Multi-Asset Risk Modeling: Techniques for a Global Economy in an Electronic and Algorithmic Trading Era

real-time.

Chapter 1

Introduction to Multi-Asset Risk Modeling—Lessons from the Debt Crisis

This introductory chapter reviews how statistical pricing and risk-forecasting models contributed to the debt crisis. For example, they gave incorrect results, underestimated risk, and mispriced collateralized debt obligations, mortgage-backed securities, and credit derivatives. Some reasons these models gave incorrect results was due to how quickly risk regimes could change due to global macroeconomic events, investor sentiment, and how ill-prepared traditional risk modeling systems (which rely on historical data) are to properly account for potential change in asset value.

We show how risk managers incorporate views and information sets across different asset classes to improve risk forecasts in rapidly changing markets. We present an overview of different types of risk and risk management products. In particular, we discuss details inherent in credit and equity risk: price risk (volatility), operational risk, country risk, default risk, company-specific risk, liquidity risk, interest rate risk, operational risk, macroeconomic factor risk. We also discuss how money managers, investors, and risk managers expand their understanding of risk and infer real-time information from derivatives and options, FX rates, debt markets, and macroeconomic changes. Finally, we discuss components of algorithmic trading, the industry’s lifeblood.

Keywords

Faulted Risk Models; Debt Crisis; Covariance Models; Volatility; Credit Risk; Liquidity Risk; Type II Statistical Error; Correlation Modeling; VIX Index; Flash Crash; Simplistic Investment Models; Static Forecasting

Financial services firms suffered significant losses brought on by one of the deepest crises ever to hit the financial services industry. As a result, risk modeling and management, loan valuation methods, capital allocation, and governance structures are shaken top to bottom. Fallout from the debt crisis continues to afflict most banks. Credit is still tight at the time of this writing. Banks, particularly those with significant levels of illiquid and difficult to sell assets, were finding it difficult to raise funds from traditional capital suppliers. In response, many institutions exited capital intensive, structured deals in an effort to deleverage and, notably, move away from international operations to concentrate on domestic business.

The debt crisis began when loan incentives, coupled with the acceleration in housing prices, encouraged borrowers to assume mortgages in the belief that they could refinance at favorable terms. Once prices started to drop, refinancing became difficult. Defaults and foreclosures increased dramatically as easy terms expired, home prices failed to go up, and interest rates reset higher. Foreclosures accelerated in late 2006, triggering a global financial crisis. Loan activities at banks froze while the ability of corporations to obtain funds through commercial paper dropped sharply. The expression a perfect storm, a once in a hundred years event, found its marker. The same could be said for Hurricane Sandy, Black Monday, and the October 1987 crash.

The CEO of Lehman Brothers said the firm was also the victim of the perfect storm—yet Lehman did not resist the lure of profits, leveraged balance sheets, and cheap credit. At Goldman Sachs, credit swaps created four billion dollars in profits. The financial modeling that helped produce results at these two investment banks were cutting edge, yet it appears that Lehman, AIG, and other profit takers amassed scant few algorithms to preserve capital protection against unexpected macroeconomic collapse.

Two government-sponsored enterprises, Fannie Mae and Freddie Mac, encouraged home sales by convincing investors that home values would rise over the long term, and that housing investments were safe. Fannie Mae and Freddie Mac created and sold Mortgage-Backed Securities (MBS), an investment considered reliable since the two firms guaranteed interest and principal payments on these loans. Originally, the two firms had strict rules governing their securities. However, during the 1990s, looser standards resulted in loan approvals to borrowers who had neither collateral nor financial strength to satisfy their obligations. Subprime loans encouraged the housing market initially, but later led to an uncontrollable housing expansion that ended up as the baseline of the financial crisis. Loan oversupply combined with the willingness of the borrowers to lend freely created an untenable outcome for lenders and borrowers.

