Risk

JULY 2001
www.risk.net

CURRENCIES INTEREST RATES EQUITIES COMMODITIES CREDIT

Markets and models

Address: 10 Hanover Square, New York, NY 10005.

T
rading desks in many product areas situated several steps away from the trad- the prices of their underliers. Traders need
at investment banks often have ing desk, who represent the firm as a models to augment their intuition, in order
substantial positions in long-term or whole rather than a particular desk, and to interpolate from the linearly changing
exotic over-the-counter derivative whose mission it is to verify the marked underlier prices to the non-linearly vary-
securities, custom-tailored to satisfy the prices of these positions. ing prices of the derivative securities. A
risk preferences of their customers. The model tells you how to manufacture the
idiosyncratic nature of these securities Derivatives, models, prices derivative from its linear underliers.
makes them relatively illiquid. For liquid securities, price verification is Options theorists and practitioners like to
As a consequence, these positions are straightforward; one simply checks that the say replicate or hedge instead of
marked and hedged using sophisticated prices assigned by the trading desks corre- manufacture, but the meaning is identical.
and complex financial models, imple- spond with those obtained from reliable ex- You can think of a derivative as a
mented as software and then embedded ternal sources. Illiquid derivative securities, mixture of its constituent underliers, much
in front-office risk management systems. our focus here, are more complex. as a cake is a mixture of eggs, flour and
These models derive prices from market A derivative securitys payout is deter- milk in carefully specified proportions.
parameters (volatilities, correlations, mined by how it depends on the prices of The derivatives model provides a recipe
prepayment rates or default probabilities, its underlying securities. So, finding the for the mixture, one whose ingredients
for example) that are forward-looking current value of a derivative security is quantities vary with time.
and should ideally be implied from the a question of relativity: what is the The derivatives model or recipe tells
market prices of traded securities. derivative security worth relative to the you how to dynamically vary the propor-
Because of their illiquidity, many of known prices of its underliers, given their tions of the mixture as time passes and the
these positions will be held for years. Yet estimated future behaviour? underlying security prices change. If you
their daily values affect the short-term The dependence of the payout on the have estimated the future stochastic
profit and loss of the banks that trade underliers values is determined solely by behaviour of the underlying securities ac-
them, and also inevitably influence the the contract, and should be unambiguous. curately, then, as their prices change and
careers and compensations of the traders But the estimated future stochastic time passes, the mixtures price will behave
and salespeople who structure, trade, behaviour of the underliers is a matter of identically to that of the derivative securi-
hedge and market them. opinion, which is where models enter. ty itself.
Increasingly, therefore, banks employ Most derivative securities are non-lin- Derivative models work best when they
autonomous firmwide risk managers, ear: their prices vary disproportionately to use as their constituents underlying secu-
rities that are one level simpler and one 5. build a computer program that incor- involving, for example, subtleties of
level more liquid than the derivative itself. porates items one through four above. price averaging, barrier crossing,
Thus, one manufactures a standard stock holidays and date conventions. In
option out of a changing mixture of stocks Derivatives price verification practice, many of the errors made in
and riskless bonds. Similarly, one manu- Simple derivatives, such as standard stock marking derivatives involve accidental
factures an exotic option, for example a or index options, usually have well-ac- or sloppy mis-specification of these
knockout call, out of standard options. cepted models that are used as a market- details in the trading systems database.
In any model, the value of the securi- quoting convention for communicating 2. Does the model provide a realistic
ty and its hedge ratios (the proportions of price information. (or at least plausible) description of
constituents in the mixture) are intrinsi- Market participants who use a broadly the evolution of the factors that
cally related to each other obverse sides accepted model are quoting their prices in affect the derivatives value? When a
of the same coin. terms of parameters (such as volatility) trader books a derivative transaction
Any model for the future behaviour of that, once entered into the model, produce into a firms risk system, he implicitly
an underlier is an attempted simplification the quoted price as output. Their use of assigns a valuation model to it.
of a reality that is likely to be much more the model does not necessarily imply that All models are simplistic. Is the
complex and unpredictable than the its assumptions about the future behaviour assigned model an appropriate simpli-
model itself. Therefore, it is not obvious of the underliers accurately reflects fication? In practice, many of the errors
which model is correct, nor even exactly their actual behaviour, though some made in marking derivatives are
what correct means. Also, once youve correspondence between assumptions the consequence of an unsuitable
picked a model, the input parameters you and behaviour must obviously exist for the assignment, or perhaps an assignment
use to describe its details are themselves model to be embraced. suitable only during the current
estimates of future security characteristics, For liquid derivative securities, one market regime. For example, an
such as volatility or correlation, whose verifies a marked price by checking that adequate model for a hybrid product
exact values are unknown. the traders inputs (such as volatility) that incorporates both equity and
When you create and use a model to agrees with those obtained from an currency risk may often justifiably
value a derivative security, you need to: observed external market. For less liquid neglect exchange rate volatility during
1. obtain a careful description of the or more exotic derivatives, one must normal, relatively low-volatility times.
securitys payout in terms of the values of verify the payout description, check the 3. Has the model been accurately
each of its underliers; plausibility of the inputs and examine the calibrated to the observed behaviour,
2. specify a model for the future behav- reasonableness of the model itself. One parameters and prices of the simpler,
iour of the underliers; must also test the computer programs that liquid constituents that comprise the
3. obtain accurate values of model incorporate the model. derivative?
parameters (volatility, for example) that To do all this, one should ask the 4. Finally, is the software reliable?
describe the behaviour of the underliers; following series of open-ended questions: Handling portfolios of complex deriva-
4. calibrate the model to these parameters 1. Does the model embody an accurate tives requires building intricate and
so that it reproduces the known prices and description of the terms of the derivatives extensive programs and systems that
assumed future behaviour of the under- payout? The function that specifies the will inevitably contain errors. Accurate
liers; and then (usually) payout can be exceedingly complex, valuation demands an organised

