 66, 202, 220; capital structure arbitrage, 68, 135–36,
entrepreneurship impeded by, 232; 174
Expected Shortfall incorporated in, Capone, Al, 23
135; leverage ratio prescribed by, 176; Carney, Mark, 168, 188
limitations of, 65–67, 173–75; opac- Castello, Francesch, 56, 160
ity of, 233; traffic-light rule in, 134; central bank digital currencies (CBDCs),
Value-at-Risk incorporated in, 76, 81 238
Basel Committee on Banking Supervi- chess, 209
sion, 64–67, 169, 199, 200, 202–3 Chicago Board Options Exchange, 123
Bear Stearns, 226 China, 154, 183; air traffic control in, 160;
behavioral economics, 102 stock market closure in, 14
bell curve, 82 China: The Bubble That Never Pops
Bentham, Jeremy, 3, 53 (Orlik), 153
Berkshire Hathaway, 137 Chinese walls, 143
Berlin, 7–8, 36 Chwieroth, Jeff, 166, 181
Berlin airlift, 99l CIA (Central Intelligence Agency), 227
Bernstein, Peter, 51 circuit breakers, in markets, 14
big data, 207 CISS (Composite Indicator of Systemic
The Big Short (Lewis), 20, 127 Stress), 79, 179, 246
Bitner, Richard, 142 clawbacks, 57

INDEX 269
clearing, 11 deflation, 31
climate change, 85 Deloitte & Touche, 63
Clinton, Bill, 15 Delta Works, 85
Coca-Cola, 154 de Neufville, Leendert Pieter, 6–7, 10,
cognitive failure, 227–32 241
collateral debt obligations (CDOs), 70, deposit insurance, 37, 38, 39, 41–42
133, 134, 139–42, 193, 212, 218 Depository Trust and Clearance Corpo-
conduits, 70–71 ration (DTCC), 115
Confessions of a Subprime Lender (Bitner), Deutsche Bank, 176
142 devaluation, 106
confidence intervals, 79 developing countries, 17–18
confirmation bias, 2 de Vres, Casper, 85
convergence trading, 123 Diamond, Douglas, 39
corporate finance, 237, 251 dinosaur extinction, 4, 113
countercyclical capital buffer, 60 diversity, in financial system: forces
Covid-19 pandemic, 8–12, 113, 114, 179, opposing, 190–91, 249–50, 251; mea-
183, 193, 232; bailouts during, 47, 162, surement of, 191; stability linked to,
234, 237, 241; banking regulation 5, 188, 190, 193, 249, 253
and, 29, 174–75; Chinese response to, divorce-and-margarine fallacy, 89–90
14; financial crisis of 2008 compared Dominican Republic, 34
with, 21, 235; financial markets dur- Dondelinger, Albert, 64
ing, 153, 234; Icelandic economy buf- dot-com bubble (1990s), 154
feted by, 156; macroprudential policy Dunford, Joseph, 181
during, 181, 187; modeling of, 91–92, Dybvig, Philip, 39
100; political response to, 15; safety dynamic replication, 119
vs. growth epitomized by, 21
Credit-Anstalt, 39 “Early Deposit Banking” (Kohn), 56
credit default swaps (CDSs), 141–42, “Econometric Policy Evaluation”
184–85 (Lucas), 107
credit rating, 44, 67, 140 econometrics, 96
Credit Suisse, 144 elasticity of demand, 118
crisis, crises: causes of, 31–34; corruption Engels, Friedrich, 203
linked to, 33–34; costs of, 45–47; fre- Engle, Robert, 86–87, 195, 197
quency of, 9, 32, 33, 50; IMF database Enron, 225
of, 9, 46; textbook example of, 10; Erdogan, Recep, 182
types of, 8 ergodicity, 95, 101, 145, 195
Crockett, Andrew, 121 Ernst & Young, 63, 133
cryptocurrency, 223, 237–38 EURISKO (artificial intelligence sys-
Cuban Missile Crisis (1962), 2 tem), 215
Cyprus, 36, 42–43, 45, 66 Eurodollars, 69
European Banking Authority, 199, 200,
data snooping, 88 202–3
de Cos, Pablo Hernández, 234, 236 European Capital Markets Union, 251

