# LO 47.4: Explain the impact of model risk and poor risk governance in the

LO 47.4: Explain the impact of model risk and poor risk governance in the 2012 London Whale trading loss and the 1998 collapse of Long Term Capital Management.
The impact of model risk has been felt significantly during two specific incidents: the 1997 collapse of Long-Term Capital Management (LTCM) and the 2012 London Whale trading loss at JPMorgan Chase (JPM). Both incidents illustrate the necessity to closely examine and vet models, and the importance of considering model risk within an organizations institutional risk governance framework.
Long-Term Capital Management
Background and Trading Strategies
LTCM was a U.S. hedge fund that existed between 1994 and 1998. The fund raised in excess of $1 billion in capital at its inception and grew rapidly over its initial years. LTCMs trading strategy relied on arbitrage positions based on market-neutral and relative-value trading. The fund began primarily as a bond arbitrage hedge fund that sought to make money by exploiting the spread differentials between bonds, including spread differences of European sovereign bonds and spread differences of corporate bonds and government Treasuries in the United States and United Kingdom. LTCM relied on a combination of extensive empirical research and advanced financial modeling to formulate bets on convergence of prices in bond markets. For example, the fund was long (bought) Spanish and Italian sovereign debt and was short (sold) German Page 120 2018 Kaplan, Inc. Topic 47 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15 sovereign debt. The strategy assumed that German sovereign bonds were overpriced relative to the weaker Spanish and Italian bonds, which were expected to increase in value with the imminent membership in the European economic and monetary union. .Another strategy was based on the expected convergence between the spreads of corporate and government bonds in the United States and United Kingdom, where spreads were expected to return to normal levels. This strategy was designed to make a profit regardless of the movement in price levels, assuming, however, that spreads moved in the appropriate direction and that correlations did not change materially. Leverage, Correlations, and Volatility LTCMs strategies were designed to generate only modest profits (around 1%). In order for the fund to generate strong performance, it needed to use extensive leverage of up to 25 times. Such leveraged positions relied on large institutional loans that were collateralized by bond investments. Shortly before the funds collapse in 1998, LTCM had capital of close to$5 billion, assets of over $125 billion, and a notional value of investments in excess of$1.25 trillion. The magnitude of LTCMs leveraged investments was unprecedented in the markets.
LTCMs strategies worked as long as positions converged as anticipated, and as long as correlations did not deviate significantly from historical levels. Volatilities were calculated based on mathematical models to be approximately in line with the risk of investing in the S&P 500. However, at the time of the funds collapse, its one-day volatility exceeded its model predicted volatility by 2.5 times, and the fund suffered losses of more than 3 times its 10-day predicted maximum loss.
Collapse and Lessons
In 1997, Asian markets experienced considerable economic and financial problems that quickly spread to several economies as contagion increased. These troubles ultimately affected Russia, which was forced to devalue its currency, the ruble, and default on its sovereign debt in August 1998. The Asian and Russian crisis triggered a flight-to-quality in European and North American markets with investors seeking the safe and predictable returns of high-quality sovereign bonds. As a result, the yields of the U.S. and German long-term sovereign bonds declined (their prices increased), while at the same time the yields on the riskier corporate bonds and riskier sovereign bonds (for example, Italy and Spain) increased (their prices fell). Credit spreads widened, volatilities increased beyond historical levels, and correlations in the market moved closer to +1 as the contagion effect of the crisis spread across markets.
With higher volatilities and dramatically widening spreads, the profits on LTCMs short positions were no longer sufficient to offset the losses on its long positions. With losses mounting, lenders demanded additional collateral. In order to meet collateral calls, LTCM had to unwind several unprofitable trades that put further downward pressure on markets given the size of the funds trading positions. At the same time, liquidity in the markets quickly began to dry up, leaving many of LTCMs market-neutral positions now directionally exposed on the long side. Ultimately, the fund became insolvent in September 1998 and was bailed out by the Federal Reserve Bank of New York in order to curb a potential global financial crisis.
2018 Kaplan, Inc.
Page 121
Topic 47 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15
LTCMs collapse highlighted several flaws in its regulatory value at risk (VaR) calculations: 1. The funds calculated 10-day VaR period was too short. A time horizon for economic
capital should be sufficiently long enough to raise new capital, which is longer than the 10-day assumption.
2. The funds VaR models did not incorporate liquidity assumptions. The assumption of
perfectly liquid markets proved to be incorrect when the fund experienced liquidity droughts.
3. The funds risk models did not incorporate correlation and volatility risks. This
weakness was especially evident when markets moved to a correlation of close to +1 and volatility increased significantly above historical and model predicted levels.
