CFA INSTITUTE RESEARCH FOUNDATION / MONOGRAPH

REVISITING THE EQUITY

RISK PREMIUM

LAURENCE B. SIEGEL AND PAUL MCCAFFREY

EDITORS
REVISITING THE EQUITY RISK PREMIUM

LAURENCE B. SIEGEL AND PAUL MCCAFFREY, EDITORS

 Statement of Purpose
The CFA Institute Research Foundation is a not-for-profit organization established to promote the development
and dissemination of relevant research for investment practitioners worldwide.

© 2023 CFA Institute Research Foundation. All rights reserved.

Neither CFA Institute Research Foundation, CFA Institute, nor the publication’s editorial staff is responsible for facts and opinions
presented in this publication. This publication reflects the views of the author(s) and does not represent the official views of
CFA Institute Research Foundation.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including
photocopy, recording, or any information storage and retrieval system, without permission of the copyright holder. Requests
for permission to make copies of any part of the work should be mailed to: Copyright Permissions, CFA Institute, 915 East High
Street, Charlottesville, Virginia 22902. CFA® and Chartered Financial Analyst® are trademarks owned by CFA Institute. To view a
list of CFA Institute trademarks and the Guide for the Use of CFA Institute Marks, please visit our website at www.cfainstitute.org.

CFA Institute does not provide investment, financial, tax, legal, or other advice. This report was prepared for informational pur-
poses only and is not intended to provide, and should not be relied on for, investment, financial, tax, legal, or other advice.
CFA Institute is not responsible for the content of websites and information resources that may be referenced in the report.
Reference to these sites or resources does not constitute an endorsement by CFA Institute of the information contained therein.
The inclusion of company examples does not in any way constitute an endorsement of these organizations by CFA Institute.
Although we have endeavored to ensure that the information contained in this report has been obtained from reliable and up-to-
date sources, the changing nature of statistics, laws, rules, and regulations may result in delays, omissions, or inaccuracies in
information contained in this report.

Photo credit: helloabc / iStock / Getty Images Plus

ISBN: 978-1-952927-35-5
 Revisiting the Equity Risk Premium

CONTENTS
Preface iv
Paul McCaffrey

Editor’s Note v
Laurence B. Siegel

Introduction: Three Decades of Equity Risk Premium Forums vi

P. Brett Hammond and Martin L. Leibowitz

EQUITY RISK PREMIUM FORUM 2021: PRESENTATIONS AND DISCUSSIONS

Edited by Laurence B. Siegel

Presentation by Roger Ibbotson: Historical Returns, Premiums, and Popularity 2

Presentation by Elroy Dimson: A Global View and American Exceptionalism 7
Presentation by Cliff Asness: The Price of the Stock Market Matters 12
Presentation by Robert Arnott: A Simple Framework for Forecasting Equity
Returns—The Devil Is in the Details 15
Presentation by Martin Leibowitz: P/Es, Risk Premiums, and the g* Adjustment 22
Presentation by Mary Ida Compton: The Equity Risk Premium in Real Life 25
Presentation by Antti Ilmanen: Understanding the Many Components
of Expected Returns 30
Presentation by Thomas Philips: Linking Expected Returns to Fundamentals 37
Presentation by Rajnish Mehra: Reflections on the Equity Premium 52
Presentation by Jeremy Siegel: Back to the Fundamentals—A Close Look
at the Equity Risk Premium 58
Presentation by William Goetzmann: 574 Years of Equity Returns from the Bazacle
Water Mill in France 67

EQUITY RISK PREMIUM FORUM ROUNDTABLE DISCUSSION

Edited by Paul McCaffrey

Part I: Bubbles, Momentum, and Winnowing 76

Part II: Risk Premium or Fear Premium? 81
Part III: Bears and Hawks, Equities and Inflation 86
Part IV: Equity Duration, Negative Real Rates Forever, Bullish on Bear Markets? 93
Part V: What Is the ERP? Negative Returns as a Hedge, Modern Monetary Theory 97

APPENDIX A: 2023 UPDATES 103

REFERENCES 107

PL Qualified This publication qualifies for 3.75 PL credits under the guidelines
Activity of the CFA Institute Professional Learning Program.
CFA Institute Research Foundation iii
Revisiting the Equity Risk Premium

PREFACE
Because it is not directly observable, the equity risk also offering a fascinating window into some of the great-
premium (ERP) is one of the great mysteries of finance. est minds in finance today. Taken together, they provide a
holistic model of how inquiries into the internal workings of
Whatever the risk-free rate happens to be, depending on the financial markets should be conducted.
the time horizon, stocks tend to generate annual returns
that are 3% to 7% higher. What accounts for such a pre- In other words, the exercise encapsulates the mission of
mium? What mechanism explains it? Is it really all about CFA Institute: “To lead the investment profession globally by
the excess risk? Or, as Rob Arnott posits later in this book, promoting the highest standards of ethics, education, and
is it more of a fear premium? professional excellence for the ultimate benefit of society.”

These questions seek to uncover the elemental forces that I plan on returning to this text often in the months and
drive the markets. To decipher the answers would, with years ahead, not only for its insights into the equity pre-
only slight exaggeration, be the equivalent of discovering mium but also for its ample doses of wit, wisdom, and
finance’s philosopher’s stone. inspiration. I know that countless readers will do so as well.
I also expect that, like me, they will be looking forward to
For this very reason, three times in the past 20 years, the next Equity Risk Premium Forum in 2031.
CFA Institute and CFA Institute Research Foundation have
assembled a roundtable of distinguished investors, other In closing, this project would not have been possible with-
financial practitioners, and academics to explore these out the singular contributions of the many roundtable par-
questions. The panel has featured an evolving cast of lumi- ticipants who generously donated their time and effort as
naries. Those who have participated in the roundtable on well as contributed their original research to this endeavor.
all three occasions are Arnott, Cliff Asness, Roger Ibbotson, Special thanks are due to both Laurence B. Siegel and
Martin Leibowitz, and Rajnish Mehra. Meanwhile, new Bud Haslett, CFA, who are respectively the Gary P. Brinson
names have been added, and their contributions are at the Director of Research and the executive director of CFA
same level of quality as the perennials. Institute Research Foundation. They shepherded this proj-
ect from conception to completion. Their diligence, com-
As these pages demonstrate, rarely have so many of mitment, and passion truly represent the best in finance
finance’s top thinkers been gathered in one place, and rarer and demonstrate why CFA Institute Research Foundation is
still is their dialogue so compelling, forthright, and incisive. such an essential pillar of modern financial scholarship.

The research and discussion that follow may not offer a Paul McCaffrey
single Eureka! moment that solves the mystery once and Editor, Enterprising Investor
for all, but they do shed considerable light on the ERP while CFA Institute

iv CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

EDITOR’S NOTE
In 2001, Marty Leibowitz organized the first Equity Risk premium back in the 1970s when I was a student at the
Premium Forum, published online by AIMR, a predeces- University of Chicago and he was my professor. I am very
sor organization of CFA Institute. Ten years later, Brett thankful he was able to attend.
Hammond, Marty Leibowitz, and I convened a similar group.
We reflected on the changes that had occurred in the Paul McCaffrey edited the remainder of this book,
previous 10 years and made new forecasts. CFA Institute consisting of the roundtable discussion that followed
Research Foundation published this work as a research the presentations.
monograph, entitled Rethinking the Equity Risk Premium.
That is enough of me. I therefore asked Brett Hammond
Then, in late 2021, at the suggestion of our executive and Marty Leibowitz to write the introduction to this
director Bud Haslett, I organized the third decennial con- book, a task that would more typically fall to the editor.
versation, which resulted in the Revisiting the Equity Their wonderful introduction comes next, followed by the
Risk Premium book that you are now reading. I chose the 11 direct presentations, discussions of the presentations,
speakers, led the discussion, and edited the part of the and the roundtable.
book consisting of the presentations and short discussions
Happy reading, and see you again in 2031!
of each presentation. Our distinguished speakers—many
of the same ones we had 10 and 20 years ago—each had Laurence B. Siegel
10 minutes to present, and then we all had 5 minutes to Gary P. Brinson Director of Research
talk about the presentation. We began with Roger Ibbotson, CFA Institute Research Foundation
because he started the investigation of the equity risk

CFA Institute Research Foundation v

Revisiting the Equity Risk Premium

INTRODUCTION: THREE DECADES OF EQUITY

RISK PREMIUM FORUMS
P. Brett Hammond
Martin L. Leibowitz

In 2001, just after the dot-com crash, a group of academ-

ics and market professionals came together under the
sponsorship of CFA Institute (known at that time as AIMR) Exhibit 1. Most Frequent Estimate
to provide estimates of the equity risk premium (ERP) over of the 10-Year Equity Risk Premium
the next 10 years. Most participants also provided essays
describing the analyses that underpinned their forecasts.1
The meeting was held at the offices of TIAA-CREF (as it was
then known), where both of us were employed at the time.

In late 2011, following the Global Financial Crisis, CFA

Institute Research Foundation convened a second risk
premium forum. The forum participants, perhaps under-
standably, did not all give numerical estimates (and there
is no record of the individual estimates provided by those
who did). Instead, or in addition, they contributed essays on
the multiple factors that should form the theoretical basis
for risk premiums.2

At a subsequent Q Group conference in October 2012, Brett

Hammond was invited to present a summary of the earlier
meetings.3 He began his presentation with the slide shown
as Exhibit 1. Source: Authors’ calculations based on a literature survey and esti-
mates from participants in the 2001 Equity Risk Premium Forum
Some attendees said this parsimonious slide was the most (AIMR 2002) and Hammond (2012).
memorable one ever presented at the Q Group. For the 2001
data, the slide relied on both a literature survey and the
following distribution of participants’ estimates, shown in
Exhibit 2. The proceedings of this third forum make up the book that
you are now holding in your hands. The distribution of this
In the decade following the 2001 forum, equity returns forum’s estimates is shown in Exhibit 3.
were terrible. The average annualized real price return was
–1.1%. Although those forecasting a zero premium came Despite radically different market environments, it is strik-
closest, none of the estimates anticipated negative premi- ing that the estimates in all three forums were so similar.
ums, much less negative total returns. They tended to be in the 3%–5% range, and notably, in
comparison to historical returns, none of them included
In contrast, in the decade following the 2011 forum, the estimates above 7% or below zero. Moreover, in the first
market turned in terrific returns, with annualized real price two cases, participants were free to choose their own
returns of around 11%. (relatively) risk-free asset benchmark, which ranged from
3-month Treasury bills to 10-year Treasury bonds or TIPS
After the first and second forum, CFA Institute Research (inflation-indexed Treasury bonds). (In the third forum, Larry
Foundation bowed to the inevitable and sponsored a third asked all of us to use the 10-year nominal Treasury bond as
forum in late 2021, under the leadership of Larry Siegel.

1 See AIMR (2002).

2 See Hammond, Leibowitz, and Siegel (2011).
3 The formal name of the Q Group is the Institute for Quantitative Research in Finance.
vi CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Exhibit 2. Distribution of 10-Year Exhibit 3. Distribution of 10-Year

Equity Risk Premium Estimates, Equity Risk Premium Estimates,
2001 2021
Frequency Frequency
7 4

6
3
5

4
2
3

2
1
1

0 0
0 1 2 3 4 5 6 More 0 1 2 3 4 5 6 More
Risk Premium (%) Risk Premium (%)

Source: AIMR (2002). Source: Authors’ calculations based on survey of participants in

the third Equity Risk Premium Forum, held in 2021.

the risk-free asset.4) It almost seems as though a 4% value

is the financial equivalent of a cosmological constant. nature of rolling 40-year returns, refreshed annually, in a
less than 100-year time period.)
The question is: Why are all three sets of estimates so
similar? One answer might be that powerful investor biases These unintended consequences of extremely high or low
tend to compress the range of estimates. forecasts and the influence of historical averages might all
lead to a more compressed range of estimates. It is hardly
On the one hand, belief in very low estimates might compel surprising that the combination of compression effects
investors to feel that prospective returns do not justify sig- and high equity volatility would lead to estimates that turn
nificant equity positions. Consequently, subsequent selling out to be wrong—often by a lot!
to reduce positions would (presumably quickly) bring val-
uations back to reasonable risk-adjusted returns. In other It also might be argued that participants knowingly or
words, a belief in a very low estimate would be inconsistent unknowingly intended the estimates to be for horizons
with a stable valuation. much longer than 10 years and that they really represent
mean returns for a distribution of future decades. For
On the other hand, a very high risk-premium estimate might example, additional work using shorter-term valuations
be interpreted as implying unrealistic or unsustainable val- and market-cycle theory might have been needed to
uation levels, or as a call for investors to move toward risk convert truly long-term estimates to 10-year estimates.
levels that far exceed their strategic risk limit. This view clearly represents a departure from the stated
10-year ground rule, however, and fails to provide practi-
In addition to investors’ valuation-based biases, estimates tioners with practical and useful information.
may also be influenced by looking back through a historical
lens. For example, an admittedly simplistic analysis of roll- These considerations lead to a somewhat different view of
ing 40-year periods beginning in 1925 reveals that the real- the ERP that focuses on the hierarchy of non-cash assets
ized median (50th percentile) return of US equities relative against which equity returns could be measured. All non-
to US bonds was 3.6%. At the 90th percentile, the realized cash assets have risk; and, in real terms, so does cash. Of
return was 3.95%. course, when comparing the return patterns of two risky
assets, it is the volatility of the difference between their
(Note how close together the median and 90th percentile prospective returns that matters. This volatility difference
rates of return are, indicating a tight distribution of 40-year depends on both the volatilities of the two assets and the
returns. This is something one might expect because of the correlation between them and is a key determinant of the

4Purchased once at the beginning of the forecast period and held through the end of it (maturity). In other words, the rate deemed to be
riskless was the yield on that bond at the beginning of the period. Another convention (the one we did not use) would be to use a hypothetical
constant-maturity portfolio of 10-year Treasury bonds as the reference (“riskless”) asset.
CFA Institute Research Foundation vii
Revisiting the Equity Risk Premium

risk premium needed to justify an allocation to the more reassurance that the fund’s key objectives will be achieved.
volatile asset. If so, should the causality be reversed? That is, do the
implicit slow-changing estimates justify the preexisting
For example, the equity premium resulting from a compar- (and also slow-changing) allocations?
ison against the mythical riskless asset, which has zero
volatility and correlation, would be misleadingly simple. In other words, could it be that this 4% “cosmological con-
That would involve, however, a likely fruitless effort to iden- stant” is not really derived solely from actual forecasts but
tify a truly nonvolatile, riskless asset while ignoring signals rather ends up being somewhat of a “goldilocks” number
contained in other risky assets that can inform compari- that comfortably fits with a variety of investor hopes and
sons with equities. To that end, 10-year Treasury bonds, institutional structures?
which have a relatively long duration, are often considered
to be a more proper base for computing a risk premium. The Of course, one of the most valuable benefits of the risk
spread of equity returns relative to a diversified portfolio of premium concept is that it encourages us to contemplate
corporate bonds also might be a viable alternative. the range of scenarios facing us at a given time and to
consider whether the offered return represents sufficient
More generally, a spectrum of both individual asset types reward. To that end, we would like to draw readers’ atten-
and well-constructed portfolios could serve as points of tion not only to the most frequent risk premium estimate
comparison for incremental equity-like investments. One but also to the full range of estimates from previous
implication is that the choice of a comparison asset or forums. In Exhibit 2, these range from zero to more than 6%,
portfolio is not straightforward and may depend on an and in Exhibit 3, they range from 1% to more than 6%.
individual’s investment goals. Another implication is that
because the available comparison assets are themselves At first glance, we might simply conclude that these ranges
risky, they may necessitate even higher risk-adjusted reflect a lack of full consensus among the expert forum
premiums. participants as well as possible methodological and defini-
tional differences in arriving at each estimate. For example,
Another problem has to do with the leverage inherent in the note that for those experts who provided estimates in
stock market, which in the aggregate varies considerably both 2001 and 2021, those offering lower versus higher
over time. Based on the well-known work of Modigliani and estimates in the first instance also did so in the second
Miller,5 expected returns from equities should be higher instance. (The identity of the participant making each
when companies are more leveraged (all other things being forecast is not revealed here, but we had that information
equal). when preparing this introduction.)

In addition, the typical equity premium estimation process Looking a little deeper, we can also see at least two more
considers a range of influences, such as real GDP, as well interesting implications for asset analysis and allocation.
as factors that affect all assets, such as interest rates and
inflation. In particular, the earnings yield is often taken as First, asset allocation analysis benefits from sensitivity
a proxy for prospective real total return. This procedure analysis that reflects uncertainty regarding asset return,
ignores consideration of the prospects for earnings growth volatility, and correlation estimates. Keeping in mind that
(which in many circumstances has been shown to be the actual returns over the past two decades have departed
dominant total return factor). Moreover, without an adjust- significantly from the most frequent risk premium estimate,
ment for the “retention effect” (earnings versus dividends rather than relying on one future scenario (that is, one risk
and share buybacks) and for new equity issuance, any premium estimate), analysts should test a compact set
simple add-on of a raw growth number also can lead to of plausible higher and lower premium estimates, such as
misleading risk premium values. those suggested by Exhibits 2 and 3, as well as covariance
estimates. Modern modeling tools and computing power
The effects of increased investor access to private markets easily support such an exercise.
of all kinds have not escaped analysts’ attention. With pri-
vate markets growing in size, different payoff structures, Second, although the third Equity Risk Premium Forum
and more limited reporting, how should we consider private required participants to use a 10-year horizon, most insti-
asset returns when estimating an equity premium? tutions and individual retirement savers are in it for the
longer haul, in many cases multiple decades. We noted that
Despite these issues and concerns, many institutional the implicit, even unconscious, influence of the long view
funds continue to base their allocation strategies on the may bias estimates toward 4%. We could add that one of
“building blocks” of assumed risk premiums. For institu- the most difficult practical problems in estimating returns
tional policy-making purposes, the net result is a portfo- is to link the short term with the long term. Specifically, if
lio whose expected return may provide a possibly false our shorter-term risk premium estimate differs significantly

5 See Modigliani and Miller (1958).

viii CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

from a long-term equilibrium estimate, it raises the approach year 10? If the latter, it isn’t obvious how we
question of how to model the time-varying nature of the should model the transition from short- to long-term esti-
premium. Of course, this concern applies to all asset mates. One approach would be to introduce a small weight
classes, not just equities. for the long-term estimate at some point and gradually
increase the weight. Another would be to assume that a
To illustrate, our 10-year risk premium estimates might period of returns below the long-term estimate will be fol-
imply projected equity returns ranging anywhere from 0 to lowed by a period of returns above the estimate to avoid
7% with a central tendency of 4% (as Exhibit 1 strongly violating the long-term equilibrium assumption. Perhaps a
suggested). Our long-term risk premium estimate, however, more satisfying approach would be simply to look at differ-
might be the same or different—say 4%, 6%, or another ent risk premium scenarios as described.
estimate. For asset allocation, should we redo the premium
estimates each year and remodel allocations accordingly? Whichever approach the analyst chooses, the results will
Or should we ignore short-term estimates for any portfolios be influenced by estimates and methods. In an era in which
with multidecade horizons and use our preferred long-term we observe macro forces acting on asset markets and
risk premium estimate for modeling purposes all the time? changing premia regimes, it might behoove the analyst not
only to use the consensus equity premium number as an
Or should we model allocations dynamically by assuming asset allocation anchor but also to test allocation sensitiv-
11% or −1% equity returns for the next few years, which- ity to different estimates and methods.
ever seems appropriate at the time, but reallocate as we

CFA Institute Research Foundation ix

REVISITING THE EQUITY RISK PREMIUM

EQUITY RISK PREMIUM FORUM 2021:

PRESENTATIONS AND DISCUSSIONS
LAURENCE B. SIEGEL, EDITOR
Revisiting the Equity Risk Premium

PRESENTATION BY ROGER IBBOTSON: HISTORICAL

RETURNS, PREMIUMS, AND POPULARITY
Roger Ibbotson: Thank you, Larry. It’s a wonderful group you The whole purpose of this study was to look at premiums,
put together here. which Exhibit 4 displays as the differences between the
rates of return. This dataset is available now from CFA
A lot of my old friends, but also a really accomplished group. Institute Research Foundation. It’s updated monthly and is
available to any CFA Institute member,1 so it’s being circu-
My work was probably the earliest here, but it didn’t come
lated again.
out of a vacuum. I was at the University of Chicago, and
Larry Fisher and James Lorie had put together the Center for What you see in these data most vividly, though, are the dif-
Research in Security Prices (CRSP). All the data were avail- ferent premiums. The dark blue line is the stock market’s total
able there. So, by writing some code, I had a great oppor- return, which shows the explosive growth of the markets.
tunity to put together the kind of data that are in Stocks, Over 95 years, $1 in the US stock market total return index
Bonds, Bills, and Inflation. Fisher and Lorie’s stock market has grown to almost $11,000. People are always astonished
data actually mentioned total returns, which was pretty to see the amount of growth you get if you can compound
unusual at the time because most people worked on price the annual rates of return over long periods of time.
indices, not total returns that included dividends.
These are nominal indexes. The inflation index grew by a
Even with dividends available on the CRSP tapes, most factor of 15, so you can divide the nominal indexes by 15
researchers treated dividends separately. You might to get real (inflation-adjusted) indexes, but the real returns
think it’s a trivial thing to add dividends and capital gains are still very large numbers. The real return on equities is
together, but that wasn’t done at the time. When invest- the premium of equities over inflation.
ment managers reported their results, they basically
reported capital gains and dividends separately as well. You can see the other premiums: bonds versus bills, bills
versus inflation, and small-cap stocks versus large-cap
We also had some data on the bond side. We have Marty stocks. We didn’t have small stocks in the original study.
Leibowitz here today, and working with Sidney Homer,
he did a lot on bond data at the time. The first release of these data came out in two Journal of
Business articles in 1976.2 At that time, we used the data
What motivated me most was the capital asset pricing not only to show historical performance but also to make
model (CAPM), because the equity risk premium came out of a forecast for the next 25 years—to the year 2000. We took
that. At the time, in the early 1970s, the CAPM was the domi- 50 years of historical data and then made 25-year future
nant model of security prices—and to implement that model, projections. Looking back from 2000, those forecasts
you had to have a measure of the equity risk premium. All turned out to be pretty close to correct.
of the researchers were talking about equity risk premiums
and other risk premiums—everything was all about risk at
the time. And we didn’t have any real measures of any of Arithmetic versus Geometric
these premiums. We had some data on stocks—the Fisher Mean Returns
and Lorie data—which were not up to date. We had some
data on bonds and other assets, but they weren’t brought Exhibit 5 shows the summary statistics of the data in
together in a form where you could look at risk premiums. Exhibit 4. Over the years, people have been confused
between geometric mean (compound annual) returns and
arithmetic mean returns.
Comparative Returns on Stocks,
Bonds, Bills, and Inflation Premiums can be measured in either arithmetic or geometric
mean terms, and the two can be very different. The difference
Please refer to Exhibit 4. Many of you have seen it— relates to the standard deviation—the bigger the standard
the “Stocks, Bonds, Bills, and Inflation” chart. This is deviation of the series, the bigger the difference between
Morningstar data at this point, because I sold Ibbotson the arithmetic and geometric mean. This relationship has
Associates to Morningstar back in 2006. become a key element in understanding asset returns.

1 In the mainland of China, CFA Institute accepts CFA® charterholders only.

2 See Ibbotson and Sinquefield (1976a, 1976b).
2 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Exhibit 4. Ibbotson SBBI: Stocks, Bonds, Bills, and Inflation, 1926–2020

Source: Data from Stocks, Bonds, Bills, and Inflation (SBBI) and Morningstar, Inc.

With highly volatile series, there can be huge differences. Components of Returns: The
The premium between small caps and large caps or, for that
matter, between stocks and riskless assets—the equity risk Riskless Rate and Risk Premiums
premium itself—differs greatly depending on whether it is
In Exhibit 6 and Exhibit 7, we break the returns on each
measured arithmetically or geometrically. We tend to talk
asset class into their component parts. In doing so, we
about it both ways.
identify different types of premiums by taking either arith-
metic or geometric differences between one asset class
Long-Term versus Short-Term series and another. The premiums include a small-cap
Riskless Assets premium, a corporate bond default risk premium, a bond
horizon premium, and a real riskless rate of interest.
We also get very different numbers for the equity risk pre-
All these premiums, plus the real riskless rate, come out of
mium depending on whether we are comparing stocks to
this analysis. To make the analysis visually clear, I some-
long-term or short-term riskless assets. All these estimates
times stack the components as in Exhibit 6. Look, for exam-
of the equity risk premium are useful—if I were making a
ple, at “cash” (Treasury bills), where the Treasury bill return
long-term forecast, I would want an equity risk premium
itself has two pieces: inflation and the real interest rate. For
that was measured relative to long-term Treasury bonds,
premiums, we can talk about either the realizations (past
and if I were making a short-term forecast, I would use the
returns) or the expectations. The current discussion is
equity risk premium relative to Treasury bills.
mostly about the expected, or future, equity risk premium.
So, in making the choice of arithmetic versus geometric
The second column or “tower” in Exhibit 6 includes the
and long versus short-term horizon equity risk premiums,
equity risk premium. This premium can be measured rela-
there are a lot of issues to address. For now, I am just defin-
tive to long-term bonds, or it can be measured relative to
ing the terms. Another issue is the starting date, which at
Treasury bills. We can put the small-cap premium or value
the time I started the study was 1926 because those were
premium on top of that. Today, of course, there is a lot of
the available data.

CFA Institute Research Foundation 3

Revisiting the Equity Risk Premium

Exhibit 5. Summary Statistics of Returns on Stocks, Bonds, Bills, and Inflation,

1926–2020

Source: Data from SBBI, Morningstar, Inc.

debate about whether those premiums even exist or what Summary

other premiums might exist, with many opinions on what
these premiums should be. I think there’s little doubt that Let me wrap up by summarizing where I think we are
there’s a liquidity premium, though, in all asset classes and going today.
situations.
There are different methods of estimating the equity risk
On the bond side, you can use the same sort of stacking premium. The historical method basically asks, “What
methods. The Treasury bond has a premium relative to do historical returns tell us about the future?” That’s the
“cash” (Treasury bills), and I call that the horizon premium, approach in Stocks, Bonds, Bills, and Inflation. Next, you’ll
referring to the time horizon of the bond. I took out the word hear from Elroy Dimson with the Dimson–Marsh–Staunton
“risk” in some of the boxes. I think it was Rajnish Mehra research on many different countries. In a related area,
who said, “that’s not necessarily a risk premium.” It really Will Goetzmann and I are currently working on some more
has to do with matching the time horizons of investors data back to 1815 for the New York Stock Exchange. Jeremy
and issuers, so the difference in yields or returns between Siegel has also done a lot of work on historical returns.
short- and long-term bonds isn’t necessarily a premium for Many of us who are here today are working in this area.
risk specifically.
The demand side is a different approach. What returns do
When you move to the right in Exhibit 7 to consider a bond investors demand for taking on the risk and other charac-
that can default, you have a default risk premium. The risk teristics of securities? The CAPM addresses that question
premium that you expect to realize is only part of the yield because it says that people are risk-averse and therefore
spread between the corporate and Treasury bonds—you demand an equity risk premium. Some of Rajnish’s major
don’t get the whole yield spread because you’ll have some work is on this topic—looking at utility curves and asking,
defaults along the way. “What are investors demanding here?”

4 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 6. Stacking Equity Premiums

Liquidity
Premium

Small
Value Value
Stock
Premium Premium
Premium
Equity Equity Equity Equity
Risk Risk Risk Risk
Premium Premium Premium Premium
Real Real Real Real Real
Interest Interest Interest Interest Interest
Rate Rate Rate Rate Rate

Inflation Inflation Inflation Inflation Inflation

Cash Large Small Value Value & Liquidity

Caps Stocks Stocks Premium

Exhibit 7. Stacking Fixed-Income Premiums

Liquidity
Premium

Default Default
Risk Risk
Premium Premium
Horizon Horizon
Risk Risk Horizon
Premium Premium Premium

Real Real Real Real

Interest Interest Interest Interest
Rate Rate Rate Rate

Inflation Inflation Inflation Inflation

Cash Treasury Corporate Less Liquid

Bonds Bonds Corporate
Bonds

I’ve been working on the demand side with a set of papers From the supply side, the question is: What cash flows
and a CFA Institute Research Foundation Monograph called does the economy supply to investors? I recently published
Popularity: A Bridge between Classical and Behavioral some work on this with Philip Straehl, looking at buybacks,
Finance.3 My co-authors are Tom Idzorek, Paul Kaplan, and because buybacks are now actually a bigger part of cash
James Xiong. They’re all from Morningstar. It says that if flow to investors than dividends.4 We definitely want to cor-
you have a preference for an asset characteristic—if you rect dividend discount models (DDMs) for buybacks. DDM
really like it—you’re going to raise the valuation of assets models are in the supply realm. Marty Leibowitz is going to
with that characteristic. The same future cash flows will talk about growth estimates, so his work would fit into the
have a higher valuation or price in the present; that means supply category.
the asset will have a lower expected return. If we don’t like
a characteristic, assets with that characteristic will have The last approach to estimating the equity risk premium
higher expected returns. is surveys, in which you might simply ask people what

3 See Ibbotson, Idzorek, Kaplan, and Xiong (2018).

4 See Straehl and Ibbotson (2017).
CFA Institute Research Foundation 5
Revisiting the Equity Risk Premium

returns they expect or think they should earn. Conceptually, dividends and having investors interpret that as bad news.
this idea is good, but the questions in the surveys tend to You can buy back or not buy back stock whenever you want.
be ambiguous. When people ask me what return I expect,
I don’t know if they’re talking about the arithmetic mean, Jeremy Siegel: Buybacks are also tax efficient.
the geometric mean, the long term, or the short term.
Roger Ibbotson: They are. We are out of time, but that’s a
I would give very different answers depending on these
great discussion.
conditions, and usually these surveys are not designed well
enough for you to know which question you’re answering. Laurence Siegel: Depending on what everyone wants to
talk about in the afternoon, we might be able to bring this
Discussion of Roger topic back.

Ibbotson’s Presentation Appendix to Roger Ibbotson’s

Robert Arnott: It will come as no surprise to you, Roger,
that I view buybacks as partly real and partly mirage. So, I’d
Presentation: Further
push back on the arithmetic of suggesting that buybacks Reading
are, sustainably on a long-term basis, larger than dividends.
You and I have already had that back-and-forth discussion Historical: What do historical returns tell us about the
in the Letters section of the Financial Analysts Journal,5 future? See:
so I’ll let it go with that.
Dimson, Elroy, Paul Marsh, and Mike Staunton. 2021. Credit
Laurence Siegel: Rob, can you summarize what you mean Suisse Global Investment Returns Yearbook 2021 Summary
by “mirage”? Edition. Credit Suisse Research Institute. https://www.
credit-suisse.com/media/assets/corporate/docs/about-us/
Robert Arnott: Buybacks are often done to facilitate research/publications/credit-suisse-global-investment-
management stock option redemption. So you noisily returns-yearbook-2021-summary-edition.pdf.
announce you’re buying back 10 million shares of stock.
Roughly concurrent with that announcement, management Ibbotson, Roger G., and James P. Harrington. 2021. Stocks,
redeems 10 million shares of stock options. The aggre- Bonds, Bills, and Inflation® (SBBI®): 2021 Summary Edition.
gate float doesn’t change. So, what we found historically is Charlottesville, VA: CFA Institute Research Foundation.
that float for the aggregate market tends to go up, not down. https://www.cfainstitute.org/-/media/documents/book/
A buyback isn’t a buyback if the float doesn’t go down. rf-publication/2021/sbbi-summary-edition-2021.ashx.

And if you go through the arithmetic on market aggregates, Demand Methods: What do investors demand? See:
as reported by CRSP, you find that dilution of shareholders
collectively across the index is the overwhelming norm for Ibbotson, Roger G., Thomas Idzorek, Paul Kaplan, and
the S&P 500, with occasional bouts of net buybacks. James Xiong. 2018. Popularity: A Bridge between Classical
and Behavioral Finance. Charlottesville, VA: CFA Institute
The net buybacks are also usually overwhelmed by net new Research Foundation. https://www.cfainstitute.
share issuance, if only by the index changing its composi- org/research/foundation/2018/popularity-bridge-
tion. If you kick out AIG and put in Tesla, for example, you’re between-classical-and-behavioral-finance.
forcing everyone holding the index to sell 1.5% of every
stock they already have in order to bring in this giant new Supply Methods: What does the economy supply? See:
company—so the aggregate float goes up, not down. Taking
Straehl, Philip U., and Roger G. Ibbotson. 2017. “The Long-
that into account, you find that indexes are diluted by an
Run Drivers of Stock Returns: Total Payouts and the Real
average of 2% a year historically. There have been bouts
Economy.” Financial Analysts Journal 73 (3): 32–52.
in the 1980s and in the mid-aughts and mid-teens (of the
current century) where buybacks for the S&P exceeded Surveys: What do investors and economists anticipate? See:
new share issuance and other forms of dilution, but…
Fernandez, Pablo, Alberto Ortiz, and Isabel Fernandez
Roger Ibbotson: I don’t think that Rob is right on this, but this Acín. 2017. “Market Risk Premium Used in 71 Countries
discussion has been in the Financial Analysts Journal. I don’t in 2016: A Survey with 6,932 Answers.” Journal of
think buybacks are going away, because they’re a much International Business Research and Marketing 2 (6):
more flexible way of paying out cash flows. There is no sig- 23–31. Updated at https://papers.ssrn.com/sol3/papers.
naling with buybacks: You don’t have the problem of cutting cfm?abstract_id=3861152.

5 See Arnott and Bernstein (2018); Straehl and Ibbotson (2018).

6 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

PRESENTATION BY ELROY DIMSON: A GLOBAL

VIEW AND AMERICAN EXCEPTIONALISM
Elroy Dimson: As a doctoral student at London Business stock market put together. For Paul Marsh, Mike Staunton,
School (LBS), I was strongly influenced by Roger Ibbotson, and me, this was the motivation for remarking that there
even though I never met him during my student days. must be something exceptional about the United States.
I discussed thesis ideas with Professor Bob Hamada, who
was visiting for a year from the University of Chicago. He We now have data on 90 markets, and of those, the 23
suggested I look at Roger’s Chicago proposal to study shown in Exhibit 8 started in 1900. For those, we have an
IPOs on the NYSE. Roger’s ideas intrigued me so much that index series that spans 121 years, except for two markets
I selected IPOs as my topic. My dissertation was on the that terminated and then restarted—namely, China and
pricing of new flotations on the London Stock Exchange. Russia. We incorporate the latter two markets into our
global index returns, so total wipeouts are still reflected in
Meanwhile, Professor Dick Brealey, head of the finance the worldwide index series.
group at LBS, convinced faculty and funders that the
school should follow Chicago’s lead by creating a research More recently, we’ve added nine more markets where we
database for financial research. That was the origin of the have 50 years or more of data. So, rather than 23 markets
London Share Price Database, which fellow student Paul from 1900, we now have a total of 32 markets with a
Marsh and I used for estimating long-term stock market history spanning at least a half-century. Then there
returns for the United Kingdom. We published our findings are another 58 countries—nearly all of them emerging
in the Journal of Business,6 the same outlet as Roger’s markets—where we have data for shorter periods. We
seminal study on the returns from US stocks, bonds, bills, remove the ones with less than 10 years of data, so we
and inflation.7 have excess returns or realized equity risk premiums in
90 markets. Exhibit 9 depicts our global returns dataset.
To conserve space, we omit markets with a relatively short
Returns on Many Markets history from this chart.
around the World Exhibit 10 shows the markets where we have a complete
After several years, Mike Staunton joined us, and we history with all asset return series starting at the same
extended the research to a series of reports and articles. time. They are shown with the usual country abbreviations
Our book, Triumph of the Optimists, published in 2002 for countries except for two: WXU is the world ex–United
by Princeton University Press, presented over a century States, and WLD just to the right of the center is the world,
of capital market history for the main asset categories.8 both measured in common currency (US dollars). Returns
Since 2002, we have published an annual volume—the are in local currency and are real (i.e., adjusted for local
Global Investment Returns Yearbook—which is an annual inflation). Equities beat inflation everywhere.
study of risk and return since 1900 on worldwide asset
The exhibit also shows bonds. We used long-term govern-
classes and factors.
ment bonds where available. As you go back in time, some
The breadth of our dataset has expanded over time. As governments didn’t issue very long bonds. And some gov-
Exhibit 8 shows, there are now 23 stock markets for which ernments, such as the United Kingdom, issued bonds with
we have return histories beginning in 1900. For all these a maturity of infinity (“consol bonds”). So, there is substan-
markets, we have not only a price index but also a measure tial variation between countries in the maturity of the bond
of income (dividend yield). series. Bonds produced, on balance, a positive real return in
local currency adjusted by local inflation. Bonds also beat
The United States was already one of the largest markets cash with one minor exception, which is Portugal, where
in the world by market capitalization in 1900. It rapidly cash did better than bonds.
became the biggest soon after that and, except for one brief
period, remained as such through to the present. For a short These differences in rates of return, or premiums, are all
interval around 1990, Japan was bigger, but that did not last. geometric differences, so they measure how much wealth-
Today, the US market has become larger than every other ier you would be in equities versus government bonds,
for example. These premiums have no numeraire, so they

6 See Dimson and Marsh (2001).

7 See Ibbotson and Sinquefield (1976a).
8 See Dimson, Marsh, and Staunton (2002).
CFA Institute Research Foundation 7
Revisiting the Equity Risk Premium

Exhibit 8. Evolution of Equity Markets since 1900

Source: Dimson, Marsh, and Staunton (2021).