On the government side, the debt crisis started with the notion that home ownership was a right. Fannie Mae was founded as part of Roosevelt’s New Deal to both purchase and securitize mortgages so that cash would be available to lend to home buyers. In 1970, a Republican congress created Freddie Mac to purchase mortgages on the secondary market, and pool and sell the mortgage-backed securities. Social and political pressure expanded the pool of home owners, mostly adding low- and middle-income families. The Equal Credit Opportunity Act prohibited institutional discrimination against minorities, while the Community Investment Act addressed discrimination in lending and encouraged financial institutions to lend to all income brackets in their communities.

In the early 80s, the passage of the Alternative Mortgage Transaction Parity Act preempted state laws by prohibiting adjustable rate mortgages, balloon payments, and negative amortization while allowing lenders to make loans with terms that obscured the loan’s cost. The Tax Reform Act prohibited taxpayers from deducting interest payments on consumer loans such as auto loans and credit cards, permitting them to deduct interest on mortgages, which sent homeowners in the millions to refinance mortgages and apply for home equity loans. As a result, household debt increased to 134% from 60% of disposable income. The elimination of Regulation Q was partially responsible for the Savings and Loan Crisis, resulting in a 40 billion dollar taxpayer loss. Despite the closing of hundreds of thrifts, lawmakers continued to deregulate the industry.

The Federal Housing Enterprises Financial Safety and Soundness Act of 1992 required Freddie Mac and Fannie Mae to devote a higher percentage of loans to support affordable housing. Banks were encouraged to do business across state lines, creating the mega banks we have today. Community banks, which had traditionally kept loans in neighborhoods, dwindled along with local loans. Behind the scenes, the banking lobby worked to repeal the Glass-Steagall Act of 1932 that separated commercial and investment banking. Gramm-Leach-Bliley, known also as the Financial Services Modernization Act of 1999, allowed financial institutions to operate as commercial and investment banks and insurance companies.

Many financial institutions ignored sustainability strategies¹ and credit guidelines originating at the Federal Reserve, Comptroller of the Currency, the Basel Committee, and other regulators years prior to the financial crisis. As an example, the office of the Comptroller of the Currency, Administrator of National Banks defined nine categories of risk for bank supervision purposes ten years before the crisis. Risks identified by the OCC included credit, interest rate, liquidity, price, foreign exchange, transaction, compliance, strategic, and reputation.²

Types of Risk

Credit Risk

Loans are the largest source of credit risk. There are other credit risk products, undertakings, and services included here such as the investment portfolio, overdrafts, letters of credit, and derivatives, foreign exchange, and cash management services. A financial institution’s credit policies and procedures define its risk profile. For example, sound credit policies and procedures deal with the following: (1) establish an appropriate credit risk environment; (2) operate under a sound credit granting process; (3) maintain an appropriate credit administration, measurement and monitoring process; and (4) ensure adequate controls over credit risk. Although specific credit risk management practices may differ among banks depending upon the nature and complexity of their credit activities, a comprehensive credit risk management program focuses on these four areas. These practices should also be applied in conjunction with sound practices related to the assessment of asset quality, the adequacy of provisions and reserves, and the disclosure of credit risk, all of which have been addressed in other recent Basel Committee documents.³ While the exact approach chosen by individual supervisors depends on a host of factors, doctrines set out in Principals for the Management of Credit Risk should have formed the basis of all bank audits. Yet, profits took rank over sustainability.

