Print this
Will Market-Neutral Strategies Survive LTCM?

Ezra Zask, principal at CP Risk Management, explains how to evaluate market-neutral trading strategies.

Market-neutral strategies and their close kin, relative-value, long-short and spread strategies, have long held out the promise of high returns, low drawdown and low volatility. Their popularity appears to be cyclical—typically, a period of crisis results in widening market spreads. This is usually followed by a period of relative market calm and stable correlations between markets, during which spread strategies appear almost too good to be true. Investors envision a hedged portfolio with rich spreads and a high likelihood of those spreads narrowing. No wonder these programs seem to be slam-dunks.

After the 1994 interest rate crisis destroyed Askin Capital Management, Long-Term Capital Management picked up the baton and touted the market-neutral approach. In a sense, LTCM was the culmination of all previous market-neutral programs. Major trading talent backed by Nobel prize-winning theorists and state-of-the-art computers were employed to scan the globe for opportunities in market-neutral and relative-value strategies. It was the most systematic and well-financed attempt ever undertaken to analyze correlations, volatilities and risk parameters in order to construct low-risk, high-return portfolios.

As we know, the fund’s results were staggeringly different from the theory. But how did other market-neutral funds fare during 1998? Was it an intrinsic flaw in market-neutral strategies or LTCM’s implementation of the strategies that spelled disaster? What will the role of these strategies be in the hedge fund universe of the future? What lessons have we learned about managing market-neutral hedge funds? Finally, what questions should investors and managers of these funds ask?

Performance of other funds

LTCM was caught in a global trend that one publication has described as the “crisis of the super-correlated credit bet.” The company invested in highly leveraged credit spreads that bet on relative improvements in the credit ratings of various instruments, including Danish mortgage bonds, Greek and Italian government bonds and U.S. junk bonds. In effect, LTCM had a one-way, undiversified bet that credit spreads would narrow. LTCM’s losses of 10 percent in June 1998 and 44 percent in August 1998 were the result of a global flight to quality that began with the Asian crisis and reached panic proportions following the Russian default last August.

The fact that many people were making the same bets became clear as other hedge funds (and many banks) reported losses. In aggregate, only 18 percent of hedge funds made money in August 1998, the worst performance since 1994. Convergence trade strategies at D.E. Shaw and Strategic Partners got caught in the same trap. So did Warren Moser’s III Global Fund, which lost on Danish mortgage securities, and Eagle Capital Management, which went under on hedged convertible-bond positions.

In aggregate, however, market-neutral hedge funds lost less than 4 percent during the fourth quarter of 1998—nothing resembling the losses cited earlier. The key difference appears to be the leverage position established by LTCM and the concentration of its portfolio on credit spreads. LTCM’s leveraged position was reportedly as high as 100:1, and its efforts at diversifying risk in equity spreads—including pair-trading of two stocks with high correlation—only worsened their losses.

How did things go so wrong?

The problem with market-neutral spread strategies stems from their deceptive internal hedge. It is relatively easy to construct and find spread strategies that have high returns and stable relationships, at least over a medium-term time frame. It is also relatively easy to construct a diverse portfolio of positions from all over the globe and from various markets. Again, for specific time periods, it is possible to find positions with low correlation, minimizing the risk of the portfolio moving in the same direction.

The next step in the process, however, is the one that normally gets funds into trouble. Convinced that their statistical models have found a portfolio with low risk, high return and limited downside, some funds go the next step and apply a degree of leverage that their models show will lead to losses within an acceptable limit. In other words, they ask the question, “What leverage can I get away with?” The answer determines the ratio.

The biggest mistake in this strategy is that it views the relationships between instruments and spreads as stable under all conditions, including financial meltdowns and tectonic shifts in financial markets. During the crisis, however, assumptions about correlations, hedges, liquidity, country default and credit risks all fell apart simultaneously.

In essence, the models used by LTCM viewed statistical relationships as mechanical and constant rather than time-bound and variable. The models thus underestimated extreme events and overestimated the inevitability of spreads behaving along historical patterns. In summary, as noted by the International Monetary Fund, the programs were short the human judgment required to implement these technologies and assess the economic and financial risks.

Where do we go from here?