Exhibit 9. The DMS Long-Term Dataset: 90 Markets, 1900–2020

Australia Australia
Austria Austria
Belgium Belgium
Canada Canada
Denmark Denmark
Finland Finland
France France
Germany Germany
Ireland Ireland
Italy DMS 23: Italy
Japan All starting in 1900 Japan
Netherlands Netherlands
New Zealand New Zealand
Norway Norway
Portugal Portugal
South Africa South Africa
Spain Spain
Sweden Sweden
Switzerland Switzerland
UK UK
US US
China (mainland) Market closure China (mainland)
Russia Market closure Russia
Brazil
India
Nine new markets Hong Kong SAR
added in 2021 South Korea
giving DMS 32 Singapore
Taiwan
Mexico
Malaysia
Thailand
Argentina
Equity returns for Chile
58 countries (56 emerging Greece
markets) giving DMS 90 Zimbabwe
We calculate excess returns for Jordan
countries with history > 1 decade Luxembourg
Philippines

1900 1950 1960 1970 1980 1990 2000 2010 2020

Source: Dimson et al. (2021).

8 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 10. Equities versus Inflation, Bills, and Bonds: 1900–2020

Source: Dimson et al. (2021).

Exhibit 11. Real Equity Returns in Local Currency and US Dollars, 1900–2020

Source: Dimson et al. (2021).

would be exactly the same if we use dollars or yen or ster- without choosing a numeraire. The annualized real return
ling to do our calculation. on the world market from a US dollar perspective was 5.2%.
If we compare that number with the return on bonds, we
If we want an anchor, we can look at the world market in have a realized premium of 3.1% annualized.
US dollars adjusted for US inflation. Whereas for individ-
ual countries we look at returns in the local currency and When reporting the equity risk premium, the more
adjust them for local inflation, we can’t look at regions common academic measure is the difference between the

CFA Institute Research Foundation 9

Revisiting the Equity Risk Premium

Exhibit 12. Relative Return of Major Markets, 20th and 21st Centuries

Source: Dimson et al. (2021).

annualized equity return and the bill return; that was well in What have we learned? The difference between the world
excess of 4% over the 121 years. So, globally, equities beat equity return and bill returns was 4.4%, and a little over
inflation, they beat cash, and they beat bonds. All by a large 3% if we draw a comparison with bonds. Equity premiums
margin, as Roger has pointed out—but it’s quite a varying do vary quite a bit between countries. Again, there are
margin! almost no currency effects over the long term. It’s the
market that matters, because we’re looking at the differ-
When I refer to the “realized premium,” I do not mean the ence between two returns measured in the same currency.
arithmetic difference that Roger highlighted, particularly in
his early work. I mean the geometric or compound annual
difference. Exhibit 11 shows these real returns, along with American Exceptionalism Revealed
the annualized rate of change in the exchange rate of each
Since 1900, the United States has had a high equity pre-
country versus the United States.
mium, 5.8%, reflecting strong equity markets. The world
excluding the United States has had a much lower equity
Currency Effects and Adjustments premium, and Europe’s is a bit lower again. The developed
world, taken as a portfolio, has done about the same as
These gaps between equity returns and fixed income the world; emerging markets, on the other hand, have
returns—that is, the equity premiums—are about the same had a low premium, 4.0%. Your perspective makes a big
whether we do the calculation in local currency adjusted difference. It also matters what period you look at: the
by local inflation or in US dollars adjusted by US inflation. last half-century or the full sample covering more than
It makes a difference year by year, but it makes almost no a century.
difference to our long-term figures.
Exhibit 12 shows how markets performed over the 20th
The currency effect, shown by the little gray bars, is the century, the 21st century through 2020, the period since
appreciation (or depreciation) of the local currency in US the Global Financial Crisis, and last year (2020).
dollars, adjusted for the corresponding inflation rates. In
other words, it is the real return on the currency from a US The United States “won” the last century. It won this com-
investor’s perspective, stated as an annual rate. petition against other major groupings such as Japan, Great
Britain, Europe, the world ex–United States, and the world.
So, the choice of market has a big impact on the historical The United States looks great. It looks great in the post–
relative returns of equities and bonds. If you have a long- financial crisis period and great over the long term, so the
term perspective, however, the currency barely matters. United States has truly been exceptional. If we teach using

10 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

the assumption that the United States represents the period and may not be representative of the future of
world as a whole, we are not teaching accurately. other places around the world, or even the United States,
of course. It is one of the stronger performers, no question
When we wrote Triumph of the Optimists, Paul, Mike, and about it.
I had a different view. The “optimists” of 1900 were inves-
tors who thought that risky securities (equities) would be But I’m still struck by the fact that your data show these
rewarding and would outperform more cautious invest- equity risk premiums are still there, everywhere around the
ments. Over the course of the 20th century, the optimists world, even if not as large as in the United States.
were vindicated, especially American optimists. We had
noted that, while we expected the stock market to continue Will Goetzmann: One of my takeaways from what you just
to provide a risk premium, it would be a smaller premium showed was that people thinking about investing in emerg-
than before. And we cautioned against extrapolating from ing markets should look carefully at whether there is any
the history of a single country (the United States). evidence of superior rewards from those markets. That is a
pretty compelling piece of evidence that your recent work
After almost a quarter of a century, we can look back and is showing. You’ve gone back and gotten a lot of those
see whether our caution was justified. And so far, despite markets that were originally missing. Yours is a much more
three bear markets and a pandemic, the United States has comprehensive view of the whole spectrum of potentially
generated higher stock market returns than non-US mar- investable markets. That was the exciting takeaway, which
kets. We discuss some explanations in our recent article, wasn’t there in your earlier work, that you and your col-
“American Exceptionalism.”9 leagues have put together nicely.

Elroy Dimson: A group called the World Federation of

Discussion of Elroy Dimson’s Stock Exchanges lists market capitalizations of different

Presentation markets. They focus not on free float, which is what we

do as soon as free-float information is available for each
Martin Leibowitz: Looking back at Exhibit 8, it reminds me market, but on total market cap. If you switch to the World
of the great article called “Survival,” by Stephen Brown, Federation of Stock Exchanges market-cap data, you
William Goetzmann, and Stephen Ross.10 I thought that was would find even more disappointment, because those
an absolutely intriguing concept. Yes, the United States is emerging markets that didn’t do so well have full weighting
exceptional, but was it just a kind of random draw where instead of the roughly 50% weighting they have with float
you had to have one exceptional country? I don’t think any adjustment.
of us really believe that, but the exhibit does show that
Jeremy Siegel: First, I want to push back on excep-
other countries have not done that well. These are not
tionalism as it relates to stock market outperformance.
terrific returns compared with what you see in the United
Outperformance is relative to expectations and is not an
States. There have been periods of real asset destruction in
absolute measure.
places like Japan and Germany and Russia.
Second, South Africa—would you call that a huge suc-
Elroy Dimson: The two big empires on the left-hand side,
cess story because it has a better return than the United
Austria and Russia, either entirely or virtually disappeared.
States? One of the most interesting graphs in the second
They do reappear, so notice the colors when they do come
edition of my book Stocks for the Long Run shows an
back. China never had a very big national stock market,
inverse relationship between stock market returns and
although it too disappeared and then came back.
real GDP growth.
But the United States has been a remarkable story. As
So, tell me about exceptionalism as measured by better
financial economists, when we think we’ve seen an anom-
stock market returns.
aly or a regularity that defies explanation, we assume the
anomaly will go away. But there’s now an increasing liter- Elroy Dimson: Yes, we reported on the inverse relation in
ature that says anomalies only partly disappear, and the Triumph of the Optimists. Since 2002, we’ve written more
success of the United States has at best partly gone away. on that topic.

Roger Ibbotson: I will admit that there’s a selection bias Laurence Siegel: As an aside, I’d point out that South Africa
with my data because I was looking at the United States. beat the United States only in local-currency terms, not in
It’s obviously been one of the best performers over the dollar terms.

9 See Dimson et al. (2021).

10 See Brown, Goetzmann, and Ross (1995).
CFA Institute Research Foundation 11
Revisiting the Equity Risk Premium

PRESENTATION BY CLIFF ASNESS: THE PRICE

OF THE STOCK MARKET MATTERS
Cliff Asness: In preparation for this, I decided to look at how Gordon-model Shiller forecast, where you just invert the
well the Shiller CAPE (cyclically adjusted price/earnings Shiller P/E and add a growth rate to it. This second way
ratio) has actually done since Campbell and Shiller’s work.11 assumes less mean reversion.
There are many measures other than the Shiller CAPE,
and later we can discuss how to make the CAPE better. I also looked at rolling out-of-sample means as a way to
Because the CAPE has become something of a lingua forecast the future. For this, you just look back 70 years
franca for discussion of the price of the stock market, starting in 1996; your forecast for the next 10 years is
I continue that. the prior 70 years’ mean; and then you just expand that
and keep continuing to use the long-term mean to make
The CAPE is often maligned, including sometimes by Shiller a forecast. And then, to get another estimate but in a
lately, for being wrong. It’s been wrong in the following cheating sort of way, I looked up the actual out-of-sample
sense: Since Campbell and Shiller first published their work mean over the 25 years from 1996 to 2021. This last way
on it, the stock market has been over its average CAPE is not fair, because this method says, “I know the average
far more often than under. As Elroy said, you expect that (in advance)—let me use that to forecast each 10-year
25 years is a long enough period to cause an anomaly to period.”
change, but it has not.
The two estimates based on the Shiller CAPE were signifi-
I look at the CAPE’s performance in a different way. In cantly better than the rolling mean. And they were slightly
the original work and in most of our papers on this, the better than the cheating mean, which surprised me a bit.
Shiller CAPE is not proffered as a great market timing This means that if you were going to forecast not the full
tool, or at least it shouldn’t be. What it should be used for out-of-sample 25 years but each 10-year period since
are longer-term predictions. Those are not the same as Campbell and Shiller’s work, you would have done better
market timing. using two rather obvious versions of their work than you
would by using the whole 25-year mean (known with hind-
sight) to forecast the subperiods. Even cheating doesn’t
Accuracy of Forecasts Made beat the two CAPE-based methods.
Using the Shiller CAPE
That doesn’t mean the criticisms of the Shiller CAPE are
I decided to see how the Shiller CAPE performed against wrong. If you are timing the market—for example, by using
some alternatives, out of sample—from 1997 on, because the CAPE to sell stocks short when that metric is high—the
1996 is about when they did their original work.12 To do CAPE method has been poor at that.
this, you use various techniques to make out-of-sample
forecasts of the equity risk premium and evaluate those Market timing is not, I would argue, what the CAPE was
various forecasts in different ways. Keep in mind that you designed to do. It was designed to give people an esti-
have only about two and a half independent 10-year peri- mate of what they might expect on equities going forward.
ods. This out-of-sample stuff is not statistics; it is scenario And for what it was designed to do, it’s done quite well.
analysis. Market timing is much harder than setting future long-term
expectations. If your forecast is for an asset return that
I look at the errors in the various forecasts and then look at is considerably less than the asset’s average return, but
some of the standard ways of evaluating how bad those still positive, and you underweight or short that asset for
errors were—the mean squared error, the average absolute 25 years, do you know what you get? Losses.
error, and the worst absolute error of the forecast—all com-
paring some forecast of 10-year returns to the subsequent So timing is different from forecasting. But I’d say that
actual 10-year returns. That is, if you have a forecasting efforts to make the Shiller CAPE better are great, as long
method, what was its mean squared error versus the actual as we use it for its intended purpose. When used in that
outcome over the out-of-sample period? way, even in its original form, it has been a pretty strong
success. I found this very interesting, and I had not looked
I look at two forecasts based on the CAPE. One is a roll- at it this way before.
ing out-of-sample regression method; the other I call a

For example, Campbell and Shiller (1998).

11

Alan Greenspan made his famous “irrational exuberance” comments based on Campbell and Shiller’s work in 1996.
12

12 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Changes in Valuation of the States is going to be better forever, that should bother
you. Do you really think an ever-increasing valuation gap is
US Equity Market versus reasonable?
Its Own History If the United States just inherently generated much higher
earnings growth, and that was 100% of the difference
It’s still hard to grasp how much starting and ending valua-
and valuations for those earnings stayed the same, you’d
tions matter.
be more tempted to say there’s something about the US
Looking at an approximately 70-year estimate, the appre- system that is exceptional on an enduring basis. There still
ciation in the Shiller CAPE is responsible for 1.3 percentage might be something, but it’s coming almost entirely from
points of the excess return of 6.5% that equities earned people being willing to pay a much higher price for US equi-
over cash. Without the Shiller CAPE appreciating, that return, ties than non-US equities, and also much higher than they
the equity risk premium over cash, would have been 5.2%. used to pay for US as compared with non-US equities.

There are two pieces of good news: (1) 5.2% is still a lot,
and (2) when you take out the appreciation in the CAPE, you Value versus Growth
actually get a more robust estimate of the equity premium. The value strategy has the same characteristics. Value has
While this second comment sounds a little mysterious, an lost money almost entirely because the dynamic portfolio
explanation of it is in my blog.13 The equity risk premium esti- of stocks that were cheap got relatively cheaper versus
mate is a little bit lower, but we know it with more certainty. the expensive ones. And, as someone with a horse in that
race, I like that better in terms of the future prospects
At any rate, valuation changes matter. A 1.3% difference
for the value factor. If value stocks had lost because the
in compound annual return over 70 years, which is the
fundamentals deteriorated, you don’t ever get that money
actual amount of time over which the CAPE appreciated at
back. If you lose from valuation differences expanding, you
that rate, adds up to a lot of money. We ignore valuation
have a decent chance of getting the money back (nothing
changes at our peril.
is guaranteed of course!).

Changes in the Valuation

of the US Equity Market versus
Discussion of Cliff Asness’s
Other Countries Presentation
Laurence Siegel: I believe that you’re right, that these
In the same article, I also looked at the United States
changes in valuation are responsible for most or all of the
against EAFE (Europe, Australasia, and Far East). Of course,
return difference between the United States and the rest of
you can only go back to about the 1970s on that. I also
the world. However, the United States might have become
looked at the value strategy.
expensive for valid reasons. Those would include an expec-
Let me focus on the United States versus EAFE because tation that high valuations in the United States will con-
it applies directly to stuff that Elroy was talking about tinue, legitimizing the current valuation.
and that Jeremy was talking about. The United States is
Setting that conjecture aside, what signals should we look
exceptional in its victory over EAFE. Over the whole period,
for to say that it’s somebody else’s turn and maybe we
from the beginning of the data, it’s still been a solid victory.
should bet against the United States?
If you want to really cherry-pick from Japan’s peak in 1989,
the US victory is gigantic.
Was the Increase in US Relative
Just about 100% of this outperformance is explained by the
CAPE of the United States going much higher, in relative Valuations Justified?
terms, than the CAPE of EAFE. This doesn’t mean the vic-
Cliff Asness: I’m not going to answer that; I’m going to
tory wasn’t real—you got the money. At any rate, the United
answer the question I want to answer, which is correlated
States has been exceptional.
to your question. In my article on this issue, I have a
But the most exceptional thing about the United States footnote saying almost exactly what Larry said: “What if
has been its leap in valuation, in price, against the world this was justified?” I ask this question in a few different
as measured by Shiller’s price to trailing 10-year earnings. areas where I looked at valuation changes. The question is
That’s a form of exceptionalism, but if you think the United whether you can expect the returns to mean revert going

See Asness (2021).

13

CFA Institute Research Foundation 13

Revisiting the Equity Risk Premium

forward. If the valuation change was justified, you don’t cities outside the United States. It seems like they have
expect valuations to mean revert. higher multiples than the United States.

If the United States tripled in value relative to other countries Cliff Asness: As someone who has only lost money on real
for good reasons, you take that money and run. But you still estate, I will accept your figures.
don’t assume it’s going to do it again over the next 30 years
or the next 70 years. So I think that’s a very good question. It Laurence Siegel: Roger, I think that real estate is more
does matter if it was justified. But it only matters for forecast- expensive in the capitals because capital cities have a
ing, going forward, which I’m going to be a coward on here. true monopoly position in many countries. In contrast, the
United States has multiple capitals, as does Germany. Berlin
If you want to get a long-term estimate of the relative is also quite cheap.
expected return of the United States, using historical data
overestimates it, whether that valuation increase was While multiple capitals may keep real estate prices under
justified or not. By simply projecting past returns forward, control, some of ours have gotten expensive, as in “London
you’re projecting that the increase in relative valuation will expensive.” Look at Los Angeles, look at New York.
happen again and again. This strikes me as a little crazy.
Thomas Philips: CAPE should never be used as a timing
But, again, it does matter a lot for whether you think tool. It really is an estimator of future returns.
(1) we’re just going to see similar returns in different places
On to your comment on changes in valuation—this goes
going forward—no more US exceptionalism, or (2) we’re
back to the mean reversion question. There really isn’t any
going to see mean reversion in returns, where US and
mean reversion. If there’s a change in valuation, there’s
non-US valuations come back toward each other and take
a change in expected returns and that’s it. Begins there,
back some of the gain from past revaluation. I’m going to
ends there. You have to think of a valuation measure such
let others take on that question.
as CAPE as an estimator of future returns, not as a basis for
projecting forward a sequence of realized returns.
Are Monopoly Effects Responsible
Bond investors are much quicker to get this than equity
for High US Relative Valuations? investors. Bonds have a printed number representing
expected return: the yield. The expected return isn’t printed
Martin Leibowitz: Cliff, quick and simple question: Do you
for equities, so it’s much easier to make that mistake
think there could be a monopoly effect here?
(regarding an increase in price as a harbinger of higher, not
Cliff Asness: Some companies in the United States are lower, subsequent returns) in equities than in fixed income.
just dominating the world with a network effect, and that It’s always equity investors who make this error.
could be part of it. I would bridge your question and Larry’s
Cliff Asness: There is nothing in there I disagree with—I’ll
and say that both of your comments are in the camp of
just reflect on that last point. For years, Rob Arnott has
today’s valuation difference being justified. I think your
made this comment. I’ve marveled at the fact that in the
story is a plausible one for the valuation in the United
bond world, if you have half the yield, people just under-
States being justified, but none of these stories forecasts
stand they will make half the money.
another tripling of relative valuation over the next 50 years
from here. Thomas Philips: That’s right.

But again, if the United States has beaten EAFE by X Cliff Asness: And in other parts of the investing world, they
percent historically—if that has all come from a justified don’t see this.
increase in valuations—as long as you don’t think that will
happen again, you don’t want to use the historical estimate. Laurence Siegel: That’s right. In the equity market, if the
You want to use an estimate that doesn’t give credit for yield falls in half, they may think they’re going to make
the one-time revaluation event. Anyway, I think that the double the money because the market just doubled to pro-
monopoly story is probably the most plausible one for why duce the lower yield. They then mentally project that price
you might pay more for US companies. In other work,14 how- performance forward.
ever, I show that it’s still unlikely that this is driving most of
what’s going on with systematic value. Thomas Philips: Or they’re using the historical mean as an
estimator because they don’t want to do the work of think-
Roger Ibbotson: In direct contrast to what’s going on in the ing about what’s really going on under the covers. It’s an
stock market where the United States has higher multiples, easy cop-out, and it really does lead you astray.
real estate seems to be much more expensive in major

See Asness (2020).

14

14 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PRESENTATION BY ROBERT ARNOTT: A SIMPLE

FRAMEWORK FOR FORECASTING EQUITY
RETURNS—THE DEVIL IS IN THE DETAILS15
Rob Arnott: A simple framework for determining forward (or at least the last 40): Yields tumbled, spreads narrowed,
returns would consist of (1) yield, (2) the norms and and valuation multiples for stocks soared.
expectations for income growth, and (3) some measure of
mean reversion. The investment firm Grantham Mayo van
Otterloo (GMO) makes what I think is a mistake by assuming
Dividend Yields and Expected
100% reversion of the valuation level to its historical mean Returns
in seven years. Empirically, that doesn’t happen. We take
a more simplistic or naïve approach of saying that maybe Dividends have always been the dominant source of real
the current spreads are a new normal. Maybe the current stock market returns. Given that, should we worry about
valuations are a new normal. Maybe they’re not, and you’re the current yield of 1.3%? Absolutely. Exhibit 13 decom-
going to get mean reversion. So, let’s just split the differ- poses the return for stocks into various components. This is
ence. Let’s assume halfway mean reversion. apropos of Roger’s work and Elroy’s work, but we’ve taken it
back to 1800 using G. William Schwert’s data.
Now, if you do that, you get an equity risk premium (ERP) of
roughly zero for the coming decade. If you don’t have mean In any event, the dividend comprises most of the return
reversion at all, there is a risk premium. But the simple for stocks, and if you take out the inflation component (the
framework provides 10-year return estimates with a stan- red line), it’s the vast majority. The blue line shows that
dard error of only 1 to 2 percentage points. real growth in dividends per share (of the S&P 500 and
predecessor indexes, and of earlier hypothetical indexes in
Where could the expectation that ERP is near zero over the periods when no published indexes existed) has been only
coming decade be wrong? Soaring earnings would help, as 11-fold over the last 221 years. And the revaluation upward
would soaring bond yields. The former is arguably unlikely (in the price/dividend ratio) has been threefold, as we see
when stimulus is tapered. Low yields help the equity risk on the orange line—it is more if you start at market lows
premium but do little to support lofty equity valuations such as those in the early 1930s.
unless buybacks are real and substantial. Empirically,
they aren’t.
Dividends per Share Have
The current pairing of high inflation and negative real bond Grown Very Slowly
yields is very dangerous; moreover, demographics offer
headwinds that are worrisome. So, I believe that an ERP My comment that dividends per share have grown only
near zero is a reasonable expectation on a 10-year basis, 11-fold in real terms in 221 years may come as a surprise.
because the headwinds are potentially quite dangerous. If Aggregate dividends have, of course, grown roughly 1,000-
you use yield, plus growth, but no mean reversion, you get fold with the macroeconomy. But, if we take away inflation,
a 2.9% real return for US equities. If you assume halfway we’re down to roughly 50-fold. Then we need to take away
mean reversion, you get a −0.8% real return, compounded “dilution.” The classic meaning of dilution is a diminished
over 10 years. That is slightly below the US aggregate share of ownership, as the shares outstanding of a given
fixed-income index, which is currently yielding less than company increase through secondary equity offerings
inflation expectations. (SEOs).

We’ve gone back over the last 30 to 50 years, depending on But, for the index investor, there’s a second—and usually
the asset class, and asked whether this simplistic method- more powerful—form of dilution from new enterprise cre-
ology would lead to pretty good forecasts. The short answer ation. Consider that over 50% of today’s S&P 500, by market
is yes. The longer answer is, the more pessimistic forecasts cap, went public less than 25 years ago. If we owned the
are likely to have a larger standard error, a larger degree S&P 500 from 25 years ago, and didn’t trade it to stay cur-
of uncertainty. The fact that the forecasts have tended rent with changes in the index, we’d now own just under
to be too pessimistic—in other words, that the outcomes half of the index, because of new enterprise creation and
were better than the range forecast—is largely a function secondary equity offerings. While new enterprises typically
of the overwhelmingly bullish tenor of the last 50 years don’t initially dilute the index investor’s dividend stream,

Thanks go to Amie Ko for her help on this presentation and exhibits.

15

CFA Institute Research Foundation 15

Revisiting the Equity Risk Premium

Exhibit 13. Dividends Have Always Been the Dominant Source of Equity
Market Real Returns

Source: Based on data from Ibbotson Associates. For more information, see Arnott and Bernstein (2002).

as they mature, they pay dividends, tacitly diluting the entrepreneurial capitalism tends to dilute existing share-
dividend income from the preexisting names. holders. You need to raise new capital from the labor
markets to fund new or growing businesses. You have the
The same effect comes into play even if we do trade to stay opposite of net buybacks—you have net new share issu-
current with the composition of the index, as we have to ance. And that net new share issuance is substantial.
sell stocks with a (typically) robust yield in order to rebal-
ance into new index holdings with a (typically) low or zero So, what we find empirically in Exhibit 14 is that whether
yield. When all these various forms of dilution are taken into you’re looking at share prices in red, dividends in purple,
account, real growth in dividend income is brought down to or earnings in brown, you’re looking at growth that largely
11-fold in 221 years for the investor who doesn’t reinvest keeps pace with per capita GDP, not aggregate GDP. There
dividends. is an exception over the last 25 years in which it has more
closely kept pace with aggregate GDP growth. But that’s
because profits have become a larger and larger share of
Do Earnings and Dividend GDP. The dilution has continued apace.
Growth Keep Pace with GDP,
We also should note that earnings do mean revert. If you
or per Capita GDP? look at trailing 10-year real earnings growth and subse-
quent 10-year real earnings growth, the correlation of
When calculated on a per-share basis (this distinction
one with the other is about −57%, as shown in Exhibit 15.
is important), earnings and dividends have generally
It’s a pretty powerful correlation. It’s not a superb predictor,
grown with per capita GDP, not with overall GDP. Why
but it’s a pretty good one.
would that be? Because, when the population is growing,

16 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 14. Earnings and Dividends Have Generally Grown

with per Capita GDP Growth

Source: Based on data from Ibbotson Associates. For more information, see Arnott and Bernstein (2002).

Exhibit 15. …And Earnings Mean Revert

Peak/Trough Earnings vs Future Real Earnings Growth, 1926 to Date
12%

8%
Subsequent 10-Year Real Earnings Growth

y = –0.0973x + 0.0347
R2 = 0.3227

4%

0%

–4%

–8%
–60% –40% –20% 0% 20% 40% 60% 80%
Real Earnings vs 10-Year Avearge

Source: Based on Ibbotson Associates data.

CFA Institute Research Foundation 17

Revisiting the Equity Risk Premium

Exhibit 16. Do Share Buybacks Supplement Dividends? For Established

Large-Cap Names, Sometimes. On the Broad Market, Rarely

Source: Based on data from the Center for Research in Security Prices (CRSP) at the University of Chicago.

The Role of Dilution in Equity Return showed in a 2005 Financial Analysts Journal article that the
untraded original 1957 S&P 500 beat the actual index by up
Do share buybacks supplement dividends for established to 0.5% per annum, depending how you handle corporate
large-cap companies? For some companies, they do; for the actions.16
broad market, rarely. Exhibit 16 shows dilution in the broad
market year by year as well as over longer rolling periods. Do Low Bond Yields Support
Dilution is measured by the amount by which the growth in
the aggregate market capitalization of the CRSP broad-cap High Equity Valuations or Growth
US equity index exceeds the growth in the index of share Beating Value?
price action of that same portfolio. That is a very simple and
very powerful way to measure the extent of dilution. The We also hear the narrative that low bond yields support
best way to think about dilution, I think, is to recognize that high equity valuations or support growth relative to value.
about 60% of the total market capitalization in the United Narratives can become self-fulfilling prophecies, but only
States today is companies that didn’t exist 30 years ago. temporarily. It’s useful to look at long-term data and ask,
Think about that for a second! “Is the relationship sound?” The old-fashioned Fed model,
which compared nominal bond yields to stock earnings
So, 50% of the market is new. If you owned an index fund yields, looked pretty good from the mid-1960s to the mid-
30 years ago and you didn’t trade it—that is, if you didn’t 2000s. But if you widen your horizons, the relationship is
dilute your existing holdings by selling the tiny bit of any utterly useless before that time period and becomes pretty
stock that is dropped from an index fund and a little bit of weak after 2006 (see Exhibit 17). And should one come
all the stocks still in the index fund, in order to buy Tesla to the conclusion that, gosh, if 2% yields are great for US
(and the hundreds of other such changes in the index over stocks, then why are 0% yields not great for European and
the last 30 years)—you would today own only 40% of the Japanese stocks? Shouldn’t they be priced at infinity?
market. So, dilution is in fact very powerful. Jeremy Siegel

See Siegel (2005).

16

18 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 17. Do Bond Yields Support High Equity Valuations?

Note: The US earnings yield series shown in this exhibit is calculated as the simple average of an earnings yield computed from Robert Shiller
online data (US Stock Markets 1871–present) and an earnings yield from the Global Financial Database. The US bond yield series is calculated as
the simple average of 10-year US bond yields from three sources: Bloomberg, Global Financial Data, and Ibbotson Associates, where available.
Source: Based on Robert Shiller online data, Ibbotson Associates data, and Bloomberg data.

It also fails back in the 1950s when you had very large normally, empirically across eight countries and spanning
Shiller earnings yields in the 7% to 10% range and 2% a half century, lead to a Shiller P/E of around 11. But we’re
bond yields. We did some work (hat tip to Marty here) at 38. So, this is a threat if the Fed has no exit strategy.
on the linkage between the level of inflation and real If the Fed can move us back to this territory in the not-
rates and the seemingly natural level of the Shiller P/E. too-distant future, then that threat can go away.
Exhibit 18 shows, based on global data spanning eight
countries, that the Shiller P/E averaged across the coun- Finally, we’re left with GDP growth. Exhibit 19 illustrates
tries, or within the eight countries and going back 40 to the dynamics of GDP growth and demographic trends.
50 years, is [in the] mid-20s. That’s not a bad Shiller P/E, In the G–8 countries, the median age is sliding upward and
but it’s highly dependent on both the real rate and the the proportion of senior citizens is soaring. In emerging
inflation rate. markets, you’re right in this sweet spot where GDP growth
is strongest when the population of young to middle-aged
Back in the 1970s, we had high inflation and a negligible adults is soaring. GDP growth is hurt when the population
real yield. That corresponded empirically to an average of children is soaring and also when the population of
CAPE of about 11. At the beginning of the 1980s, we had senior citizens is soaring.
high inflation and a high real rate, the Volcker effects, which
corresponded to a Shiller P/E of about 13. Then we came In the emerging markets, the population of children is fall-
into this sweet spot: the 1990s, 2000s, 2010s, where we ing and the population of senior citizens is rising—but not
were kind of circling around this benign sweet spot of mod- fast, yet. The next 20 years will change that. The median
erate inflation, moderate real yields. This dot that’s off the age is right in this sweet spot. So, you have a little bit of
chart to the left is the last 12 months, where inflation has a headwind from demography. That headwind turns into
been over 5% and real yields have been nearly −4%. a potentially major headwind for stocks in the coming
15 to 20 years because, as the population of senior citi-
If you extend these lines, you’ll see that this current zens soars, the impact on average stock market returns
observation point (5% inflation and −4% real yield) would is quite large.

CFA Institute Research Foundation 19

Revisiting the Equity Risk Premium

Exhibit 18. Do Real Yields and Inflation Rates Have a Bearing

on Equity Valuation Multiples?

Source: Based on data from Global Financial Data. For more information, see Arnott, Chaves, and Chow (2017).

Exhibit 19. GDP Growth Slows with a Growing Population of Children

and Retirees

Note: The solid black line is the equal-weighted country weighting scheme. The dotted line is the square root of GDP weighting scheme. The gray-
shaded area is the 95% confidence interval.
Source: Based on data from the United Nations, the Penn World Table, and Global Financial Data.

20 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Discussion of Robert Arnott’s that retail investors are getting more and more sensitive
to tax efficiency.
Presentation So, for all these reasons, I think you sort of have the wrong
model here. Not that DCF is the wrong method—DCF is the
Are Buybacks a Substitute right method; it’s how you apply it that would essentially
get you the right kind of answer here.
for Dividends?
Rob Arnott: My only pushback is that you’re completely
Roger Ibbotson: I should respond right away, because
ignoring the dilutive effect from new enterprise creation. If
one of the topics you’ve talked about is a zero expected
you include that, the S&P 500 has had average dilution over
return arising from the fact that dividend yields are so low
the last 30 years of a little over 1% per year, meaning that
and growth has been so low. Both of those points, though,
net buybacks have been negative. There were times when
change when you put buybacks into the picture. In 1980,
buybacks were net positive—2005 to 2008, in particular.
there was a change in the US law to allow a safe harbor
for buybacks. Before that, it was basically considered Exhibit 16 desperately needs to be brought up to date
insider trading. So, after 1980, there was a general shift because I last updated it in late 2017. Back then, the nar-
in companies paying dividends versus paying out cash in rative was that buybacks are huge. No; net of new share
other ways. issuance, net of changes in the index and the dilution that
comes from new enterprise creation, net buybacks were
If you look at a traditional discounted cash flow (DCF)
only about 1%, and that was a little bit of an abnormal
model where the return is equal to the dividend yield plus
outlier. So, I would push back on your results.
dividend growth, it’s sort of wrong on two points: first, if the
dividend yield is artificially low, and second, if measured Roger Ibbotson: You might have noticed that the actual
growth in dividends over, say, the last 40 years since 1980 paper we wrote did include share issuance.
has been really low because you started with a higher yield
and moved to a lower yield, the DCF expected return is also Rob Arnott: It included share issuance, but it didn’t include
very low. But neither of these inputs is good for forecasting new enterprise creation and changes in the index itself,
because they are both downward biased. which dilutes shareholders in a big way. If you own that
static portfolio from, let’s say, 30 years ago that consti-
But once you put the buybacks in, total cash flow payouts tutes just 40% of the S&P, then you’ve missed out on the
have been similar to what they’ve always been. What’s hap- growth associated with those new enterprises coming
pened is that buybacks have replaced dividend yields as into the picture. It’s a very, very big missing component.
the preferred form of cash flow to the investor. One reason I have a world of respect for your legacy, your work on
is that it’s far more flexible for a corporation to pay out cash long-term returns; I don’t have a world of respect for that
when they want to, rather than on some regular schedule. particular paper.
They don’t have to worry about cutting their dividends in
bad times. If they cut out the buyback, it’s no big deal. And Roger Ibbotson: Okay—that whole debate is written up in
it’s much more tax efficient to not pay dividends, given letters to the Financial Analysts Journal.17

See Arnott and Bernstein (2018); Straehl and Ibbotson (2018).

17

CFA Institute Research Foundation 21

Revisiting the Equity Risk Premium

PRESENTATION BY MARTIN LEIBOWITZ: P/Es,

RISK PREMIUMS, AND THE g* ADJUSTMENT
Martin Leibowitz:18 I’m not trying to solve the issue of risk premium of 4.4%, which is consistent with the assumed
premiums as much as trying to shed light on two different 7.1% total return.
angles of it. Let’s look at Exhibit 20. It is a very simple way
of looking at total return in terms of dividend yield or, in So that’s one way of adjusting risk premium calculations
view of the comments that were just made, cash flow yield: based on an earnings yield so that they are more consis-
the flow or yield that accrues to investors without price tent and make more sense. If you just say that the earn-
appreciation—look at it that way. Then you add the growth. ings yield minus the riskless rate is the risk premium, it
leaves out a growth measure of some sort. And the growth
Suppose a stock has a P/E of 22 and sells at a price such measure is not the growth of dividends. That would be an
that its expected return, k, is 7.1%, a rate commensurate overstatement. It should be this 2.5% value-added growth,
with the stock’s risk. The 22 P/E provides an earnings yield g*, not the 5.5% growth in dividends or growth in earnings.
of 4.6%. We apply, say, a 35% payout ratio to that earnings
yield to arrive at a dividend yield of 1.6%. If you add to that The key point is that using the earnings yield alone as a
a 5.5% growth rate, which is shown as the growth assump- return estimate can fail to fully account for levels of earn-
tion in the exhibit, you get back to the total return of 7.1%. ings growth that reflect the economic value added, or
The other bars in the left panel of Exhibit 20 can be viewed franchise value, of the firm—that is, earnings growth higher
as the two components of the 7.1% total return. than that predicted by the retention rate times the return
on capital. To provide theoretically consistent long-term
Now, that 7.1% can be viewed from a different angle. Look return projections, an adjustment term must be added
at the right panel of the same exhibit, which starts with the to the earnings yield. That term is the g* we have been
earnings yield instead of the dividend yield and then tries discussing.
to build up to the 7.1% expected return in a different way.
This “net growth” approach contributes to a more con-
The P/E of 22 gives you a 4.6% earnings yield, but that sistent estimate of equity risk premiums. Perhaps most
yield is not the expected return because it does not fully importantly, when these premiums are based on the earn-
account for the firm’s growth. If you were to add the total ings yield alone, this adjustment helps compensate for the
5.5% growth rate to the earnings yield, you would get resulting underestimation.
10.1%—far above the 7.1% expected return. To match the
7.1% return, a “growth-like” number of 2.5% would have to
be added to the 4.6% earnings yield, as shown in the left
Probability of Beating a Given
bar in the right panel of Exhibit 20. Yield Target
Exhibit 22 shows another way of looking at the equity risk
The Missing 2.5% premium.