London Whale
Background and Trading Strategy
JPMorgan Chase & Company (JPM), along with its principal banking subsidiary JPMorgan Chase Bank, is a U.S. financial company and one of the largest derivatives traders in the world. JPM garnered international headlines when in the first half of 2012 it sustained losses in excess of $6 billion due to risky synthetic credit derivatives trades executed by a trader, called the London Whale, in its London office. The London trading desk belonged to JPMs Chief Investment Office (CIO), which was responsible for managing the banks excess deposits. The CIO was tasked with keeping the banks risk level down and prudently managing the banks$330 billion in excess deposits. Instead, the CIO used the deposits to engage in high- profit potential, high-risk derivatives trading strategies. In 2006, the CIO began a new series of synthetic credit derivatives trading strategies within its Synthetic Credit Portfolio (SCP). Trading focused less on hedging risk and more on earning profits from short positions.
Risk Culture, Model Risk, and Operational Risk
The CIO used various risk metrics for its trading activities, including VaR limits and credit spread widening limits.
In 2011, the CIO was instructed to reduce the banks risk-weighted assets (RWA) in order to reduce regulatory capital requirements. Instead of the common practice of selling high risk assets, the CIO instead launched a trading strategy to offset its outstanding short positions by taking long positions in synthetic credit derivatives. This resulted not only in an increase in the portfolios risk and size, but it also put the portfolio in a net long position, which reduced the hedging protection provided by the SCP.
Concurrently, in early 2012, and in response to breaching its own internal VaR limits as well as the banks VaR limits, the CIO adopted a new VaR model which lowered its calculated VaR by 30%. The revised model allowed the CIO to remain within its VaR limit and at the same time engage in more higher-risk trading activities. However, the bank failed to seek regulatory approval of the new model. In addition, there were manual
Page 122
2018 Kaplan, Inc.
Topic 47 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15
and calculation errors when implementing the model, which led to greater model and operational risk for the bank. Ultimately, the revised VaR model was reversed later in 2012 and the previous model was reinstated.
By 2012, the SCP was losing money on its strategies. In order to minimize its reported losses, the CIO changed its derivatives valuation practices from using midpoint prices (prices at the midpoint of the bid and ask) to using more favorable prices within the bid-ask spread during each day. As the losses in the SCP strategy increased, JPMs counterparties began to dispute the CIOs values, which led to frequent collateral disputes. Ultimately, JPMs positions soured and the bank lost close to \$6.2 billion.
The losses from the London Whale trade and the subsequent investigations revealed a poor risk culture at JPM. Risk limits were routinely downplayed or ignored, limit breaches were disregarded, and risk models were altered to favor riskier trading activities.
2018 Kaplan, Inc.
Page 123
Topic 47 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15
Ke y C o n c e pt s
LO 47.1 Model risk becomes important when quantifying the risk exposures of complex financial instruments, including exotic or synthetic derivatives and structured products. Model risk can give rise to losses from model errors, errors in assumptions, carelessness, fraud, or intentional mistakes. These errors can lead to undervaluing risk, overvaluing profit, or both. Six common model errors include: 1. Assuming constant volatility.
2. Assuming a normal distribution of returns.
3. Underestimating the number of risk factors.
4. Assuming perfect capital markets.
3. Assuming adequate liquidity.
6. Misapplying a model.
LO 47.2 Implementation error could occur when models that require complex simulations are not allowed to run a sufficient number of runs. This may result in incorrect output and therefore an incorrect interpretation of results.
For model implementation, considerations include frequency of refreshing model parameters, including volatilities and correlations. Correctly estimating parameters (durations, volatilities, and correlations) is challenging, however, implementing a model with input errors will result in inaccurate results.
Common valuation and estimation errors include: 1.
Inaccurate data.
2.
Incorrect sampling period length.
3. Liquidity and valuation problems.
Page 124
2018 Kaplan, Inc.
Topic 47 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15
LO 47.3 Model risk can be mitigated either through investing in research to improve the model, or through an independent vetting process. Vetting consists of six phases: 1. Documentation.
2. Vetting the soundness of the model.
4. Benchmark selection.
5. Health check and stress testing of the model.
6.
Incorporating model risk into the risk management framework.
LO 47.4 Long-Term Capital Management (LTCM) was a U.S. hedge fund that used arbitrage strategies to exploit spread differentials between bonds, including spread differences of European sovereign bonds and spread differences in corporate bonds and government Treasuries. LTCMs strategy was to make predictable, low returns and then amplify them using extensive leverage.
The collapse of LTCM in 1998 highlights three important lessons: 1. Utilizing a 10-day VaR period as a proxy for the time horizon for economic capital is too short. A time horizon is needed that is sufficiently long enough to model the time to raise new capital.
2. The funds VaR models ignored the possibility that liquidity may decline or even
completely dry up in periods of extreme stress.
3. The funds risk models ignored correlation and volatility risks. Specifically, the fund
did not account for stressed scenarios with material rises in volatility or an increase in positive market correlation as contagion risk spread across international economies.