Effective management of multi-asset, particularly loan portfolio, credit risk requires that the board and (loan) administration understand and control an institution’s risk profile, and preserve credit culture and portfolio integrity. To accomplish this, bankers must have a thorough knowledge of the portfolio’s composition and its inherent risks. They must understand the portfolio’s product mix, industry and geographic concentrations, average risk ratings, and other aggregate characteristics. They must be sure that the policies, processes, and practices implemented to control risks of individual loans and portfolio segments are sound and that lending personnel adhere to them.⁴

Interest Rate and Market Risk

Market risks arise from adverse movements in market price or rates, for instance, interest, foreign exchange rates, or equity prices. Traditionally, management and regulators concentrated strictly on credit risk. In recent years, another group of assets have come under scrutiny: assets typically traded in financial markets. The level of interest rate risk attributed to the bank’s lending activities depends on the composition of its loan portfolio and the degree to which the terms of its loans expose the bank’s revenue stream to changes in rates. As part of the risk management process, banks typically identify exposures with heightened sensitivity to interest rate changes, and develop strategies to mitigate the risk: interest rate swaps, for example.

Liquidity Risk

Liquidity risk embodies the likelihood that an institution will be unable to meet obligations when due because assets cannot be liquidated, required funding is unavailable, or specific exposures cannot be unwound without significantly lowering market prices because of market disruptions brought on by a macroeconomic shock. Despite rigorous models and risk management controls, financial institution exposure from tail risk can accumulate. For highly leveraged financial institutions, cumulative exposure from tail risk can threaten survival in a stressed environment, as many banks learned too late. Capital position has a direct bearing on an institution’s ability to access liquidity and survive a debt crisis. Weak liquidity might cause a bank to liquefy assets or acquire liabilities to refinance maturing claims. As part of liquidity planning, a bank’s overall liquidity strategy should form an important measure of sustainable management.⁵

Price Risk

Exposures originated for sale as part of a securitization or for direct placement in the secondary market carry price risk while awaiting packaging and sale. During that period, the assets should be placed in a held-for-sale account, where they must be repriced at the lower of cost or market.⁶

Foreign Exchange Risk

Foreign exchange risk is present when a loan or portfolio of loans is denominated in a foreign currency or funded by borrowings in another currency. In some cases, banks enter into multi-currency credit commitments that permit borrowers to select the currency they prefer to use in each rollover period. Foreign exchange risk increases if bankers do not hedge political, social, or economic developments. In addition, results can be unfavorable if one currency becomes snarled in stringent exchange controls or experiences wide exchange-rate fluctuations.

Transaction Risk

The level of transaction risk depends on the capability of information systems and controls, the quality of operating procedures, and the capability and integrity of employees. Significant losses in loan and lease portfolios have resulted when information systems failed to provide adequate statistics to identify concentrations, failed to document auditing failures, expired facilities, or when stale financial statements led to model breakdowns. Banks have incurred losses because they failed to perfect or renew collateral liens; to obtain proper signatures on loan documents; or to disburse loan proceeds as required by the loan documents.⁷

Compliance Risk

Lending activities encompass a broad range of compliance responsibilities and risks. By law, a bank must observe limits on its loans to a single borrower, to insiders, and to affiliates; limits on interest rates; and the array of consumer protection and Community Reinvestment Act regulations.⁸ A bank’s lending activities may expose it to liability for the cleanup of environmental hazards.

Strategic Risk

Inappropriate strategic decisions about underwriting standards, loan portfolio growth, new loan products, or geographic and demographic markets will threaten a bank’s sustainability in a debt crisis. It is for this reason that regulators are taught to be especially focused when auditing new business and product ventures. These ventures require significant planning and careful oversight to ensure the risks are appropriately identified and managed. As we know, many banks extended consumer loan activities to subprime borrowers with dire consequences. The following was written 10 years prior to the debt crisis: "Do banks understand the unique risks associated with this market, can they price for the increased risk, and do they have the technology to service this market? Moreover, how will they compete with the nonbank companies who dominate this market? Both bankers and examiners need to decide whether the opportunities outweigh the strategic risks. If a bank is considering growing a loan product or business in a market saturated with that product or business, it should make sure that it is not overlooking other lending opportunities with more promise. During their evaluation of the loan portfolio management process, examiners should ensure that bankers are realistically assessing strategic risk."⁹

Reputation Risk

Negative publicity regarding business practices will cause a decline in the customer base, costly litigation, or revenue reductions. Inefficient loan delivery systems, failure to meet a community’s credit needs, and lender-liability lawsuits will compromise reputations.