This 2.5% figure serves as a kind of net growth add-on that The curves show the risk premium needed for equi-
trues up the earnings yield to the stock’s expected return. ties to provide a 40%, 50%, 60%, and 80% probability of
This net growth term, g*, plays a key role in estimating the having returns greater than the riskless rate. Exhibit 22
equity risk premium. assumes that equities have a 16% volatility. The lower
curve shows that for an 80% probability of success over
The equity risk premium is typically portrayed as the
a 10-year period, the required premium is 4%. Over a five-
expected return over the yield on some “riskless asset,”
year period, again using a simple normal distribution
but one sometimes sees this premium computed as the
model, the required premium rises to 5.6%. This is in the
earnings yield less the riskless rate. Suppose we have a
range of numbers that we’ve been hearing throughout this
riskless rate of 2.7%, as shown in Exhibit 21. (I wrote these
conversation.
materials in 2018, three years ago, when a 2.7% yield was
more realistic.) In our example, if we were to subtract that If you try to get a little more true to the real world—forget
2.7% yield from our earnings yield of 4.6%, we would get about the normal distribution and instead use a simu-
a risk premium below 2%, which is pretty bleak. If we add lation that takes into account volatility drag, lognormal
the 2.5% net growth term back in, however, we get a risk

I’m speaking as part of my new consulting organization with the snappy title of Advanced Portfolio Studies, rather than as a Senior Advisor
18

of Morgan Stanley.
22 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Exhibit 20. Return as Dividend Yield + Growth or Earnings Yield +

“Growth Adjustment”

Source: Morgan Stanley Research.

Exhibit 21. Earnings Yield and Risk Premium Are Trued Up by Adding g*

Source: Morgan Stanley Research.

CFA Institute Research Foundation 23

Revisiting the Equity Risk Premium

Exhibit 22. Simulation-Based Risk Premium versus Probability of Beating

Any Yield

Source: Morgan Stanley Research.

effects, and so forth—you need a 6.5% risk premium to can’t capture all of it. Some of it must be reinvested to put
have an 80% probability of beating the riskless rate over a in place new assets to support the expanded business.
five-year period. You can actually derive that from first principles using the
Edwards–Bell–Ohlson equation. And you can get an expres-
While these seem like reasonable numbers, the probabil- sion for your “net growth.”
ity of success is only one among many relevant factors.
Nevertheless, this concept of a probability of success over Martin Leibowitz: Yes, you can. And Jim Ohlson’s work is
some time horizon seems like a useful first-order consider- important. But just in terms of the basic DDM, it comes out
ation for the risk premium discussion. that net growth, g*, is
b
g* = × (r − k ),
Discussion of Martin (1 − b)
where b is the earnings retention rate, r is the company’s
Leibowitz’s Presentation return on capital, and k is the market rate of return or
opportunity cost of capital. The key thing is to look at the
Thomas Philips: Marty, I should note that what you call “net
spread over the market rate, r – k, because that’s what
growth” can be written as growth times one minus book-
generates added value.
to-price. Basically, you get earnings growth, but investors

24 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PRESENTATION BY MARY IDA COMPTON:

THE EQUITY RISK PREMIUM IN REAL LIFE
Mary Ida Compton: I’m the lone practitioner here, in a that the clients can avoid huge surprises. The consul-
room full of scholars. When we (in the consultant com- tant would love to agree with the actuary to unify these
munity) think about the equity risk premium, we think goals, but we’re constrained by the realities of the market.
about the way it applies in real life. I work for Alan D. Biller & The consultant’s returns are forward-looking rather than
Associates. We are a consultant to Taft–Hartley plans. They backward-looking, so clearly we will not always or even
are governed by boards of trustees made up of company usually agree with the actuaries.
management and members of labor unions. They are not
investment professionals, and they hire us to develop their So the clients are trusting me in my role as a consultant.
investment policy statements. So that means that we need We meet with trustees at least quarterly and continually
to have some asset allocation guidelines. educate them on relevant topics. We don’t have much
time to talk to them—they have a lot of paperwork to deal
Asset allocation for these clients is an optimization prob- with and the investment report is just one aspect of their
lem with some asset class constraints. There’s a target meeting preparation. Capital market assumptions are an
rate of return and the need to minimize risk at that level of even smaller part of that. Between meetings, we email
expected return. It’s pretty basic. internally generated research pieces with data from aca-
demics, economists, and the likes of Bloomberg; and we
But here’s the rub: For the last decade or two, the truly appreciate the work that all of you in this forum are
target rate of return has been increasingly difficult to doing so we can use it as it filters into our capital market
achieve at modest levels of risk. Actuaries dictate the assumptions.
target investment return. We don’t get to decide what
that is. They expect the plan to meet this target return At every quarterly meeting, comments are tossed into
over the long term, and they base that number on his- the discussion, like those advertisements they used to
torical market performance, which is problematic if have in the movies where they would show you, for a
markets are declining. There’s a long time lag between fraction of a second, a picture of a Coca-Cola and you’re
what markets do and how quickly actuaries will change supposed to “subliminally” get thirsty and go out and
their numbers. buy one. In that way, we try to remind the trustees of
market issues impacting their plans. We do it in conversa-
tions between meetings, dinners, phone calls, and even
Conveying Reasonable Capital golf outings.
Market Assumptions to Clients
We’re trying continually to talk to individual people with
To help our clients, we try to set reasonable expectations responsibility for the plan and understand how they think,
for the future, so they can judge how much money there and how we can best convince them of our capital market
will be to distribute to the pensioners. The trustees want assumptions. This is a relationship business.
the higher actuarial return. They use that in their DCF model
to apply to their future liabilities. They would like to make
those liabilities appear smaller, so that both management
Capital Market Assumptions
and labor can contribute less to the pension plan. And at the Present Time
should the company ever consider selling itself, it’s better
to have the lower apparent level of liabilities on its balance It’s no surprise that 10-year projected returns have been
sheet. So, those are their incentives. declining for at least 15 years, probably longer, but nobody
remembers before that. Exhibit 23 shows our capital
The actuary wants to keep the trustees happy and not get market assumptions. We construct them by surveying
fired. So, they’re reluctant to lower the actuarial rate, which 25 to 50 asset managers and gathering all of their
would cause everybody a lot of pain in the short run while expected 10-year projections. We do two things: (1) We
preventing pain in the long run if the lower rate is econom- look at the median number, and (2) we look at a Winsorized
ically correct. mean.19 And then we average those two.

Then there’s the consultant who, at least in our case, wants We throw out wildly obvious outliers. What you see in
to provide the most accurate prediction of the future so the exhibit is after these adjustments.

A Winsorized mean is a type of average that removes the influence of the most extreme outliers. See Dodge (2003).
19

CFA Institute Research Foundation 25

Revisiting the Equity Risk Premium

Exhibit 23. Capital Market Assumptions in 2021: 10-Year Projected Returns

Are Down

Note: Eq = equities; RE = real estate; FI = fixed income; HY = high yield; Agg = aggregate.
Source: Based on data from Alan D. Biller & Associates.

Conversations with the Client that caused losses in fixed income and spooked some
equity investors. High equity valuations and rising interest
So, let’s go into a meeting. The client will say, “Why are rates have created volatility that is likely to persist, so we
these expected returns so low? Are you kidding me?” And want to get that idea in there, too.
we say, “No, sorry, we’re not kidding you. This is reality.
The other thing we do is to pose questions being asked by
You’ve heard about how high the prices of equities have
the media and the general public. Is inflation here to stay?
gone. The prices have increased faster than expected
Can additional government spending be a positive cata-
earnings can support.”
lyst, or could higher taxes be a negative catalyst? We then
Then we provide some exhibits to illustrate that. You see describe the trajectory of current variables that influence
how low bond yields are. Savings accounts pay peanuts. markets, using pictures, graphs, and whatever else is nec-
Certificates of deposit don’t get you much more. Mortgage essary to make the point.
rates are super low. These are facts they can really relate
We talk about the Fed having more impact on the markets
to. They know them from their real lives; they understand
than it used to. That is something that people need to pay
these concepts.
attention to and look at.
They also kind of understand the concept of an equity risk
So, that was one meeting. Outside the meeting, consul-
premium, but we don’t ever call it that. But we do convey
tants have to use multiple approaches because we’re
the idea that, if bond yields are down at 1%, you can’t
dealing with a group of individuals who are all different
have stocks up at 12% for the next 10 years. If stocks
from one another. Each of them has different expecta-
had an expected return of 12%, people would all put their
tions. There’s macroeconomics, demographics, the idea
money in stocks and then the stock market would go way
of equity returns building on bond returns. As a result,
up and then drop off a cliff. They kind of understand that
consultants think of their job as an art form rather than a
concept, too.
science. They are there to translate complex information
At a recent meeting, one of the consultants talked about into a language that their clients can understand and use
the recent increase in the 10-year Treasury yield and how to make decisions.

26 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

The Endless Zero Interest but we work hard to make sure that clients are not going to
come up short by taking liquidity risk and getting nothing
Rate Policy for it.

I seem to be able to use Exhibit 24 no matter what year it is. We can also expand the opportunity set of investments.
This is from 2017, but the slide is not out of date. Interest Some clients are pretty traditional, pretty conservative.
rates are near zero. We’ve been telling clients that interest You can move into “core-plus” bonds from plain old core
rates are near zero for years, and we’re glad they’re still bonds. You can go into mid-cap equities. You can go into
listening to us. international equities. We make the case that international
will give you diversification, and we never know when the
Back in the meeting, the client may ask: “Okay, now what United States is going to underperform. It will happen; we
do we do?” Maybe they say, “We just had this amazing just don’t know when. So maybe you should be prepared
year—that should help us pad the future.” Or maybe they for that. (See Exhibit 25; note that the country and regional
say, “The world has changed—see those recent [very returns are presented as cumulative excess returns over
positive] market returns?” We get all those comments. the MSCI World benchmark, which may be an unfamiliar
space.)
We first respond that you shouldn’t take a recent year of
excellent performance as a windfall. It is more conservative Some clients don’t like to hear that. They say they’re not
to consider it as eating into future years’ performance—so going international, that it’s just not happening. We have
be aware. Second, don’t take the last couple of years of to respect that, and it becomes a constraint that we have
excellent performance as an indication of future perfor- to live with. We do make plainly clear that capital market
mance. This soaring market is not a new normal. We can assumptions, ours or anyone else’s, don’t always come
help you get additional return, but it’s going to involve true. So we show them Exhibit 26, which shows realized
taking some risks. returns versus the expected returns based on our capital
market assumptions over time. We’ve been humbled, to say
Ways of Enhancing Expected the least. But the capital market assumptions help guide
the discussion. They’re pretty good directionally—they’re
Return not so good during market shocks.
You can take on additional liquidity risk by going into To sum up, the most important thing we can do is to
private capital: private equity and private debt. We see educate our clients to help them keep their expectations
a lot of flows going in that direction. It’s a little scary, realistic.

Exhibit 24. What’s Happening Now: Interest Rates Near Zero

Source: Based on data from FRED (Federal Reserve Bank of St. Louis).

CFA Institute Research Foundation 27

Revisiting the Equity Risk Premium

Exhibit 25. Nominal Total Returns on US, Non-US, and World Equity Markets
A. Nominal Total Return Indexes of 10 Leading Non-US Countries,
in US Dollars, Expressed Relative to MSCI World Index, 1970–2020
Return US$
4.00

2.00

1.00

0.50

0.25

0.10
1970 1976 1982 1988 1994 2000 2006 2012 2018

Australia Canada France Germany

Italy Japan Netherlands Spain
Switzerland UK

B. Nominal Total Return of US and World ex-US, in US Dollars,

Expressed Relative to MSCI World, 1970–2020
Return US$
4.00

2.00 World Ex-US

1.00

US
0.50

0.25

0.10
1970 1976 1982 1988 1994 2000 2006 2012 2018

Note: Panel B uses same scale as Panel A.

Source: Based on data from MSCI.

28 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 26. Realized versus Expected Returns, 60% Global Equities/

40% Core Bonds, 1980–2019
Average Annual Return (%)
20
18
16
14
12
10 Expected Return at Start of Period
8
6
4
Realized Average Annual Return
2
0
01/80–12/89 01/86–12/95 01/92–12/01 01/98–12/07 01/04–12/13 01/10–12/19

Source: Based on data from Bloomberg L.P. and Alan D. Biller & Associates.

Discussion of Mary Ida have been coming down from implausible levels and are
finally getting to something semi-sensible.
Compton’s Presentation Mary Ida Compton: It’s also interesting that our expected
Roger Ibbotson: I can’t help pointing out that actuaries returns are much lower than those of our competitor con-
are not economists. For political reasons, their numbers— sulting firms. I think they like to say, “Oh, sure, we can get
discount rates or expected returns—are way too high; you 8%.”
they’re playing some sort of game where they have to go
Martin Leibowitz: Way back during the 1980s, the actuar-
along with that.
ies actually had kind of a standard discount rate of 4%. They
More to the point, I think that Miller and Modigliani’s kept it there—they didn’t change it much at all—and this
argument—that assets and liabilities are separate—is the was at a time when interest rates soared, as we all know,
relevant way to look at this. The discount rate for the lia- until rates reached 15% on long-term bonds. And actuaries
bility is determined by the risk of the liability, not of the still kept the discount rate at 4%, okay? That was their con-
assets—and, if you are really making a commitment to pay servative approach to things.
these liabilities, the liabilities are a riskless asset to the
What one could do, of course, was to buy long-term
beneficiary. So, the riskless rate would be the discount rate.
bonds—“defease” or immunize the pension portfolios—
It would not be the return on the stock market or a portfolio
and, all of a sudden, the cost of that strategy would be
of different assets.
far, far less than your notional liabilities based upon the
I’ll leave it at that, realizing that those kinds of argu- 4% discount rate. Some firms, like Salomon Brothers and
ments are not what you need, Mary Ida, to handle your a couple of others, did this, and it was a win–win for
client. everybody, except possibly the actuaries.

Mary Ida Compton: We’d love it if they’d listen to you, Roger! As a result, corporations and their balance sheets greatly
improved. The beneficiaries benefited from much sounder
Robert Arnott: I was struck by the fact that, relative to your decisions in terms of the assets that are supporting
forward-looking returns, the actual subsequent returns their needs. Needless to say, some of the firms that were
were underwater during most of the last 25 years, as involved in the transaction process also did reasonably
shown in Exhibit 26. This is all very familiar territory to us. well. So, actuaries can be wrong in two directions!
Folks anchored on wonderful returns in the 1980s and
1990s and forecasted 10% annual returns when stocks Laurence Siegel: I appreciate Marty’s historical perspective
were yielding 1% in dividends, so a 10% overall return was that there was a long period where the actuaries were
utterly implausible. The return forecasts made by actuaries too low, not too high.

CFA Institute Research Foundation 29

Revisiting the Equity Risk Premium

PRESENTATION BY ANTTI ILMANEN:

UNDERSTANDING THE MANY COMPONENTS
OF EXPECTED RETURNS
Antti Ilmanen: Earlier presentations initially focused on simple long-run way of looking at the payout yield through
historical average returns. Then we moved to the fuller an earnings lens without getting involved in the buyback
picture of time-varying expected returns. I’ll be very much revolution.
in the second camp in my comments here.
The second number in the DDM equation, the growth rate,
Exhibit 27 shows 120 years of expected real returns on is the 1.5% real earnings-per-share (EPS) growth rate.
US stocks and bonds using some simple measures. I want Later on, I’ll have an exhibit on each of yield and growth
to emphasize that this is not a world of compressed pre- and at that point I’ll go into details.
miums right now; the premium between equity and bond
expected returns is pretty normal. What is unusual is that For the mean reversion term, I’m assuming no mean rever-
both series are near their historical low points: second or sion over the forecast horizons, so that’s different than
first percentile. This scenario contrasts sharply with what what Rob Arnott, for example, was showing. The logic
we saw 20-some years ago during the dot-com boom, behind my choice is that there have been some structural
when bonds had 4% real returns. So, everything now seems changes, and we may return to that. But there’s also a
expensive compared to its own history, and there is no practical reason why I’m not including mean reversion.
easy solution to that problem for anybody. That is offered If you look at the dark blue line in Exhibit 27, it’s quite
just as background. time-varying. If I assume mean-reverting valuations,
that would imply an implausibly wide historical range
and an even lower expected equity real return today.
Estimating the Inputs for the
Dividend Discount Model Components of the Yield
So, how did I come up with the equity expected returns? on Equities: Buybacks, Dilution,
I did it from the perspective of the dividend discount model
(DDM): yield plus growth and plus or minus mean reversion.
and All That
For yield, I’m estimating the real yield at 1.4% by averaging On the question of yield, there has already been some
two measures: (1) dividend yield and (2) half of the recip- debate here between Rob Arnott and Roger Ibbotson.
rocal of the Shiller PE. The second measure is meant to be a Exhibit 28 shows some data from the Straehl and

Exhibit 27. Simple Expected Real Return of the US Equity Market

and 10-Year Treasuries, 1 January 1990–30 September 2021

30 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 28. US Annual Dividend Yield and Other Parts of the

Net Total Payout Yield, 1900–2020

Source: Straehl and Ibbotson (2017), extended by author.

Ibbotson study.20 I really like this study, much more than not EPS growth, in this modern version of the DDM.
Rob did. We start with a long history of dividend yields, and Conveniently, there has recently been not much difference
then the red line here adds buyback yields. Those are gross between aggregate earnings and EPS growth.
buybacks. In the spirit of what Rob was saying, you should
then add the negative contribution from new share issu-
ance. I’ve included both of those in the exhibit.
Is the 1.5% EPS Growth Estimate
Too Low?
If you look at the last 20 years of these data, which are for
the S&P 500 constituents, you can see buybacks and new That may have been a little complicated. I have assumed,
share issues largely canceling each other out. In the olden and I keep assuming, a 1.5% real growth rate for EPS. This is
days, we thought that dividend growth was subject to dilu- in the spirit of what Bernstein and Arnott did 20 years ago,
tion of between 1.5% and 2% per year, with dilution caused looking at the historical growth rates of GDP per capita, div-
by the growing number of shares and measured as the idends per share (DPS), and EPS.21 Looking at the 120-year
spread between the per-share and aggregate growth rates. history in Exhibit 29, you can see that all those numbers
Nowadays, that spread seems to be much closer to zero. are 1.5% or a little higher. Importantly, these are real com-
pound growth rates (geometric means). If we looked at
In other words, dilution has largely gone away because new nominal rates and arithmetic means, instead of 1.5% we’d
issues are offset by buybacks. This means you are taking see 7.5% real EPS growth—not so stingy.
these two factors into account correctly in the calculation
of the broad payout yield (dividend yield plus buybacks I looked at different time periods. Over the last 40 years,
minus new issuance), so you don’t need to also adjust all these numbers are a little higher. DPSs have grown at
for dilution in the growth term of the DDM equation. The a particularly high rate over the last decade and have had
growth term should then be aggregate earnings growth,

Straehl and Ibbotson (2017). This exhibit, like most others in this section, is from my latest book (Ilmanen 2022).
20

See Arnott and Bernstein (2002).

21

CFA Institute Research Foundation 31

Revisiting the Equity Risk Premium

Exhibit 29. US Economic, Dividend, and Earnings Growth, 1900–2020

Notes: RGDP = real gross domestic product (showing economic growth). RDPS = real dividends per share. REPS = real earnings per share. DPS and
EPS on the S&P 500 index and its predecessors.
Source: Based on data from Robert Shiller’s website (www.econ.yale.edu/~shiller/) and FRED.

the highest growth rate of any of these variables over that L’Her et al. start with the total nominal return, and then
40-year history. split that into (1) real return and (2) inflation and currency
effects. Next, in the second step in Exhibit 30, you use
I’d be open to thinking about 2% real compound growth the DDM approach of splitting real return into the dividend
instead of that 1.5%, but one reason why I haven’t is the yield, real DPS growth, and valuation change.
evidence that Elroy and his coauthors have gathered on
markets outside the United States. They have looked at Then they go further. In the next step down in Exhibit 30,
DPS growth in many other countries, and that variable has they split up real DPS growth into (1) real economic
averaged near zero in real terms. So, again, I stick with that growth per capita and (2) what they call the DPS gap. The
1.5% estimate and I don’t try to do anything fancier. In our DPS gap is the difference between DPS growth and eco-
practical capital market assumptions, we do perform a nomic growth. In the subsequent step, they subdivide the
more complex calculation (time-varying, country-varying), DPS gap into sales-per-share gap (SPS) and some inter-
but that’s not based on empirical evidence because there esting economic measures: change in the payout ratio and
aren’t enough 10-year earnings growth numbers for a seri- change in profit margin. The last step apportions the SPS
ous study. gap between net buybacks and relative dynamism of the
listed sector.

The Deep Decomposition The point is that you can use simple mathematical tricks,
of Equity Returns such as using logs and adding and subtracting terms,
to come up with any decomposition of stock returns you
Exhibit 30 is a bit of a detour. I wanted to give a shout-out want—but they have tried to come up with something eco-
to the literature on deep decomposition of stock market nomically interesting. At the bottom of Exhibit 30, you can
returns. We’ve all done some versions of this, and the study see that they’ve applied this method on many countries,
that I’ve seen take it further than any other is by L’Her and using 20-plus years of data, to decompose stock market
his coauthors.22 He refers to the components of return as returns into these pieces.
“net buybacks and seven dwarfs.”
Overall, I think this is an interesting path to explore in the
future—to try to better understand equity returns in the

See L’Her, Masmoudi, and Krishnamoorthy (2018).

22

32 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 30. Decomposing Stock Market Returns into More Granular Pieces
Nominal Return
(USD)

rusd

Real Return Inflation and

(local currency) Currency Return
r inf + fx

Change in Real DPS

Dividend Yield
Multiples Growth

dy p/d dps

Real Economic DPS Gap

Growth per Capita

gdp pc dps gap

Change in Change in
SPS Gap
Payout Ratio Profit Margins

sps gap payout margin

Net Buybacks Relative

Dynamism

nbb rd

Inflation
Price-to- Real per Nominal
Net Relative Payout Real and
Dividend Dividend Capita Margin = Total LC = Total
Buybacks, Dynamism, Ratio Currency
Yield, dy Change, GDP Growth Return in
nbb rd Change Return, r Return,
p/d Growth USD
inf + fx
All Mkt Avg 2.9% –0.3% 2.1% –2.2% 0.5% 0.5% 1.2% 4.7% 1.9% 6.6%
US 1.9 0.4 1.4 –1.8 2.4 0.7 1.2 6.1 1.9 8.0
China 2.6 1.0 8.2 –26.5 14.9 –0.9 1.3 0.7 2.8 3.5

Source: AQR; L’Her, Masmoudi, and Krishnamoorthy (2018).

past and looking forward by linking equity returns to their a big valuation gap between private and public equity but
fundamental determinants. that it narrowed hugely in the mid-2000s when the endow-
ment model was popularized for the first time. Since then,
it has been pretty narrow.
Private Assets
There’s another argument that investors may be fine with
I want to address private assets very briefly. As shown in
private assets not earning a premium over public ones.
Exhibit 31, there’s a pretty good case to be made that there
While investors would prefer to get an illiquidity premium
was a bigger gap between private and public equity returns
for private assets, they also like the smoothness of private
in the past than there will be in the future. The gap will
asset valuations so much, for reporting and compensation
probably be much more modest, unless private asset fees
purposes, that they may not insist on such a premium.
fall a lot. The time-series chart shows that there used to be

CFA Institute Research Foundation 33

Revisiting the Equity Risk Premium

Exhibit 31. Equity Premia in Private and Public Equity: Expected Returns
on the Two Asset Classes May Not Differ Much
A. Ex Ante Valuation Gap between US Public and Private Equity, January 1998–September 2018

B. Fair Illiquidity Premia in Private Assets May Be Offset by Investors’ Paying for the Lack of Mark-to-Market

Notes: Increasing popularity of private equity since the mid-2000s may have reduced the prospective return edge over public equity. P/E man-
agers’ gross “alpha” is partly offset by high fees. Institutions’ required returns may balance illiquidity and smooth returns. Despite its popularity,
investable P/E AUM remains much smaller than public equity (about $5 trillion versus $100 trillion).

Discussion of Antti Ilmanen’s Antti Ilmanen: The Shiller CAPE is a really good starting
point. You then ask whether you can do something better.
Presentation First you should ask: Is it for timing, is it just for planning
purposes, and so forth? An interesting aspect of the cur-
Thomas Philips: If you had to make a choice between using rent situation is that it is not a world of compressed premia.
Shiller’s CAPE or a decomposition method to make a fore- The premia are pretty standard, but everything is expen-
cast, which would you choose and why? sive. So our question is whether we expect all these expen-
sive valuations, including CAPE, to cheapen in the future.
34 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

I do think that there is a link to bond yields: If you start But that’s just for the next decade. US public pension plans,
to see real bond yields rising, all asset classes, including which use such optimistic expectations, say somewhere in
equities, will be in more trouble. So, CAPE is a good start- their fine print that they have a 40-year horizon. With such
ing point, but depending on the purpose, I would look at a long horizon, we can look beyond the next decade during
other information as well, including what’s happening in which we have negative real bond yields and so on. The
bond land. subsequent 30 years could be better. Horizon matters.

Roger Ibbotson: In the Straehl and Ibbotson paper,23 we did Martin Leibowitz: As one who has sat on a number of
use total payouts to predict market returns, and they actu- investment boards of major universities—from Harvard, with
ally were a little better at it than CAPE. the biggest endowment, all the way down to the Institute
for Advanced Study, with probably the smallest endowment
Antti Ilmanen: I do have a bit of a favorite, called equity of a prestigious organization, and along with Cliff on the
share. It is equity market cap as a percentage of the total of board of the University of Chicago, that wonderful bastion
all asset classes, and in past timing exercises, it performed of intellectual volcanic eruptions—none of them has a
clearly better than CAPE. It looks at the prospects for equi- spending rate below 4%. They’re almost all in the 4%, 4.5%,
ties relative to, say, bonds, and so it has missed the richen- 5%, sometimes 5.5% range.
ing of all asset classes together. That’s been a good thing
for backtests because, again, those low expected returns These numbers are the basis of extensive and very
have not yet been converted into low realized returns. But, thoughtful discussions that, I think, are eventually deter-
for relative purposes, I would look at equity share. mined by one factor: The institutions need the money. They
cannot tolerate a lower spending rate. And many of them
Laurence Siegel: Equity share misses two concepts. One have a spending formula that is not only based upon a
is absolute as opposed to relative valuation, as you said. combination of forecasts but heavily weighted toward what
The other is the preference of governments and corpora- they did last year.
tions for issuing debt. Governments have gone hog wild
in spending, so they have had to float a lot of debt to Laurence Siegel: Having also worked for a very large
finance it. That issuance drives equity share below what endowed institution, the Ford Foundation, I can see the
it otherwise would be. problem. If you stop paying your professors, you’re going to
have bigger problems than a slightly shrunken endowment.
Antti Ilmanen: Yes. As always, those types of metrics Sometimes, as Jack Bogle said, you just have to budget for
depend both on the performance of each series and on a shrinking asset base in real terms.
issuance. Maybe performance has dominated and the issu-
ance part is just noise, but I don’t know. Empirically, equity Cliff Asness: I’d like to comment on Marty’s comment, lest
share has been helpful at forecasting, beyond the relative he or I ever get accused of being a perma-bear. Marty and
performance of equities and bonds. I share the distinction of basically being fired from the
investment committee at the University of Chicago for
Laurence Siegel: Interesting. saying we should not sell stocks relatively close to the
2009 low, and, if possible, that we should buy them. So, you
Thomas Philips: Do you calculate equity share using
can take even a hallowed institution with brilliant people on
just corporate debt and corporate equity, or do you use
a committee, and still get panicky, herd-following behavior.
overall debt?
Marty and I were saying, “We don’t know if stocks will go
Antti Ilmanen: I’ve seen both versions, and it looks fine lower, but they’re much lower than they were before; who
either way. I’ve also seen versions that throw in real estate should own stocks now, if not us?” And we lost that fight
or other assets. and our jobs.

Elroy Dimson: If you were talking to a standard endowment, Martin Leibowitz: Yes, but it saved me a lot of donations.
not one that has special access and special skills, what
Thomas Philips: That’s very human, a universal problem.
would you regard as a sustainable level of spending, one
It’s not just the University of Chicago.
that would maintain the real value of the assets?
Cliff Asness: The other guy who agreed with Marty and
Antti Ilmanen: The type of exercise that you see here
me on the University of Chicago investment committee
suggests a real return of less than 2% for a 60/40 portfo-
was David Booth, who didn’t get fired. When you lose an
lio, so I think 2%. For the next decade, a 5% real return or
argument where the school’s biggest donor is on your
spending rate is a pipe dream. Two percent is realistic; you
side, that’s impressive. That reveals how hard it is to go
could get 3% with some useful strategies on top of 60/40.
against bad thinking. This was the opposite of the current

See Straehl and Ibbotson (2017).

23

CFA Institute Research Foundation 35

Revisiting the Equity Risk Premium

discussion about low expected returns being ignored—for Now, stop and think—does that really make sense in a
once, we were screaming that expected returns were world where any kind of forecast, any kind of judgment, any
better than normal. People don’t like to buck the trend in kind of sense of valuation matters? If the stock market is
either direction; whatever has been going on will continue. booming, you’d think what you should do is not only sell it
back to your original allocation but go below that, because
Thomas Philips: I’m going to make a comment on the flip presumably the market is less attractive than when you
side of this. I’m in India, taking care of my parents. There’s made your original allocation. And vice versa—if the market
a venture capital boom out here. It’s like the United States is down and you’re in a situation where you think the eval-
in the late 1990s or China in the 2010–2015 period, and uations are really attractive, should you go back to your
people think this is going to go on forever. The idea that it previous allocation? No, you should go beyond that. But no
might not is a difficult conversation to have at either end. one does. I can assure you that no one does that.
It’s not one-sided.
Thomas Philips: Yes, we’re human. We’re all humans.
Martin Leibowitz: People tend to stick with their asset allo-
cation no matter what. Institutional investors will choose an
allocation and rebalance back to it, in many cases almost
mechanically.

36 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PRESENTATION BY THOMAS PHILIPS:

LINKING EXPECTED RETURNS TO FUNDAMENTALS
Thomas Philips: In my mind, the right way to address the Marty Leibowitz, and Stan Kogelman (LLK).25 It can be
questions being raised in this meeting is to first weave succinctly summarized as: “Over a 2Dmod-year horizon, the
economic insights into simple models of expected return, return of a constant duration zero-coupon bond is close to
and to then estimate these models using robust statistical its initial yield.”26 (Dmod means modified duration.) A 2Dmod-
techniques, because the signal-to-noise ratio in finance year horizon means a time horizon or holding period equal
is truly awful, as you can see from Exhibit 32. (It’s actually to twice the modified duration of the bond. Remarkably,
easier to communicate with Voyager than it is to predict the theorem applies surprisingly well in practice to coupon
financial markets!) So, forget ordinary and generalized bonds as well as to portfolios of coupon and zero-coupon
least squares, the Gauss–Markov theorem, the consump- bonds that are rebalanced back to their initial duration
tion CAPM, and above all, forget TensorFlow, which is the monthly or even annually, provided that the holding period
modern-day equivalent of “plastics.” is shortened by one year to 2Dmod − 1 years. So, if the mod-
ified duration of a constant duration bond portfolio is five
The way I calibrate my thinking about expected returns is to years, its return over a nine-year period starting at time t
start by asking whether I should think about them in nomi- is close to its initial yield at time t. Exhibit 33 tests this
nal, real, or excess return terms. (By “excess return” I mean principle against the subsequent realized returns27 of the
the return in excess of cash.) And, for reasons I don’t fully Bloomberg Barclays US Aggregate Index, which has a fairly
understand, nominal returns are far more predictable than stable duration, over a long period.
real returns and excess returns. So, I work out everything in
nominal terms and then translate between nominal and real Remarkably, the LLK theorem works even better for the Agg
quantities using inflation breakevens, which are traded.24 than it does for the US Treasury Index. At any rate, LLK is
a great way to calibrate your thinking about the expected
return of bonds. So, my estimate of the expected return of
Expected Returns 10-year US Treasuries is 1.6% per year, compounded.
on Bond Portfolios
Does it work in credit? Yes, it does, as Exhibit 34 illustrates
Let me start with bonds. I’m going to appeal to a very using a short-term, high-grade credit factor constructed
beautiful theorem originated by the late Terence Langetieg, using a long position in the ICE BofA one- to three-year

Exhibit 32. Signal-to-Noise Ratio in Finance versus Other Applications

Application Signal-to-Noise Ratio

Capital markets 0.5

Deep space communications 1–10

AM radio 100

FM radio 1,000

Good audio amplifier 10,000–100,000

Very low-noise laboratory amplifier 100,000–10,000,000

24Breakeven inflation is the difference between the nominal yield of a nominal Treasury bond and the real yield on a maturity-matched
inflation-protected Treasury bond (a TIPS bond). It is the market’s expectation of future inflation. A long breakeven position can be created by
taking a long position in a US TIPS bond and a maturity-matched (ideally a cash flow timing matched) short position in a US Treasury nominal bond.
25 See Langetieg, Leibowitz, and Kogelman (1990).
A constant-duration zero-coupon bond is a hypothetical bond that is continually renewed so that its duration stays constant. Think of this as
26

continually selling the currently held bond and buying a new bond with the same duration using the proceeds of the sale.
27 That is, the return of the index from time t to time t + 2Dmod − 1.
CFA Institute Research Foundation 37
Revisiting the Equity Risk Premium

Exhibit 33. Bond Portfolio Return Predictions from LLK Model Compared
with Realizations, 1976–2020

Note: LLK = Langetieg, Leibowitz, and Kogelman (1990).

Source: Based on Bloomberg data.

Exhibit 34. The Short-Term, High-Grade Credit Factor

40% 7%

35%
6%

30%
5%

25%
4%

20%

3%
15%
Plot Area

2%
10%

1%
5%

0% 0%
2/1/1988 2/1/1992 2/1/1996 2/1/2000 2/1/2004 2/1/2008 2/1/2012 2/1/2016 2/1/2020

Cumulative Yield Differential (LHS) Cumulative Return (LHS) Spread vs. Treasuries (RHS)

Source: ICE Data Indices LLC. All rights reserved.

38 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

AAA-A index and a short position in the ICE BofA one- to relationship is close to linear, except that it’s a bit concave
three-year Treasury index. The LLK model works remarkably when prices are depressed (i.e., when CAPE is low and
well, but you do have to worry about defaults and index 1 CAPE is high). I don’t understand why; the reason may
construction rules that force bonds out of an index if their relate to real economic activity, such as restructurings
rating drops below some limit. In this case, the index in that start to become popular when prices are depressed.
question is a AAA-A short-term index;28 and if a bond gets Despite not fully understanding the reason for the convex-
downgraded to BBB+ or worse, it gets pushed out of the ity in the relationship between valuation ratios and realized
index. The issuer doesn’t always default, and the bond returns, I’m going to exploit it when I predict the returns of
often pays out par at maturity (but not always—Lehman the S&P 500.
bonds didn’t pay out in full)—but you, as an index investor,
don’t get to participate in its recovery.
Noise Reduction by Eliminating
You can see some losses getting locked in during the 2008 the Worst Quarter Each Year
Global Financial Crisis, but even so, this is a decent way to
calibrate your thinking about credit premia. What do I think Given all the attention that CAPE has attracted in the media,
the short-term, high-grade credit factor is going to give us? it’s worth thinking about the economics of what Campbell
About 0.3% per annum. and Shiller are actually doing when they average 10 years
of real earnings to compute the denominator of CAPE.
In my view, the 10-year average does three things:
Expected Returns on Equities
1. It reduces noise in earnings (the 10-year average is a
I now come to equities. Many years ago, I derived the fol-
simple linear filter),
lowing expression for the expected return of the stock
market, starting with the Edwards–Bell–Ohlson equation 2. By adjusting for inflation, it makes past earnings more
and making some simplifying assumptions:29 readily comparable to present earnings,31 and
3. It gives you a better perspective on the earning power
E[Et +1 ]  B 
E[r]Market = + g×1− t  of the S&P 500 over an economic cycle.
Pt  Pt  (1)
E[St +1 × Profit Margint +1 ]  B  Ádám Kóbor, who works at NYU’s endowment, and I
= + g×1− t .
Pt  Pt  reduce noise in earnings in a very different way that also
eliminates the need for any inflation adjustment (see
In the first line of Equation 1, the expected return of the Exhibit 36).32 We find that the worst quarter’s earnings
market is the forward earnings yield plus a scaled growth in each year are by far the noisiest, and if you discard
term that is functionally the same as Marty’s g*. (Marty the worst quarter, add up the remaining three quarters’
might have a different perspective on that term.) In effect, earnings, and multiply the resulting three-quarter sum by
not all of the growth in earnings accrues to investors— four-thirds, you get a remarkably good, unbiased, low-noise
some of it must be reinvested in assets needed to support predictor of next year’s earnings. Think of this as a simple
the growth. You can also write the expression in terms nonlinear filter that is applied to quarterly earnings. The
of sales and profit margins, as I did in the second line of first line of Exhibit 36 summarizes the statistics of annual
the equation, and that ties in nicely with Warren Buffett’s (i.e., sum of all four quarters) earnings. The volatility of this
emphasis on the market cap-to-GDP ratio as a powerful series is about 33% per annum—it has lots of skew, and the
predictor of future equity market returns.30 excess kurtosis is 24, so the total kurtosis is about 27.

We can proxy forward earnings yields very roughly with By throwing out the worst quarter’s earnings each year
1 CAPE (Shiller’s 10-year CAPE), which is the current price (the E3 series), volatility comes down by about 45% and
divided by the average of the most recent 10 years’ real skew and excess kurtosis dissipate almost completely.
earnings. We can also proxy expected returns with realized You could throw out the two worst quarters and multiply
returns—though you need to be careful when doing so; the sum by 2 (E2) or keep only the best quarter and mul-
changes in expected returns can drive realized returns far tiply by 4 (E1), but there’s little benefit to throwing out
away from expectations. You can see in Exhibit 35 that the more than the worst quarter. Additionally, look at line 5

An AAA-A (often pronounced “six-A”) index is one that consists only of bonds rated AAA, AA, or A.
28

See Philips (1999).