270 INDEX
European Central Bank, 22, 31, 42; financial liberalization, 35–36

quantitative easing by, 45; risk dash- financial repression, 66
board of, 78, 88, 92, 127, 214; signals Financial Risk Forecasting (Daníelsson),
missed by, 79–80, 88 196
European Commission, 42, 43 Financial Services Authority (FSA), 40,
European Union, 66 41, 64, 182
evergreening, 49 Financial Stability Board (FSB), 8, 169,
EVT (riskometer), 172–73 184, 185, 234
exchange rates, 77, 170–71 financing, types of, 102–3
exchange traded funds, 185 Finland, 46, 47
expected shortfall, 75, 135, 197–200, 243 Flash Boys (Lewis), 217
exponentially weighted moving average flooding, 84–85
(EWMA, riskometer), 131–32, 172, Food Bank for New York City, 193
197, 198 Food Standards Agency (UK), 161
externalities, 163 football, 192
extreme value theory (EVT), 85, 198 Foster, Norman, 110
Fragile by Design (Calomiris and
fallacy of composition, 4–5, 227, 246 Haber), 161–62
false resilience, 242–43, 245–46, 252 France, 38–39, 49
Fama, Eugene, 256n4 (chapter 5) Franz Ferdinand, archduke of Austria,
Fast Food Nation (Schlosser), 130 10–11, 120, 193, 234
fat-tail risk, 82–85, 231 French, Fama, 154
Federal Aviation Administration (FAA), French and Indian War, 6–7, 120
161 Friedman, Milton, 32, 148–49, 194
Federal Deposit Insurance Corporation Friedrich II, king of Prussia, 7–8
(FDIC), 38, 39 Fusarium Oxysporum, 189–90
Federal Reserve, 16, 32, 125, 149–50; dur-
ing Great Depression, 148–49, 152 Galvão, Ricardo, 128
fiat money, 30, 238 gambling, 55
Financial Conduct Authority (FCA), 54, Gamestop crisis (2021), 11, 115
64, 159, 211 GARCH (riskometer), 143, 172, 198, 201
financial crisis of 2008, 1, 4, 8, 9, 15, Gauss, Johann Carl Friedrich, 82
127, 152, 203, 213, 225, 232; causes of, General Motors, 163
67, 133, 153, 188; Covid-19 pandemic General Theory of Interest, Employment
compared with, 76, 81; credit default and Money (Keynes), 96, 98, 101,
swaps linked to, 184–85; European 112–13, 158
Central Bank’s blundering linked Georgiou, Andreas, 120, 149
to, 79–80, 88; government policies Germany, 49, 61–62, 183, 251
linked to, 16–17; Lehman Brothers Giffen, Robert, 118
bailout and, 47; liquidity during, 32, Giffen gods, 118
149, 170, 178; regulatory shift follow- gilts, 45, 151
ing, 29; subprime mortgages linked Gissurarson, Hannes, 98–99
to, 16, 19–20, 88, 133, 141, 153, 180, Glass-Steagall Act (1935), 148, 191
183, 241 Global Alpha fund, 137–38

INDEX 271
Global Financial Systems (Daníelsson), ICBC (Industrial and Commercial Bank

135, 164 of China), 176–77
Goldman Sachs, 23, 137 Iceland, 36, 44; financial crisis in (2008),
Goodhart, Charles, 3, 20, 60, 64, 66, 22–23, 24, 37, 42, 149, 153, 155–56,
101, 105, 107–8 220
Goodhart’s law, 3, 105–8, 216 IKB Bank, 70–71
Goodwin, Rupert, 129 Iksil, Bruno, 139
Google, 154, 209 India, 52–53
Great Depression, 1, 8, 28, 178; bank Indonesia, 46
regulation during, 143; bank runs inflation: bailouts linked to, 164; central
during, 31, 38, 41; Federal Reserve’s banks’ focus on, 182; excessive preoc-
passivity during, 148–49, 152; margin cupation with, 4, 157–58; intractabil-
buying linked to, 116; market failure ity of, 31; misconceptions surround-
linked to, 97, 98; money supply col- ing, 69; in 1970s, 38, 70; recent ebb
lapse during, 32; volatility during, 86 in, 179; sterilization to counter, 171;
Great Moderation, 16, 69–70, 103, 139, unemployment vs., 105–6, 107
225 ING, 177
Greece, 42–43, 47, 66 injuries, in football and lacrosse, 192
Greensill Capital, 144 input–output model, 99, 100–101, 102
Greenspan, Alan, 16, 82, 186 insurance companies, 184–86
Gros Michel banana, 189–90 interconnectedness, 8–9, 11–12, 15, 241
Groundhog Day (film), 26 internal ratings-based (IRB) method,
G20 group, 49, 53, 65, 169, 184 196
International Monetary Fund (IMF), 8,
Haber, Stephen, 161–62 42–44, 101, 225
Hagan, Patrick, 130 invariance, 228
Haldane, Andy, 210 investment horizons, 75
Hamburg, 7, 36 Ireland, 27, 42, 44, 47, 165
Hayek, Friedrich August von, 94, Italy, 45, 49, 62, 166
97–99, 101, 114, 187, 214 It’s a Wonderful Life (film), 29–30, 31
hedge financing, 102–3
hedge funds, 122–23, 127 James, Kevin, 245
Hellwig, Martin, 59 Japan, 47, 48–49, 183, 251; banking crisis
Henderson, James, 101 in, 65, 152–53, 186–87
Herstatt, Jowan David, 61 JP Morgan: during Covid-19 pandemic,
Herzberg, Elaine, 169 236; in London Whale scandal,
high-frequency trading, 217 130–31, 146; Morgan Stanley split
Holder, Eric, 25 from, 148; solvency of, 59, 176, 177;
Hong Kong, 183 Value-at-Risk developed by, 80–81,
Honohan, Patrick, 46–47 195, 196, 197
HSBC, 25, 159
Human Compatible (Russel), 207 Kaizen, 193, 195, 225
hurricanes, 82 Kaku, Michio, 208
Hurricane Sandy, 193 Keating, Charles, 38