Risk management procedures explained

Because the use of models requires an intricate interaction strating that the model captures the relevant features of both the
between mathematics, software, systems and common sense, it is market and product it addresses, and should show that it has
useful to have procedures in place for the validation, modification been adequately designed, built and tested. The report should
and maintenance of models produced in each product area. also address the numerical accuracy and stability of the
algorithms and software, and emphasise those domains where
Model control inaccuracies or approximations become unacceptably large. This
All new derivatives models should be tested and certified by some- documentation will provide a long-term corporate memory of the
one other than the developer. Alterations to models and software principles and implementation of the models.
should be logged and documented.
Rational documented model assignment
Quantifying a desks model risk When a trader books a deal and assigns a valuation model to a
Each product area should use its front-office risk management sys- complex OTC derivative security, there should be an explanation of
tem to report the absolute value of the sensitivity of each position why the particular model was chosen, the simplifications it
in their portfolio to volatility, interest rates, or other key determi- involves, and why they are currently justifiable.
nants of derivative value. These sensitivities can be summed and
grouped into bins, with each bin corresponding to one of the range Input testing
of models used to value derivative products. Ranking the bins by The simplest test of prices is to report the extent of the correspon-
decreasing sensitivity directs attention to those models whose dence between the input parameters in the desks models and the
impact on the desks risk is greatest. same parameters obtained from external market sources. The
fraction of the desks sensitivity to its key derivative parameter (say,
Model validation volatility) that has been successfully compared with external
The models responsible for the greatest sensitivity of the portfolio sources is a useful, if somewhat naive, quantitative measure of the
should be documented and validated. The developers and users scope of verification.
of a particular model should produce a written report demon- However, input testing can be complex. Some parameters are
 Risk management procedures explained (contd)