29

See Buffett and Loomis (2001).

30

Adjusting by revenues would be even better and can be shown to be equivalent to averaging profit margins over the past 10 years. See Philips
31

and Ural (2016).
See Philips and Kóbor (2020).
32

CFA Institute Research Foundation 39

Revisiting the Equity Risk Premium

Exhibit 35. Empirical Relationship between Valuation Ratios and Subsequent

Realized Returns, 1925–2010 (subsequent returns until 2020)

Notes: The subsequent return is the realized return from time t (the time at which the CAPE is computed) to t + 1 (one-year subsequent return)
or from t to t + 10 (10-year subsequent return). The last year that can be included in a subsequent return computation (either t + 1 or t + 10) is
2020, because that is the last year for which we have realized return data.
Source: Based on data from Robert Shiller’s website (www.econ.yale.edu/~shiller/).

40 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 36. Discard Worst Quarter to Fix Bias/Sigma/Skew/Kurtosis:

Statistical Properties of Filtered and Unfiltered S&P 500 Earnings,
31 December 1925–31 December 2020
_
Earnings Series NObs X σx Min Max Skew Kurtosis

Et+1/Et 95 1.088 0.331 0.22 3.43 3.5 24.2

E3t+1/E3t 95 1.064 0.184 0.49 1.53 −0.4 0.7

E2t+1/E2t 95 1.063 0.176 0.53 1.47 −0.3 0.5

E1t+1/E1t 95 1.061 0.170 0.65 1.48 0.0 0.2

Et+1/E3t 95 0.998 0.214 0.19 1.49 −0.9 1.8

Et+1/E2t 95 0.956 0.212 0.17 1.38 −1.0 1.6

Et+1/E1t 95 0.915 0.206 0.17 1.33 −1.0 1.4

OEt+1/OEt 32 1.068 0.174 0.60 1.47 −0.6 0.8

OE3t+1/OE3t 32 1.066 0.152 0.69 1.38 −0.5 0.0

OE2t+1/OE2t 32 1.068 0.145 0.71 1.32 −0.7 0.2

OE1t+1/OE1t 32 1.068 0.143 0.71 1.28 −0.8 0.2

OEt+1/OE3t 32 1.031 0.168 0.55 1.34 −0.8 0.6

OEt+1/OE2t 32 1.011 0.162 0.53 1.27 −1.0 0.8

OEt+1/OE1t 32 0.991 0.156 0.51 1.22 −1.1 1.1

Source: Philips and Kóbor (2020), using data from Robert Shiller and S&P Dow Jones Indices.

of Exhibit 36—there’s essentially no bias in our forecast I can obtain two independent forecasts for the 10-year
of next year’s earnings using E3. The simple “throw out return of the S&P 500 from two models—one based on
the worst quarter” trick results in a very good forecast of earnings to price (earnings yield), and another based on
one-year forward earnings, certainly much better than a sales to price—and then combine the two forecasts to get
forecast based on CAPE. Interestingly, we don’t see any real my final forecast for the forward looking 10-year return of
improvement when applying our trick to operating earnings the S&P 500. I’ll include a quadratic term in each of the
(the various OE series). models to capture the empirical concavity that I see in the
data, and let’s see where we come out.

Comparing Earnings-to-Price Exhibit 37 shows the results. I plot actual 10-year real-
and Sales-to-Price Models of ized returns against out-of-sample forecasts made using
expanding-window robust regressions, of the 10-year
Expected Return return of the S&P 500 using the filtered earnings yield
and the sales-to-price ratio. The x-axis displays the pre-
But our “throw out the worst quarter” trick doesn’t give us
dicted return, and the y-axis displays the corresponding
a feel for what earnings might have been if we averaged
realized return.
over an economic cycle. I’m going to address this using
the sales-to-price ratio, because sales times profit margin In-sample, the fits (which I have not displayed to minimize
equals earnings. For any given level of earnings, profit mar- visual clutter) are decent, even though they are noisy. The
gins must be low when sales are high and vice versa, and out-of-sample predictions, however—which are what really
competitive forces in an open economy, often driven by matter to investors—are awful: The relationships appear
new entrants, will tend to induce some degree of reversion linear (but with the wrong slope) when returns are low
in profit margins.

CFA Institute Research Foundation 41

Revisiting the Equity Risk Premium

(i.e., when valuation ratios are high), but there is an enor- but even so, the quadratic term has certainly cleaned
mous flat region when return forecasts are high (i.e. when things up.
valuations are low), say from a predicted return of about
8% per annum on up. The dashed line has a slope of 1 and an By the way, the quadratic term also cleans up CAPE (see
intercept of 0, and perfect forecasts would plot right along Exhibit 39). With the quadratic term added, CAPE’s out-of-
the line. Plotting the data in this way gives us a quick feel for sample predictions are noticeably better.
how good (or bad) our predictions are. We can get formal and
set up a Mincer–Zarnowitz framework here, but I’m not going Combining the Earnings-Based
to go down that road—a picture paints a thousand words.
and Sales-Based Models
Guess what happens when we add a quadratic term to
our out-of-sample expanding window robust regressions? The way I’m going to make a forecast in practice is to take
As you can see from Exhibit 38, the flat spots clean up, an earnings-based model, quadratic, robust regression,
and the points plot roughly parallel to the dotted line! Not filtered earnings, and a sales-based model, again linear and
perfect, but very good. Revenues look better than earnings, quadratic, out-of-sample robust regressions, and combine
their forecasts. Even with (naïve) equal weights for the two

Exhibit 37. Out-of-Sample Predictions Made Using E3t/Pt and St/Pt Aren’t Great

(continued)

42 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 37. Out-of-Sample Predictions Made Using E3t/Pt and St/Pt Aren’t Great
(continued)

Notes: All dashed lines have slope = 1, intercept = 0. Dates on the plot are 1970–2010. I use data for sales and earnings from 1946. The first
25 years are used to build an initial model, and my first true out-of-sample forecast starts in 1970.
Source: Philips and Kóbor (2020), using data from Robert Shiller and S&P Dow Jones Indices.

models, forecasting ability is good, as shown in Exhibit 40. You can see a bias in that there are a lot of points that plot
It is possible to refine the weights further and weight the above the dashed line. That’s a feature, not a bug, of our
forecasts in inverse proportion to their forecast error vari- forecasts, because it’s generally recognized that realized
ance, and we’ve done that in Philips and Kóbor (2020; not returns were higher than expected returns over the period
shown).33 we studied on account of a decline in expected returns.34
So, it’s a good thing, not a bad thing—in other words, it’s to

This combination of forecasts has proven so effective in a variety of settings that economists actually refer to it as the Forecast Combination
33

Puzzle! See, for example, Claeskens, Magnus, Vasnev, and Wang (2016).
For more on this, see Philips (1999).
34

CFA Institute Research Foundation 43

Revisiting the Equity Risk Premium

be expected in a good model over this period—that we have their long-term average of about 6% of revenues. An equally
more points above the line than below it. weighted average of the two forecasts is 2.2% per annum.

If, instead of equally weighting the forecasts, I weight them

Interpretation of the Philips in inverse proportion to the variance of their forecast errors,
and Kóbor Results I get an expected equity return of about 2.3%. That’s almost
the same as what I get using a simple (equally weighted)
Let me put these graphs in perspective. If I just use average. Using CAPE and CAPE-squared, the predicted
filtered earnings-to-price and its square, I’m predict- 10-year return of the S&P 500 is about 2.45% per annum.
ing 6% per annum returns for the next 10 years. That’s A quick peek into the Federal Reserve’s FRED database tells
essentially saying the very high profit margins of today me that breakeven inflation is about 2.6% per annum. So,
are going to persist. If I use sales-to-price and its square, my forecast of expected real return on equities is negative,
I get −1.6% per annum for the next decade. In effect, the but the equity premium is ever so slightly positive because
model is telling me that profit margins are going to decline my forecast of bond returns from the 10-year Treasury yield
from their current level of about 12% of revenues toward is 1.6%. Exhibit 41 summarizes these forecasts.

Exhibit 38. …But a Quadratic Term Perks Up the Models Nicely

(continued)

44 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 38. …But a Quadratic Term Perks Up the Models Nicely (continued)

Notes: All dashed lines have slope = 1, intercept = 0. Dates on the plot are 1970–2010. I use data for sales and earnings from 1946. The first
25 years are used to build an initial model, and my first true out-of-sample forecast starts in 1970.
Source: Philips and Kóbor (2020), using data from Robert Shiller and S&P Dow Jones Indices.

Moreover, there’s a big question that I’ve left unanswered. Mean Reversion in Equity Returns
I have handled nonlinearities in the relationship between
forecasts and realizations in one way (adding a quadratic or Equity Risk Premiums
term) but have not provided any evidence to suggest that
On the mean-reversion question, I’m not a fan of mean
it’s the right way. Are there other ways to do it? Surely there
reversion, either in returns or in valuation ratios. In Exhibit 42,
are! If so, what is the best way, and why?
I don’t see either CAPE or interest rates reverting to some his-
Another question to which I have no answer: Why are torical mean. In fact, I think that there has been a structural
nominal returns more predictable than real returns? Is it shift in the mean of both series after 1980: The mean CAPE
because inflation is volatile and injects additional noise has risen, and interest rates have simultaneously declined.
into real returns? I don’t know. Also, will profit margins stay
I think most of what people think of as mean reversion in
elevated? Again, I don’t know, but I suspect they’re going to
returns is just the result of a change in expected returns
converge toward their long-run average, which is about half
or interest rates, which results in a one-time shock to
their current level. Competition is a fierce force.

CFA Institute Research Foundation 45

Revisiting the Equity Risk Premium

realized returns—a decline in expected return results in entirely by the steady decline in the bond’s yield (which is
above-average returns during the transition, and vice versa. a good proxy for its expected return). And over the entire
Forward-looking returns will, of course, reflect the new level 10-year life of the bond, the return is exactly 10% per
of expected return, and if this differs from the prior expected annum, which is the same as its initial yield.
return, the new realized returns will, on average, be lower (or
higher) than the old realized returns. It’s easy to see what
appears to be mean reversion in realized returns (i.e., high
Mean Reversion in Value
returns followed by low returns and vice versa)—and to fail to and Growth
grasp that this is driven entirely by a shift in expected return.
That said, some things are indeed mean reverting. In fact, I’m
It’s easy to visualize this phenomenon for a 10-year going to show you an example where mean reversion is real
zero-coupon bond that is issued with a yield of 10% and and exists for sound theoretical reasons. Even so, mean rever-
whose yield declines by 1% each year, so that its yield sion can disappear for an extended period without warning.
when it finally matures is 0%. The realized return of the
bond in each of the first five years of its life exceeds its My example involves mean reversion in the per-share earn-
initial yield of 10%, and then falls below 10% in each of the ings of growth and value indexes. Style indices are recon-
subsequent five years. The return of the bond appears to stituted by the index provider every year. Growth and value
be mean reverting, but it’s not—the shift in returns is driven indices, in particular, are reconstituted so that each has

Exhibit 39. The Quadratic Term Helps CAPE As Well

(continued)
46 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Exhibit 39. The Quadratic Term Helps CAPE As Well (continued)

Notes: All dashed lines have slope = 1, intercept = 0. Dates on the plot are 1970–2010. I use data for sales and earnings from 1946. The first
25 years are used to build an initial model, and my first true out-of-sample forecast starts in 1970.
Source: Philips and Kóbor (2020), using data from Robert Shiller and S&P Dow Jones Indices.

half of the capitalization of the market immediately after the earnings of the value index implode, then recover and
reconstitution. Rebalancing induces mean reversion in per- catch up with those of the growth index. The long-term
share earnings and equalizes their long-term growth rate.35 growth rate of per-share earnings is about the same for
It also equalizes the long-term price return (not the total both styles.
return) of the two indices.
But when I pull the data window forward to 2021, the
Exhibit 43 shows the situation I observed when I first pattern looks very different: The reliable divergence-
noticed mean reversion in style-index earnings around followed-by-convergence pattern that is so evident in
2001. The per-share earnings of the growth index are much Exhibit 43 disappears after December 2006.36 From 2007
more volatile than those of the value index. You can see to 2021, the earnings of the value index grew much more

For a proof, see Philips (2002).

35

The graph is not shown but is available from the author at [email protected] and can also be found in Martin, Philips, Stoyanov, Scherer,
36

and Li (forthcoming 2024).

CFA Institute Research Foundation 47
Revisiting the Equity Risk Premium

slowly than those of the growth index. In fact, the earnings 2006. I don’t have a good explanation for why the historical
of the growth index grew substantially faster than nominal pattern went awry. But even with a very strong rebalancing
GDP, which is clearly unsustainable! force attempting to equalize the growth rate of earnings,
markets can go in directions that are unexpected—and that
A close look at both time series of earnings makes clear run diametrically opposite to what theory predicts—for a
that the earnings of the value index dipped very sharply long, long time.
during the Global Financial Crisis in 2008 (no surprise, as
banks were hugely overrepresented in the value index), I’m not a fan of the mean reversion story because it’s so
as well as during the COVID-driven blip in 2020, but in both easy to misinterpret changes in expected return as move-
cases, they did not rapidly recover and catch up to the ment toward some nonexistent historical mean—especially
earnings of the growth index as they had so reliably done when your analysis is data driven, and you look only at
in the past. In short, the mean reversion that my theory realized returns and don’t calibrate your thinking using a
predicted just hasn’t happened for 15 years! reasonable theoretical framework.

This divergence in earnings growth explains a big chunk

of the underperformance of value relative to growth since

Exhibit 40. …And Combined Forecasts Are by Far the Best

(continued)

48 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 40. …And Combined Forecasts Are by Far the Best (continued)

Source: Philips and Kóbor (2020), using data from Robert Shiller and S&P Dow Jones Indices.

Exhibit 41. Summary of Equity and Bond Expected Return and Equity Risk
Premium Forecasts
• My current views on nominal expected returns for the next 10 years:
■ S&P 500 via E3/P and (E3/P )2: 6.0% per annum
■ S&P 500 via S/P and (S/P )2: −1.6% per annum
1
■ weighted average of these two forecasts: 2.3% per annum
σ E2
■ S&P 500 via CAE/P and (CAE/P )2: 2.45% per annum
■ UST10: 1.6 % per annum, 10-year Breakeven inflation: 2.6%

• My expected real returns are negative, but the equity premium is slightly positive.

CFA Institute Research Foundation 49

Revisiting the Equity Risk Premium

Exhibit 42. Is There Mean Reversion in Expected and/or Realized Returns?

Source: Based on data from www.econ.yale.edu/~shiller/data/ie_data.xls.

Exhibit 43. Growth and Value Index EPS Do Mean Revert until 2000:
Per-Share Earnings Growth, 31 December 1976 to 31 December 2000

Source: Philips (2002).

50 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Discussion of Thomas Thomas Philips: That has been true historically, but there is
no known reason for it to continue in perpetuity.
Philips’s Presentation Rajnish Mehra: So, if you multiply these numbers together,
1 1
Rob Arnott: Just a quick observation: Empirically, mean × , that will give you, as a ballpark number, gross return
3 3
reversion in returns is weak. on capital of about 11%, that is, 1 divided by 9. Depreciation
Thomas Philips: It probably doesn’t exist at all. It’s really is 6 to 7 percentage points, so you’re left with a stable
a reflection of moving from one expected return regime number for return on capital. If you look at growth account-
to another. The transition induces a realized return that is ing, you’re going to get a return on capital of 4% or 5%,
different from the return that you expected. It’s not that assuming those numbers are stable.
returns are mean reverting; it’s that expected returns are
Thomas Philips: But they have not been stable since
unstable and move around a lot.37
1991—profit margins for the S&P 500 have quadrupled,
Rajnish Mehra: I think the planning horizon has a lot to do from about 3% of revenues to about 12% of revenues.
with whether there is mean reversion. If you look at data
Rajnish Mehra: The capital-output ratio was very stable
at a daily, monthly, or even yearly frequency, you might
in the economy. This is one of Nicholas Kaldor’s “stylized
not see it, but if you look at a lower frequency like five or
facts.”39 The share of output going to capital is, or used
seven years, there is a good capital-theoretic reason why
to be, pretty stable. In my talk I’ll discuss what has gone
these low-frequency returns should be mean reverting. The
wrong since 2007. But these relationships, on which most
capital-theoretic reason has to do with the business cycle.
of our macroeconomic intuition is based, held up to 2007,
The capital-output ratio is very stable—it’s 3 or 3.5; and the
and after that you’ve seen a huge change.
share of output going to capital is about a third.38

For evidence of instability in expected returns, see Philips (1999).

37

The capital output ratio is usually expressed as the ratio of GDP to capital employed in the economy; it is typically about 3. Its recipro-
38

cal, capital/GDP, is thus about one-third (although it has decreased in recent years). The share of output going to capital comes from the
Cobb–Douglas production function.
In 1957, the economist Nicholas Kaldor listed six “stylized facts” that he said described the dynamics of economic growth. Much effort in
39

macroeconomics has gone toward either confirming or overturning them. See Kaldor (1957).
CFA Institute Research Foundation 51
Revisiting the Equity Risk Premium

PRESENTATION BY RAJNISH MEHRA:

REFLECTIONS ON THE EQUITY PREMIUM
Rajnish Mehra: I’m going to discuss something totally premium is at least partially a risk premium, size and value
orthogonal to what has been previously presented here are not. My argument is based on the premise that a gen-
today. But it is relevant, especially to what Elroy Dimson uine risk premium is invariant to whether or not I know that
said about American exceptionalism. the premium exists.

Is the Equity Premium The Size Premium

a Risk Premium? The size premium was documented by Rolf Banz and Marc
Reinganum at about the same time as we wrote our equity
Empirically, we observe several factor premia—for example,
premium puzzle paper.41 In the fall of 1979, Myron Scholes
the Fama–French three-factor model identifies three: the
had invited me to visit the Center for Research in Security
equity premium, the size premium, and the value premium.
Prices (CRSP), and Ed Prescott was visiting the Economics
I want to address the question: Are these factor premia
Department at the University of Chicago. We worked in
a premium for risk? If they are, we can ask a second
Fischer Black’s old office on the top floor of the business
question—how much of the factor premium is a risk pre-
school, now known as Booth. Many of you may remember
mium? For example, Ed Prescott and I documented that
the suite of offices on that floor, including Jim Lorie, Jon
only about 1 percentage point of the equity premium is a
Ingersoll, Eugene Fama, and Myron Scholes.
premium for bearing systematic risk—hence, the “Equity
Premium Puzzle.”40 Let’s look at Exhibit 44.

Let me just share some thoughts on this, and then we can You can see the dramatic size premium in the exhibit; I
discuss it. haven’t put up the t-statistics or any other details, but the
key finding in Rolf Banz’s 1980 work was that the size pre-
Textbook finance characterizes the equity, size, and value
mium was a huge 8.3% per year (of small- over large-cap
premia as risk premia. I will argue that, while the equity
stocks).42 You’d do anything for that!

Exhibit 44. The Size Premium before and after It Was Documented in 1980
Annual Mean Value Weighted Returns (%)

Period Small Firms Large Firms Size Premium

1927–1979 18.81 10.51 8.30

1927–2020 16.64 11.79 4.85

1980–1989 15.01 17.79 −2.78

1990–1999 15.96 19.31 −3.35

2000–2009 10.29 1.25 9.04

2010–2019 13.35 14.36 −1.01

1980–2020 13.84 13.43 0.40

Note: “Value-weighted” means capitalization-weighted. Returns are arithmetic means.

See Mehra and Prescott (1985).

40

See Banz (1981); Reinganum (1981).

41

See Banz (1981). Banz’s definition of small cap was small indeed—the bottom quintile, by count, of New York Stock Exchange stocks sorted by
42

capitalization each year. Later research revealed a smaller size premium (over the same historical period) for stocks that were in the intermediate
quintiles.
52 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

But you couldn’t do anything about it. Buying small-cap the value premium was discovered in 1990. Just looking at
stocks was not, up until 1980, an actionable decision rule. the realized returns, it is apparent that the value premium
Once you got to know about it in 1980, it became action- “disappeared” once it became a part of our information set.
able. After that, the premium just isn’t there, and the pre- Changes in expected stock returns are hard to measure,
mium for the entire 1927–2020 sample (including the period however, and we need another decade of data to make a
where it was so large) is statistically indistinguishable from definitive statement about the value premium.
zero. That fact leads me to conclude that the small-cap pre-
mium is not a risk premium. It was a premium. But once it
was in everyone’s information set and became tradeable, it
Persistence of the Equity Premium
disappeared. The risk is still there, but the premium is not. On the other hand, if you look at the equity premium as
shown in Exhibit 48, it’s as stable as it ever was. Knowledge
Exhibit 45 is the illustration that you would normally see
about the existence of the premium did not eliminate it. The
in books documenting the differential returns of small and
persistence of the equity premium is considerably different
large stocks. (The use of an arithmetic rather than logarith-
than what you see with the value or the size premium. This
mic scale exaggerates the difference, which is what many
is consistent with it being a risk premium.
of these book authors want to do.)

But I think Exhibit 46 is what you really want to show. This Mean Reversion in Equity Returns
starts in 1980, and there is no big difference between the
returns of large versus small stocks. The other point I want to talk about is whether the equity pre-
mium is mean-reverting and perhaps predictable. The profes-
Something similar happened with the value premium (see sion’s view on this topic has shifted over time. The prevailing
Exhibit 47). For the sake of this analysis, I’m assuming that paradigm in the 1960s and 1970s (the halcyon days of the

Exhibit 45. Cumulative Total Returns on Small- and Large-Cap Stocks,

1927–2020

Source: Based on data from Kenneth French’s website (https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/).

CFA Institute Research Foundation 53

Revisiting the Equity Risk Premium

Exhibit 46. Cumulative Total Returns on Small- and Large-Cap Stocks,

1980–2020

Source: Based on data from Kenneth French’s website (https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/).

Exhibit 47. The Value Premium before and after It Was Documented in 1990
Annual Mean Value Weighted Returns (%)

Time Period Growth Firms Value Firms Value Premium

1927–1989 11.27 17.59 6.32

1927–2020 11.88 15.86 3.98

1990–1999 20.34 17.57 −2.77

2000–2009 1.01 8.26 7.25

2010–2019 15.67 12.65 −3.04

1990–2020 13.13 12.34 −1.78

efficient market hypothesis!) is best characterized by a In the 1990s, there was a paradigm shift in whether stock
quote from Fama: “This paper has presented strong and volu- returns are predictable or not. In their 1988 paper, Fama and
minous evidence in favor of the random walk hypothesis.”43 French took a very different position: “There is much evidence

See Fama (1965).

43

54 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 48. The Equity Premium before and after 1979

% Real Return on Market % Real Return on Riskless % Real Premium
Time Period Index Mean Security Mean Mean

1889–2020 8.2 1.3 6.9

1889–1978 6.98 0.8 6.18

1980–2020 9.6 1.5 8.1

Exhibit 49. Market Value to GDP Ratio and Subsequent Average Seven-Year
Equity Return, 1947–2020

Note: Data are for the United States.

that stock returns are predictable”44—in other words, they are relationships to indicators of fundamental
not a random walk. And then, in John Cochrane’s presiden- value, … Thus … when stock prices are very high
tial address to the American Finance Association, he said, relative to these indicators … [they] will … fall in
“All price-dividend ratio volatility corresponds to variation in the future to bring the ratios back to more normal
expected returns. None corresponds to variation in expected historical levels.46
dividend growth, and none to ‘rational bubbles.’”45
Let me show you some empirical evidence regarding equity
The implicit underlying belief is that the predicting variables return predictability. Exhibit 49 shows the ratio of US
(dividend-price ratios, earnings-price ratios) follow a station- equity market capitalization to GDP along with subsequent
ary process that reverts to some unspecified normal value. seven-year returns.

Campbell and Shiller succinctly summarize this view: This relationship held up well until the Global Financial
Crisis. Looking at market value to GDP, it was a stationary
It seems reasonable to believe that prices are series up to 2007. After that, however, it has no longer
not likely ever to drift too far from their normal

See Fama and French (1988).

44

See Cochrane (2011).

45

See Campbell and Shiller (1998, p. 11).

46

CFA Institute Research Foundation 55

Revisiting the Equity Risk Premium

been so, as shown by applying the standard test for non-

stationarity. A lot of our economic intuition was based on
Discussion of Rajnish
this earlier relationship. I entirely agree with the earlier Mehra’s Presentation
presenters that this is not a market timing strategy, but it
does give you an idea of what the average equity return is Jeremy Siegel: The ratio of equity market capitalization
going to be. to GDP is often called the Warren Buffett indicator. It’s his
favorite indicator. I have often criticized it. Until the last
In 2007, there was a structural shift in the economy. Real 20 years or so, about 7% of the profits of the S&P 500 were
interest rates become negative; currently the entire term from foreign sales. Now 40% to 45% of profits are from for-
structure of real interest rates up to 30 years is nega- eign sales. So, to compare US market cap to just US GDP is
tive, implying a negative marginal product of capital. Any not an apples-to-apples comparison.
assessment of the equity premium after 2007 must take
into account these negative real interest rates. Rajnish Mehra: I’m looking only at domestic operations
here. I’m not looking at foreign equity.
One plausible explanation is that the equity premium went
up after the Global Financial Crisis. If you take the historical Jeremy Siegel: I’m not talking about foreign companies. I’m
(1929–2020) relationship between the market value/GDP talking about profits of US firms and the market capitaliza-
ratio and subsequent equity return and extrapolate from tion of US firms. Isn’t that what you’ve used in Exhibit 49?
it as shown in Exhibit 50, then the expected compound
annual return on equities over the next five years is −5%. Rajnish Mehra: I have used domestic corporations, yes.
I wouldn’t have too much faith in that forecast, however,
Jeremy Siegel: Yes, domestically housed corporations,
because of the structural change.
but they’re getting their profits from abroad, when they
To sum up my views on the mean reversion story, I think didn’t before.
that it was valid until about 2007, but something changed
Laurence Siegel: Rajnish, when you use the term “domestic
in the economy around that time and we’re out of that
operations” it suggests that you’ve broken out the foreign
paradigm.

Exhibit 50. Market Value/GDP Ratio and Subsequent (Next Five Years)
Average Equity Return, 1929–2020

56 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

operations of US-based companies. I don’t think you’ve done Rajnish Mehra: I think that’s the most likely scenario. That
that, have you? the risk premium has gone up is consistent with the fact that
real expected returns have become smaller and maybe gone
Rajnish Mehra: No, I have not. What I’m saying is that the negative.
market capitalization of listed domestic corporations is not
the full market value of all businesses in the United States. But there are other stories that are floating around. There is
an excellent paper by Farhi and Gourio called “Accounting for
Laurence Siegel: I am aware of that argument and agree with Macro-Finance Trends: Market Power, Intangibles, and Risk
it—that the market cap of a stock market index misses a lot of Premia.”47 They present evidence on the trends affecting
privately held companies, sole proprietorships, and so forth. some key macroeconomic and finance variables, focusing on
I think Jeremy is saying something different, which is that six groups of indicators. I think the most plausible scenario
the S&P itself, holding that constant, has become more of a is an increase in the risk premium, but one has to solve this
global index over time as its constituent companies became puzzle jointly with other observations. You can’t just pick one
multinationals. part of it—you must address the fact that the risk-free rate
has declined so much and yet the return on equity has not
Rob Arnott: Rajnish, in looking at the past returns and past
declined. Why is that so? These are hard issues, and we don’t
linkages with the linkage breaking down since 2007, I think it
have enough data after 2009 to resolve them.
is strictly a function of what Cliff was alluding to earlier, which
is revaluation. The valuation ratio has soared. A revaluation
alpha should never be part of our forward-looking expected
risk premium.

See Farhi and Gourio (2018).

47

CFA Institute Research Foundation 57

Revisiting the Equity Risk Premium

PRESENTATION BY JEREMY SIEGEL: BACK TO THE

FUNDAMENTALS—A CLOSE LOOK AT THE EQUITY
RISK PREMIUM
Jeremy Siegel: Many of you are familiar with Exhibit 51, published an article about the problems with GAAP earn-
which covers 1802–2020, a period of more than 200 ings (those based on generally accepted accounting prin-
years. The compound annual real return on equities has ciples) and believe a tight definition of operating earnings,
been 6.8%; my forward prediction is between 4.5% and 5% with expensing for options, are much better indicators of
per year from current valuations. Bonds have earned a real market value.48 This is especially true since GAAP earnings
return of 3.6%, and bills have had a real return of 2.6%; real definitions have changed dramatically over the years.
returns on fixed-income assets will almost certainly be
negative in the future. I have also written about the “aggregation bias,” which
causes the CAPE ratio based on the sum of the earnings
Then there’s gold and the dollar. The dollar is measured of S&P 500 companies to provide a very distorted view of
against consumer goods in the United States (in other the valuation of the market in recessions because it aggre-
words, the inverse of the Consumer Price Index). Notice gates losses on individual stocks with profits on other
that the decline of the dollar has had no effect on the real stocks on a one-for-one basis. Bob Shiller agrees that it’s a
return on stocks as would be expected since stocks are bias and, along with the changes in GAAP earnings defini-
real assets. There are short-term effects of inflation on tions, has biased the CAPE ratio up dramatically since the
the stock market but no appreciable long-term effects, Financial Crisis and yielded overly bearish projections.
as theory would suggest.
Another source of bias in the CAPE ratio is the shift to buy-
backs from dividends, giving the earnings yield an upward
The CAPE Ratio through History tilt, so that the 10-year past average—which the CAPE ratio
employs to determine the price-to-earnings (P/E) ratio—is
The CAPE ratio has gone through many transformations
biased downward compared to historical data. This can be
since Bob Shiller’s original article in the 1990s. I have

Exhibit 51. Real Total Returns on Major US Asset Classes, 1802–2020

Source: Siegel (2014), with updates to 2020.

See Siegel (2016).

48

58 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

corrected by adding back share repurchases to dividends, The S&P 500 measure of operating earnings expenses
which is called the total return CAPE and which Shiller options and all sorts of other items that could be capital-
has done. ized: It’s a very conservative look at earnings. The current
estimate of next 12 months’ S&P 500 operating earnings is
In his products with Jeffrey Gundlach’s DoubleLine Capital, $211, so that index is now selling at a 21 P/E. The expected
Shiller uses another definition of CAPE, the “relative return return is then 4.6%, or 1 divided by 21.
CAPE,” which measures the P/E relative to the last 20 years.
This approach sharply reduces the CAPE ratio from his orig- Now, you might ask, is this cyclically adjusted? I don’t
inal and all later formulations. know where we are in the cycle. We had a short recession
last year after a long expansion; I’m going to be agnostic
Most recently, Shiller has pivoted to the “excess return and say we are midcycle so no adjustment needs to be
CAPE,” which measures valuation relative to interest rates. made. The real expected equity return of 4.6% is more than
For years, Bob has told me that interest rates do not affect 5 percentage points above the real yield on TIPS, which is
P/Es in the historical data, but perhaps he has changed now about −1%. Currently the equity risk premium is 5.6%—
his mind. All these transformations have reduced the mag- that is, [4.6%–(−1%)]—almost double the 3.2% historical
nitude of the CAPE ratio and made the market appear less average (6.8% stocks minus 3.6% bonds). This is something
“overvalued.” we should think about.

Another important question: Why should the normal P/E for

Epicycles the market today be the 150-year historical average of 15?
Are we like the ancients putting epicycles on the geocen- The cost of a diversified portfolio has declined radically
tric model of the solar system, trying to force the CAPE ratio over the last 50 years. This is something that I pointed out
to do something it cannot do? I’ve talked to smart investors 20 years ago. I said that the ability to get a fully diversified,
who have been following CAPE and reducing their equities cap-weighted efficient portfolio at virtually zero cost did
over the last decade. They are not happy with their results. not exist through the 19th and first three-quarters of the
20th century.
Clearly the CAPE ratio has been an extremely poor predictor
over the past decade. But even over the last 40 years, the How high would transaction costs have been from 1870
CAPE ratio has predicted that the market has been “over- through 1970 to get a fully cap-weighted indexed portfolio,
valued” about 95% of the time. One flagrant example is considering high brokerage commissions and wide bid
that the market became overvalued according to the CAPE and ask prices, to keep a portfolio balanced? Perhaps 1%
ratio in May 2009, when the S&P 500 was around 900, less to 1.5% per year? Today the cost to the investor of holding
than one-fifth of the current level. Clearly the overvaluation an index fund is 1 or 2 basis points. Those low costs make
signaled by CAPE in the years immediately following the it possible to obtain a risk return trade-off that is far supe-
Financial Crisis has been one of the worst predictions in rior to what investors could receive in the 19th or early
forecast history. 20th century. That means that the average P/E should rise
accordingly.
Over the last 150 years, the single-year P/E has averaged
about 15. That implies an “earnings yield” or expected So, should the P/E be 20? 22? That would lead to a 5%
real return of one-fifteenth or 6.7%. That has been almost to 5.5% expected forward-looking real return. The histor-
exactly the long-term real return on equity. It is simple and ical real return on stocks is 6.8%, but once you subtract
direct. Obviously, we need to be mindful of dips and booms transaction costs, you get this lower level. So, a 20–22 P/E
in profits at business cycle peaks and troughs, but CAPE today with costless indexing is about the 15 P/E of the
based on GAAP earnings is in my opinion fraught with too 1870–1970 period.
many problems to be a useful predictor.
Behind the Earnings Estimates
Valuation and Current Return Returning to the S&P earnings estimates reported in Exhibit
I’m now going to provide a current forecast and thoughts 53, I think these are underestimates. The “beats” this year,
on what earnings estimate to use. earnings beating averages of analyst expectations, have
been prodigious. In Exhibit 54, the percentage of firms
Exhibit 52 shows the P/E based on historical 12 months reporting upside surprises is at a high.
operating earnings of the S&P 500 since 1954. The
median value has been 17.3, but the trend is upward, as
will be explained shortly. Yesterday the S&P 500 was at
Back to Buybacks
4500. Stocks are selling for 21 times next year’s earnings. I want to push back strongly on Rob Arnott’s argument
Exhibit 53 shows various measures of earnings on the that buybacks are not substitutes for dividends. They are.
S&P 500 as of a recent date.
CFA Institute Research Foundation 59
Revisiting the Equity Risk Premium

Exhibit 52. P/E of the S&P 500 Based on 12-Month Trailing Operating Earnings,
March 1954–June 2022

Exhibit 53. Current S&P Earnings Estimate

Date 7 October 2021
S&P 500 closing level 4399.76
Dividend yield (12 months) 1.35%

Earnings Estimates (bottom up)

One-Quarter One-Quarter Twelve-Month Twelve-Month

Operating Reported Operating Reported
Earnings per Earnings per Operating Reported Earnings per Earnings per
Quarter Ending Share ($) Share ($) Earnings P/E Earnings P/E Share ($) Share ($)

31 December 2022 57.00 52.81 20.22 21.30 217.59 206.57

30 September 2022 55.32 51.36 20.83 21.97 211.19 200.23

30 June 2022 54.03 49.30 21.56 22.77 204.07 193.21

31 March 2022 51.24 53.11 21.77 22.88 202.09 192.30

Source: Based on data from the S&P Global website (www.spglobal.com).

60 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 54. Current S&P Earnings Estimates

Note: Percentage of S&P 500 companies that reported earnings above or below the consensus estimate at the time of the earnings report.
Source: Based on I/B/E/S data from Refinitiv; Yardeni Research Inc.

Exhibit 55. Lower Dividends, Higher EPS Growth

Reported EPS Real Dividend Dividend Real Capital Real Stock Payout
Growth Growth Yield Gains Returns Ratio

1871–2021 2.04% 1.57% 4.29% 2.57% 7.1% 57.2%

1871–1945 0.67% 0.74% 5.31% 1.32% 6.8% 66.8%

1946–2021 3.41% 2.38% 3.28% 3.82% 7.4% 49.0%

1929–2021 2.34% 1.77% 3.89% 3.46% 7.6% 53.7%

As Exhibit 55 shows, over 1871–1945, a very long period, Rob, you talked about new companies causing dilution.
there was very little real EPS growth or real per-share I published a paper that tracked the stocks that were in the
dividend growth. Since 1946, the dividend yield has gone original S&P 500 when it was first constituted in 1957.49
down about 2 percentage points and real earnings growth
has gone up about 2.7 percentage points. They come close Rob Arnott: I remember that paper. I accepted that paper
to balancing each other out. Expected real stock returns (for the Financial Analysts Journal).
stayed about the same from before 1946 to after; the divi-
Jeremy Siegel: The portfolio of original stocks beat the
dend payout ratio went down and EPS growth went up.
actual, continuously reconstituted S&P 500. So, do you
Tax considerations aside, pure theory tells you there is an need these new stocks that were put into the S&P 500 to
exact one-for-one trade-off between buybacks and divi- get the overall market return? Not between 1957 and the
dends. I believe the long historical data confirm this. date of my paper (2006).

See Siegel and Schwartz (2006).