272 INDEX
Kerviel, Jérôme, 143 219; in less democratic countries, 183;

Keynes, John Maynard, 94, 96–99, 101, winners and losers under, 182
112–13, 125, 158, 236 Maginot Line, 157
King, Mervyn, 20 Magnus, George, 153
Klingebiel, Daniela, 46 Malaysia, 183
Knight, Frank, 94, 95, 97, 99, 101 Malkiel, Burton, 125–26
Kohn, Meir, 56 Mandelbrot, Benoit B., 256n4 (chapter
K-shaped crises, 8 5)
Kurzweil, Ray, 209 Manias, Panics, and Crises (Aliber), 153
margarine-and-divorce fallacy, 89–90
Laeven, Luc, 9 margin buying, 116–18, 179
Leeson, Nick, 217–18 margin calls, 117, 124
Lehman Brothers, 47, 71, 78, 175, 226 Markowitz, Harry, 193–95, 197
Leland, Hayne, 118–19 mark-to-market feedback, 14
Lenat, Douglas, 215 Martin, William McChesney, Jr., 149–50
Lenin, Vladimir, 209 Marx, Karl, 58–59
Leontief, Wassily, 99, 100, 102 McNamara, Robert, 100
leverage, 15, 58, 116 mean–variance model, 194–95
leverage ratio, 59, 136, 176 Mencken, H. L., 89
Lewis, Michael, 20, 127, 217 Mendeleev, Dmitri, 139
LIBOR, 23–24, 159 Merton, Robert C., 122
linear programming, 99–100 Mexico, 69, 149
liquidity, 30, 31, 112, 164, 241; during microprudential regulation, 191; by
crisis of 2008, 32, 149, 170, 17832, 149, artificial intelligence, 211, 216, 218–19,
170, 178; difficulty of measuring, 244 222; Basel Accords aimed at, 173, 174,
Little Britain (television program), 208 177
Lloyds Bank, 174 The Milkmaid (Vermeer), 8
loan-to-value ratio, 180, 183 Millennium Bridge, 109–12, 114, 116, 117
London, 12–14, 15, 35, 55, 79 Minority Report (film), 168
London School of Economics, 20 Minsky, Hyman, 3, 94, 102, 103–4
Longerstacy, Jacques, 81 Minsky effect, 69–70, 104, 186, 193, 216
Long Term Capital Management Mises, Ludwig von, 96
(LTCM) crisis (1998), 9, 122–23, model risk, 91–92
124–25 Moldova, 34
Lucas, Robert, 107, 236 momentum investing, 124
Luxembourg, 36, 37, 62, 64 A Monetary History of the United States
(Friedman and Schwartz), 32
MA (riskometer), 172 Monetary National Income Analogue
machine learning, 207 Computer (MONIAC), 106
Macrae, Robert, 21, 144, 173, 183, 212 monetary policy, 20, 105, 158, 179, 181
macroprudential regulation, 98, 169, money market funds, 69
174, 177–78, 184, 226, 246; artificial monoculture, 190
intelligence ill-suited to, 212, 216, Monte Carlo experiments, 90–91