Investors interested in pursuing market-neutral and relative-value strategies will have to perform a greater level of due diligence. The only chance for breaking the cycle of boom and bust for these programs is to incorporate the mistakes of the past and integrate them into a coherent checklist of questions to ask when contemplating an investment in these programs.

Question 1: What are the assumptions behind the models used to establish the spread positions? All models make dramatic assumptions and simplifications in order to capture reality in a set of numbers. These assumptions include the distribution of risk (normal or non-normal), direction of markets, stability of patterns of association and so on. An investor needs to understand the precise nature of these assumptions, their impact on the program’s risk management and, most important, the fund’s contingency plans in case their assumptions prove to be wrong.

Some funds apply a degree of leverage that their models show will lead to losses within an acceptable limit.

Question 2: What is the rationale for the spread? This is a key question that is rarely asked. However, every spread should have a solid story that includes information on what, exactly, is being spread apart. Further, the spread needs a rationale, whether statistical, economic or technical. For example, the rationale for credit convergence spreads needs to be based on one of two arguments: the mean reversion position argues that spreads will converge or diverge based on their historical relationship; the economic fundamentals argument posits that economic factors will lead to changes in credit risk and, in turn, to spread convergence or divergence. For example, the rationale for convertible arbitrage positions is based on a precise mathematical relationship between convertible bonds and shares, since bonds are convertible into shares.

Credit convergence trades are more difficult. To paraphrase Myron Scholes, convergence trades need a date with destiny, or a time when two securities will converge and there will be a high probability that the bet will pay off. If the spread does not meet this hurdle, it is incumbent on the manager to provide the reasons he or she expects the spread to succeed and the underlying assumptions behind this spread.

Question 3: What risk management measurement and strategies are used by the fund? A wide variety of strategies are now available to hedge fund managers. Investors should make sure that, at a minimum, a value-at-risk approach is used. In addition, portfolios need to be subjected to rigorous stress testing and scenario analysis to simulate the effect of financial crises on the portfolio. Finally, a contingency plan based on these analyses should be articulated that includes such worst-case assumptions as reduced liquidity, increased margin calls and an increase in the correlation between instruments.

Question 4: How is the human factor integrated into the decision-making process? If the financial crises of the past few years should have taught us anything, it is that models can only provide the framework for analysis. At the end of the day, the human factor needs to be incorporated into decision-making. A combination of statistical analysis, experience and imagination is needed to take into account historical data when projecting future risks. Investors should carefully determine the experience of fund managers and the integration of their experience in decision-making.

Zask can be reached at zask@chipar.com.

Table 1: Market-Neutral, Relative-Value and Spread Strategies
Although these terms are used interchangeably, they are, in fact, distinct strategies with widely different structures, risks and management strategies.

Relative-value or spread trades: The most basic spread trade consists of a long position in one instrument offset by a short position in another. (A portfolio consists of multiple spread positions.) The portfolio profits if the manager’s relative valuation leads to a rise in the value of the long position(s) and/or a decline in the value of the short position(s)—in other words, as the spread widens or narrows. Relative-value strategies differ from “directional” investment, where the gain or loss comes from the rise or fall in price of a single asset rather than the relative movement of two or more assets. At least in theory, this makes them less volatile.

Types of Spreads: Spreads are viable in all markets: equity, fixed income, currency and commodity. In general, they fall into the following types:

Type Examples
Intracommodity Calendar spreads
Yield-curve spreads
Intercommodity Junk bond/Treasury spread
Currency cross-rate spreads
Intermarket Equity/interest rate spreads
Soy/corn spreads
Cattle arbitrage
Source product Crack spreads (oil)

Market-neutral: Market-neutral strategies are a special subcategory of spreads. They are constructed so that a portfolio’s assets are relatively insensitive to the movements of major markets—whether equities, interest rates, currencies or commodities. In contrast to relative-value strategies, in which two or more offsetting assets are meant to neutralize or reduce risk, market-neutral strategies attempt to reduce risk to a wider market movement.

Remunerating Performance

Kevin Dowd, a professor at the University of Sheffield, explains how to use a variation of the Sharpe ratio to compensate traders for their profits.