49

CFA Institute Research Foundation 61

62
Exhibit 56. S&P Profit Margins: Mean Reversion?
Revisiting the Equity Risk Premium

CFA Institute Research Foundation

Source: Based on data from David Bianco, DWS Americas.
 Revisiting the Equity Risk Premium

In recent years, the new stocks added to the index have not being included in book value, despite being tremen-
done better. So, I’m not going to say that if I repeat the dously valuable, and thereby making some value stocks
experiment, I’m going to get that same result today. But look like growth stocks when you use price-to-classical-
it’s not a given, empirically or theoretically, that you must book as the choice variable. That could be one reason why
have the “new” stocks to get a return that approximates the apparent value premium has declined. I don’t agree
the index. with everything in that paper, and I’m sure we’re going to
talk about it in discussion, but you raised a valid issue.

Profit Margins
The Changing Correlation between
I think the issue of profit margins, which Raj mentioned,
is important. Exhibit 56 shows historical profit margins on Stocks and Bonds
the S&P 500. Currently the large profit margins are almost
We have not talked enough about the collapse of real rates
all in technology stocks, which have a profit margin of 25%.
around the world, which is unprecedented and the biggest
Is this phenomenon likely to mean revert? If you take out
surprise for macroeconomists over the last two decades.
technology and communication services, which are really
Exhibit 57 shows this for US TIPS.
tech, you have 50% of the market where profit margins are
much lower. The margin increase has been mostly in those Exhibit 58 illustrates that the massive decline in real rates
two sectors. is a worldwide phenomenon.

We can debate whether this concentration of profits is per- Exhibit 59 shows that the correlation between the S&P 500
manent or not. Rob, your last paper, which touched on that and 10-year Treasuries has changed from positive to neg-
topic, was really good.50 You talked about intellectual capital ative. John Campbell and Luis Viceira, among others, have

Exhibit 57. Ten-Year US TIPS Yield, 14 February 1997 to 9 September 2022

Source: Based on data from FRED.

See Arnott, Harvey, Kalesnik, and Linnainmaa (2021).

50

CFA Institute Research Foundation 63

Revisiting the Equity Risk Premium

Exhibit 58. Estimates of the “Natural” Real Rate of Interest,

Five Major Countries/Regions, 1999–2021

Exhibit 59. Trailing 12-Month Correlation of S&P 500 and 10-Year

US Treasury Bonds, 1969–2021

Source: Clarida (2019).

64 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Exhibit 60. Annual Rate of Money (M2) Growth vs. Inflation, 1868–2020:
150-Year Record Broken with 2020 Increase

talked about this.51 Fed Vice-Chairman Richard Clarida gave the money supply grow as fast as it did last year (2020).
an excellent address in Zurich about this issue, in which he I said in July 2020 we were going to have rapid inflation
claims that more than 3 percentage points of the decline in next year, and continued excessive money supply growth
real term premium is caused by the change of the correla- augurs badly for inflation in the future.
tion between the 10-year Treasury and risk assets.52

We can speculate about whether that negative correlation Discussion of Jeremy

is going to persist, but the impact of the real rate collapse
that has already taken place is enormous. This forum is
Siegel’s Presentation
about the expected equity risk premium—the difference Rajnish Mehra: Jeremy, would you agree one of the more
between expected returns on stocks and fixed-income plausible explanations for lower expected returns in the
assets (bills and bonds). So, the forecast for bonds is tre- future is that the realized risk premium has been much
mendously important to what we’re discussing. higher than expected?

Upcoming Inflation Jeremy Siegel: I expect real equity returns to be lower by

about 2 percentage points per year than their historical
I want to end my presentation with a comment about how average. I’m not sure there’s any mean reversion.
much inflation we’re going to have. Exhibit 60 is an update
of Milton Friedman’s long-term chart of the growth of the
money supply. In more than 150 years, we have never seen

See Campbell, Sunderam, and Viceira (2017).

51

See Clarida (2019).

52

CFA Institute Research Foundation 65

Revisiting the Equity Risk Premium

Rajnish Mehra: Isn’t it possible that you have these low and −1% on TIPS because the yield on the 10-year TIPS
expected returns on real assets because people are really is known. On nominal bonds, because there’s going to be
scared or our risk aversion is very high? much more inflation, the real return will be −2% or −3%
or −4%.
Jeremy Siegel: I don’t see it that way. If people are scared,
then prices should be low, causing expected returns to be Roger Ibbotson: If you got a big jump in the risk premium,
high. But expected returns are low. you’re going to have an immediate drop in the market.

In addition, the correlation between stocks and bonds has Jeremy Siegel: If you have a jump in expected real returns,
turned negative, after being positive for decades. There are you’re absolutely right.
many reasons for this change. If you use any beta model
to analyze this situation, when you change the correlation Roger Ibbotson: So, you can’t justify this big rise in the
of two major assets from positive to negative, you change stock market from a rise in the risk premium.
the expected return dramatically. Any hedge asset has a
Jeremy Siegel: No—if expected real returns go up from the
negative real expected return. And now, US Treasuries are
current level, the stock market goes down.
viewed as the hedge asset of the world, and they’re bought
for that attribute. That wasn’t happening in the 1960s, Laurence Siegel: All other things being equal, yes of course.
1970s, and 1980s.
I have a great deal of concern about extremely low or neg-
The real return on fixed income has dropped far more dra- ative interest rates being contractionary, although they’re
matically, in my opinion, than the real return on equities intended by central banks to be expansionary.
going forward. So, I predict a 4.5% real return on stocks;

66 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PRESENTATION BY WILLIAM GOETZMANN:

574 YEARS OF EQUITY RETURNS FROM
THE BAZACLE WATER MILL IN FRANCE
Laurence Siegel: As Monty Python used to say, “And now I was in graduate school, and Roger and I were talking
for something completely different.” Will is going to tell us about early stock companies. I thought the Dutch East
about the long run. We all think we’ve been looking at the India Company was the earliest one, in 1602. Roger said he
long run, but Will really is. had heard there were some earlier companies in southern
France—water mill companies. I don’t know where he heard
Will Goetzmann: It’s amazing to see so many people who about them, but it took me a decade—I had to wait until
have spent decades of their careers collectively and sep- the Internet became a practical tool for me to go and trace
arately studying the equity premium and making a huge down this source.
difference in the way people invest their money. So, it’s
great to reconnect with everybody and also to see so many Over the last decade or so, my colleagues David Le Bris
extreme pessimists and optimists in the same group. and Sébastien Pouget, both in Toulouse, have been mining
the archives of that city for incredible information about
the very earliest corporations. They were full corporations
The Oldest Joint-Stock Company with publicly traded shares, limited liability, juridical entity,
For something a little bit different, as Larry said, I pres- annual shareholder meetings—anything you think a corpo-
ent 574 years of equity returns. I am continuing to work ration should be, these companies had it before 1400.
actively on this ancient company that started in 1372 and
The building in Exhibit 61—actually, the foundation of this
another one that started in 1374. This quest began when
19th century building—was the location for one of those

Exhibit 61. Honor del Bazacle

• Earliest documented corporation 1372–1946

• Same business activity for ages (grain milling up to 1888, electricity afterwards)

• Dividends and prices from 1371 to 1947 (very rich data from 1520 onwards)
Source: Photo credit: Getty images/yvon52.
CFA Institute Research Foundation 67
Revisiting the Equity Risk Premium

companies, called the Bazacle Company. It used the same don’t have registers for the period before about 1500. After
technology, hydro power, for nearly 600 years—first to mill that time, you can see that annual dividends were very vola-
grain and eventually to generate electricity. David was able tile. There are also some negative dividends; those are calls
to find a rich vein of dividend and price information for this on shareholder capital to make up the difference between
company from about 1530 onwards. It was a pretty big income and expenses for a given year. The big capital calls
company, and Toulouse was famous through the centuries are in times when there were huge floods, the mills were
as the big market for grain in southern France. The mills that knocked out, and the company had to raise more capital
emerged were a significant part of the business of the city. to build them back. At such times, you had a choice as a
shareholder: Either come up with the capital or hand the
We’ve been able to collect transfer prices for shares in shares back to the company—that’s the limited liability part.
these companies from shareholder registers. We also
have dividend information for long stretches of time. Eventually these shares were listed on the Paris stock
Interestingly, dividends were paid in grain until the late exchange in the 19th century. They traded there as public
1700s, which investors could easily convert to cash in companies until 1946.
the Toulouse market. We used prices from this market to
express dividends and share prices in grams of gold or Exhibit 63 shows Bazacle Company prices and dividends
silver. There were negative dividends, which I’ll discuss; in livres Tournois (Tours pounds, a currency in use in France
and the companies had de facto limited liability, which in the Middle Ages). Prices are in red and dividends are in
makes it fun to argue about the origins of corporate gover- blue. We also show some moving averages. It is extremely
nance and related matters. pleasant to see that the prices and dividends do move
together, suggesting that maybe there’s some rationality
to the whole process of asset pricing. Prices may actually
Dividends Are Everything represent expectations about future dividends.

Exhibit 62 shows the dividend series. The paucity of the We’ve been talking about the equity premium. Over the
dividend data in the early years reflects the fact that we whole time period that we studied the Bazacle Company,

Exhibit 62. Dividends per Share of the Bazacle Company, Year by Year,
1372–1946

Note: Missing data prior to 1526 are shown as zeroes. This does not mean dividends were zero.
Source: Le Bris, Goetzmann, and Pouget (2019).
68 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Exhibit 63. Bazacle Share Prices (red) and Dividends (blue), 1532–1920

Source: Will Goetzmann. For details, see Le Bris et al. (2019).

capital appreciation was de minimis. We have the whole

history of the company—we know when it started, and we
Exhibit 64. Summary Statistics of
know when it was purchased at a low price by the French Real Dividend Yields on Bazacle
government that nationalized the generation of electricity
in the 1940s.
Company Stock, 1372–1946
Most of the returns, then, came from dividends. Inflation Dividend Yield
over that long time period is not easy to calculate, but we
Real Mean Std. Dev.
do have accurate prices for various goods stated in terms
of silver. Based on this information, our best estimate of 1372–1946 5.16% 7.55%
the dividend yield in real terms is just over 5%, as shown in
Note: For details, see Le Bris et al. (2019).
Exhibit 64. That is not too far out of line with all these dis-
cussions we’ve been having—“real 5%” might be a little more
optimistic than some of you. But that’s what you got. This given that we didn’t have to deal with the problem of stock
estimate could be said to contain survival bias, because prices wandering off to infinity. We built a simple model,
we picked the oldest company, but the company did die in first describing the dividend process using an autore-
1946, so there is a sense in which it didn’t really survive. gressive moving average model. The autoregressive part
is unsurprising: Dividends tended to be positively related
A Medieval Test of a Modern Asset to each other from year to year. The moving average part,
however, suggests that there was—I hate to use these
Pricing Model words, but I’m going to use them—some amount of mean
reversion. These dynamics suggest that dividends were
Because there was not much capital appreciation, we were
partially forecastable.
able to perform some estimations of asset pricing models,
CFA Institute Research Foundation 69
Revisiting the Equity Risk Premium

rate is difficult when there is no riskless rate starting in

1372. So, we used Parisian municipal bonds and other
Exhibit 65. Asset Pricing Model proxies. Nevertheless, the estimated premium may also
for Bazacle Company Stock reflect risk premium variations in the bond rate.

• Does price reflect expected future dividends? Term Structure of the Equity
• Dividends autocorrelated, moving average
Premium
• Build a model and estimate
Scholars have long been interested in the term structure
• Results:
of the equity premium.53 The equity risk premium must
1. Dividends ARMA(1,1) = [0.80, −0.35] price uncertainty that can happen in both the near term
2. Expected dividends explain prices: and the distant future. Near-term fluctuations are mostly
15% to 45% of variation. a function of current, stationary risks. Long-term risks, like
3. Cannot reject pricing model uncertainty around climate change, may command their
own premium.

Our model also allows us to estimate a term structure of

We next use an asset pricing model to see whether the
the Bazacle Company equity risk premium, which we show
predictable future changes in dividends were reflected
in Exhibit 66. The benefit of using the Bazacle Company is
in the price of Bazacle shares. The answer is yes. There
that the six centuries of data mean that both short-term
were shocks, like wars and famines, that also affected firm
and long-term risks were realized in our sample.
output. This model does a pretty good job of explaining
changes in prices (see Exhibit 65). This makes it a bit of We find that much of the premium for holding shares was
a novelty, a rational asset pricing model that we actually associated with risk in the near term. Perhaps because of
can’t reject. the long-term mean-reverting tendency of grain prices, the
premium at longer horizons is much less than that of the
The model also allowed us to estimate an equity risk pre-
near term.
mium over 574 years. Calculating a premium over a riskless

Exhibit 66. Term Structure of the Risk Premium

Source: Le Bris et al. (2019).

See, for example, van Binsbergen and Koijen (2017).

53

70 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

This downward-sloping equity risk premium term struc- it refused to die. I would moderate my return extrapolation
ture of the Bazacle Company is interesting because by noting that it is the only company, that we know of, that
current theories about the equity risk premium imply an has done that. It is a real survivor.
upward-sloping term structure.54 Our findings are more
consistent with term-structure estimates using dividend Will Goetzmann: That is true.
strips.55 A natural question to ask of our term structure anal-
Antti Ilmanen: Will, you said there wasn’t much, if any, div-
ysis is why the very long-term risk pictured in Exhibit 66
idend growth over this long history. We know that output
is near zero. One answer may be due to the survival of the
was growing at close to a zero pace until about 1800 and
company itself. Milling was an essential technology for cen-
then sped up to 1% plus, per year. Do you see anything like
turies. Even when the Bazacle Company converted to gen-
that in the dividend growth data?
erating hydroelectric power, it continued to generate profits.
In fact, the company that acquired the firm in 1946 is now Will Goetzmann: There were periods of technical innova-
partially privatized, and you can now effectively invest in it. tion. For example, in the 1300s, I think output was much
lower. Sometime before 1500, there was a huge jump in the
So, get that 5% stuck in your head. It’s not too far off from
company’s technology. It may be related to them rebuild-
the equity risk premium that we have measured since 1926
ing the dam. In Exhibit 61, you can see there’s a dam that
using US stock market data. My coauthors and I collected
had to be built through the collective efforts of investors
centuries of archival data from a unique firm that allowed
contributing to it. That was one big jump. Then, of course,
us to estimate an asset pricing model that reflected risks
there was the transition to electricity generation, which is
both near term and long term. The model suggests that
another technological change.
equity investors from the Middle Ages to the modern era
were not entirely foolish. The prices they set reflected the So, technological innovation happened in fits and starts.
value of expected future dividends. The Bazacle site itself is close to the Toulouse School of
Economics. If you’re ever there, it is worth a tour. The foun-
Laurence Siegel: …and the risk of those cash flows.
dations of the building date back to the 14th century.

Discussion of Will Jeremy Siegel: The good thing is that, with such a long time
series, you have a terminal price. The price was probably
Goetzmann’s Presentation very depressed because the company was nationalized.
The government probably paid very little for it—just a year
Jeremy Siegel: Gold has appreciated in real terms at almost after World War II ended—but the terminal price doesn’t
1% per year. If these dividends were paid in gold or precious matter that much to the annualized real return when the
metals, you might add almost a percentage point to the period is that long.
real yield, which would move it closer to 6%.
Will Goetzmann: There was one book written about this
Will Goetzmann: That may be true. We did have a gold series company and published in, I think, 1954 by a French legal
as well, but I think we converted everything into silver. historian. It got maybe six citations in its whole history up
until about 10 years ago. I commissioned and worked with
Jeremy Siegel: I’m just saying that 6% is really right in the
a translator, and we translated it into English and published
ballpark of what Elroy was saying the risk premium was
it with Yale University Press.56 It’s an extraordinary story,
around the world.
and we keep finding new things to write about it.
Laurence Siegel: I think there is some survival bias
Jeremy Siegel: It’s interesting to compare it with World War
because, first, the company is even older than you said.
II Germany and Japan. Japan had 90% of its capital bombed
When Roger Ibbotson and I first looked at this company
out and disabled. Yet Elroy and others have gotten return
(without traveling to France), we noted that it had been
data for Japan starting in 1900, and they’re only a little bit
functioning as a water mill for a couple of hundred years
lower than the world return.57 Germany and Austria also
before 1372. We didn’t have stock prices, but it was a going
have returns spanning the war period, through a total
concern, a business. And then in 1946, it was acquired by a
destruction and a rebuilding, and equities in Germany and
government—it’s not quite fair to say that it died; it just had
Austria maintained a premium over every other asset in
a new owner. So, this company really survived every possi-
those countries.
ble catastrophe that the world could have thrown at it, and

See Bansal and Yaron (2004).

54

See van Binsbergen and Koijen (2017).

55

See Sicard (2015).

56

The Japanese returns had much more risk, because of the near collapse (but not total collapse!) of the Japanese market in World War II.
57

CFA Institute Research Foundation 71

Revisiting the Equity Risk Premium

Separation of Investors There’s no inflation index that actually works; they gave up
on collecting data of that period.
from Their Investments in Wartime
Laurence Siegel: In our book draft, we use exchange rates,
Laurence Siegel: The problem was that you couldn’t hold which were published continuously. The exchange rate of
onto your claims, to your shares. If you were taking the last the reichsmark versus the dollar is a proxy for inflation.
boat out of Hamburg to go to Britain or the United States,
you might have carried the share certificates with you Elroy Dimson: That is the only solution.
but were unlikely to get any money for them. I think a lot of
Martin Leibowitz: Will, if my memory is correct, about 25
investors lost everything in Germany and Austria, but the
years ago, you and Steve Ross and Steve Brown published
indexes didn’t. The indexes came back stronger. So, there’s
the paper called “Survival.”59 Right?
a wedge between any individual’s experience and the mar-
ket’s collective experience. Will Goetzmann: Yes.
Will Goetzmann: All of us working in this area have had to Martin Leibowitz: If you were writing that paper now,
figure out what to do about those difficult periods when given what we’ve talked about, what would you change,
the markets broke down and people were separated from if anything?
their capital. It’s a heroic effort to put these pieces together.
We do the best we can, but we know that it’s impossible to Will Goetzmann: You know I love history, so I couldn’t put
do it perfectly. this company down. But could we draw conclusions about
5% going forward for the whole world? We simply can’t.
Thomas Philips: Have you seen the Dimson, Marsh, and
Staunton data showing that Austria did much worse than That is the insight: We are prisoners of the history that
Germany during World War II? survived for us to study it. It’s really important for us to
recognize that. When we talk about premia for things that
Laurence Siegel: Yes. What happened? were discovered in the past, “P-hacking” is now the term
in academia for this, so the conditioning process is really
Elroy Dimson: We use all of the data from 1900s to the
crucial. Steve Ross wrote a paper that Stephen Brown and
current time. Apart from Russia and China, there’s only one
I always loved—in fact Stephen and I wrote a paper about
country that we are unable to bridge, and that is Germany
Ross’s paper. 60 The Ross working paper, called “Regression
in its first hyperinflation. We can bridge it in the 1948
to the Max,”61 was about the belief that during a bubble,
hyperinflation but not in the 1922–1923 hyperinflation.
we should see more autocorrelation in prices and that
We had a then-doctoral student, now a Stockholm autocorrelation identifies it as a bubble. In fact, that pattern
professor, collecting the data for Austria over a long period, may be misleading.
so we got data for Austria that way. We missed two years
What Steve said is that whenever you identify an internal
for Germany. There are ways around that: One is to define
maximum in a price series retrospectively, you’re going to
history with hindsight, as Global Financial Data does—if
see something that looks like autocorrelation preceding the
they know that there is trouble coming up, they then
maximum. In other words, it’s easy to call a bubble ex post.
switch to another data source to bridge it. That leaves me
This is relevant today, because so many novel investments
very uncomfortable.
like cryptocurrency have suddenly soared in value. A stan-
Laurence Siegel: You may want to talk to Tom Coleman, dard methodology to test for a bubble relies on autocorrela-
with whom I’ve written a paper that may turn into a tion of similar metrics.62 Ross’s insight is that these tests
book.58 He has a lot of data for Germany during the first may not work well.
hyperinflation.
So, anyway, I think the insight of conditioning biases is
Elroy Dimson: It is possible. Basically 1922–1923 was dif- really useful. I wouldn’t change much in the paper; there
ficult because nobody had any vehicles that could move is a little bit about the equity premium and kind of a spat
around fast enough to collect the prices that were going about how big it could get, which I might revisit if really
up so many hundreds or thousands of percent at the peak. pushed.

See Coleman, Oliver, and Siegel (2021).

58

See Brown et al. (1995).

59

See Brown and Goetzmann (2018).

60

See Ross (1987).

61

See Phillips, Shi, and Yu (2015).

62

72 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Martin Leibowitz: If I’m thinking correctly, Toulouse is at Elroy Dimson: It sounds to me like a family business in that
one end of the Canal du Midi, and it was built in something sense.
like 1617, wasn’t it?
Laurence Siegel: Or a private equity investment.
Will Goetzmann: That sounds right.
Mary Ida Compton: Yes, maybe.
Martin Leibowitz: It was an amazing engineering feat at
that time, and I think it was very much used for commer- Will Goetzmann: It’s an interesting business because, as
cial transport between the Atlantic and the Mediterranean the company matured, we got a lot of information about
coast. Could that have had any impact on the company in who the investors were because they listed their profes-
terms of its long survival? sions in the registers. Very few of them were bankers, but
a large and increasing chunk of the shares were owned
Will Goetzmann: That’s a good point. Another early com- by religious institutions—institutional investors. That led to
pany, a company chartered by Jean-Baptiste Colbert, built frictions of various sorts, because, if a church owned, say,
the Canal du Midi.63 The Canal du Midi passes right near the 30% of the equity in the company, they did not have 30% of
Bazacle, so the canal must have been a way for barges to the votes. It was not one share, one vote; it was one share-
transport grain. So, yes, the canal and the mill must have holder, one vote. So large investors didn’t dominate in terms
been really closely connected. of control.

Mary Ida Compton: It’s amazing that the company actu- And there were transaction costs. If you wanted to sell
ally got money out of their equity investors on an ongoing your share, you had to have a big dinner for every share-
basis. Could you imagine that happening today—if you holder. Those were expensive dinners. The church never
bought equity shares in a company and the company said, sold their shares, so people were saying that it was unfair
“We need more money”? for the church to never have to throw dinners for everybody.
There was a big discourse about that, never resolved.
Will Goetzmann: We have a theory that the occasional
negative dividends solved a Jensen and Meckling agency Martin Leibowitz: The Church, I presume, was tax exempt.
problem. Bazacle investors did not leave “free cash flow” in
the firm for managers to exploit.64 Will Goetzmann: I think so.

Robert Arnott: You could have secondary equity offerings. Elroy Dimson: And it probably had a large holding in TIPS, in
We’ve seen several bubble companies take advantage of the sense that the tithing of people’s income was a hedge
this year’s wild valuations to issue new shares and get against increasing labor costs; in effect zero exposure to
some cash to do whatever with. people’s human capital.

Mary Ida Compton: Yes, but never is a public equity investor Will Goetzmann: That’s an interesting wrinkle.
asked to fork over more money.
This marks the completion of the Equity Risk Premium
Robert Arnott: Welcome to the world of partnerships. Forum 2021: Presentations and Discussions.

Jean-Baptiste Colbert (1619–1683) was (among other positions) controller-general of finances under King Louis XIV of France.
63

See Jensen and Meckling (1976).

64

CFA Institute Research Foundation 73

REVISITING THE EQUITY RISK PREMIUM

EQUITY RISK PREMIUM FORUM ROUNDTABLE

DISCUSSION
PAUL MCCAFFREY, EDITOR
Revisiting the Equity Risk Premium

PART I: BUBBLES, MOMENTUM, AND WINNOWING

Editor’s Introduction I’d recommend emerging markets value, but more
broadly I’d recommend diversification.”
Don’t gamble! Take all your savings and buy some He then said, “What do you think of Tesla and Bitcoin?”
good stock and hold it till it goes up, then sell it.
If it don’t go up, don’t buy it. I replied, “They’re very speculative; they’re very frothy.
If you want to go for it, go for it, but don’t put any money
—Will Rogers into those that you can’t afford to lose.”
In the first part of the Equity Risk Premium Forum So, three months later he came to me and said, “Dad, I put
Roundtable Discussion, the participants conduct an the money half in Bitcoin and half in Tesla.” At the end of
in-depth and freewheeling exploration of the bubble phe- 2020 he sent me his account statement, and it showed
nomenon in finance, touching on the momentum factor and +382% for the year. He asked, “Dad, how’d you do?” I said,
the winnowing process, or how certain emerging and frothy “I’m pretty happy with my 12%.”
sectors grow more concentrated over time, separating into
a small number of winners and many more losers. It’s awfully interesting to see that what we regard as
“bubbles” can go much, much further and last much
Indeed, the various bubbles of the past 75 years have longer than most people realize. My favorite example is
tended to echo one another in their development and out- the Zimbabwe stock market during the hyperinflation in
come and have all taught the same lesson, as Laurence the first six weeks of summer 2008. Suppose you saw this
Siegel observes. hyperinflation in Zimbabwe and said, “Get me out of here.
In fact, I’m going to take a short position. I’m going to short
“After the automotive bubble, we had bubbles in aviation
Zimbabwean stocks, and I’ll do it on a safe, small part of my
and radio; then, in the 1960s, the electronics boom; and
portfolio—2% of the total.”
various others later on,” he says. “You can always look back
and say that the bubble was justified because of one great The Zimbabwe stock market, in local currency terms, then
company that is still prospering, like IBM or Boeing. But did rose 500-fold in six weeks as the currency tumbled 10-fold.
you want to hold the index of that industry? Probably not.” So, in dollar terms, it went up 50-fold, meaning that you
just got wiped out. A 2% short position became a 100%
Of course, then the question becomes how to distinguish
short position. Eight weeks later, the currency had fallen
the Googles from the AltaVistas, the Fords from the
another 100-fold and the market basically dropped to zero
Studebakers, early in their development. Unfortunately,
and stopped trading. So, you would have been right, but
there is no easy formula.
you would be bankrupt. These bubbles are very, very inter-
“The basic message may be,” as Martin Leibowitz observes, esting. It is very dangerous to bet against them except in
“you want to have that company which can, if I can use the modest ways.
term, compound its success.”

And, of course, even the company that can compound its

Momentum Investing, Pro and Con
success can’t do it forever—maybe not even for very long. Martin Leibowitz: In the short-term factor studies that
“Palm Computing was briefly worth more than General people have done, one of the factors that keeps cropping
Motors,” Robert Arnott says. “Disruptors get disrupted.” up—with the heaviest weights—is momentum. This is very
curious: Why should momentum have that kind of empha-
Roundtable sis in these types of analysis? If the market is efficient,
would you really expect that momentum would be such a
Robert Arnott: Funny anecdote: My eldest son is some- powerful force? I think there’s an explanation for it, but it
what of an entrepreneur, and he came to me in late 2019 certainly raises eyebrows.
and said, “Dad, I’ve got a quarter million I want to invest.
Where should I invest it?” I answered, “You’re in tech, so Robert Arnott: We published a paper entitled, “Can
don’t invest it in tech. You’ll want to diversify. Your reve- Momentum Investing Be Saved?”1 This was a deliber-
nues all come from the United States, so you want inter- ately ironic title, because how can something that works
national diversification; invest outside the United States. possibly need saving? Well, it works in the sense that, if
you buy stocks that have gone up historically, they keep

1 See Arnott, Kalesnik, Kose, and Wu (2017).

76 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

going up—but the effect has a very short half-life, three strategy, so there’s something to the analogy between
months or less. The stocks stop going up after about six or momentum (in stocks) and the return pattern of options.
eight months on average, and then they give it all back and
then some, which means that you’d better have a sell disci- How many of those left-tail events occur is the variable that
pline or you’re in trouble. drives everything. If you see one 2009-style momentum
reversal every 100 years—and, at that magnitude, that’s
That’s why momentum and value aren’t at odds with about what we’ve seen—momentum is fine. Every once in a
one another. Value says to buy antimomentum stocks. while, it gets killed, but it’s fine. If you see three in the next
Momentum says to buy momentum stocks (obviously). The 100 years, it could wipe out the premium. So, momentum
former is right in the long term, and the latter is right on a investing is a bet that the next 100 years will look like the
very short-term basis. (Cliff Asness is far more expert on last 100. See Exhibit 67.
momentum trading than I am, so maybe he’ll comment.)
One last observation would be that standard momentum, Momentum works a lot better in combination with a value
wherein you build the portfolio using the last 12 months’ strategy that not only uses value as a metric but that also
return other than the last 1 month, has not added value updates the prices fairly frequently—at least at the same
since 1999. So, you got 22 years of slightly negative returns, frequency as momentum so that they’re highly negatively
overwhelmingly driven by the momentum crash in 2009. correlated. I wrote some material on the momentum crash
in 2009 in which I showed that if you combined momentum
Laurence Siegel: I think Cliff would admit or confirm that with value, this was actually not a very tough period for our
momentum can’t really work indefinitely. firm (AQR). It wasn’t a great period, but it wasn’t all that bad
because value did so well. So, it’s a classic case of evaluat-
Cliff Asness: These are all facts. We knew before the 2009 ing something in isolation versus in a portfolio. If I were to
reversal, the momentum crash, that it has a bad left tail. trade only momentum, I would be somewhat terrified. Not
Like anything that is asymmetric or option-like, that risk everything we do has a Sharpe ratio that lets us sleep well
is present. Option replication is essentially a momentum every night.

Exhibit 67. Monthly Returns on Momentum (top third of stocks by trailing

return) vs. Antimomentum (bottom third) Strategies, 1825–2020

Notes: Trailing return: previous 12 months except for previous one month. L/S denotes long–short portfolios of top third minus bottom third, with
and without adjustment to make portfolios industry neutral. Momentum uses the last-year return with a skip month.
Source: Mikhail Samanov, Two Centuries Investments, using data from Goetzmann, Cowles, and Fama–French studies. Industry-neutral return
series since 1968 from AQR.

CFA Institute Research Foundation 77

Revisiting the Equity Risk Premium

But momentum alone? The left tail has been too bad. You can I think that’s a stretch. But it’s overreaction or underreac-
make money for a long, long time like some people are now, tion. The market cannot be completely efficient if you can
and—no one believes it now—they can lose it really, really make money with momentum trading.
fast. Momentum is part of a process that’s also looking for
cheap and, in a different vein, high-quality stocks. We think Cliff Asness: I’ve heard all the efficient-market explana-
the long-term evidence is still very strong about that overall tions for momentum. I’m fine with it either way. As I’ve said
process, but momentum alone is and should be terrifying. many times, I don’t care if our premiums are risk premi-
ums or behavioral premiums. I’ve just never bought the
Laurence Siegel: I’ve tried to describe momentum as: efficient-market explanations.
You look at what stocks have gone up and you buy them
because you’re betting that other people are looking at the Laurence Siegel: What are these explanations?
same data and they’re also going to buy them. Obviously,
Cliff Asness: There are a few. One of them is really bad and
there has to be a point where that game is over.
is still brought up. It’s that momentum is an estimate of the
Cliff Asness: There really doesn’t have to be, Larry. One of expected return. Eleven or 12 months of returns are the
the themes of this talk is that people can keep doing stupid return people expect. So, of course, on average it should
things way longer than we ever thought they could. predict. I studied this as part of my dissertation. I showed
both analytically and through simulations that it does
There are two main explanations for momentum, and predict, but you get a 0.2 t-statistic over 100 years.
they’re amusingly opposite. One is your version, which is
essentially overreaction: You’re buying something because Estimates of the expected return based on one year of his-
it has gone up. You are using no fundamental knowledge torical data are incredibly noisy. Then you have to ask why
whatsoever. you are using one instead of five years, because five-year
returns have a reversal aspect to them and should lead to
The other is underreaction. Yes, you can laugh at finance a better estimate. Other explanations are a little bit more
when it has two competing theories that start with the philosophical—they use real option theory to say that the
opposite word. Underreaction is very simple: Fundamentals Nasdaq was fairly priced at 5000 in the year 2000. Perhaps
move, and so do prices, but prices don’t move enough. You there were states of the world where the Nasdaq was really
would expect this latter effect from the anchoring phenom- worth 25,000! This explanation says that momentum
enon in behavioral finance. wasn’t irrational; it just didn’t pay off because the stocks
turned out not to be worth those prices. But there was a
My personal view: It’s very hard to disentangle these expla- chance. I’ll never say the chance was zero, because we’re
nations because I think both are true and one or the other all statisticians on this forum, and we’d all recoil from giving
dominates at different points in time. On this panel, it’s 0% or 100% odds to anything; we don’t issue guarantees.
controversial to say this, but I think this is a very bubble-ish But I come fairly close to guaranteeing that the tech bubble
time. The overreaction version of momentum is dominating. was net irrational. It got Amazon right.
In more normal times, with more typical value spreads and
nothing too crazy, momentum makes a lot of its money
because people don’t react enough, particularly when Back to Bubbles
changes in fundamentals are revealed.
Laurence Siegel: The tech bubble has been like every other
Momentum even changes your philosophical view of bubble. It’s rational to expect one company to win and all
markets, because overreaction is a disequilibrium strategy. the others to go away; we just don’t know which com-
And, to the extent any of us care about whether we’re help- pany the winner will be. We had at least 1,900 automobile
ing the world, if momentum is overreaction, then momen- companies (not all at the same time) in the early part
tum investing is hurting the world. It is moving prices of the 20th century. Now, we have two and a half in the
farther away from fair value than they already are. On the United States.
other hand, if momentum is underreaction, then momen-
Cliff Asness: Two and a half?
tum investing is fixing an inefficiency caused by people not
reacting early enough; it moves prices toward fair value, Laurence Siegel: I can’t decide if Chrysler is a domestic
toward equilibrium. or a foreign company.

One of my holy grails is to disentangle this question: When After the automotive bubble, we had bubbles in aviation
is one effect driving momentum, and when is the other? and radio; then, in the 1960s, the electronics boom; and
And I would like it to be of practical use, which we all know various others later on. You can always look back and say
is not always the same as disentangling it successfully. that the bubble was justified because of one great com-
pany that is still prospering, like IBM or Boeing. But did you
Roger Ibbotson: Some people have tried to explain momen-
want to hold the index of that industry? Probably not.
tum as if it were consistent with efficient markets, although

78 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Robert Arnott: A few years back, we tried to come up with a price, you would have said that those assumptions aren’t
definition of the term “bubble” that could actually be used plausible. Well, guess what? They exceeded it. They’re the
in real time. Cliff, having written “Bubble Logic,”2 would only one.
probably be very sympathetic to this effort. What we came
up with is this: If you’re using a valuation model, such as Cliff Asness: To be interesting, any of these conversations
a discounted cash flow (DCF) model, you’d have to make has to be about a portfolio. There may be individual stocks
implausible assumptions—not impossible assumptions, but that I would say are ridiculous, but you can never feel
implausible ones—to justify current prices. And, as a cross- nearly as strongly about one stock as about a portfolio. One
check on that first part of the definition, the marginal buyer company could invent the cure for male pattern baldness
has zero interest in valuation models. or figure out how not to fog up your glasses when you’re
wearing a COVID mask. These are two of the most lucrative
To apply this method to Apple, you’d have to use aggressive possible inventions. The exception, clearly, should not drive
assumptions but not implausible ones. So, it’s not a bubble. the rule.
To apply it to Tesla—I debated Cathie Wood about three
weeks ago at the Morningstar conference, and I asked what Robert Arnott: Correct.
her sell discipline was, and she said “We have a target price
of $3,000. You get there, if you assume 89% growth over Winnowing—How Industries
the next five years and valuation pari passu with today’s
FAANG stocks at the end of the five years.” And I had to with Many Companies Become
grant that her analysis was mathematically correct. Concentrated over Time
What I didn’t say, because I had been told by my host to Laurence Siegel: What I was saying about the electronics
play nice, was—gosh—89% compounded for five years bubble, the airline bubble, and all the others is that you
is 25-fold growth. Do you really think that Tesla will be don’t want an index of those companies—you want the
25 times its current size in five years? Amazon grew to winner. You had no idea who that was going to be until after
14 times the size it was 10 years ago, and that company the battle is over.
is a stupendous growth story.

So, you can use techniques in real time to gauge a bubble.

Where it gets really squishy is that you can’t use it to value Exhibit 68. Title Page of 1720
Bitcoin. But you couldn’t use it to value the US dollar either.
Pamphlet about the South Sea
Old Bubbles Bubble
Will Goetzmann: So, Rob, I’m going to show you Exhibit 68.

This is a book, or pamphlet, published by Archibald

Hutcheson in 1720 during the South Sea Bubble. Your strat-
egy is exactly the strategy he took. He said, “What assump-
tions do you have to make about the South Sea Company’s
profits in order to justify the price levels of that company’s
stock?” I think you just followed the footsteps of somebody
who called that particular bubble before it burst.

Robert Arnott: That’s pretty good.

Roger Ibbotson: In the Louisiana Purchase, they actually

did achieve the profits needed to justify the bubble price of
the Mississippi Company. It’s just that shares in the com-
pany didn’t provide the ownership rights to them.

Robert Arnott: The implausible part of the definition leaves

room for the exception that proves the rule. Amazon wasn’t
bubbling to new highs in 2000. It was cratering after 1999,
but it was trading at crazy multiples even so. If you asked, Source: William Goetzmann.
in 2000, what assumptions would justify the then-current

See Asness (2000).