INDEX 273
Moore, Gordon, 208 Phillips curve, 106

Moore’s law, 205, 208 Ponzi financing, 102–3
moral hazard, 22, 37, 151, 162, 163, 233, Popper, Karl, 203–4
235–37 portfolio insurance, 118
Morgan, John Pierpont, 148, 241 Portfolio Selection (Markowitz), 194
Morgan Stanley, 148 Portugal, 42, 54
MSCI Inc., 197 Powell, Jerome, 181, 247
Murray, Bill, 26 precautionary principle, 112, 241
The Mystery of Capital (Soto), 28 prices, 95, 102, 114–16, 118
Price Waterhouse, 63
Netherlands, 36, 84–85, 153, 177 Prince, Chuck, 34–35
network effects, 12 principal-agent problem, in financial
New York City, 15 trading, 143
Nifty Fifty stocks, 153, 154 private equity, 246
Nixon, Richard M., 157 procyclicality, 67, 193, 194, 212; Basel
Nomura, 124 regulations and, 175; financial mono-
normal distribution, 82, 84 culture linked to, 191, 202, 203, 213,
Northern Rock, 20, 39–41, 80, 226 236, 249; macroprudential regulation
Norway, 47, 183 and, 179, 181, 186–87; of regulatory
The Number That Killed Us (Triana), 81 laxity, 156
Prudential Regulation Authority (PRA),
Occupy Wall Street, 27 64
offshore banking, 42–44 public transport, 54–55
online banking, 42, 214 put options, 113–14, 118–19
The Open Society and Its Enemies (Pop-
per), 203 quant funds, 137–39
options, 133–34 quantitative easing, 15, 45, 238
Organisation for Economic Coopera-
tion and Development (OECD), 66 Rain Man (film), 209
Orlik, Thomas, 153 A Random Walk Down Wall Street
ossification, 202, 226 (Malkiel), 126
Overend & Gurney (O&G), 150–51 Rato, Rodrigo, 23
RBS (Royal Bank of Scotland), 174
Panama disease, 190 RBS-ABN AMRO, 175
Panopticon, 3, 53–55, 71, 216 real estate, 3, 10, 103, 153, 182–83
Pareto, Vilfredo, 85 reasoned probability, 102
Pascal, Blaise, 74, 225 Rebel Ideas (Syed), 227
PayPal, 238 recession, 31, 141
Peltzman, Sam, 192 Reconstruction Finance Corporation, 41
pensions, pension funds, 58–59, 76, 199, Reform Act (1832), 28
228–29, 231, 245 RegTech, 211
Petrov, Stanislav, 2 Regulation Q, 69
Phillips, Bill, 105–7 regulation vs. supervision, 52–53

274 INDEX
regulatory capture, 161 Schumpeter, Joseph, 163

relative value trading, 123 Schwartz, Anna, 32, 148–49
religion, 55 scientific socialism, 203–4
reserve requirements, 30, 33 secular decline, 231, 232
Richelieu, Armand Jean du Plessis, duc Securities and Exchange Commission
de, 22 (SEC), 217
risk: aggregating, 244–45; aversion to, securitization, 20, 80
57, 160, 161, 252; difficulty of measur- self-driving cars, 209–10, 217, 219
ing, 3, 4, 57, 72–77, 82–85, 92, 126, sell-on-loss rules, 119
127, 129, 179, 186, 196, 231, 241, 244; September 11 attacks, 227
endogenous, 3–4, 67, 68, 109–27, Seven Years’ War, 6–7, 120
214, 225, 228, 241–42; exogenous, Shackle, George Lennox Sharman,
3–4, 68, 113, 114, 195, 214, 226, 228, 101–2, 107
252; fat-tail, 82–85, 231; growth shadow banking, 8, 236
linked to, 17, 21, 25; incentives for, Sharpe, William F., 121
56; latency of, 76–77; manipulable Sharpe ratio, 121
measurements of, 108, 127, 133–36, Shin, Hyun Song, 66, 113, 121
143, 146; perceived vs. actual, 121; sys- shipping, 150
temic, 2, 8–10, 16, 18, 20, 55, 177–78, significance testing, 79
184; uncertainty vs., 95–96, 195 silos, 230
“Risk, Uncertainty and Financial Stabil- Singapore, 36, 42, 183, 248–49
ity” (Goodhart), 107–8 Skalweit, Stephan, 6
risk competition, 192 Smith, Adam, 58, 126
RiskMetrics, 197, 211 Socialism (Engels), 203
risk theater, 169, 172, 175, 185, 188 Soto, Hernando de, 28
risk-weighted assets, 59 South Korea, 37, 225
Robinhood trading platform, 115 sovereign debt crisis (2010s), 16, 42–44
Rodgers, Kevin, 159 sovereign wealth funds, 231, 245, 248
Roosevelt, Franklin D., 38, 149 Soviet Union, 2, 69, 89, 97, 99, 188, 209
Rothschild, Nathan Meyer, 126, 249 Spain, 22, 23, 153, 165, 177; bank overex-
Royal Bank of Canada (RBC), 177 pansion in, 42; corporate finance in,
Rubinstein, Mark, 118–19 237, 251; traffic enforcement in, 54
Rumsfeld, Donald, 100 Spanish flu pandemic (1918), 190
Russel, Stuart, 207 speculative financing, 102
Russia, 42–43, 124, 183 speed limits, 29
spurious correlation, 89
Sanrio Company, 68 stagflation, 157, 179, 236
Santander, 165 Stalin, Joseph, 209
savings-and-loan (S&L) crisis (1980s), Stamp, Josiah, 101
10, 37–38 Standard & Poor 500 index, 77, 83,
Schlosser, Eric, 130 88–89, 119, 123, 131
Schneier, Bruce, 169 sterilization, 171
Scholes, Myron, 122–23 Stewart, James, 29–30