One of the most important issues in managing derivatives trading is that of performance evaluation: how to evaluate past performance, with a view toward remuneration. Clearly, managers want to reward traders for earning high returns. If they reward only high returns, however, managers effectively encourage traders to take large amounts of risk, since traders can usually only earn higher returns by taking greater risks. Remuneration policies that reward returns but ignore risks—such as profit-related bonuses—therefore encourage more risk-taking than is in the firm’s best interest.

How do we evaluate the performance of different traders and take into account not just the returns earned, but also the risks assumed to obtain those returns?

There are a number of other rules, but most of these have their own problems. For example, the risk-adjusted return on capital (RAROC) measure—the ratio of return over the value-at-risk—attributes excessively large risk-adjusted returns to safe investments. Similarly, the information ratio—the ratio of return over the standard deviation of return—is known to be unreliable.

What is needed is a remuneration policy that encourages traders to seek profits, but also allows for the risks they take in doing so. But how do we evaluate the performance of different traders (or different trading portfolios) and take into account not just the returns earned, but also the risks assumed to obtain those returns? Put more simply, how do we adjust returns for risk?

Perhaps the most common answer to this problem is to use the Sharpe rule, first suggested by William Sharpe in 1966. Suppose we have to rank a number of past investments in terms of their risk-adjusted returns. These investments might be individual investments, or whole portfolios, such as those amassed by different traders over some period of time. The Sharpe rule says that we should rank the performance of these investments by means of their Sharpe ratios, where the Sharpe ratio for each investment is the ratio of its excess return—its return over and above some benchmark return—divided by the standard deviation of its excess return. This benchmark could be some stock market or sector index, the cost of capital or some other measure of the opportunity cost of funds invested.

To use the Sharpe rule, we select a benchmark, measure the various returns after the event and construct the excess returns and standard deviations. We then construct the Sharpe ratios and rank the investments accordingly. A high Sharpe ratio is good because it implies high return, low risk, or both; a low Sharpe ratio is bad because it implies a low return, a high risk or both. The investment with the highest Sharpe ratio therefore has the biggest risk-adjusted return, the investment with the second-highest Sharpe ratio has the second-biggest risk-adjusted return and so forth.

A limitation in using the Sharpe rule, however, is that it is only reliable if various investments have returns that are uncorrelated with the return on the existing portfolio. Once we have non-zero correlations—most particularly, once there are investments with different degrees of correlation with the existing portfolio—the Sharpe ratio can become unreliable and give incorrect risk-adjusted rankings.

For instance, suppose we are ranking two investments, A and B. Investment A generated a slightly higher excess return than B, but both had the same standard deviation of excess returns. Its slightly higher return therefore gives A a slightly higher Sharpe ratio than B. Using the Sharpe rule, we would determine that A had a slightly higher risk-adjusted return than B.

Now suppose that A had a return that was strongly positively correlated with our existing portfolio, while B had a return that was strongly negatively correlated with our existing portfolio. Investment A might be a speculative position that significantly increased our overall portfolio risk, while B might be a hedge position that decreased our overall risk. The net result is that A generated a slightly higher return than B, but A increased our overall risk exposure and B decreased it. It is now clear that investment B generated the higher risk-adjusted return, even though A has a slightly higher Sharpe ratio. The Sharpe rule is therefore unreliable, because it ignores the differing correlations of the two investments with our existing portfolio. Since it ignores these correlations, the Sharpe ratio tends to understate the risk associated with investment A and overstate the risk associated with investment B. A Sharpe-ratio ranking may no longer give us a reliable indication of which investment was best.

Fortunately, this problem has a straightforward solution. Since the traditional Sharpe rule compares different investments but ignores their correlations with our existing portfolio, we need to find some way to capture those correlations within the alternatives we wish to consider. The solution is simply to redefine those alternatives to include our existing portfolio. Instead of comparing A and B and ignoring our existing portfolio, we compare A plus the existing portfolio—let’s call this portfolio A*—against the alternative of B plus the existing portfolio, which we can call B*. Next, we apply the earlier Sharpe rule to portfolios A* and B*, and rank them accordingly. If A* ranks higher than B*, we can conclude that A had a higher risk-adjusted return than B, and vice versa.

The sharpe rule is only reliable if various investments have returns that are uncorrelated with the return on the existing portfolio.