2

CFA Institute Research Foundation 79

Revisiting the Equity Risk Premium

Cliff Asness: Yes. But if you buy 10 stocks, equally weighted, and you hold
the shares and never rebalance, all you need is one winner.
Elroy Dimson: At the end of the 19th century, there was a
very large number of automobile companies. How do you Cliff Asness: But, Jeremy, that means that three years later
know which automobile company is going to survive? you’re willing to own a very concentrated portfolio and ride
it for a very long time. You have to be plausible in framing
Robert Arnott: You don’t. the experiment. Who’s really going to own that portfolio in a
few years?
Elroy Dimson: The canal companies had a hot issue period.
But people didn’t realize that within 30 years there would Jeremy Siegel: But that is what a lot of people do.
be railways all over the country.
Cliff Asness: And the people who do that are ex ante idiots.
Robert Arnott: And the rail companies in the 1840s— Ex post, we laud some of them.
same thing.
Robert Arnott: Jeremy, you won’t necessarily be ahead.
Martin Leibowitz: There were 5,000 different models of In 2000, Cisco was the largest market-cap company on the
automobiles in the early part of the 20th century. planet for about a nanosecond. Since then, its sales have
grown 12% per year for 21 years; its profits have grown
Robert Arnott: Just try getting spare parts for them these
13% per year for 21 years. That’s stupendous growth, yet its
days. [Rob Arnott is an antique car collector. Ed.]
share price is below the year 2000 peak by a goodly margin.
Martin Leibowitz: Exactly. Does a company—one that sur-
So just figuring out which company is going to be the survi-
vives and prospers and has the ability to sell stock and the
vor and winner doesn’t always help. It depends on what you
ability to dominate the market—does that produce a certain
pay. Of the top 10 market-cap tech names in the year 2000,
kind of momentum that is real and carries forward up to a
10 (all of them) underperformed over the next decade. Even
certain point? I would think so.
though the overall market is up hugely since 2000, 9 of the
Elroy Dimson: That’s a question for strategy professors. 10 top-cap tech stocks declined on average over the next
two decades. The one winner was Microsoft, and it beat
Robert Arnott: That holds true until the disruptor gets the S&P by only about 1.5% to 2% per annum over the last
disrupted. Palm Computing was briefly worth more than 21 years.
General Motors. It was also briefly worth more than the
company that spawned it, 3Com. And its product, Palm Martin Leibowitz: So, the basic message may be that you
Pilot, was ubiquitous. Two years later, the BlackBerry want to have that company that can, if I can use the term,
crowded it out, and six years after that, the iPhone was compound its success.
invented. Disruptors get disrupted. How many web search
Laurence Siegel: Yes.
firms were disrupted sequentially, one after another,
after another? Robert Arnott: Or, as Will Rogers famously said, “Investing is
simple. You just buy what’s going up.”
Cliff Asness: I’m still using AltaVista.
Martin Leibowitz: “If it don’t go up, don’t buy it” is the exact
Laurence Siegel: You’ve always been a little different, Cliff.
quote.
Jeremy Siegel: If you continuously rebalance, portfolios of
a few big winners and lots of losers tend to go toward zero.

80 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PART II: RISK PREMIUM OR FEAR PREMIUM?

Editor’s Introduction dominant in the discussion of asset pricing because the
key idea is preferences.”
Suppose a Roman centurion had invested one In terms of what makes up the premium, Rajnish Mehra
drachma at 4% and this compounded in a totally wonders, “How much of it is a risk premium and how much
safe way over the years. of it is due to other factors?”
—Sidney Homer Mehra concludes that about 1.5% of a 6% premium comes
down to risk. The rest is the result of other factors.
Yes, suppose he did. How much would the centurion’s
descendants have today?

We don’t have the exact answer, but decades later, Rob

Roundtable
Arnott ran a similar calculation. “In a footnote, I hypothe-
sized one ounce of gold, which at the time was $350 an Do Most Stocks Underperform
ounce,” he says. “So, 1/350th of an ounce of gold back at
the birth of Christ growing at 5% [would have resulted in] a
the Market?
sphere of gold as large as the earth’s orbit around the sun.” Roger Ibbotson: Hendrik Bessembinder has a paper that
says most of the return comes from just a few companies.3
One could say that Arnott’s result shows the incredible
power of compound interest, but that would be wrong. Robert Arnott: Oh, that is so true.
What it really shows is the absurdity of compounding up
any growth rate for very long periods of time. Antti Ilmanen: And potentially so misleading. That study
should be a paean to diversification, not for concentrated
We thus must be careful about extrapolating too far into stock picking. It is often read the other way.
the future from an estimate of the equity risk premium, or
of the riskless interest rate unless that number is zero, or Laurence Siegel: What Bessembinder said is that only 4%
about any other growth rate. With that as a caveat, let’s of the shares in the market outperformed Treasury bills.
return to our task of estimating the equity risk premium for But because you don’t know in advance which 4% are going
the next 10 years. to generate all the return, you have to diversify. If someone
misinterprets his paper as saying you should concentrate,
What explains most stock returns? And, is a “risk premium” they really don’t understand the paper.
an accurate description of what equities provide? In the
second installment of the Equity Risk Premium Forum Cliff Asness: I’ve skimmed the paper, and I think he looks
Roundtable Discussion, panelists address these questions at compound returns. You would expect most companies to
and consider the components that make up the equity risk underperform the average when measured by compound
premium as well as why the phenomenon is so large. returns, because companies leave the sample when they
die or have near-death experiences, driving the compound
They start by noting that only a few stocks account for return down, potentially toward zero.
most of the returns of an equity index. Indeed, according to
a study by Hendrik Bessembinder, a paltry 4% of the stocks I’m not denying for a second that the distribution of
in the market outperformed Treasuries over his study long-period returns is going to be asymmetric and that a
period. So just a handful of companies are responsible for few winners drive a lot of the market’s overall return. I sus-
the equity premium. pect he is painting too negative a case. If we had a large
number of equally good stocks, their average geometric
But is that premium really a risk premium? “For at return (which I think is what he looks at) would be con-
least 20 years I’ve been an advocate of the notion that siderably lower than the return on the portfolio. And you’d
we shouldn’t call it a risk premium,” Rob Arnott says. say that most of these stocks underperformed the market
“We should call it a fear premium.” when that really wouldn’t be true.

“I’ve used the word ‘popularity,’ which includes all kinds Roger Ibbotson: I think that if you measure each company
of premiums, whether they are risk or non-risk,” Roger as though that’s all you held, you’d get that geometric
Ibbotson says. “And I think that risk has become too

3 See Bessembinder (2018).

CFA Institute Research Foundation 81
Revisiting the Equity Risk Premium

return. And most of them would underperform. So, I think Jeremy Siegel: I wrote a paper called “The Nifty-Fifty
it’s appropriate to measure the geometric return. Revisited,”4 where 20 years later I said that if you had
bought the Nifty 50 on an equally weighted basis at the
Cliff Asness: I’m not denying that, Roger. And I’m not saying peak of the mania back in 1972, you actually did as well as
it’s not a good point. But it is a less surprising point than it the S&P, but you became concentrated again. Interestingly,
is often painted as. again, a few winners compensated for most of the stocks
being losers—but everyone depicts that period as “crazy
The Weird Behavior of a Geometric- prices.” But, depending on how you played it, you did as
well in those crazily priced stocks as in the S&P 500. You
Mean Stock Index Futures Contract just didn’t have as much diversification. It’s the same phe-
nomenon we were talking about with Cliff earlier.
Jeremy Siegel: The Value Line index, traded on the Kansas
City Board of Trade, was a geometric-mean index, and it was Roger Ibbotson: What we have been discussing also
the first stock futures index. It preceded the S&P 500. The related to the increasing concentration of wealth. As entre-
futures contract sold at a premium until a great economist preneurs, you are putting all your bets on your particular
that we all know, Fischer Black, determined that geometric enterprise, and most of them fail. The few winners stand
indexes will always underperform. He worked at Goldman out, but that is not the typical entrepreneur’s experience.
Sachs, and he had them take huge multibillion dollar posi-
tions against the premium in the Value Line contract. He Robert Arnott: That’s the best argument for a very low cap-
made a lot of money for the firm because the premium ital gains tax, because if you are taxed on your winnings
finally disappeared, and it traded at the discount that he and get no rebate on your losings, you sharply diminish the
talked about. Eventually they just delisted the entire index. incentives for risk-bearing.

Elroy Dimson: There’s a neat little article that Jay Ritter Martin Leibowitz: Rob, if you look at personal portfolios, not
wrote about that. of the ultrarich but of more typical well-to-do investors, you’ll
find that, after the kind of run-ups that we’ve had, people
Jeremy Siegel: Yes, I was at Wharton with Jay. I was the have held onto their stocks. They have an aversion to paying
one pulling the January effect off the Kansas City Value Line capital gains taxes, so their portfolios have become extraor-
contract. dinarily concentrated. One can only wonder whether society
wants to encourage that kind of risk-taking.
Laurence Siegel: I never thought that the market
could possibly be deep enough for a firm the size of
Goldman Sachs to make real money off of the Value Line Are Stocks Used to Smooth
geometric-mean effect.
Personal Consumption?
Jeremy Siegel: I’m not saying how much money. Fischer
Rajnish Mehra: Larry had asked me to comment on the
made some off of Jay Ritter. He made some off of me,
macro models from which an equity premium can be
although I started disentangling myself from that pretty
derived. I wanted to bring up one point here, which is that as
early, because I realized that someone knew something on
wealth concentration increases, stocks are not being used
the other side of the trade. But I don’t know the magnitude
to smooth consumption over time. People are holding on to
for Goldman Sachs back then.
their stocks, passing them from generation to generation.
Fischer was the one who taught everybody. He figured it
If stocks are not used to smooth consumption, then the
out. This should be at a discount. It should never be at a
valuation of stocks has very little to do with the real econ-
premium. For three years, it was at a premium, and it was
omy. The macro models are only valid if you have some
the only stock index that was trading.
substitution over time. I have a paper showing that if you
Roger Ibbotson: When Fischer Black was working at don’t have any substitution of current for future consump-
Goldman Sachs, he’d come up with an idea once a year, or tion but the stocks instead go into bequests, the value of
something like that. Some of the other people there were the market is going to be detached from the underlying
wondering what was going on with him—this guy hardly economy.5 Any thoughts on that?
works. But when he had an idea, it was always a really
Laurence Siegel: I think that, on the margin, stocks are
impactful one. He wasn’t your typical investment banker.
being used to smooth consumption within one person’s
Several speakers at once: No!!! lifetime. The average person who owns stocks has several

4 See Siegel (1995).

5 See Constantinides, Donaldson, and Mehra (2007).
82 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

hundred thousand dollars in capital. They’re going to retire you were trying to assess the productivity of the society
on that as best they can, with the added benefit of Social when you look at the demographic profiles.
Security. They do not have any other way to smooth con-
sumption other than to follow the classic lifecycle model. Arnott: In fact, if you look at Japan, you find that Japan’s
per-working-age adult GDP has, in the last 25 years, grown
If you only look at ultra-high-net-worth people, they are roughly pari passu with Europe’s per-working-age adult
different—but there are not many of them. GDP, which has grown roughly pari passu with the United
States. The difference in growth rates is entirely a function
Rajnish Mehra: But more and more are not using stocks to of the demographics.
smooth consumption. There used to be a time when people
were using it.
A Deranged Market Hypothesis?
Jeremy Siegel: It’s always been that way. Franco Modigliani
pioneered the lifecycle hypothesis after Friedman’s perma- To interject something that will sound like it’s out of left
nent consumption model. That is how people behaved. But, field but has a very direct bearing on the risk premium
then as now, there were people for whom stocks were just dialogue—for at least 20 years, I’ve been an advocate of
a means to build wealth and pass it on. the notion that we shouldn’t call it a risk premium. We
should call it a fear premium. Many of you may remember
Today, something like 50% of the population owns some David Hirshleifer’s famous thought experiment in 1999, in
stocks directly or through their pension funds. That number which he said (paraphrasing): Suppose a school in Chicago
used to be 10% or 15%, and they skewed richer than they had come up with the Deficient (or Deranged) Market
do now. So, I would actually say that more people are using Hypothesis and Bill Blunt (i.e., not Bill Sharpe) at Stanford
stocks for consumption smoothing in retirement than they had come up with DAPM, the Disorderly Asset Pricing
used to. Model, they would be declared to be the most validated and
proved set of hypotheses in the social sciences.6
Laurence Siegel: Jeremy’s right. We used to be a defined
benefit pension plan world, where middle- to upper-middle- He was joking, but he meant that if your starting point was
class people—the only people that count on the margin in market inefficiency, you could find ample proof of that, just
this exercise—had their consumption smoothed for them by as many efficient-market types say it’s well documented
corporations, unions, and governments. Now you have to that the market is efficient. If it had been called a fear
do it yourself. You’re going to sell stocks when you get low premium from the beginning, the value effect would be
on money, now more than ever. The fact that a few people expected—not as a risk factor, but because buying loathed
have so much money that they’ll never need to sell stocks and feared companies is scary. The size effect would be
isn’t really relevant. expected but relatively weak, because buying small com-
panies that are not widely understood engenders a little
Mary Ida Compton: Also, we’re all living longer. more fear than buying well-established companies.

Laurence Siegel: Wait until you’re 90! You’ll sell stocks to Roger’s liquidity factor would be expected. Long-horizon
provide for your consumption. mean reversion would be expected. Even momentum would
be expected, based on fear of missing out. If we thought of
Robert Arnott: Demography is a very big factor in lifecycle
the equity premium as a fear premium—if we had the luxury
consumption smoothing. We have an enormous roster of
of going back 60 years and labeling it a fear premium—a lot
baby boomers, right here on this forum, who are looking
of the so-called anomalies that we’ve talked about would
to convert assets accumulated during a working lifetime
not be anomalies at all. They would be totally reasonable
into goods and services late in life. And there is a propor-
and expected.
tionally smaller roster of prospective buyers than in past
generations. Roger Ibbotson: I think that the fear premium is an inter-
esting concept, and I’ll give it some thought. I’ve used the
Laurence Siegel: Millennials are a larger generation than we
word “popularity,” which includes all kinds of premiums,
are, and eventually they’ll be rich enough to buy our stocks.
whether they are risk or non-risk. And I think that risk has
But it may be a long wait.
become too dominant in the discussion of asset pricing
Robert Arnott: And it’s proportional relative to the past. because the key idea is preferences.

Martin Leibowitz: Rob, I think this question also connects We started out with the Capital Asset Pricing Model, where
with your graph on demographics. If I’m thinking correctly, you only are afraid of one thing, one kind of risk. Ultimately,
we generalize it to include many dimensions of risk, but we

6 See Hirshleifer (2001).

CFA Institute Research Foundation 83
Revisiting the Equity Risk Premium

want to generalize it even further, to non-risk characteris- The evidence that nobody gets those returns is that we’re
tics. For example, I don’t think of liquidity (actually the lack not all rich. (See my introductory article in this book, “Why
of it) as a risk, even though the literature talks about liquid- Aren’t We All Rich?”) From time to time, almost everyone
ity risk. You can conceive of a liquidity factor, but that factor has cash flow needs, emergencies, times when you need
does not make liquidity a measure of risk. Illiquidity may to withdraw from the market or at least can’t contribute to
be a source of fear. However, there are a lot of preferences it. As Jeremy has said, you spend the “income,” but income
that go beyond fear. is a legal concept denoting whatever is subject to the
income tax. More likely you spend your market “profits” in
But I agree with you, Rob, that fear encapsulates a broader whatever way your mental accounting defines “profit.” So,
notion than risk as we measure it. It’s an interesting con- the vagaries of human life make it impossible to realize a
cept, but it might not be general enough. 5%, 6%, 7% equity premium.

Martin Leibowitz: On that score, I’m reminded of an event

The Equity Premium “Puzzle” that took place when Sidney Homer and I were writing
Literature: Why is the ERP So Big? Inside the Yield Book.10 It goes back to the 1960s and
early 1970s. After we had written the book, Sidney asked
Jeremy Siegel: I’d like to address Raj’s original article,7 me a question. He said, “Suppose a Roman centurion had
which asks, “Why is the equity risk premium so big?” invested one drachma at 4%, and this compounded in a
Everyone has twisted and turned, used the Von Neumann– totally safe way over the years.” He asked me to calculate
Morgenstern utility function, and done various other things what that total amount would be today.
to get an answer. Does anyone here have an explanation
with which they feel satisfied for why the equity risk pre- This turned out to be a very difficult problem because
mium is so large and persistent and universal? standard calculators couldn’t do the math. Even using a
computer didn’t work. I had to use logarithms, and when I
Rajnish Mehra: I’ve tried to give some answers. I think got the answer, it turned out to be far more than the total
the borrowing constraint stuff that I did with George wealth of the world at that point in time.
Constantinides and John Donaldson is one answer.8 If
younger people can’t borrow to buy enough equities to hedge Laurence Siegel: I’m calculating it while you speak…
their future income uncertainty, and older workers have
mostly resolved their income uncertainty, then (as we wrote), Martin Leibowitz: The next question was an even better
question. Sidney asked, “What happened to it all?”
fluctuations in [the] consumption [of older work-
ers] occur from fluctuations in equity income. At Laurence Siegel: Where did all the money go? Of course,
this stage of the life cycle, equity income is highly I would say that all that wealth was never created in the
correlated with consumption. Consumption is first place. The idea of investing a drachma at 4% for 2,000
high when equity income is high, and equity is no years is a thought experiment that has never been put into
longer a hedge against fluctuations in consump- practice—and won’t be.
tion; hence, for this group, it requires a higher rate
Jeremy Siegel: People consumed the dividend. The growth-
of return.9
of-a-dollar, or drachma, calculation assumes that we invest
And this middle-aged group is the dominant, price-setting the dividend.
group in the equity market. So this market segmentation story
Laurence Siegel: Consumption!
is, I think, a reasonable explanation for equity prices that are
low enough to provide, on average, a high rate of return. Jeremy Siegel: You consume the dividend.

Martin Leibowitz: Consumption, wars, pandemics.

Why High Equity Returns Have
Not Made Everyone Rich Jeremy Siegel: No. Just consume the dividend. You don’t
need any of that other stuff.
Laurence Siegel: Some decades back, I wrote that the
equity market is much riskier than it looks from the Laurence Siegel: It’s 2.6 × 1034 drachmas.
Ibbotson chart because nobody gets those returns.

7
See Mehra and Prescott (1985).
8 See Constantinides, Donaldson, and Mehra (2002).
9 See Constantinides et al. (2002, p. 271).
10 See Homer and Leibowitz (2004).
84 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Robert Arnott: I did that as a thought exercise in one of explain the equity risk premium have already taken into
my Journal of Portfolio Management papers.11 In a footnote, account your point about the market being riskier than
I hypothesized one ounce of gold, which at the time was what we see.
$350 an ounce. So, 1/350th of an ounce of gold back at
the birth of Christ growing at 5% and the outcome was a Martin Leibowitz: What’s the problem with just looking at
sphere of gold as large as the earth’s orbit around the sun. the issue of moving from a riskless asset into a risky asset
and asking the question: What level of premium does it
Laurence Siegel: And if you add a few more millennia and take to achieve a sufficiently satisfactory level of success—
go back to the days of the Pyramids, the sphere of gold of beating that base level over a typical relevant invest-
might be larger than the universe. ment period like 5 years or 10 years?

Elroy Dimson: If you look at Victor Haghani’s website, you Roger Ibbotson: It’s not too high.
see where he asks, “Where are the billionaires?” He used
the long-term data that we’ve been discussing to work out Martin Leibowitz: When you do that, you get numbers of
how many billionaires there ought to be if it weren’t for all 4% to 6%, which is in the range of the numbers we’ve been
those who are siphoning it all off. talking about. So that is not unreasonable in terms of how
people would think about making the move from riskless
And of course, there would be many more billionaires than too risky.
there actually are.
Rajnish Mehra: So, Marty, let me set the stage a little
Laurence Siegel: What website is this, please? bit. What’s happening is that we’re observing a pre-
mium, 6.5%. That’s an observation. Now, you try to come
Rajnish Mehra: Elm Partners. up with a model that is consistent with other observa-
tions in insurance literature, other macro models, other
Elroy Dimson: Victor Haghani was one of the LTCM part-
possible estimates of risk aversion, and so forth. That
ners, who started up another firm to look after the modest
model, which is consistent with other observations and
amount of wealth that he still had.
with macro, generates a risk premium of only about
Elroy Dimson: He’s done that in a TEDx Talk as well.12 It’s 1% or 1.5%.
very amusing. But the problem is that what he was mod-
The question is, why such a big difference between the
eling was somebody who never spends any of it. If people
observation and the model answer? There’s no dispute
behaved like that, there would be loads and loads of bil-
about the size of the realized premium. But how much
lionaires, but they would be worse off than somebody who
of it is a risk premium and how much of it is due to other
doesn’t have any money at all. They both end up having
factors? That is something that I wanted to bring up today
spent nothing, but the Victor Haghani clients would have
in a serious way. How much of this 6.5% is a premium for
spent their time also worrying about how things are going.
bearing risk itself?
Laurence Siegel: The billionaires wouldn’t really be worse
Once the existence of a premium is known—once it is in the
off because they would have a nonexpiring option to stop
information set—it must persist if it is a genuine risk pre-
being misers and live a little, but the point you’ve made is
mium, because the risk continues to be there. If it’s a factor
indeed very funny.
premium, it does not have to persist. All factors come into
and go out of fashion. People will say, “value is working.”
Back to Asking Why the Equity So, at that stage, there may be a value premium; or “size is
working,” or “momentum is working,” or “accruals are work-
Premium Is So Large ing.” So, I wouldn’t say that those are risk premiums; those
Jeremy Siegel: Larry, I want to go back to your point that are factor premiums.
the market is actually riskier than we perceive. Raj’s origi-
The question is: Is this premium that we observe for equi-
nal model is a model of consumption maximization under
ties a risk premium? We have several theories that address
uncertainty, with risk and all the rest, and it can’t derive
the question, and some of them would say that not all of
the premium. There are some variations of his model where
the 6% is a risk premium. They say part of it is a risk pre-
you have a minimum amount of consumption, and so on.
mium and the rest is a premium for other things.
But the standard models that have been derived to try to

See Arnott and Bernstein (1997).

11

12Victor Haghani’s TEDx Talk “Where Are All the Billionaires? Why Should We Care?” is available at www.youtube.com/watch?v=1yJWABvUXiU. It is
also accessible from the website of Haghani’s firm, Elm Partners: https://elmwealth.com.
CFA Institute Research Foundation 85
Revisiting the Equity Risk Premium

PART III: BEARS AND HAWKS, EQUITIES

AND INFLATION
Editor’s Introduction Roundtable
The problem with timing … is that there just aren’t
enough data points to prove anybody can do it. Does Equity Risk Decline with
So why bother? You’re just shooting yourself in Longer Investor Time Horizons?
the foot.
Martin Leibowitz: We’ve been talking about “the” risk pre-
—Mary Ida Compton mium. Will Goetzmann pointed out, though, that over the
course of time, the risk premium has declined, depending
When the Equity Risk Premium Forum was held in October
on whether you invest for 40 years or 400. The idea of the
2021, inflation had only just begun to spike and was far
risk premium as being a term structure is very important.
from reaching its recent highs. Nevertheless, the par-
Because what premium you would demand if you’re invest-
ticipants expected it to increase further thanks to all
ing for 1 year will be different from when you’re investing
the expansive monetary policy. In the third part of the
for 5 years or, say, 100 years. We would expect that to
Roundtable Discussion, they delve into the inflation out-
be a declining curve. That’s very important, because the
look as well as the shortcomings of the Shiller cyclically
investors can choose their time horizon, just as they can in
adjusted price earnings (CAPE) ratio and the prospects
bonds. Over a long time horizon, the risk that is relevant for
of a low-return environment in the years ahead, among
them may be much less.
other topics.
Rajnish Mehra: No, Marty, that is not correct. You’re assum-
CAPE’s principal weakness is as a timing mechanism,
ing mean reversion. If you have an IID (independent and
according to the panelists. “The more times you respond to
identically distributed) process, then horizon shouldn’t
a CAPE signal in a given period,” Elroy Dimson explains, “the
matter. The result that Will got is precisely because there
lower your long-term return is going to be.”
is a mean-reverting component in the dividend structure.
Of course, that doesn’t mean the CAPE isn’t useful, even If you have mean reversion, Marty, you are 100% cor-
essential, it’s just a question of applying it in the proper rect. Risky assets will look less risky over time. But if the
contexts. As Cliff Asness remarks, “When you are forecast- returns are IID draws, then the time horizon wouldn’t make
ing poor 10-year returns, even if you don’t explicitly say to a difference.
underweight equities, sometimes that’s what it sounds like.
Jeremy Siegel: That is true, but I’m making one correction.
But we should remember that CAPE is not that good for
You have to have a degree of risk aversion over 1 for that.
that. The forecast is, nevertheless, important.”
You need two conditions for getting a higher equity alloca-
As for the outlook for equities over the next decade, the tion for longer periods: mean reversion and risk aversion
panelists were decidedly bearish. greater than 1.

Jeremy Siegel posed the question, “How many here think Robert Arnott: Mean reversion has been a lively topic today.
the next 10-year equity returns are going to be below the It is weak on a short-term basis, which is one reason the
long-run average?” CAPE is such a lousy predictor of 1-year returns. But, on
longer horizons, it’s pretty good. Jeremy, you’ve written
All agreed. about this, where 30-year S&P volatility, when annualized,
is distinctly lower than the volatility of 1-year returns. This
They weren’t optimistic about inflation, either. comes from the fact that there is mean reversion over long
horizons. For example, 10-year real returns for US stocks
“The Fed balance sheet has exploded to record levels,”
have a −38% serial correlation with subsequent 10-year
Roger Ibbotson notes. “Where is that money going to go?”
earnings; and 10-year real earnings growth has a −57%
Much of it went into real estate and the stock market. correlation with subsequent 10-year earnings growth.
But goods and services weren’t immune. And don’t bet That means there is mean reversion. But it acts over a long
on that changing any time soon, at least according to enough horizon that most people think that returns are IID.
Jeremy Siegel.
Will Goetzmann: I spent the first 10 years of my early
“I think we’re going to have 20% inflation in the next two or research career on the weakness of the mean reversion
three years,” he says. “Not each year, but cumulatively.” evidence. But then the 2013 Nobel Prize award cited
86 CFA Institute Research Foundation
 Revisiting the Equity Risk Premium

Bob Shiller’s work demonstrating the predictability of That doesn’t mean that you’d change your stock/bond
stock returns. The evidence is always a bit marginal, and it allocation even if you buy his model. He seemed to imply
depends on your assumptions and on where you get the that it did. I pointed out that that parameter uncertainty
data. And, as Goyal and Welch have shown, sometimes it would be true of every asset. Furthermore, even TIPS are
sort of falls in the statistically significant zone, and some- not risk-free, as they adjust with a lag and would suffer
times it kind of falls out of it.13 It depends on when you’re greatly in hyperinflation. Every asset has that same extra
doing your measurement. So, it’s a bit of a chimera to say degree of uncertainty, what’s called parameter uncertainty.
that we know for sure. I’m not entirely convinced that you
would bet your wealth on this reversion process.
Noise
Antti Ilmanen: When I look at the literature, I see evidence
I also want to mention one thing in response to what Rob
of mean reversion over time horizons from 3 years up to
said about using fear for value investing. All you need is
15 years. It’s similar to business cycles having turned from
a noisy market, where there are shocks to prices away
4-year cycles into 10-year cycles. We have many questions
from equilibrium, plus or minus, to have value “work.” There
on structural changes. The evidence is really fuzzy, and
might be more than just noise in the market, but noise is all
usable or actionable evidence is almost zilch because of all
you need. Prices just flip up and down. This has nothing to
this horizon uncertainty.
do with sentiment, overreaction, underreaction, or anything
By the way, I wanted to comment earlier on mean reversion like that—just price movement unrelated to fundamentals.
in a different context, not about the premium but about And that will yield a value premium, I believe. That’s it.
the riskiness of stocks being related to the time horizon. You don’t need anything else.
There is a counterargument by Pastor and Stambaugh that
equity risk doesn’t decline with horizon.14 When you take Does CAPE Work Internationally?
into account parameter uncertainty—the fact that we don’t
know how big the equity premium is—their analysis sug- Elroy Dimson: Paul Marsh and I tried the Shiller CAPE on a
gests that risk in equities doesn’t decline with the time large number of different countries. We took all of the coun-
horizon and, if anything, rises with it. tries that had data from 1900 onwards. Of course, we don’t
have P/Es. I doubt that earnings in the United States from a
century ago are comparable to earnings calculated today,
Visualizing Returns over Time: but they’re better than earnings figures for other countries,
Trumpets and Tulips which we don’t have at all. What we do have is dividends,
and those numbers are reliable.
Roger Ibbotson: Even if returns were IID, what you would
get, of course, is a lognormal spreading out of wealth out- In the United States, we can look at the relationship
comes over time—multiplied by the square root of time. And between the Shiller CAPE10 and what could be called
the compounded return is divided by the square root of the Shiller CAPD10 (cyclically adjusted price/dividend
time. So, you get two entirely different shapes, depending ratio). D10 is dividends averaged over a cycle of 10 years.
on whether we’re talking about the compound return or just We extrapolate from that relationship to get a pseudo-CAPE
your ending wealth. Over time, ending wealth spreads out for all the countries.
in the shape of a tulip. The compound annual return, in con-
trast, is averaging out and looks more like a trumpet. We created some trading rules to move away from equities
when the Shiller pseudo-CAPE is telling you that you should
The tulips and trumpets apply only if returns are IID. If be out of equities. For almost all countries, the trading
there’s some other sort of return pattern, then the shapes strategy reduces your long-term return from that country.
will be different. It reduces the return even though it sometimes tells you to
get out of equities for moderately good reasons, because
on balance equities give you a premium, and you missed
Coping with Parameter Uncertainty the premium. So, the more times you respond to a CAPE
Jeremy Siegel: Antti, I want to return to what you said about signal in a given period, the lower your long-term return is
Stambaugh. Parameter uncertainty also applies to bond going to be.
returns—you don’t know what the parameters are for the
Laurence Siegel: Doesn’t part of this result from build-
real risk-free rate, either.
ing in a 10-year lookback as Bob Shiller did? That seems

See Goyal and Welch (2003).

13

See Pástor and Stambaugh (2012).

14

CFA Institute Research Foundation 87

Revisiting the Equity Risk Premium

awfully long to me. The world was a very different place Elroy Dimson: Because the aggregate value wasn’t there.
10 years ago. Or did you look at different periods? I know the British figures better than I know them for other
countries. The proportion of equities was something like
Elroy Dimson: We did it with 1, 2, 5, and 10 years. The con- 15%, and the rest was fixed income.
clusions are the same. The Shiller signal is an inaccurate
signal, and the number of times that it takes you out of Laurence Siegel: Maybe you just use the aggregate supply
equities and into something else—typically cash, whatever of securities as the benchmark.
the lower-risk alternative would be—the more costly it is
to pursue the Shiller strategy. So, I’m not as convinced as Elroy Dimson: You could do that. We didn’t. I think that
most of you are that CAPE works. I think CAPE maybe works would lead you in the same direction.
in the United States.
Martin Leibowitz: Even in the United Kingdom, the fixed-
When we looked at different countries, it only really worked income market was mostly government bonds.
in the United Kingdom. In the United Kingdom, it took you
Elroy Dimson: Outside of the United States, there are no
into equities in late 1974 when share prices were very
markets with a long-term history for corporate bonds.
depressed, and then in the first six weeks or so of 1975,
the stock market doubled. In that one instance, CAPE pro- Will Goetzmann: Another thing, though, is that if you’re
duced a very large benefit. But that’s an anomaly—it’s one flipping between cash and stocks, it’s not the same risk
observation. as a 50/50 portfolio. The reason is that the probability of
an overall portfolio decline of 20% is larger for the flipping
What Is the Right Benchmark for strategy than for the 50/50 strategy, because the flipping
strategy is sometimes all equities and the 50/50 strategy is
Testing a Stock-Bond Switching always diversified. So, a benchmark that is 50/50 or 60/40
Strategy? is not the same risk profile at all. If you’re concerned about
the magnitude of losses, you’re facing a higher chance of
Robert Arnott: Elroy, the relevant benchmark for a switching something extreme happening if you’re flipping.
strategy would be a static mix strategy. Not equity returns,
but a balanced portfolio return that matches your average Elroy Dimson: Yes. This was not a Journal of Finance paper.
equity exposure. It appeared in our yearbook in 2013.15 People were inter-
ested in it. One would do much more if this was an aca-
Elroy Dimson: No, that’s using hindsight. We roll forward, demic paper.
and we had alternative strategies that only used either
past data or contemporaneous data from other markets. Robert Arnott: I’m guessing more practitioners read it than
If you know what is going to happen—if you know what read the Journal of Finance.
the unchanging passive strategy would be—then Shiller
Will Goetzmann: If you improve on this, it might be worthy
wins hands down. But that’s not what we looked at.
of the Financial Analysts Journal, Elroy.
Laurence Siegel: I think he’s saying your benchmark should
Elroy Dimson: If I do a few more like that, I might get tenure.
be of comparable risk, so it should match the amount of
[Dimson has been tenured since the late 1970s. Ed.]
equity exposure on average over time in your CAPE strategy
up to that point, whether it’s 50/50 or 60/40 or some other
fixed mix. Will We Be in a Low-Return
Elroy Dimson: No, not at all, Larry. For most of the historical Environment?
period, having anything close to half your money in equi-
Cliff Asness: The discussion seems to have regressed
ties would have been so crazy that nobody would have
to CAPE or some similar measure as a timing tool again.
imagined doing it. You have to use data that exist at the
I want to reemphasize something I said in my presentation.
point of measurement, and then model that going forward.
I think CAPE has been an empirical failure for timing. It has
You can’t take a peek at what’s going to happen in the next
still been a success if all you want to know is whether
century and conclude that 60/40 is a plausible asset mix.
you expect the next 10 years to be better or worse than
Laurence Siegel: Why do you think it was crazy to have half average.
your money in equities if there was a supply of equities that
Robert Arnott: Very much so.
would have allowed you to do that?

See Dimson, Marsh, and Staunton (2013).

15

88 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Laurence Siegel: I agree that CAPE is a tool for forecasting, Martin Leibowitz: On the other side of that coin: How often
not timing—but some people will use long-term forecasts have you heard the argument that “I have to be in stocks
as a timing tool, although they should not. because bonds don’t give me any return”?

Cliff Asness: Larry, if you remember, I also said we’ve all Mary Ida Compton: A million times.
been guilty of that. When you are forecasting poor 10-year
returns, even if you don’t explicitly say to underweight Martin Leibowitz: When will that argument be false?
equities, sometimes that’s what it sounds like. But we
Laurence Siegel: When the expected return on stocks is
should remember that CAPE is not that good for that. The
lower than the expected return on bonds.
forecast is, nevertheless, important. If you’re a pension
plan and expecting 2% instead of 6% on stocks in the next Jeremy Siegel: You’re right.
10 years, that information might be relevant to you.
Robert Arnott: That was the case in the year 2000.
Laurence Siegel: No kidding.
Jeremy Siegel: That was about the only time.
Cliff Asness: It helps you answer questions like, how much
do you have to save? How much can you spend? It is an Robert Arnott: Mary Ida’s task is very challenging. Any sort
important number. It’s just not an important number for of timing mechanism is going to be suggesting buying
deciding when to get in and out of the market. when equities are fiercely out of favor, unloved, cheap—and
will suggest trimming when they’re relatively fully priced
Jeremy Siegel: But what happens if you say that stocks are and people are comfortable with them. So, for far too many
going to return less, but bonds will return much less? institutional investors, that sort of information, while useful,
is not actionable.
Will Goetzmann: Then Mary Ida has a problem when she
talks to her clients. Mary Ida Compton: The problem with timing, which we
never do, is that there just aren’t enough data points to
Laurence Siegel: She sure does.
prove anybody can do it. So why bother? You’re just shoot-
Jeremy Siegel: That means you go into stocks. They’re ing yourself in the foot.
going to return less, but you go into stocks.
Laurence Siegel: Mary Ida faces a situation that I believe
Mary Ida Compton: It’s a strategic asset allocation decision, most of us don’t, which is that her clients have fixed
not a tactical one. Stick with it over the long term, but what liabilities. As individual investors, we can adjust our con-
you as a pension plan sponsor are going to have to do is sumption to the varying fortunes of our portfolios, but a
suck it up and put some more money into the fund. pension fund really can’t. They have to come up with out-
side money. Moreover, the fortunes of markets and of pen-
Cliff Asness: Yes, you’re exactly right. When expected sion plan sponsors are correlated. When the market’s down,
returns on everything are low, and you don’t have the the company is usually also not doing well. It really puts
ability to know when those low returns will be realized, you in a terrible situation. You are supposed to earn some-
you simply lower your expectations. thing like 7% to meet your pension obligations, but there’s
nothing to buy that has an expected return of 7%.
Laurence Siegel: That’s what Jack Bogle said: Budget for it.
Cliff Asness: If you literally have a subsistence level of
Cliff Asness: It’s important to note that saying “returns on required return that is considerably higher than any reason-
an asset will be lower than normal” is different than saying able portfolio’s expected return—and it’s true subsistence,
“they have a negative expected return.” So, when we say like you have to make it or you die—you are forced to do the
stocks will be worse than bonds, do we mean that stocks opposite of most of our instincts. You’re forced to take more
have a negative expected return? If you actually believe risk when risk is not being very well rewarded. While that’s
that, you should underweight them or short them. a real-world problem for some, it is not the optimal strategy.