Faulted Risk Models

The most disastrous business failures in the 2008 financial crises arose due to unpreparedness and the lack of the imagination needed to plan for the next inevitable economic crisis. The Chief Financial Officer of Goldman Sachs commented to the Financial Times in August 2007: We are seeing things that were 25-standard deviation moves, several days in a row. Putting that quote in perspective, a 7.26 sigma daily loss would emerge once in every 14 billion years: the age of the universe. The problem here is not 7.26 sigma, but rather within formula-rich, badly behaved models failing the human test: profits above strategic long-term planning, haphazard development, programming errors, and a foggy notion of systemic risk. Since the financial crisis, educators and practitioners investigated failed deals to figure out how the deal was analyzed, why it was accepted, and, sadly, what was ignored. Even if basic tenets of best practice financial modeling had been followed, models programmed to maximize short-term profit at the expense of sustainability planning self-destructed in an economic crisis.

For that reason, 2008 might well be recalled as the year stress-test modeling failed.¹⁰ (Modeling) failed those institutions who invested hoping it would transform risk management. Failed the authorities who had relied – perhaps over-relied – on the signal models provided about financial firms’ risk management capabilities. And, perhaps most important of all, failed the financial system as a whole by contributing, first, to the decade of credit boom and, latterly, the credit bust.¹¹ Perhaps above all, after every major disaster stress testing addresses regulatory objectives, which are not necessarily going-concern objectives. Regulatory objectives relate to the preservation of the financial system, prevention of systemic problems, and minimizing taxpayer cost; going-concern objectives, on the other hand, must address the ongoing integrity of an institution.¹² When tested against real stress—the crisis itself—large segments of the financial system panicked, and a large number failed.

Against that backdrop, now is as good a time as any to review model failure from a regulatory perspective: During the crisis, value-at-risk (VaR) models severely underestimated the tail events and the high loss correlations under systemic stress. VaR model workhorse for assessing risk in normal markets did not fare well in extreme stress situations. Systemic events occur far more frequently and the losses incurred during such events have been far heavier than VaR estimates have implied.¹³ Value at Risk (VaR) is based on normality and linked to historical statistical relationships. Under the normality assumption, the probability of large market movements is largely underestimated and, more specifically, the probability of any deviation beyond 4 sigma is near zero. In the real world, 4-sigma events do occur, and they certainly occur more than once every 125 years, which is the supposed frequency of a 4-sigma event (at a 99.995% confidence level) under the normal distribution. Even worse, the 20-sigma event corresponding to the 1987 stock crash is supposed to happen not even once in trillions of years. VaR failures led the Basel Committee to encourage banks to focus on rigorous stress testing that captures risk beyond the normality assumption, that is, extreme tail events and the requisite sustainability strategies.

Financial Models Breaking Down in the Equity Markets

A funny thing happens when we model: things that are supposed to happen, do.

The deficiencies of financial models and frequency of the so-called extreme outliers in the financial industry have been dissected over and over. These include, for example, the 1987 Program Trading Crash, the Internet/tech bubble of 2003, the Sub-Prime and Debt Crisis of 2008–2009, and the May 2010 Flash Crash. Mathematical models have had a proven history of breaking down, especially when they were needed the most. To be fair, in some situations the reason behind modeling failures was the onset of extreme and unprecedented events: e.g., events not observable or even imaginable prior to the event. Possibility of the extreme was rarely quantified. However, in innumerable other cases, these mathematical and financial models behaved badly because analysts made errors within an ill-conceived model structure, measured inappropriate historical look-back periods, and applied incorrect statistics. Analysts and quants blamed model breakdowns on regime shifts, structural change, or a black swan event. After eons of financial research and empirical studies, predicting when events will take place or even uncovering them at the onset is an extremely difficult undertaking; analysts learn of these events only after they have already taken place, and harm has occurred.