not single numbers, but one-dimensional curves of interest rates, or be exhibited over the life of the trade due to transactions costs,
even multi-dimensional surfaces of implied volatilities. hedging error and model risk. Ultimately, such analyses should be part
Some more complex input parameters (for example, default of the desks own front-office valuation system.
correlation, spread volatilities or default recovery rates) have no reliable
external market sources because of the lack of liquid trading instru- Periodic comprehensive model review
ments. For these, one must estimate their most reasonable implied Immature derivatives markets often display prices that are consistent
values indirectly, by means of theoretical models that link their values with the usual Black-Scholes assumptions. Then, as markets mature
to those of other traded instruments. and participants gain experience of the supply, demand and shocks
that their underliers and derivatives can experience, prices start to
Comprehensive price verification reflect these realities more accurately.
Nothing is better than a completely independent check of price and For example, ever since the 1987 stock market crash, equity index
hedge ratios. A strategist knowledgeable about the market, but organi- derivatives have displayed a skew in which out-of-the-money puts trade
sationally separate from the trading desk, should start with the at much higher Black-Scholes implied volatilities than out-of-the-money
confirmed trade details and build an independent model to describe calls. During the past 10 years, skews of this type have become preva-
product and the market, then calibrate the model, and finally provide lent in most derivatives markets, from swaptions to gold options.
an estimate of value and hedge ratio. This is the most thorough and Skew is inconsistent with most simple and widely used options
valuable way to check prices. It provides a truly autonomous verification models. A major challenge in determining the accuracy of illiquid exotic
of value and, equally important, highlights the sensitivity of price and derivatives prices is the battle to develop realistic models that can be
hedge ratio to the choice of model. calibrated to reflect the skew of standard options, and which can then
be used to value the exotic ones. For listed equity index options, liquidi-
Valuation adjustments for transaction costs, hedging error ty is good and the standard skew is easy to observe. In swaptions mar-
and model risk kets, accurate information on the skew is much more difficult to obtain.
It is never clear what profit or loss will result from hedging a derivative It is therefore generally advisable to periodically revisit entire
security to its expiration. Markets will move in unexpected ways, some- derivatives markets and their models, to examine existing approaches
times intensifying transactions costs and often dismantling what may and perhaps develop new ones.
have seemed a reasonable hedging strategy. Since it is never clear exactly which model (for example, for the skew)
These effects are rarely captured by the conventional models used is correct, it is a good idea to investigate the effect on pricing and hedg-
in front-office trading systems. Therefore, for illiquid positions, it is ing of a variety of plausible models that can be calibrated to the mar-
important to estimate the adjustments to the conventionally marked ket, and so understand how sensitive the desks marks are to the par-
values that can occur as a result of long-term hedging. ticular model they use, and the assumptions it makes about the future.
Monte Carlo models can simulate both underlier behaviour and a
traders hedging strategy, to create distributions of the resultant profit Pre-review is better than verification
or loss of the entire portfolio. These distributions can be used to deter- Price verification suggests a check done after the deal is closed, but
mine a realistic adjustment to the trading desks conventional marks the most effective way to verify prices is in advance, in order to provide
that can be withheld until the trade is unwound and their realised independent analysis and confirmation on pricing and hedging before
profit or loss determined. deals close. Divisional risk managers and desk heads should seek to
These adjustments for hedging vary, as they should, as an option have the prices, hedges and risks of complex deals reviewed by inde-
moves in or out of the money, or as volatility changes. Monte Carlo pendent analysts in advance of closing. Assurance in advance is better
analysis offers a good sense of the variation in portfolio value that will than verification (or lack thereof) afterwards.

framework for the control and testing traders cannot be too knowledgeable of the group specialises in a particular
of the development, alteration and release about the assumptions behind models product area, ranging from swaps
of models and code. and their limitations. Nevertheless, it is through equity, currency and commodity
wise to provide independent price derivatives, to more recently developed
A multi-faceted approach to verification, delivered from the vantage markets such as credit derivatives.
price verification point of a firm as a whole, whose Importantly, the group also contains
A derivative security that cannot be interests may not always coincide with several ex-traders with a strong quantitative
bought or sold in a liquid market has no those of an individual trader. background who bring familiarity with
unambiguously correct value. Since it Some of the methods recommended in markets and their practices.
may have to be held and hedged until it the box on page 50 are derived from Because of the subtleties of models
expires, its realised value will depend on practices developed within the Deriva- and the unattainablity of correctness,
the future behaviour of its underliers tives Analysis group in the firmwide risk a multi-faceted approach to price
and their markets, as well as on the department at Goldman Sachs. This verification is best. One should selective-
details of the hedging strategies adopted group collaborates with controllers and ly span the valuation process at all
by the traders managing the portfolio. the divisional and firmwide risk managers levels of detail. The methods described
A firms first, broadest and most to provide independent price and in the box cover a broad spectrum,
efficient line of defence is the quality of model verification for selected complex moving progressively from coarser
the traders, model builders and software or illiquid derivative securities. to finer scales of resolution.
developers in each product area. The group consists mostly of quantita-
Emanuel Derman is grateful to Bob
Ultimately, one relies on them to be tive derivatives strategists, similar to those
Litzenberger, Noel Donohoe, Slim Bentami
careful, honest and responsible. employed in building and using models for and Leon Tatevossian for insightful advice
In particular, education is important; derivatives trading desks. Each member and remarks