But if you believe that stocks have a healthy positive Sometimes people skip a step and end up saying that their
risk premium, just half of the normal amount—and if you expected return on stocks is 11%. Sometimes Wall Street
underweight them now and overweight them later on when strategists do this. They engage in a kind of magic presti-
they’re more attractive, you could still make money (if the digitation where they say to themselves, “I’ve explained to
timing signal is any good). Underweighting a positive hurts you why holding stocks is justified; justified means normal;
you, but overweighting a positive helps you more. This is a normal means 11%.” [Crowd laughter.]
very long game…
That last step is not right. You have to accept the lower
Robert Arnott: …and it will be wrong at times. expected return on both stocks and bonds. I think some
people forget that bonds now have very low yields and

CFA Institute Research Foundation 89

Revisiting the Equity Risk Premium

that you add the equity risk premium to that low number. have to, or when some event forces you to. So, the contin-
You don’t get 11%. uation of a strategy in an institution, and in individuals as
well, has inertia—in other words, a bias to the strategy that
is already being pursued. That observation has a power
Reaching for Yield—in Bonds beyond just the theory that you should maintain a cer-
and in Stocks tain allocation over time. Never mind the theory. There is a
behavioral imperative that forces people in an institution to
Roger Ibbotson: Clearly, this happens in the bond market maintain a consistent strategy.
because people reach for yield when spreads are really
tight. Obviously, that is a time when lower-quality bonds are And in fact, in some ways, even for an institution it doesn’t
not giving much payoff for the extra risk, but at such times, make sense, because, as we were saying earlier, if they
bondholders actually start buying more of them. You may had a belief that the original allocation was based upon
see something analogous in the stock market: When the some set of risk premium assumptions, then if the market
equity risk premium is low (signifying less payoff for risk), changes radically, wouldn’t you think that if those risk pre-
Mary Ida’s clients may want her to invest more in equities, mium assumptions change radically, there should be some
not less, because that’s the only way they have a possibil- corresponding shifts in the allocation? No, they typically
ity of meeting their goals. seem to be rebalanced back to the same allocation they
had a year before, two years before, three years before, four
Mary Ida Compton: What happens in reality, though, is that years before. One of the most amazing behavioral phenom-
when they realize they’re going down the tubes—instead of ena is that allocations are amazingly stable over time.
moving out on the risk axis to get potentially higher returns,
they stick all the money in something that’s very stable,
like core bonds. The problem with the risk assets is that Is It Time to Pray?
their volatility is high, and the client doesn’t want to take a
Elroy Dimson: There was another solution to that in 2008.
chance on being underwater three years earlier than they
I was, like many of you, invited to a number of conferences
would have been otherwise. They may assume that the
about what should we do as this crisis unfolded. When
pension fund will go under in five or 10 years and say to
things looked really bad and one of the fund managers
themselves, “We got a death sentence and we’re just going
asked the audience, “What should I do?” somebody piped
to hunker down and pay out what we can, and we know
up and said “Pray.”
it’s only going to last for five years.” They face a weird set
of incentives. Jeremy Siegel: I’d like to ask a very informal poll. How many
here think the next 10-year equity returns are going to be
The Psychology of Investing below the long-run average? I certainly do. Is there anyone
here who doesn’t? Or are you uncertain?
in Terrible Markets
Cliff Asness: I agree, they will be below. [Crowd nods
Elroy Dimson: Don’t these people need some personal agreement.]
advice, as well as advice about management of the pen-
sion fund? Jeremy Siegel: Okay, so everyone. Here’s the harder
question. How many here believe the equity risk premium—
Mary Ida Compton: Emotional advice? You mean psycho- the title of this decennial conference—is going to be lower
logical advice? than its historical value? Let’s say it’s 3.5% expressed on a
compound basis, or 4%?
Elroy Dimson: To work longer. And maybe at a slower pace.
Mary Ida Compton: That’s the historical level?
Mary Ida Compton: Well, the jobs may not be there.
Robert Arnott: On a 20-plus year basis, I would say no—it
Elroy Dimson: You have to get your mind around that.
will be higher.
Cutting your expenditure on holidays or lowering your cost
of living in some other way. You’ve got to adjust to it. Jeremy Siegel: Okay, 20 years, given what we’re facing in
bonds, with TIPS yields being −1.
Martin Leibowitz: The mentality is this: If you find yourself
in dire straits, you invest with some hope that the market Robert Arnott: I get it.
will somehow bail you out. You just continue doing what
you’re doing in the short run and postpone deciding to cut Jeremy Siegel: I’m just wondering: if TIPS go to −2, you got
back on expenses. a capital gain on TIPS.

So, a change of strategy is something that is not done Robert Arnott: Right, but if the Shiller P/E just goes halfway
casually. It’s done very reluctantly—either when you back to historical norms, that costs you about 4% per year.

90 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Jeremy Siegel: If, if. The Effects of Negative Real Yields,

Robert Arnott: I know, I know. Monetary Expansion, and the
Exploding Fed Balance Sheet
Bond Duration and Equity Duration
Robert Arnott: Jeremy, whether you’re looking at forecasts
Cliff Asness: I agree with Rob on that one, because one of nominal stock returns against nominal 10-year bonds or
thing I think is missing is the bond duration is far shorter real stock returns against real 10-year TIPS, you’re talking
than what you might call equity duration. The time path about the same thing. But you’re describing stocks as a
also matters with interest rates. If they shot up today and real return vehicle, which they are. I’ve always advocated
stayed there for 10 years… another interpretation of dura- comparing stocks with TIPS; that was the basis for my
tion is the breakpoint on expected return. You make more, paper in 2001, “Death of the Risk Premium.”17
not less, because they went up from the investment.
By the way, I’m glad the topic of bonds came up because
Jeremy Siegel: I disagree with you. The duration of TIPS is I’m reminded of Larry’s earlier comment about negative
less… I mean, a 30-year TIPS is less than that of the stocks, real yields perhaps actually doing damage. Knut Wicksell,
because the stocks give a dividend 1.5 to 2 real. The real is the Swedish economist, not as widely known as he should
negative on the TIPS. The duration is actually higher. be, postulated a natural rate of interest for bonds, and it
was not negative. Of course, back then, the gold standard
Cliff Asness: I was thinking, I said 10-year, Jeremy. was everywhere, so he was really talking about real yields.
Positive real yields serve as a speed bump to discourage
Jeremy Siegel: Well, 10-year I’d have to think about and
malinvestment, misallocation of resources, and propping
work the math. But go to the 30-year TIPS and it’s still
up of zombie businesses. If the speed bump is too high—if
like −1.
you have 5% real yields—then it stops all kinds of things,
Robert Arnott: But no, it’s not. including a lot of good ideas. If the speed bump is negative,
then you’re proactively encouraging malinvestment.
Jeremy Siegel: It’s minus what?
What’s interesting about malinvestment is that it can be in
Robert Arnott: −0.2, I think. the private or the public sector. So-called modern monetary
theory (MMT) is all about facilitating government spending
Jeremy Siegel: Okay, but… as if government spending is inherently a good thing. But
government spending often involves a great deal of waste.
Martin Leibowitz: What if you don’t?
So, I would argue that negative real yields will damage
Jeremy Siegel: There’s something called index risk when long-term growth, not help it.
you go 30 years out. But I’m not going to talk about that.
Roger Ibbotson: I certainly agree with that. Was it you, Rob,
Martin Leibowitz: Relative to real rates, the duration is very or was it Jeremy who had a slide about the money supply
different. affecting the Fed balance sheet?

Cliff Asness: But they’re … Jeremy Siegel: It was me.

Jeremy Siegel: I’m talking about real assets—we should be Roger Ibbotson: The Fed balance sheet has exploded to
talking about real rates. record levels. It really rose after the financial crisis, but then
it kept on going and now, with this COVID crisis, has truly
Martin Leibowitz: The duration of stocks with respect to exploded. At the same time, the deficit has exploded, too,
real rates is very long. to record levels relative to GDP. It’s higher than in World
War II. So, you have all that money flying into the economy—
Jeremy Siegel: Yeah, really, really long, but not with stocks this is the supply of money. Where is that money going
that are giving you large dividends. So that’s a positive 2% a to go? It could go into inflation, although not that much of
year, like a consol paying 2% a year.16 it has so far. It has to go somewhere.

16
A consol (short for “consolidated annuity”) is a bond that never matures but continues to pay the coupon rate “forever.” It can be sold to another
investor. The British government began to issue consols in the mid-1700s and paid their coupons for more than 250 years. The last ones were
redeemed (called) in 2014. Nothing is forever.
See Arnott and Ryan (2001).
17

CFA Institute Research Foundation 91

Revisiting the Equity Risk Premium

Jeremy Siegel: There is a lot of inflation, and there in the kind of inflation that Milton Friedman described. It’s
will be more. inflating the bond and stock markets.

Roger Ibbotson: I think it’s been going into the stock and Jeremy Siegel: It’s totally inflating consumer goods. I think
bond markets. we’re going to have 20% inflation in the next two or three
years. Not each year, but cumulatively.
Jeremy Siegel: Both.
Cliff Asness: Cumulative is not that bad.
Laurence Siegel: And real estate.
Jeremy Siegel: But at 7% a year?
Roger Ibbotson: I think that’s what’s inflating the markets.
It’s not inflating consumer goods as you typically would Cliff Asness: I’m joking!

92 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PART IV: EQUITY DURATION, NEGATIVE REAL RATES

FOREVER, BULLISH ON BEAR MARKETS?
Editor’s Introduction “Bonds are so unattractive that individual and institutional
portfolios have become massively overweighted in equi-
ties and much riskier than they used to be,” Laurence
You’re better off with all stocks if you have
Siegel says. “At the Ford Foundation, we got down to 9% in
a 50- or 100-year horizon. If you are 100% in
fixed income at one point. This allocation, if widespread,
stocks, you would be much better off if the stock
increases the amount of risk in society tremendously.”
market crashed. You [would] never buy another
penny’s worth. You just reinvest whatever cash That risk is often made manifest in bear markets. But the
you might get—the dividend flow. Your final real panelists stress that such down markets are not wholly
returns are going to be so much higher. negative. Depending on the investment horizon, they can
be critical opportunities.
—Jeremy Siegel
“If you’re spending in the near term, you really hate
How much chocolate would you give up today to ensure
bear markets,” Rob Arnott says. “If you’re saving now for
that you had a supply of chocolate tomorrow?
expenditures in the future, you really love bear markets.”
That might seem an odd question, but the panelists explore
Love or hate, the reaction tends to come down to age:
it as a way to better understand negative interest rates in
The old hate bear markets, whereas the young love them.
this section of their roundtable discussion. What the phe-
nomenon of negative rates comes down to is a bet that the “Unless you care a lot about your bequests,” Cliff Asness
future will be worse than the present and that locking in adds, “in which case we’re all essentially immortal.”
what you are going to receive tomorrow, even though it is
less than you have today, is rational.
Roundtable
“If times are good today and times are really bad tomorrow,”
Rajnish Mehra says, “then I would give up a lot of stuff
today to get something guaranteed tomorrow in a very, Back to Equity Duration
very bad state of the world where the marginal utility is
Antti Ilmanen: I want to revert to the duration question.
very high. [Under such conditions] I will get negative rates.
If you are talking about 30-year versus 10-year TIPS, that’s
But that’s not a world I want to live in.”
very easy. But equity duration is harder. We can debate
Before delving into that, the participants first consider whether the duration is 15 or 25 or some other number.
equity duration, shrinking bond allocations, and the appeal
Martin Leibowitz: Back in the old days, when actuaries
of bear markets, among other topics.
were actually computing equity duration, they came out
The loose monetary policies adopted by central banks since with numbers like 25 and 50. The way they did it was to just
the Global Financial Crisis have influenced equity duration. look at the dividend discount model and note that the flows
The excess cash that flowed into the economy needed a are back-ended. So, they came up with a very long duration,
place to go, and it flooded into stocks and other risk assets assuming essentially a deterministic model.
and raised their valuations. That seemed to work for a while
If you actually looked at the sensitivity of equity returns
to sustain economic growth.
to changes in interest rates, however, you find that dura-
“So, you will have higher returns in the near term,” Martin tions were about three to four years. The relationship
Leibowitz says. “But over the longer term, it’s going to be was extremely sloppy; it had a terrible R-squared. So,
disastrous. As a result, you definitely have a term premium, I think that to try to identify a clear-cut, short-term rela-
a term structure regarding the effects of very low interest tionship between movements of equities and short-term
rates. And, unless you’re riding the short-term wave, it’s not movements of bonds or bond yields is a fool’s errand.
a happy prospect.”
Antti Ilmanen: I think you have to use equities’ own dis-
As interest rates have fallen, investors have shrunk their count rate in this exercise.
allocations to fixed income. This, too, comes with potential
Roger Ibbotson: I mentioned the growth in the money
consequences.
supply. I don’t quite see this as a present value calculation.

CFA Institute Research Foundation 93

Revisiting the Equity Risk Premium

Yes, with negative interest rates, the present values would How Should Perpetual Institutions
soar, but I think it’s an overpricing actually, because the
money has to go somewhere, so it goes into the equity and Invest?
bond markets. I classify the high prices as overpricing and
Jeremy Siegel: Endowments and foundations are long-term
not just the present value calculation that is high due to a
money, so why is that wrong?
low discount rate.
Martin Leibowitz: That’s their theory. It’s not necessarily
Martin Leibowitz: Well, go back to what Rob Arnott was
wrong, but it leads to a level of risk that is tough to
talking about earlier in terms of malinvestment. If you have
stomach...
low rates and lots of money, you will invest in more things
and, as you point out, Roger, you will raise valuations Laurence Siegel: ...such as in the crash of 2008.
of equities and all risky assets. So, you will have higher
returns in the near term. Antti Ilmanen: How about 2008? Mary Ida has written
about it.
But over the longer term, it’s going to be disastrous. As a
result, you definitely have a term premium, a term structure Martin Leibowitz: There is an argument that if you had an
regarding the effects of very low interest rates. And, unless infinite horizon, you’d be 100% in stocks.
you’re riding the short-term wave, it’s not a happy prospect.
Laurence Siegel: Or 200% stocks.

Itsy-Bitsy, Teeny-Weeny Bond Mary Ida Compton: Even better.

Allocations Laurence Siegel: I was being sarcastic. We don’t have an
infinite horizon, because we have liabilities or payouts or
Laurence Siegel: I’m also concerned about the supply side
whatever, and then we die. Say no to leverage.
of the economy. Negative real rates in the 1970s were
called financial repression.18 They discouraged savings and Martin Leibowitz: ...if you had a long enough horizon,
impinged on growth. In some ways, financial repression and if you had enough resources to meet the margin calls.
has come back even worse than in the 1970s. Private,
profit-seeking actors are reluctant to buy the bonds, but
the governments need the money. So, the central banks The Beauty of Bear Markets
are buying them with newly created money. (If You’re a Buyer)
Bonds are so unattractive that individual and institutional Jeremy Siegel: You’re better off with all stocks if you have
portfolios have become massively overweighted in equi- a 50- or 100-year horizon. If you are 100% in stocks, you
ties and much riskier than they used to be. At the Ford would be much better off if the stock market crashed.
Foundation, we got down to 9% in fixed income at one You would never buy another penny’s worth. You just
point. This allocation, if widespread, increases the amount reinvest whatever cash you might get—the dividend flow.
of risk in society tremendously. Your final real returns are going to be so much higher.
Martin Leibowitz: Larry, do you know what the average Laurence Siegel: We all want to buy low, but that train
fixed-income holdings of endowment and foundation seems to have left the station.
portfolios are these days?
Jeremy Siegel: We want to anticipate the change in the
Laurence Siegel: I haven’t seen it lately, but 9% would be discount rate if there’s going to be one. That’s what you
my best guess. want to do.
Martin Leibowitz: It’s under 11%—even if you include cash Robert Arnott: Peter Bernstein and I wrote a paper in the
as fixed income. mid-1990s entitled “Bull Market, Bear Market, Should You
Really Care?”19 The point of the paper was: If you’re spend-
Laurence Siegel: I would call cash fixed income.
ing in the near term, you really hate bear markets. If you’re
Martin Leibowitz: The most important aspect of that is not saving now for expenditures in the future, you really love
what its return will be but the fact that it’s nonequity. bear markets. But very few people think that way.

See McKinnon (1973); Shaw (1973).

18

See Arnott and Bernstein (1997).

19

94 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Jeremy Siegel: That’s actually age-related. If you’re very Negative Real Interest Rates
young, you love bear markets. If you’re really old, you love
the bull market. Forever?
Cliff Asness: Unless you care a lot about your bequests, Then, the risk-free rate gets determined by the risk aversion
in which case we’re all essentially immortal. of individuals and how much they want to guarantee a cer-
tain amount of consumption in the future, even if achieving
Jeremy Siegel: That’s right. that guarantee means a negative real return. They can’t do
any better, there is no other, better investment to have; if
Cliff Asness: The conversation about two minutes ago, you want a certain return in the future, it could very well be
“What if you can just stick with this, no matter what?” negative because there’s nothing economically to stop it.
becomes a Saint Petersburg paradox. If you could keep the
same level of investment in a risk asset—or, in the case of Rajnish Mehra: The riskless rate could very well be neg-
the Saint Petersburg strategy, doubling down—you would ative; that is fine; it is not under dispute that it could be
eventually win. But we all know that that’s not how the negative. But is this an equilibrium phenomenon?
world works.
Jeremy Siegel: Yes, it could be negative forever and you can
Martin Leibowitz: Gambler’s ruin. get a very nice steady state as long as the rate of return on
risky capital is positive.
Cliff Asness: Gambler’s ruin will occur at some point.
Laurence Siegel: What’s the logic behind that?
Do Real Interest Rates Represent Jeremy Siegel: That the real return could be negative
the Marginal Productivity of Capital? forever?

Rajnish Mehra: What is your best explanation for the nega- Laurence Siegel: Yes.
tive real term structure? How will this end? How long will it
Jeremy Siegel: Because of the amount of risk aversion.
persist? Do you think this could be an equilibrium?
If I want to guarantee chocolates tomorrow, to get 1.9
Jeremy Siegel: Theory says the real term structure should chocolates tomorrow, I may have to give up two today.
be negative. You’re hedging against changes in the dis-
Laurence Siegel: Isn’t there a liquidity trap? Just buy the
count rate for your future consumption if you build these
two chocolates today and eat them tomorrow. Storage
intertemporal models. The real term structure appeared
costs cannot be that high.
to be positive before we had TIPS yields to measure it,
because our then-measure of the real term structure mixed Rajnish Mehra: But it’s a terrible economy to live in, Jeremy.
nominal risk and inflation risk. That biases you toward It assumes that the economy is getting worse and worse
an upwardly sloped real term structure. But many macro and worse.
models produce only inverted real yield curves. And some
of those models, given risk aversion and age and other fac- Jeremy Siegel: If someone guarantees me some quantity
tors, give you a negative long-term real return. of chocolate tomorrow, and if I know I will need it—I’ll take
1.9 chocolates in exchange for giving up two chocolates
Rajnish Mehra: But if I weren’t living in a world of certainty, today. That’s totally a long-run equilibrium.
I wouldn’t get real negative rates, would I?
Rajnish Mehra: The only time you’re going to get nega-
Jeremy Siegel: No, in a world of uncertainty, you would still tive real rates is when the marginal utility is increasing
get negative real rates. over time.
Rajnish Mehra: Just start with a simple world of certainty. Jeremy Siegel: No, I disagree. You’re going to have a nega-
Rates should reflect the marginal product of capital. tive real “sure” (riskless) rate forever. In a growing model, in
a steady-state model, in a perfectly fine long-term model.
Jeremy Siegel: The rate of return on risky capital...
Rajnish Mehra: Wouldn’t it be a declining economy?
Rajnish Mehra: Just hold on there for a second. Cut out
the risk; we’re talking about TIPS. Their yield would be the Jeremy Siegel: No, not at all.
marginal return on capital.
Laurence Siegel: This is a little above my pay grade, but at
Jeremy Siegel: No, it isn’t. It’s absolutely not. In my opin- some level, you only know what you learned in school. If I
ion, the return on physical capital is the unlevered equity defer my consumption voluntarily, I should get something
return, which is definitely positive. Unless you think there’s for it. Somebody else gets to use what I’m not using for that
a lot of mean reversion, as Rob does, it’s positive. amount of time, and I should be able to charge for that use.
CFA Institute Research Foundation 95
Revisiting the Equity Risk Premium

Jeremy Siegel: No. In an uncertain world, you have positive Roger Ibbotson: I agree with Jeremy because most of your
real rates only if there’s a storage technology that enables investments involve risk and have positive payoffs.
you with zero cost to have goods today stored so that you
have the same goods tomorrow. That is impossible, so you Jeremy Siegel: Absolutely.
absolutely have negative rates in equilibrium.
Laurence Siegel: You have to put so many weird conditions
Robert Arnott: Jeremy, I will never give up two bars of choc- on the economy and on behavior to get negative real rates
olate today for 1.9 tomorrow. Never. in equilibrium.

Rajnish Mehra: If there is a famine tomorrow—if times Roger Ibbotson: It’s just the risk-free rates—the real risk-
are good today and times are really bad tomorrow—then I free rates—that are negative. I don’t see any reason why
would give up a lot of stuff today to get something guaran- they couldn’t continue over indefinitely.
teed tomorrow in a very, very bad state of the world where
Jeremy Siegel: It can continue for 100 years.
the marginal utility is very high. Then I will get negative
rates. But that’s not a world I want to live in. Laurence Siegel: It’s an interesting question.

Laurence Siegel: That’s a declining economy. You can get Roger Ibbotson: But I don’t think it’s such an interesting
negative rates in equilibrium in a declining economy, but question that we have to talk about it indefinitely.
not in any other state. We had less storage technology in
the past than we do now; and we got positive real rates Laurence Siegel: It’s interesting to me because I just said
from thousands of years in the past until 13 years ago. the opposite and, if I’m wrong, I want to know why.
Which is the rule and which is the exception?

Jeremy Siegel: You can get a negative risk-free rate of inter-

est in a growing economy. I’m not saying that it will happen.
But it can be an equilibrium in a growing economy.

96 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

PART V: WHAT IS THE ERP? NEGATIVE RETURNS

AS A HEDGE, MODERN MONETARY THEORY
Editor’s Introduction is largely negative, with Rob Arnott pointing to data sug-
gesting that, far from the redistributive effect envisioned
by MMT proponents, MMT policy does quite the opposite
There’s one aspect of MMT that I have some sym-
in practice.
pathy for—the notion that what we spend money
on is far more important than how we finance it. “The implication is that, if you pursue MMT, you’re going to
be enriching the people that you’re ostensibly looking to
—Cliff Asness
‘milk’ with the intent of enriching the poor and the working
In the final part of their roundtable discussion, panelists class,” Arnott says.
return to the nature of the equity risk premium and home
For his part, Cliff Asness gives a nuanced, contrarian take.
in on what it is and what it isn’t. Jeremy Siegel sums up his
view: “It should always be the difference between what- “The one good point in MMT, which they don’t stress
ever the real riskless return is, positive or negative, and the enough, is this,” Asness contends. “If the government did
return on risky equity.” much less and charged zero tax rates, so that there was a
big deficit, the libertarian in me would think that’s a good
From there, they consider the utility of negative nominal
world. And if the government spent a ton of money and
yields and why the correlation between stocks and bonds
fully financed it with taxes, I might think that’s a bad world.
has flipped from positive to negative over the last sev-
I think MMT does make that distinction.”
eral decades and flipped back to positive recently. On the
former, negative nominal yields on fixed-income assets Before wrapping up, the panelists revisit their forecasts in
do serve a purpose: If the investor wants to consume later the 2001 and 2011 Equity Risk Premium Forums of what
or hedge against a future economic collapse, the fixed the equity risk premium would be, and once again consider
income is still a good deal. These rationales should not, why it is so large and whether it really is a risk premium.
however, distract from the sea change that the emergence
of negative nominal yields represents. It is instructive to compare all these forecasts to the
actual results after the fact. The speakers calculated a
“Over 3,000 years of history,” Laurence Siegel observes, realized equity risk premium relative to a 10-year Treasury
“nominal yields have always been positive until the last 12 bond of 11.73% for the 10 years from 30 September 2011
or 13 years. Has the hedge property overtaken the invest- to 30 September 2021, and 2.88% for the 20 years from
ment property of fixed-income assets, suddenly, for the 30 September 2001 to 30 September 2021.
first time?”
So, what accounts for the size of the premium?
The consensus among participants is that the hedging
component has indeed eclipsed the investment property “One possibility would be that stocks are perceived as
as the primary reason for holding bonds. being much riskier than they are,” Roger Ibbotson says.

As for changes in the correlation between stocks “It could be the Tversky–Kahneman loss aversion explana-
and bonds, Antti Ilmanen attributes it to two kinds of tion,” Jeremy Siegel observes. “It is a behavioral explanation
uncertainty. for why there’s such a high risk premium. People react
asymmetrically to losses versus gains.”
“Stocks and bonds tend to be driven by growth and infla-
tion,” he observes. “When there is more growth uncertainty, “My theory is that we’re all listening to bad news and con-
stocks and bonds tend to move in opposite directions, stantly bombarded with anxieties about the world coming
so we’ve had negative stock/bond correlation for the last to an end,” Will Goetzmann remarks. “We know that those
20 years. Before that, there was, relatively speaking, more emotions make people really worried about stock market
inflation uncertainty and we tended to have positive stock/ crashes.”
bond correlations.”
As to which explanations best fit the data, panelists will
The panelists also address modern monetary theory (MMT), likely revisit that question at the fourth Equity Risk Premium
which “seems to have taken over the government and the Roundtable in 2031. So stay tuned.
Fed,” according to Roger Ibbotson. Their collective take

CFA Institute Research Foundation 97

Revisiting the Equity Risk Premium

Roundtable going to be the sum of three terms. The first term will be
the time preference, the rate at which we prefer to con-
sume today rather than tomorrow. That’s about 1% per year.
Back to the Equity Risk Premium
The next term is the growth rate of consumption times the
Jeremy Siegel: Well, Roger, what is meant by the equity risk inverse of the elasticity of intertemporal substitution. In a
premium? I don’t think it matters whether the reference growing economy, the consumption growth rate is posi-
asset is long- or short-term bonds. tive (historically about 2%). The elasticity of intertemporal
substitution is about a half or a third or something in that
Martin Leibowitz: If bond returns are prospectively ballpark, implying a coefficient of relative risk aversion
negative, shouldn’t the risk premium be measured against about 2 or 3.
positive returns?
The third term is –0.5γ2σ2, where γ (gamma) is the coeffi-
Jeremy Siegel: No. It should always be the difference cient of risk aversion and, σ2 the variance of the growth
between whatever the real riskless return is, positive or rate of consumption (about 0.00123). Unless you become
negative, and the return on risky equity. Always. extremely risk-averse with a risk aversion parameter of
45 or 50, this third term will be negligible, and the first
Martin Leibowitz: If someone is investing and they want to
two terms will dominate—so, normally, the risk-free rate
get a positive return, bonds would not be a consideration.
increases as your risk aversion goes up. It will start declin-
Jeremy Siegel: Yes, they would. It’s their hedge. What do ing only if you become extremely risk-averse,21 resulting in
you mean—just because the return is negative, it doesn’t a negative real return even when the growth rate of con-
do anything? sumption is positive.

Martin Leibowitz: Negative returns are not an exciting This is Fischer Black’s solution to the equity premium
hedge. puzzle, by the way. His solution, in private conversation,
was that you have a risk aversion of 45. In such a case, you
Jeremy Siegel: They’re not exciting, but they’re absolutely a can solve everything. Why? Because the risk-free rate will
hedge. A lot of hedges have a negative expected return. become very small and may become negative.

Roger Ibbotson: If you want to consume later instead of Roger Ibbotson: You have a preference to consume later
earlier, because we are planning for some future thing, instead of now.
you’ll get a negative real interest rate.
Rajnish Mehra: You can just use constant relative risk
Robert Arnott: This whole discussion hinges on whether aversion. That’s not going to change. I could cook up an
there is a zero-return alternative to the negative-return example, but that will be inconsistent with everything you
risk-free asset. know—the risk aversion will come out to be so high that
you would not get out of your bed every day.
Jeremy Siegel: There is not. If there were a storage
technology, there would be.
Nominal Fixed Income as a Hedge
Robert Arnott: Stuff it under your mattress. The return on
that will be zero in nominal terms. But a lot of governments
or Insurance
around the world are trying to replace currency with some- Jeremy Siegel: There’s another reason why you might have
thing else. negative equilibrium real rates. That is government reaction.
If things collapse and prices go down as in a great depres-
Jeremy Siegel: Paul Samuelson wrote that famous article
sion, nominal assets are the best assets to hold. They
about money having a zero nominal return. Remember?
become a negative-beta asset. That’s why I talked about
Long-term equilibrium with and without social contrivance
the negative correlation between bonds and risky assets
of money, the forced equilibrium.20 But, the truth is, as
that will prevail if things go bad. That would cause people
you’re saying, Rob, money gives you a zero nominal return
to hold more bonds. How much they hold has to do with
in an inflationary environment. It is a negative real return, so
the perception of whether those nominal assets are in fact
you have no zero real return alternative.
effective risk hedges or not.
Rajnish Mehra: Jeremy, let me just continue one second
Laurence Siegel: They become an insurance asset.
more and then we’re done with it. The real rate of return is

See Samuelson (1958).

20

See Equation 18 and the discussion following it in chapter 1 of Mehra (2008).

21

98 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Jeremy Siegel: Yes. An insurance asset, as you know, will The Flip from Positive to Negative
very often give you a negative return. When nominal assets
are perceived as an insurance asset, which has happened Stock-Bond Correlation
at various times in history, one could ask why—maybe the
Antti Ilmanen: I don’t think that issue is directly related
concern is default by the government, money not being
to the equity premium. There are other things that I can
redeemed in gold properly.
talk about.
When everything is priced in money, and the concern is
I want to say something quickly on the stock/bond
about another financial crisis or a pandemic crisis or what-
correlation. We have a nice story on why the sign flipped
ever, prices of goods and services and real assets decline,
from positive to negative 20 years ago. Stocks and bonds
and bonds do extremely well. Nominal fixed assets do
tend to be driven by growth and inflation. When there is
extremely well. They take on a really negative beta, which
more growth uncertainty, stocks and bonds tend to move
I think gives them a tremendous hedging ability. I think
in opposite directions, so we’ve had negative stock/bond
trillions of dollars’ worth of demand are generated to hold
correlation for the last 20 years. Before that, there was, rel-
that asset.
atively speaking, more inflation uncertainty and we tended
Laurence Siegel: Some form of money or bonds has always to have positive stock/bond correlations. So, we are waiting
had that hedge property. Yet over 3,000 years of history— to see if those relative uncertainties flip again.
as you and Sidney Homer showed, Marty—nominal yields
Laurence Siegel: The stock-bond correlation was nega-
have always been positive until the last 12 or 13 years. Has
tive from the mid-1950s to the mid-1960s. I think there
the hedge property overtaken the investment property of
was growth uncertainty then, but relatively little inflation
fixed-income assets, suddenly, for the first time?
uncertainty.22 That supports your story, Antti.
Jeremy Siegel: Yes.
Jeremy Siegel: I think you’re right. The correlation flip is also
Laurence Siegel: Why? related to the fact that, when you have supply shocks, you
will have a positive correlation between stock and bond
Antti Ilmanen: It changed 20 years ago. Before that, there returns. By the way, I’m not talking about the constrained
was rarely a negative correlation between stock and bond supply situation that is happening right now; that is very
returns. specific to current news. I mean oil shocks and other more
typical shocks—you’re going to have that positive correla-
Jeremy Siegel: Let me tell you an interesting story. A lot tion. The reason is that supply-shock inflation is bad for the
of people analyze the VIX equity volatility index. I was economy, so stocks and bonds go down together. You get a
confused about why there was so much demand for VIX positive beta on long bonds.
assets, and then someone told me, “We love VIX assets
because they’re negatively correlated with the stock If the stocks are more demand-related, caused by financial
market.” And I said, “Yes, but do you know that, if you hold crises or pandemics or something else like that, then you
them, they’re going to deteriorate by 5% to 10% a year every tend to get a more negative correlation. The difference, as
single year, all the time?” They didn’t really understand that. I mentioned earlier, is enormous. Go through the math and
see what that does to real yields. It depresses them tre-
So, I gave a lecture about government bonds being mendously. So, I agree with you—the correlation changed,
negative beta assets. One money manager came to me and I think it had to do with supply shocks versus demand
and said, “Jeremy, I had $3 billion in VIX products for the shocks in a macro system.
negative correlation. Why don’t I try to get a positive nomi-
nal return, even if it’s only 1%, by holding long-term nominal Martin Leibowitz: Rob, does this observation relate to the
US government bonds instead?” And he did that. He said, P/E smile that we’ve talked about so much in the past?
correctly, “Forget about those VIX assets. Bonds are so
much better, even though they give negative returns.” Robert Arnott: I think it does, but spell out to me with what
you mean by the question.
Cliff Asness: Jeremy, I very much agree with you, but we
should acknowledge that not everyone on earth agrees Martin Leibowitz: As real rates go up beyond a certain
that long-volatility assets have a negative expected return. point, P/Es start to come down as the high real rates
Antti Ilmanen has gone quite a few rounds with Nassim become a constraint on growth, first naturally and then
Taleb on this very issue. Fed-induced. As real rates go lower, you find yourself in a
situation where, beyond that tipping point, the prospects

See, for example, Ilmanen (2003).

22

CFA Institute Research Foundation 99

Revisiting the Equity Risk Premium

for equity growth or economic growth are sufficiently dour Robert Arnott: We (at Research Affiliates) have a draft
that the correlation goes in the other direction. paper that Chris Brightman wrote a year ago, and he hasn’t
published it because he was worried about upsetting cli-
Robert Arnott: I think that’s exactly right; Exhibit 69 ties ents in the middle of the COVID pandemic. The paper shows
into that. While you described it as a smile, it’s more a direct link between deficits and corporate profits. That
of a frown. is to say, a trillion dollars of deficit spending goes hand in
hand with a trillion dollars of incremental corporate profits
Martin Leibowitz: Yes, it is a frown.
over the next four years. This relationship has a theoretical
Robert Arnott: The peak multiples are found at moderate basis that would take too long to get into right now. The
levels of inflation—1% to 2%—and moderate real rates, implication is that, if you pursue MMT, you’re going to be
2%, 3%, maybe even 4%. The multiples fall off pretty sharply enriching the people that you’re ostensibly looking to “milk”
from there. So, a lot of this variability in multiples hinges with the intent of enriching the poor and the working class.
on central bank policy. And, in an MMT world, I’m not sure
Laurence Siegel: I think most of us knew that. We just
the central bankers are likely to be pursuing policies of
couldn’t prove it. I’d love to read Chris’s paper.
anything other than moderate to high inflation and negative
real rates. Cliff Asness: That’s the verdict on quantitative easing for
10 years now. There’s one aspect of MMT that I have some
Modern Monetary Theory sympathy for—the notion that what we spend money on is
far more important than how we finance it.
Roger Ibbotson: Does anybody here have a positive opinion
about MMT? It seems to have taken over the government Robert Arnott: Yes.
and the Fed. Does anybody think there’s something positive
Laurence Siegel: Yes.
to that?

Exhibit 69. Does MMT Pose a Threat to the ERP? Only If Fed Has No Exit Strategy

Source: Research Affiliates; RAFI Indices.

100 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Cliff Asness: The one good point in MMT, which they don’t Jeremy Siegel: And, by the way, I would say that bonds did
stress enough, is this: If the government did much less and much better than everyone predicted.
charged zero tax rates, so that there was a big deficit, the
libertarian in me would think that’s a good world. And if the Roger Ibbotson: Definitely.
government spent a ton of money and fully financed it with
Jeremy Siegel: Stocks and bonds both exceeded expecta-
taxes, I might think that’s a bad world. I think MMT does
tions over the last 10 years.
make that distinction. I just then make every policy choice
opposite from them. Martin Leibowitz: My recollection—I could be wrong, and
you’ll correct me on this, Larry—was that the numbers
Robert Arnott: The level of taxation is not the taxes we pay.
ranged from a 0% risk premium up to around 6%, with
It’s the money that we spend. Because whatever is spent
an average of 3.5% to 4%. It’s very interesting how those
is either coming out of tax revenues or pulled out of the
forecasts correlate with a lot of the numbers we’ve been
capital markets through running deficits and increasing the
bouncing around today, with very different types of expla-
debt. The money is being pulled out of the private sector
nations for how we got there.
in both cases. So, spending sets the true tax rate and is
what’s disturbing about a $3 trillion to $5 trillion deficit. Laurence Siegel: Marty, those were the forecasts in the
2001 forum, the first one. In the 2011 forum, the estimates
Elroy Dimson and Robert Arnott thanked the group and
were all very close to 4%.
said goodbye.
Looking at the 2001 (20 years ago) forecasts, the lowest
Martin Leibowitz: Why don’t we plan on getting together in
was Rob’s and it was zero. But these were not 20-year fore-
100 years and see how good our projections were?
casts; they were 10-year forecasts. The highest forecast
Jeremy Siegel: I think we should get together 10 years was that of Ivo Welch, but the highest forecast from among
from now. those present today was Roger’s. Congratulations, Roger.