Emanuel Derman¹⁴ provided an overview of how models have broken down over the years and how incorrect application of these models helped lead to the financial crisis of 2008–2009. He traced the usage of quantitative models and corresponding improper insight beginning with the mortgage markets (2007) through the end of the financial crisis and directly questioned why bankers put so much faith in these models in the first place. The fundamental problem: We put too much reliance on models without adequate questioning and understanding.

Amir E. Khandani and Andrew Lo¹⁵ investigated the sudden devaluation of several quantitative long/short equity hedge funds. During the first week of August, many of these funds experienced unprecedented losses, and it has been postulated that this was due to a coordinated deleveraging of similarly constructed portfolios caused by a high correlation across quantitative strategies. The authors found evidence that the hedge funds began unwinding their portfolio position in July 2007 through the first week in August, after which there appeared to be a reduction in market-making activity, resulting in a liquidity crunch. One of the key takeaways from this paper is that even if the quantitative approach was properly structured with accurate model parameters, the models still broke down because the funds failed to consider the effect of using incorrect input variables in the model. That is, the strategies across many firms may have been highly correlated, with the majority of these firms selling shares at the same time, resulting in a large reduction in market liquidity. This in turn caused a drop in market price (temporary market impact) due to the liquidity discount required to offset the portfolio positions.

The models may have only considered the holdings of each individual fund and concluded that there was adequate market liquidity to offset positions during the specified timeframe. But if the models considered the effect of the aggregated sell-off, then the fund strategies may have been different. In short, the August 2007 Quant Meltdown (as it has become known) may have been caused by a gross understatement of the systematic market risk and the highly correlated strategies, e.g., inaccurate model input data. Bernard Donefer¹⁶ noted many aspects of trading algorithms and the data that they rely on to make execution decisions. Incorrect data, either due to bad calculations, simple data errors, or even untested trade code can dramatically affect the way electronic trading algorithms will behave in the market place in real-time. Modeling errors can cause a strategy to be highly profitable if we are lucky or to suffer a disastrous loss if we are unlucky. And if we are on the unlucky side of the trade, it may even threaten the funds viability and potentially force the company out of business (e.g., algorithm (algo) trading errors of August 2012).

Empirical analysis conducted over the 2008–2009 financial crisis revealed several real-world examples where models have broken down due to various reasons and have caused numerous issues. Some of these observations are described ahead. The financial crisis of 2008–2009 was accompanied by numerous unprecedented events (sub-prime defaults, company bankruptcies, etc.), and notably a large number of measured extreme price-movement days (Kissell, 2009). For example, over the 189 day period from July 2008 through March 2009, many financial professionals have reported that there were 44 days where the SP500 price index moved three standard deviations or more, i.e., 23% of the days experienced a three standard deviations move or more! However, this extreme price-movement period is highly exaggerated and was calculated using volatility computed over the period 1990 to June 2008, where volatility was 18.6% annualized and 1.17% daily. But even a short-term volatility measure, such as a 66-day rolling measure, results in a large number of calculated extreme price-movement days. For example, this model results in extreme price movements occurring 7 days or 4% of the 189 day period, which is still extremely high. A data point should only exceed a three standard deviations movement 0.27% of the time. This relates to approximately 1.5 days in a 250 trading-day year. Models that were using any historical volatility measure to gauge potential extreme price movement during the financial crisis were significantly underestimating the true potential of price swings and significantly underestimating the potential of extreme price movement and price surprises. Incorporation of the VIX volatility index into current volatility estimates, however, dramatically improves estimation accuracy, especially during periods of rapid price movement and regime shifts. Application of our VIX volatility adjustment (see Chapter 3) finds only one three standard deviations price-movement day in this period, which is consistent with statistical theory. Models that were relying on a historically based volatility measure (either long-term or short-term) during this period were significantly underestimating the likelihood of extreme price