INDEX 275
Stumpf, John, 57 unemployment, 33, 105, 107

subjective probability, 102 United Kingdom, 9, 49; corporate
successful general’s syndrome, 157, 158 finance in, 150, 251; deposit insurance
supermarket sales, 207 in, 41–42; financial sector in, 36–37;
supervision vs. regulation, 52–53 monetary policy in, 182; Northern
supply and demand, 118 Rock failure in, 20, 39–41, 80, 226
Sweden, 26, 47, 48, 49 U-shaped crises, 8
Switzerland, 11, 49, 170–73, 177 usury, 55
Syed, Matthew, 227 Uthemann, Andreas, 212
systematically important financial insti-
tutions (SIFIs), 161, 175–76 Valencia, Fabián, 9
systemic risk, 2, 8–10, 16, 18, 20, 55, Valenzuela, Marcela, 104, 245
177–78, 184 Value-at-Risk, 75, 91, 134, 142, 143, 243;
in Basel regulations, 76, 81; manipu-
tail risk, 82, 92, 134, 190, 228–29, 231, 243 lability of, 130–33; misapplication
Taipei 101 building, 67–68 of, 76, 133, 145; JP Morgan’s devel-
taxation, 44, 180 opment of, 80–81, 195, 196, 197; as
technological singularity, 209 volatility alternative, 81
Tesobonos, 69 Varley, John, 23
t-GARCH (riskometer), 172–73, 201 Vayanos, Dimitri, 21
Thailand, 36, 37, 183 Venezuela, 15, 34, 183
Tolstedt, Carrie, 57 Vermeer, Johannes, 8
“too big to fail” problem, 25, 34 Vietnam War, 100
Tourre, Fabrice, 23 Viniar, David, 138
Toyota, 193 volatility, 2, 14, 75, 83–84, 134; cluster-
traffic laws, 29, 52, 54 ing of, 86; crises linked to, 104–5;
traffic-light rule, in risk measurement, 134 diversity vs., 190; fat-tail risk inversely
tranches, 140–42 related to, 231
transfer of responsibilities, 202 Volatility Index (VIX), 123–25
Triana, Pablo, 81 Volcker, Paul, 70
Truman, Harry S, 73, 130 Voltaire, 225
Trump, Donald, 15, 184 Volvo Cars, 5, 169–70, 175, 227, 235,
Tucker, Paul, 181 V-shaped crises, 8
Tukey, John W., 78
tulip mania, 153 Walker, Andrew, 166
Turing test, 209 Walmart, 207
Turkey, 69, 182 Washington Consensus, 60–61, 170
2001 (film), 222 The Wealth Effect (Chwieroth and Wal-
Typhoon Soudelor, 68 ter), 166
The Wealth of Nations (Smith), 58
Uber, 169 Weatherstone Dennis, 80
UBS, 4, 133, 134, 146, 176, 218 Wellington, Arthur Wellesley, Duke
Unelected Power (Tucker), 181 of, 27