Elroy Dimson: I’ll put it in my diary. Roger Ibbotson: Whoever was highest won. There was
nothing especially prescient about my forecast. Also, we
Laurence Siegel: We are going to do it in 10 years. should repeat that these were 10-year forecasts made
20 years ago. Apparently, Larry doesn’t have the 2011
Mary Ida Compton: Good! forecasts handy.

Laurence Siegel: 100 years is a good idea too. Laurence Siegel: No, I don’t. I’m sorry.

Jeremy Siegel: Was mine 4.5% or 5%? I forget.

Reflecting on Our Past Forecasts
Laurence Siegel: Jeremy, yours was 3% to 4%.
Rajnish Mehra: Larry, after the last forum in 2011, you sent
an e-mail with everybody’s forecast for the equity premium. Martin Leibowitz: What was Roger’s?

Laurence Siegel: It was an e-mail with all the forecasts Laurence Siegel: 5%.
from 2001, so we could compare our then-current (2011)
forecasts to the old ones (2001). I don’t have a record of Martin Leibowitz: That was the highest?
the forecasts from 2011. But I do remember that Brett
Hammond gave a talk at the Q Group in 2011, where he said Laurence Siegel: Ivo Welch gave 6% to 7%.
that all the 2011 forecasts were very close to 4%.
Martin Leibowitz: Okay.
Roger Ibbotson: I missed the last forum because of
Antti Ilmanen: Did we specify what maturity bond?
a snowstorm, but I think markets exceeded almost
everybody’s expectations. Laurence Siegel: A 10-year bond.

Laurence Siegel: It sure did. Jeremy Siegel: What is the right answer?

Roger Ibbotson: So, it doesn’t matter what we said. Mary Ida Compton: Do you mean, what actually happened?
Whatever the forecasts were, the market did better.
Jeremy Siegel: What was the last 10 years’ realized equity
Laurence Siegel: That’s right. risk premium, and what was the last 20 years’ realized
premium?
Roger Ibbotson: The person who had the highest
estimate won. Mary Ida Compton: I have the 10-year numbers here.
For the 10 years ended September 2021, the S&P 500

CFA Institute Research Foundation 101

Revisiting the Equity Risk Premium

returned 16.63%, compounded annually. Long Treasuries Will Goetzmann: My theory is that we’re all listening to bad
returned 4.39%. news and constantly bombarded with anxieties about the
world coming to an end. We know that those emotions
Laurence Siegel: So the realized 10-year equity risk pre- make people really worried about stock market crashes.
mium from 30 September 2011 to 30 September 2021 was
1.1663 There’s plenty of evidence of that. In a paper I’m working
− 1 = 11.73%.
1.0439 on with Bob Shiller, we look at earthquakes in the region
Over the 20 years from 30 September 2001 to where people are making their market forecasts. They get
1.0951 more pessimistic, and they think there’s going to be a crash
30 September 2021, it was − 1 = 2.88%.
1.0644 when they find out that there has been a local earthquake.
That is a pretty thin margin over bonds, and the highest So, I think that this issue is behavioral and not necessarily
forecaster wouldn’t have won, but we didn’t ask for 20-year easily modeled.
forecasts in 2001—so there is no winner, and no loser.
Jeremy Siegel: But you’re also saying that we’ve been
Roger Ibbotson: So I guess I didn’t win. heavily bombarded with bad news for 150 years?

Laurence Siegel: Actually, Roger, you did win because Ivo Will Goetzmann: I think the most recent period is the most
Welch isn’t here. For 2001–2011, you had the highest fore- extreme example. People have been talking down the
cast of the people who are here, and the actual return was market for the last decade, and the market has been doing
much higher than the highest forecast. pretty well.

Cliff Asness: My forecast for the next time is 1 basis point Mary Ida Compton: People love that kind of stuff; they cling
above the highest forecast. to it. It’s on the media, it’s on social media, it’s in the news-
papers. Remember the Y2K problem? Was that crazy or
Everyone: [Laughter] what? I know people who liquidated their equity portfolios
because they were afraid of the Y2K problem.
Afterthoughts: Good News Jeremy Siegel: You’re talking about being bombarded
and Bad News over the last 10 years with negativity. You’re writing a
paper with Bob Shiller, whose CAPE ratio is exactly the
Roger Ibbotson: Before we close, I want to address reason why people have been bombarded with negative
Rajnish’s comment about the premium for equities not news. The CAPE ratio was on the cover of the Economist
being a risk premium. I’m trying to think of what the pre- magazine twice.
miums could be for. One possibility would be that stocks
are perceived as being much riskier than they are. Is that a Will Goetzmann: One time I was in a bus for one of these
possibility? National Bureau of Economic Research conferences on
behavioral finance, and Bob Shiller and Dick Thaler were
Laurence Siegel: Yes, that’s a possibility. both on the bus. One of them was saying, “I’m 100% in
stocks.” And the other one says, “I’m 100% out.”
Roger Ibbotson: Or there’s a really extreme tail risk that
people price in? And they both had great theories supporting their decision,
right? So, what am I supposed to do?
Jeremy Siegel: It could be the Tversky–Kahneman loss
aversion explanation. It is a behavioral explanation for why Laurence Siegel: And they both have Nobel Prizes, so they
there’s such a high risk premium. People react asymmetri- both must be right.
cally to losses versus gains.
I want to thank our 11 extremely distinguished speakers,
Mary Ida Compton: True. plus everyone else who helped organize this forum and
made it happen.

102 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

APPENDIX A: 2023 UPDATES

Because of the rapidly changing financial environment in the next decade. These are the main drivers behind
since the time of the forum, we asked the 11 participants our equity expected return being well above what it was
to provide updated estimates of the equity risk premium. last year.
Here is the request:
Far too many investors ignore the role of starting valuations
Given that many changes have taken place since in determining their capital market expectations for the
this forum was held in late 2021, what is your next decade. When valuation multiples have soared, past
best estimate of the ERP now? (Single-point returns look fabulous, but forward returns are impaired.
best estimate, geometric mean, 10-year horizon, Our central expectation of a fair value CAPE multiple for the
US equity total return in excess of the 10-year S&P 500 in 10 years is 23.4, while our expectation for the
Treasury yield, which was 1.82% on 3 February yield of a 10-year bond in 10 years is 3.1% (0.6% above our
2022 and 3.54% today.) expected inflation rate in 2033), both well below current
levels. Mean reversion in valuation multiples will likely erode
We also asked the participants to elaborate on their stock market returns (albeit far less than a year ago), while
answers if they felt like it—not required. the bond yield expectation has almost no impact on the
bond market return.
Here, we present their answers in the order in which they
spoke in the forum. Many thanks to the participants for their Exhibit A1 shows the changes in our expectations broken
prompt and revealing responses. out by component.

Roger Ibbotson
As you know, I am mostly a very long run forecaster.
Exhibit A1. Change in Expectations
But given the rise in rates, I am lowering my ERP to 5%.
S&P 500 10-Year Return Feb 2022 Apr 2023

Elroy Dimson Dividend yield 1.4% 1.7%

Real dividend growth 1.2% 1.4%

For strategic asset allocation, we learn little—and nothing
statistically significant—from recent annual performance. Inflation 2.6% 3.2%
So, my estimates have not changed. I remain at around
Mean reversion –3.2% –2.2%
a 3.5% premium relative to Treasury bills. You asked for a
premium relative to bonds, so an ERP of 2.5% relative to Nominal 10-year return
10-year Treasuries is still in the right ballpark.
No mean reversion (IRR) 5.3% 6.4%

Cliff Asness 50% mean reversion 2.0% 4.1%

Real 10-year S&P return

Please reduce my estimate by 1%.
No mean reversion (IRR) 2.7% 3.1%
[It was 4% on February 3, 2022, so his current estimate is
3%. —Ed.] 50% mean reversion −0.6% 0.9%

10-Year Treasury return

Rob Arnott Bond yield 1.8% 3.5%

My answer 14 months ago is shown following my current Yield roll-down 0.1% –0.2%
answer. Since that time, we’ve seen an unprecedented
Nominal 10-year return 1.9% 3.3%
bear market in bonds and a reasonably ordinary equity
bear market. Bear markets boost prospective returns. Real 10-year bond return −0.7% 0.1%
Specifically, the equity dividend yield has risen by 0.4%
(1.7% versus 1.3%) over the period, while the starting CAPE ERP with no mean reversion 3.4% 3.1%
ratio has fallen 9 points (28 versus 37). Meanwhile, our ERP with mean reversion 0.1% 0.7%
10-year expectations for inflation remain elevated, consis-
tent with our expectations for “higher for longer” inflation

CFA Institute Research Foundation 103

Revisiting the Equity Risk Premium

In valuing a 10-year bond over a 10-year horizon, we look to would cost us about 3.2% per annum because of
model a “close to constant maturity” bond by assuming a falling valuation multiples. That takes us to 2.0% per
string of 1-year holding periods of 10-year bonds. This adds annum as a 10-year return expectation, or a 0.1% ERP.
a roll return over the next year. Two major changes alter our So, with mean reversion halfway to historical norms, the
bond return expectations: (1) With an inverted yield curve, ERP is 0.1%. If this world of negative real yields and nose-
the roll yield swings negative; and (2) with inflation expec- bleed valuation levels is a long-term sustainable “new
tations up materially, the real return on bonds has more normal,” I’d expect a 3.4% ERP. But I think our aging demo-
of a headwind from inflation, which is a tailwind for the graphic increases the likelihood of the former. So, let’s go
nominal return for stocks. with that number, 0.1%.
One might expect, then, that because bond yields have
moved much more than stock market earnings yields,
our ERP is down. If we ignore prospective mean reversion
Marty Leibowitz
toward historical norms, that is indeed the case; on an I don’t have a fresh new ERP estimate, but my current
internal rate of return basis, the ERP has softened slightly thinking is that the risk premium needs to include a term
from 3.4% to 3.1%. Adjusting for mean reversion, however, that represents some component of the expected growth
we find the opposite: Because mean reversion won’t erode over the relevant horizon.
equity returns as much as was the case at recent peaks
in CAPE ratios (and because mean reversion doesn’t much This growth term should ideally estimate the real growth
affect 10-year bond returns), the ERP, with mean reversion after deducting the capital cost of such growth. (See my
toward historical valuation norms, has improved from article with Stan Kogelman and Anthony Bova, “P/E Ratios,
0.1% to 0.7%. Risk Premiums, and the g* Adjustment,” Journal of Portfolio
Management, April 2019.)
[I later asked Rob to modify his answer, if needed, to reflect
the fact that the other participants, responding to a later To neglect this admittedly hard-to-quantify term is analo-
version of my request, compared the expected equity gous to confounding a discount bond’s current yield with
return to the 10-year Treasury bond yield—that is, the its yield to maturity.
expected return on a 10-year bond bought today and held
Given this point, with the standard risk premium (based on
to maturity—not the expected return on a bond portfolio
the earnings yield) appearing now to be quite modest, and
managed to have a roughly constant 10-year maturity,
with the current growth outlook being arguably more muted
which could be different. He responded, “With an inverted
than usual, the combination of these two terms suggests
yield curve, the difference in expected returns between
that the today’s risk premium is actually rather low on a
a buy-and-hold 10-year Treasury and a rolling 10-year
historical basis.
Treasury portfolio is not material, not more than about
0.2%.” —Ed.] Of course, the more fundamental question is whether this
risk premium is sufficient to justify the risk inherent in
My 2022 Response: equities.

• 10-year T-Bond: 1.82% yield. Steady state, that’s the

return. Mary Ida Compton
Because the duration is less than 10 years, rising On average, I think the ERP is 5%. Thinking about the
yields would boost (not lower) this return. If inflation markets between now and 10 years from now, I expect
is 2.6% over the coming decade, and the real yield the US equity market to have outperformed today’s 10-year
migrates toward its historical norm of 1.6%, then bond yield by about 3% per year.
income will be reinvested at steadily higher rates,
pushing the return up by about 0.1% (yep, it moves I believe in mean reversion, which contributes to my
the needle that much) to around 1.9%. expectation of a 5% ERP on average over long periods.
I think the ERP is tied to the economic cycle, which
• S&P 500, steady state return of 5.2%. Yield is 1.4%, our
includes as subcomponents the growth cycle and the debt
expected inflation is 2.6%, and real growth in dividend
cycle. The inflation trajectory presents a layer on top of the
income (and earnings) is 1.2%. So, without any mean
cycles to influence the ERP for the coming 10 years.
reversion toward historical valuation norms, the equity
return is 5.2% and the ERP is 3.4%. My estimate is primarily driven by the increase in rates
Returning from a current Shiller P/E of 36.9 to a still- for two reasons. One is that the discount rate is higher,
expensive level of 27 (halfway to the long-term which affects the attractiveness of equities. The second
historical median; for what it’s worth, I agree that the is that the market can choose to get less volatile decent
long-term median is way too low in today’s world) (better than we’ve had in recent memory) returns from

104 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

fixed income for the first time in ages, so I expect higher Our prediction game focuses on the S&P 500, as did most
allocations to that asset class. This will decrease the of the ERP Forum discussion. Let’s look a bit more broadly.
demand for equities. I also believe there is a recession I’d consider the S&P 500 to still be mildly on the rich side,
coming, and I’m not sure of its magnitude. Although the whether compared to its own history or compared to
jury is still out on the impact that will have on equities for 10-year TIPS or nominal Treasuries. Some pockets of US
a decade (you could argue they will go up because Jerome equity markets are clearly cheaper (have higher starting
Powell will be happy and cut rates, or you could argue they yields and likely higher prospective returns), given the
will go down because of decreased demand), I think the wide valuation spread between so-called value stocks
recession will be a headwind. and growth stocks. Likewise, many equity markets outside
the US are cheaper and seem to have higher prospec-
I consider a different lens that reinforces my opinion. tive returns for the coming decade, especially emerging
I believe we have recently peaked on the growth rate, in part markets.
because debt is no longer nearly free, so we’re headed for
slower growth, detracting from the 5% average. At the same Finally, describing the current environment as one of com-
time, equities seem to be pricing in fairly strong growth. This pressed risk premia seems most apt for private equity
disconnect also detracts from the 5% average. So—bottom and many other private illiquid asset classes. These are
line—3%. Thanks for the opportunity to contribute. essentially long-duration assets, yet their valuations have
not responded much to the 2022 rise in the riskless part of

Antti Ilmanen their discount rate from –1% to +1% (or, stated differently,
to a less benign funding environment). Even before 2022,
increasing inflows into private equity had brought valua-
My updated ERP estimate is 3%.
tions of this asset class close to those in public markets,
The most important reason for reducing my estimate by while fees remained high, thus suggesting that one should
0.5% is that Treasury yields rose by 1.7% while equity not count on an illiquidity premium, net of fees, in private
yields rose at best half of that (e.g., the simple cyclically markets.
adjusted earnings yield, inverse of the CAPE ratio, rose from
roughly 2.75% to 3.5%). Broadly speaking, we have shifted
from a world where most long-only assets were expensive
Tom Philips
because of their low common real riskless discount rate— My estimate of the ERP has declined.
and any premia beyond riskless rates were ordinary—to a
world where the real riskless discount rate has normalized My current estimate of the nominal return of the S&P 500
to a positive level. For example, the 10-year TIPS yield has for the next decade is 3.62% per annum, and so my esti-
risen by two percentage points, from about −1% to a bit mate of the equity premium relative to the 10-year Treasury
more than 1%. Non-bond assets, in contrast, now have yield of 3.54% is 0.08%. I see that you rounded our earlier
compressed risk premia. estimates to the nearest 0.1%, which makes my current
estimate of the equity premium for the next decade 0.1%.
Based on those simple yield changes, I could have
reduced my estimate to 2.5%, but I round up my estimate The change in my estimate is driven by the following
to 3% for two reasons: (1) Our broader capital market esti- factors:
mates, which use somewhat fancier inputs, give a 4%
expected real return for US equities and 1% for Treasuries. 1. the rise in interest rates from 1.82% to 3.54%,
(2) Although my estimate does not reflect mean-reverting increasing the expected return of bonds,
valuations, there clearly was more potential for that to be 2. the decline in the S&P from 4501 to 4130, contributing
a negative consideration when the CAPE ratio was in the to an increase in the expected return of stocks, and
high 30s than today when it is in the high 20s. As an aside,
3. the decline in the profits of the S&P 500 over the past
I suspect that the lowest estimates in this survey from
year (from $197.87 in 2021 to $172.75 in 2022), con-
early 2022 will now be revised upward the most if they
tributing to a decrease in the expected return of stocks.
were predicated on mean-reverting valuations.
The combination of the second and third factors has raised
Clearly, inflation uncertainty is elevated despite the recent
my expected return for stocks from 2.6% in early 2022 to
decline in headline inflation rate. Changes in the inflation
3.62% today, but the expected return of bonds has risen
environment would influence my estimate more, but the
even faster.
impact of higher inflation in the next decade is unclear—it
could boost nominal equity returns a little but more likely Although my estimate is a point estimate using cur-
would reduce real returns. So the net impact on the equity rently available data using the same methodology that
premium is unclear, even if I were to get the inflation I used in February 2022 (see below), I think there is a
forecast right.

CFA Institute Research Foundation 105

Revisiting the Equity Risk Premium

better-than-even chance that the realized equity premium I expect the Fed to gradually let rates revert back to their
will actually be negative, for two reasons: historical norm.

1. I expect corporate profit margins to continue their

decline from their 70-year high on account of an Jeremy Siegel
increase in protectionism, wages, and government
spending that will likely be funded by an increase in Here’s my view on the subject. The current P/E of the
taxes, and market is just short of 20, and about 20 is what I believe
is equilibrium. Yes, it is higher than the historical P/E, but
2. I expect the prospective equity premium to increase because of lower real rates, more-liquid markets, and the
over the decade, further depressing the realized ease of total diversification, I believe the equilibrium P/E
premium over that time span. has risen over time. This would be my base case—no mean
reversion here. (I have even convinced super-bear Jeremy
Note: My February 2022 number was based on the formula
Grantham of that, although he may not agree with my
1/(Residual variance weighted average of two expected
number!)
return estimates)—one estimate based on filtered earnings,
the other based on revenues. A P/E of 20 translates into a 5% earnings yield and hence a
5% expected long-run real return on equities.
Rajnish Mehra For bonds, inflation-compensated bonds (TIPS) are yielding
just over 1%.
The biggest challenge in making a forecast at a 10-year
horizon is Fed policy, which has distorted markets This makes the equity risk premium, computed in com-
since the global financial crisis. Historically, MV/GDP pound returns, between 3.5% and 4%—very healthy in
(that is, the market capitalization of US equities divided my opinion!
by same-year US GDP) was a stationary time series that
was mean-reverting. Unfortunately, it is no longer so.
(The MV/GDP time series no longer passes the test for Will Goetzmann
stationarity; at a later date, I may revisit the issue with
data going back to 1929.) I’m sticking to 5% based on the very long-term data.

In light of this, using ratio analysis (P/E, MV/GDP, etc.) and [Recall that Professor Goetzmann presented data on the
predictive regressions to estimate the equity premium is world’s oldest public traded company, Honor del Bazacle
unlikely to be very informative. At a 10-year horizon, my (also known as Société des Moulins de Bazacle), going
estimate is revised down to 4.5%–5%, largely because back 574 years. —Ed.]

106 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

REFERENCES
AIMR. 2002. Equity Risk Premium Forum. www.cfainstitute. Bessembinder, Hendrik. 2018. “Do Stocks Outperform
org/-/media/documents/book/rf-publication/2002/rf- Treasury Bills?” Journal of Financial Economics 129 (3):
equity-risk-premium-forum-2001.pdf. 440–457.

Arnott, Robert D., and Peter L. Bernstein. 1997. “Bull Market? Brown, Stephen J., and William N. Goetzmann. 2018.
Bear Market?” Journal of Portfolio Management 24 (1): “Stephen Ross’s Contribution to Ex Post Conditioning and
26–29. Survival Bias in Empirical Research.” Journal of Portfolio
Management 44 (6): 42–46.
Arnott, Robert D., and Peter L. Bernstein. 2002. “What Risk
Premium Is ‘Normal’?” Financial Analysts Journal 58 (March/ Brown, Stephen J., William N. Goetzmann, and Stephen A.
April): 64–85. www.researchaffiliates.com/documents/ Ross. 1995. “Survival.” Journal of Finance 50 (3): 853–73.
FAJ_Mar_Apr_2002_What_Risk_Premium_is_Normal.pdf.
Buffett, Warren, and Carol Loomis. 2001. “Warren
Arnott, Robert D., and William J. Bernstein. 2018. “‘The Long- Buffett on the Stock Market.” Fortune (10 December).
Run Drivers of Stock Returns: Total Payouts and the Real https://money.cnn.com/magazines/fortune/fortune_
Economy’: A Comment.” Financial Analysts Journal 74 (1). archive/2001/12/10/314691/.

Arnott, Robert D., Denis B. Chaves, and Tzee-Man Chow. Campbell, John Y., and Robert J. Shiller. 1998. “Valuation
2017. “King of the Mountain: The Shiller P/E and Ratios and the Long-Run Stock Market Outlook.” Journal of
Macroeconomic Conditions.” Journal of Portfolio Portfolio Management 24 (2): 11–26.
Management 44 (1): 55–68.
Campbell, John Y., Adi Sunderam, and Luis M. Viceira. 2017.
Arnott, Robert D., Campbell R. Harvey, Vitali Kalesnik, and “Inflation Bets or Deflation Hedges? The Changing Risks of
Juhani T. Linnainmaa. 2021. “Reports of Value’s Death Nominal Bonds.” Critical Finance Review 6 (2): 263–301.
May Be Greatly Exaggerated.” Financial Analysts Journal
77 (1): 44–67. Claeskens, Gerda, Jan R. Magnus, Andrey L. Vasnev, and
Wendun Wang. 2016. “The Forecast Combination Puzzle:
Arnott, Robert D., Vitali Kalesnik, Engin Kose, and Lillian J. A Simple Theoretical Explanation.” International Journal
Wu. 2017. “Can Momentum Investing Be Saved?” Working of Forecasting 32 (3): 754–62.
paper, Research Affiliates (14 October). Available at https://
ssrn.com/abstract=3099687. Clarida, Richard. 2019. “Monetary Policy, Price Stability,
and Equilibrium Bond Yields: Success and Consequences.”
Arnott, Robert D., and Ronald J. Ryan. 2001. “The Death of Speech at the High-Level Conference on Global Risk,
the Risk Premium.” Journal of Portfolio Management 27 (3): Uncertainty, and Volatility, Zurich (12 November).
61–74. www.federalreserve.gov/newsevents/speech/
clarida20191112a.htm.
Asness, Cliff. 2000. “Bubble Logic, or How to Learn to Stop
Worrying and Love the Bull.” Working paper, AQR (1 August). Cochrane, John H. 2011. “Discount Rates.” Journal of
www.aqr.com/Insights/Research/Working-Paper/Bubble- Finance 66 (August): 1047–108.
Logic-Or-How-to-Learn-to-Stop-Worrying-and-Love-the-Bull.
Coleman, Thomas S., Bryan J. Oliver, and Laurence B.
Asness, Cliff. 2020. “Is (Systematic) Value Investing Dead?” Siegel. 2021. “Puzzles of Money, Inflation, and Debt:
AQR (blog; 8 May). www.aqr.com/Insights/Perspectives/ Applying the Fiscal Theory of the Price Level.” CFA
Is-Systematic-Value-Investing-Dead. Institute Research Foundation (29 November). www.
cfainstitute.org/en/research/foundation/2021/
Asness, Cliff. 2021. “The Long Run Is Lying to You.” AQR puzzles-of-inflation-money-debt.
(blog; 4 March). www.aqr.com/Insights/Perspectives/
The-Long-Run-Is-Lying-to-You. Constantinides, George M., John B. Donaldson, and Rajnish
Mehra. 2002. “Junior Can’t Borrow: A New Perspective on
Bansal, Ravi, and Amir Yaron. 2004. “Risks for the Long Run: the Equity Premium Puzzle.” Quarterly Journal of Economics
A Potential Resolution of Asset Pricing Puzzles.” Journal of 117 (1): 269–96. www.academicwebpages.com/preview/
Finance 59 (4): 1481–509. mehra/pdf/Junior%20QJE.pdf.

Banz, Rolf W. 1981. “The Relationship between Return Constantinides, George M., John B. Donaldson, and Rajnish
and Market Value of Common Stocks.” Journal of Financial Mehra. 2007. “Junior Is Rich: Bequests as Consumption.”
Economics 9 (1): 3–18. Economic Theory 32 (1): 125–55.

CFA Institute Research Foundation 107

Revisiting the Equity Risk Premium

Dimson, Elroy, and Paul Marsh. 2001. “UK Financial Market Ibbotson, Roger G., and Rex A. Sinquefield. 1976b. “Stocks,
Returns 1955–2000.” Journal of Business 74 (January): 1–30. Bonds, Bills, and Inflation: Simulations of the Future (1976–
2000).” Journal of Business 49 (3): 313–38.
Dimson, Elroy, Paul Marsh, and Mike Staunton. 2002.
Triumph of the Optimists: 101 Years of Global Investment Ilmanen, Antti. 2003. “Stock–Bond Correlations.” Journal of
Returns. Princeton, NJ: Princeton University Press. Fixed Income 13 (2): 55–66.

Dimson, Elroy, Paul Marsh, and Mike Staunton. 2013. “Mean Ilmanen, Antti. 2022. Investing amid Low Expected Returns:
Reversion.” In Credit Suisse Global Investment Returns Making the Most When Markets Offer the Least. Hoboken,
Yearbook 2013, 17–27. Zurich: Credit Suisse Research NJ: Wiley.
Institute. https://credit-suisse.com/media/assets/
corporate/docs/about-us/research/publications/credit- Jensen, Michael C., and William H. Meckling. 1976. “Theory of
suisse-global-investment-returns-yearbook-2013.pdf. the Firm: Managerial Behavior, Agency Costs and Ownership
Structure.” Journal of Financial Economics 3 (4): 305–60.
Dimson, Elroy, Paul Marsh, and Mike Staunton. 2021.
“American Exceptionalism: The Long-Term Evidence.” Kaldor, Nicholas. 1957. “A Model of Economic Growth.” 
Journal of Portfolio Management 47 (7): 14–26. Economic Journal 67 (268): 591–624.

Dodge, Yadolah, ed. 2003. “Winsorized Estimation.” Langetieg, Terence C., Martin L. Leibowitz, and Stanley
The Oxford Dictionary of Statistical Terms. Oxford, UK: Kogelman. 1990. “Duration Targeting and the Management
Oxford University Press. of Multiperiod Returns.” Financial Analysts Journal 46 (5):
35–45. www.cfainstitute.org/en/research/financial-ana-
Fama, Eugene F. 1965. “The Behavior of Stock Market lysts-journal/1990/duration-targeting-and-the-manage-
Prices.” Journal of Business 38 (1): 34–105. ment-of-multiperiod-returns.

Fama, Eugene F., and Kenneth R. French. 1988. “Dividend Le Bris, David, William N. Goetzmann, and Sébastien Pouget.
Yields and Expected Stock Returns.” Journal of Financial 2019. “The Present Value Relation over Six Centuries:
Economics 22 (1): 3–25. The Case of the Bazacle Company.” Journal of Financial
Economics 132 (1): 248–65.
Farhi, Emmanuel, and François Gourio. 2018. “Accounting for
Macro-Finance Trends: Market Power, Intangibles, and Risk L’Her, Jean-François, Tarek Masmoudi, and Ram Karthik
Premia.” Brookings Institution (13 September). www.brook- Krishnamoorthy. 2018. “Net Buybacks and the Seven
ings.edu/bpea-articles/accounting-for-macro-finance- Dwarfs.” Financial Analysts Journal 74 (4): 57–85.
trends-market-power-intangibles-and-risk-premia.
Martin, R. Douglas, Thomas K. Philips, Stoyan Stoyanov,
Goyal, Amit, and Ivo Welch. 2003. “Predicting the Equity Bernd Scherer, and Kirk Li. Forthcoming 2024. Portfolio
Premium with Dividend Ratios.” Management Science Construction and Risk Analysis. New York: Springer.
49 (5): 639–54.
McKinnon, Ronald I. 1973. Money and Capital in Economic
Hammond, P. Brett. 2012. “Rethinking the Equity Risk Development. Washington, DC: Brookings Institution Press.
Premium.” Presentation at Q Group (October).
Mehra, Rajnish, ed. 2008. Handbook of the Equity Risk
Hammond, P. Brett, Martin L. Leibowitz, and Laurence Premium. Amsterdam: Elsevier.
B. Siegel. 2011. Rethinking the Equity Risk Premium.
Charlottesville, VA: CFA Institute Research Foundation. Mehra, Rajnish, and Edward C. Prescott. 1985. “The Equity
Premium: A Puzzle.” Journal of Monetary Economics 15 (2):
Hirshleifer, David. 2001. “Investor Psychology and Asset 145–61.
Pricing.” Journal of Finance 56 (4): 1533–97.
Modigliani, Franco, and Merton H. Miller. 1958. “The Cost of
Homer, Sidney, and Martin L. Leibowitz. 2004. Inside the Capital, Corporation Finance and the Theory of Investment.”
Yield Book. New York: Bloomberg Press. American Economic Review 48 (3): 261–97.

Ibbotson, Roger G., Thomas M. Idzorek, Paul D. Kaplan, Pástor, Luboš, and Robert F. Stambaugh. 2012. “Are Stocks
and James X. Xiong. 2018. Popularity: A Bridge between Really Less Volatile in the Long Run?” Journal of Finance
Classical and Behavioral Finance. Charlottesville, VA: 67 (2): 431–78.
CFA Institute Research Foundation.
Philips, Thomas K. 1999. “Why Do Valuation Ratios Forecast
Ibbotson, Roger G., and Rex A. Sinquefield. 1976a. “Stocks, Long-Run Equity Returns?” Journal of Portfolio Management
Bonds, Bills, and Inflation: Year-by-Year Historical Returns 25 (3): 39–44.
(1926–1974).” Journal of Business 49 (1): 11–47.

108 CFA Institute Research Foundation

 Revisiting the Equity Risk Premium

Philips, Thomas K. 2002. “The Source of Value.” Journal of Sicard, Germain. 2015. The Origins of Corporations: The
Portfolio Management 28 (4): 36–44. Mills of Toulouse in the Middle Ages. New Haven, CT: Yale
University Press.
Philips, Thomas K., and Ádám Kóbor. 2020. “Ultra-Simple
Shiller’s CAPE: How One Year’s Data Can Predict Equity Siegel, Jeremy J. 1995. “The Nifty-Fifty Revisited: Do Growth
Market Returns Better Than Ten.” Journal of Portfolio Stocks Ultimately Justify Their Price?” Journal of Portfolio
Management 46 (4): 140–55. Management 21 (4): 8–20.

Philips, Thomas K., and Cenk Ural. “Uncloaking Siegel, Jeremy J. 2005. “Perspectives on the Equity Risk
Campbell and Shiller’s CAPE: A Comprehensive Guide Premium.” Financial Analysts Journal 61 (6): 61–71.
to Its Construction and Use.” Journal of Portfolio
Management 43 (1): 109–25. Siegel, Jeremy J. 2014. Stocks for the Long Run, 5th ed.
New York: McGraw Hill.
Phillips, Peter C. B., Shuping Shi, and Jun Yu. 2015. “Testing
for Multiple Bubbles: Historical Episodes of Exuberance and Siegel, Jeremy J. 2016. “The Shiller CAPE Ratio: A New Look.”
Collapse in the S&P 500.” International Economic Review Financial Analysts Journal 72 (3): 41–50.
56 (4): 1043–78.
Siegel, Jeremy J., and Jeremy D. Schwartz. 2006. “Long-Term
Reinganum, Marc. 1981. “Misspecification of Capital Asset Returns on the Original S&P 500 Firms.” Financial Analysts
Pricing: Empirical Anomalies Based on Earnings Yields and Journal 62 (1): 18–31.
Market Values.” Journal of Financial Economics 9 (1): 19–46.
Straehl, Philip U., and Roger G. Ibbotson. 2017. “The Long-
Ross, Stephen A. 1987. “Regression to the Max.” Run Drivers of Stock Returns: Total Payouts and the Real
Unpublished manuscript, Yale University. www.research- Economy.” Financial Analysts Journal 73 (3): 32–52.
gate.net/profile/Stephen_Ross3/publication/247378278_
Straehl, Philip U., and Roger G. Ibbotson. 2018. “‘The Long-
Regression_to_the_max/.
Run Drivers of Stock Returns: Total Payouts and the Real
Samuelson, Paul A. 1958. “An Exact Consumption-Loan Economy’: Author Response.” Financial Analysts Journal
Model of Interest with or without the Social Contrivance of 74 (1).
Money.” Journal of Political Economy 66 (6): 467–82.
van Binsbergen, Jules H., and Ralph S. Koijen. 2017. “The
Shaw, Edward S. 1973. Financial Deepening in Economic Term Structure of Returns: Facts and Theory.” Journal of
Development. New York: Oxford University Press. Financial Economics 124 (1): 1–21.

CFA Institute Research Foundation 109

 Named Endowments
CFA Institute Research Foundation acknowledges with sincere gratitude the generous contributions of the Named
Endowment participants listed below.

Gifts of at least US$100,000 qualify donors for membership in the Named Endowment category, which recognizes
in perpetuity the commitment toward unbiased, practitioner-oriented, relevant research that these firms and
individuals have expressed through their generous support of CFA Institute Research Foundation.

Ameritech Miller Anderson & Sherrerd, LLP

Anonymous John B. Neff, CFA
Robert D. Arnott Nikko Securities Co., Ltd.
Theodore R. Aronson, CFA Nippon Life Insurance Company of Japan
Asahi Mutual Life Insurance Company Nomura Securities Co., Ltd.
Batterymarch Financial Management Payden & Rygel
Boston Company Provident National Bank
Boston Partners Asset Management, L.P. Frank K. Reilly, CFA
Gary P. Brinson, CFA Salomon Brothers
Brinson Partners, Inc. Sassoon Holdings Pte. Ltd.
Capital Group International, Inc. Scudder Stevens & Clark
Concord Capital Management Security Analysts Association of Japan
Dai-Ichi Life Insurance Company Shaw Data Securities, Inc.
Daiwa Securities Sit Investment Associates, Inc.
Mr. and Mrs. Jeffrey Diermeier Standish, Ayer & Wood, Inc.
Gifford Fong Associates State Farm Insurance Company
John A. Gunn, CFA Sumitomo Life America, Inc.
Investment Counsel Association of America, Inc. T. Rowe Price Associates, Inc.
Jacobs Levy Equity Management Templeton Investment Counsel Inc.
Jon L. Hagler Foundation Frank Trainer, CFA
Long-Term Credit Bank of Japan, Ltd. Travelers Insurance Co.
Lynch, Jones & Ryan, LLC USF&G Companies
Meiji Mutual Life Insurance Company Yamaichi Securities Co., Ltd.

Senior Research Fellows

Financial Services Analyst Association

For more on upcoming CFA Institute Research Foundation publications

and webcasts, please visit www.cfainstitute.org/research/foundation.
 CFA Institute
Research Foundation
Board of Trustees
2022–2023

Chair Frank Fabozzi, CFA* Kingpai Koosakulnirund, CFA

Aaron Low, PhD, CFA The Johns Hopkins University CFA Society Thailand
LUMIQ Carey Business School
Lotta Moberg, PhD, CFA
Vice Chair Margaret Franklin, CFA William Blair
Joanne Hill, PhD CFA Institute
Maureen O’Hara, PhD*
Cboe Vest LLC
Bill Fung, PhD Cornell University
Amra Balic Aventura, FL
Dave Uduanu, CFA
BlackRock
Heinz Hockmann, PhD Sigma Pensions Ltd
Aaron Brown, CFA Lovell Minnick Partners LLC
Kurt D. Winkelmann
City of Calgary
Punita Kumar-Sinha, CFA Navega Strategies, LLC
Kati Eriksson, CFA Infosys
Danske Bank

*Emeritus

Officers and Directors

Executive Director Secretary
Bud Haslett, CFA Jessica Lawson
CFA Institute CFA Institute
Gary P. Brinson Director of Research Treasurer
Laurence B. Siegel Kim Maynard
Blue Moon Communications CFA Institute
Research Director
Luis Garcia-Feijóo, CFA, CIPM
Coral Gables, Florida

Research Foundation Review Board

William J. Bernstein, PhD Paul D. Kaplan, PhD, CFA Krishna Ramaswamy, PhD
Efficient Frontier Advisors Morningstar, Inc. University of Pennsylvania

Elroy Dimson, PhD Robert E. Kiernan III Andrew Rudd, PhD

Cambridge Judge Business Advanced Portfolio Management Advisor Software, Inc.
School
Andrew W. Lo, PhD Stephen Sexauer
Stephen Figlewski, PhD Massachusetts Institute Allianz Global Investors Solutions
New York University of Technology
Lee R. Thomas, PhD
William N. Goetzmann, PhD Alan Marcus, PhD Pacific Investment Management
Yale School of Management Boston College Company

Elizabeth R. Hilpman Paul O’Connell, PhD

Barlow Partners, Inc. FDO Partners
 ISBN 978-1-952927-35-5

Available online at www.cfainstitute.org 9 781952 927355