In 2012, JPMorgan Chase (JPM) and its Chief Investment Office (CIO) sustained severe losses due to risky synthetic credit derivatives trades executed by its London office. The losses from the London Whale trade and the subsequent investigations highlighted a poor risk culture at JPM, giving rise to both model and operational risks across the firm. Risk limits were routinely ignored and limit breaches were disregarded.
2018 Kaplan, Inc.
Page 125
Topic 47 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15
C o n c e pt Ch e c k e r s
1.
2.
3.
4.
3.
A risk analyst for a mid-sized bank believes that two common errors in model building include the assumption of constant volatility of returns and the assumption of a non-normal returns distribution. The analyst is correct with regard to the assumption(s) of: A. volatility of returns only. B. non-normal returns distribution only. C. both volatility of returns and non-normal returns distributions. D. neither volatility of returns nor non-normal returns distributions.
Which of the following scenarios is the best example of a model error? A. Assuming a non-normal distribution of returns. B. Assuming perfectly liquid markets. C. Assuming variable distribution of asset price. D. Assuming imperfect capital markets.
The chief risk officer (CRO) of a European corporation recommends increasing the length of the sampling period in order to minimize model risk. However, increasing the length of the sampling period will most likely: A. B. diminish the power of the statistical test. C. put higher weight on obsolete information. D. diminish the relevance of old data.
increase estimation errors.
Gamma Investments, LLC (Gamma) uses monthly model vetting to mitigate potential model risk. Gammas managers recently accepted the use of a model for valuing short-term options on 30-year corporate bonds, but rejected the same model to value short-term options on three-year government bonds. The managers also frequently test proposed analytical models against a simulation approach. These model vetting techniques are examples of which of the following vetting phases?
Accepting/rejecting a model A. Health check of the model B. Soundness of a model C. Health check of the model D. Soundness of a model
Testing models against simulation Stress testing Stress testing Benchmark modeling Benchmark modeling
Which of the following flaws in Long-Term Capital Managements (LTCM) value at risk (VaR) calculations were most evident following its collapse in 1998? I. The calculated 10-day VaR period was too short. II. The funds VaR model assumed strong positive correlation. A. I only. B. II only. C. Both I and II. D. Neither I nor II.
Page 126
2018 Kaplan, Inc.
Topic 4 7 Cross Reference to GARP Assigned Reading – Crouhy, Galai, and Mark, Chapter 15
C o n c e pt C h e c k e r An s w e r s
1. A The analyst is correct with respect to the assumption of volatility of returns only. Another
common model error is the assumption of a normal distribution of returns. Market participants frequently make the simplifying assumption in their models that asset returns are normally distributed. However, empirical research shows that returns tend to be non- normally distributed.
2. B Six common model errors include: (1) assuming constant volatility, (2) assuming a normal
distribution of returns, (3) underestimating the number of risk factors, (4) assuming perfect capital markets, (5) assuming adequate liquidity, and (6) misapplying a model.
3. C Adding more observations to the model reduces estimation errors and improves the power of statistical tests. However, it gives greater relevance to old and potentially stale data and puts greater weight on obsolete information which may now be irrelevant.
4. D Accepting the model for one use but rejecting it for another (inappropriate) use is an example of vetting the soundness of the model. In other words, the model vetter (in this case the risk managers) should ensure that the mathematical model reasonably represents the asset being valued.
Testing a proposed analytical model against a simulation approach or a numerical approximation technique is an example of benchmark modeling.
Health check of the model ensures that the model contains all of the necessary properties. Stress testing a model uses simulations to check the models reaction to different situations.
5. A LTCM s collapse highlighted several flaws in its regulatory VaR calculations. The fund relied on a VaR model that: (1) used a 10-day horizon, which proved to be too short to sufficiently model the time to raise new capital, (2) did not factor in liquidity risk (in other words, it assumed markets were perfectly liquid), and (3) did not incorporate correlation and volatility risks, where in fact markets exhibited strong positive correlation during periods of stress in 1997 and 1998.
2018 Kaplan, Inc.
Page 127
The following is a review of the Operational and Integrated Risk Management principles designed to address the learning objectives set forth by GARP. This topic is also covered in:
Ri s k Ca pi t a l A t t r i b u t i o n a n d Ri s k -Ad j u s t e d Pe r f o r ma n c e M e a su r e me n t
Topic 48
E x a m F o c u s
This topic covers the application of the risk-adjusted return on capital (RAROC) approach to the allocation of economic capital. The application of a hurdle rate for capital budgeting decisions as well as an adjusted version of the traditional RAROC approach is also presented. For the exam, know the differences between economic capital and regulatory capital, and be able to compute RAROC for capital budgeting as well as adjusted RAROC. Also, be familiar with the qualitative concepts discussed, such as reasons for using economic capital to allocate risk capital, the benefits of RAROC, and best practices in implementing the RAROC approach.
R i s k C a p i t a l
, E c o n o m i c C a p i t a l
, a n d R e g u l a t o r y C a p i t a l