276 INDEX
Wells Fargo scandal, 57 Yankelovitch, Daniel, 100

Why Aren’t They Shouting? (Rodgers), Yellen, Janet, 104
159 yield curve, 70
Why Xi’s China Is in Jeopardy (Magnus),
153 Zer, Ilknur, 104, 245
Wirecard, 62, 220 Zhang Heng, 73
World Bank, 46 Zhou, Chen, 90, 134
WorldCom, 225 Zigrand, Jean-Pierre, 21, 121
World War I, 157 Zimbabwe, 183
zombie banks, 48–50
Xiaochuan, Zhou, 103–4 zoning, 180

The Illusion of Control Why Financial Crises Happen, and What We Can (And Can't) Do About It

Uploaded by

Copyright:

Available Formats

The Illusion of Control Why Financial Crises Happen, and What We Can (And Can't) Do About It

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Illusion of Control Why Financial Crises Happen, and What We Can (And Can't) Do About It

Uploaded by

Copyright:

Available Formats

The Illusion of Control

Y8019-Danielsson.indb i 2/22/22 10:54 AM

New Haven & London

Y8019-Danielsson.indb iii 2/22/22 10:54 AM

Copyright © 2022 by Jón Daníelsson.

Yale University Press books may be purchased in quantity for educational,

Set in Galliard type by Newgen North America.

Library of Congress Control Number: 2021948855

This paper meets the requirements of ansi/niso z39.48-1992 (Permanence of

Y8019-Danielsson.indb iv 2/22/22 10:54 AM

1 Riding the Tiger 1

4 The Risk Panopticon 51

5 The Myth of the Riskometer 72

6 Ideas Matter: Risk and Uncertainty 94

7 Endogenous Risk 109

8 If You Can’t Take the Risk, Change Riskometers 128

9 The Goldilocks Challenge 147

10 The Risk Theater 168

11 The Uniformity, Efficiency, and Stability Trilemma 189

12 All about BoB: Robots and the Future of Risk 205

Y8019-Danielsson.indb v 2/22/22 10:54 AM

13 The Path Not to Take 223

14 What to Do? 240

Y8019-Danielsson.indb vi 2/22/22 10:54 AM

Several friends and colleagues made invaluable contributions to the

Y8019-Danielsson.indb vii 2/22/22 10:54 AM

Y8019-Danielsson.indb ix 2/22/22 10:54 AM

Figure 1. Credit: Copyright © Ricardo Galvão.

Almost every economic outcome we care about is long term.

Y8019-Danielsson.indb 1 2/22/22 10:54 AM

Y8019-Danielsson.indb 2 2/22/22 10:54 AM

retirement not being as comfortable as hoped for, are easy to under-

Y8019-Danielsson.indb 3 2/22/22 10:54 AM

financial system. The opposite of endogenous is exogenous. When an

Y8019-Danielsson.indb 4 2/22/22 10:54 AM

Y8019-Danielsson.indb 5 2/22/22 10:54 AM

Figure 2. Credit: Lukas Bischoff/IllustrationX.

Leendert Pieter de Neufville was only twenty-one in 1751 when

Y8019-Danielsson.indb 6 2/22/22 10:54 AM

Y8019-Danielsson.indb 7 2/22/22 10:54 AM

to mistrust the Prussian authorities. The money stopped flowing, and a

Y8019-Danielsson.indb 8 2/22/22 10:54 AM

Y8019-Danielsson.indb 9 2/22/22 10:54 AM

Y8019-Danielsson.indb 10 2/22/22 10:54 AM

great powers of Europe amid raised expectations of an impending war. This

Y8019-Danielsson.indb 11 2/22/22 10:54 AM

Y8019-Danielsson.indb 12 2/22/22 10:54 AM

Figure 3. The financial system network in 1774. Source: M. Postlethwayt,

Y8019-Danielsson.indb 13 2/22/22 10:54 AM

London to the real economy as commercial credit evaporated. In the be-

Y8019-Danielsson.indb 14 2/22/22 10:54 AM

It’s the Politics, Stupid

Y8019-Danielsson.indb 15 2/22/22 10:54 AM

the market. Not surprisingly, many commentators question why we si-

Y8019-Danielsson.indb 16 2/22/22 10:54 AM

Y8019-Danielsson.indb 17 2/22/22 10:54 AM

management and volatile politics. They depend on a small number of

We Had Forgotten History by 2008

Y8019-Danielsson.indb 18 2/22/22 10:54 AM