This new rule will give the same rankings as the traditional Sharpe ratio if both investments are uncorrelated with our existing portfolio, but can give differing rankings in the more usual cases where this zero-correlation assumption does not hold. In such cases, our new Sharpe rule will still be reliable, even though the traditional one will not be. We thus have a new version of the Sharpe rule—a generalized Sharpe rule—that overcomes the main limitation of the traditional Sharpe rule and is valid regardless of the correlations of different investments with our existing portfolio.

Implementation issues

Of course, there still remain a number of implementation issues, but these are more or less easily dealt with:

  • We would have to choose a benchmark. However, selecting a benchmark is nothing new, and we would usually have to select some benchmark anyway, even if we did not use any form of Sharpe rule.
  • We would need data to implement any Sharpe rule, but we already have the data needed (that is, past observations of various returns).
  • We have to decide when and how frequently we wish to evaluate performance. Perhaps the most convenient approach is to do so at the end of each day, with each investment, portfolio and trader evaluated in light of the institution’s overall portfolio at the start of each day. In other words, we start each day by telling each trader (or asset manager, or whatever) what the overall portfolio is, and then apply the generalized Sharpe rule at the end of the day to evaluate his or her performance.
  • Finally, having produced a ranking of risk-adjusted returns by each unit, we need to apply some remuneration schedule to it to determine what the actual remuneration for each unit should be. My advice would be to apply a simple schedule, such as a linear one.

Kevin Dowd can be reached at kdowd@sheffield.ac.uk.

Taking the Stress Out of Stress Testing

Shankar Nagarajan, vice president of risk management advisory at Bankers Trust, explains why worrying about triple-twists in the yield curve may not be enough.

Another crisis and another round of hand-wringing. The aftermath of the Asia-Russia-flight-to-quality-LTCM crisis has given rise to new soul-searching among risk managers. Did we anticipate this? Did we plan for it? Are our models good enough? Did we stress-test our models for these scenarios? The finger-pointing has begun and heads have started to roll. And now a recent poll has even managed to elicit mea culpas from many risk managers who have not stress-tested twists in their volatility curves or correlations. Is all this a bit overdone?

Surely, most financial institutions put their models through standard recommended stress tests by now—parallel moves, twists in yield curves, spread widening, currency depreciation and so on. And it is not unreasonable to expect scarier scenarios such as volatility twists and correlation shocks to be tested as well. As the debate turns to the technical minutiae of stress testing, it is well worth keeping a little perspective.

What is risk?

This question may sound a bit silly in the current environment, but it is not. Clearly, one can conjure up any number of scary events (individual and joint) that may have potential profit-loss consequences, but not all of which are worth testing for. Implicit in any stress-testing scenario is an assumption about the likelihood of an event and its relevance to the firm’s portfolio. Most objective measures of riskiness rely on empirical evidence, as does our subjective perception of risk. This means that past events are considered likely, while we have no clue about hypothetical events that have never happened. Fundamentally, we have no way to judge whether, for example, there is a risk of the sun not coming up tomorrow—until, of course, it finally doesn’t. Until Russia defaulted, there had never been an instance of a sovereign defaulting on domestic debt, but not on foreign debt, as historically most sovereign defaults were the other way around. Needless to say, few risk management models even perceived it as a risk. The point is, despite all the usual 20-20 recommendations, there is no objective way to identify a small set of likely, scary and relevant scenarios.

You have run through the scary scenarios. So what? What is the firm to do if a stress test reveals that the capital needs to be doubled or certain large (and currently profitable) positions need to be cut in half to survive the stress event? The typical response of the senior management is, “We now know how poorly we will fare in this event, but how likely is this event in the first place?” The answer, usually, is small. The more extreme the stress event, the smaller the likelihood. Which bank wants to triple the capital or exit a major market segment on the basis of a minuscule likelihood of bad events? The most likely action: yet another risk report gets filed away. This brings us to the next point.


Rather than test for hypothetical scenarios, which may elicit big yawns from management committees, many risk managers, perhaps cautioned by Voltaire, test for past crises. Technically, the advantages of a historical crisis scenario (such as the 1998 meltdown) over individual stress events (such as spreads widening by 100 basis points) are that the former captures correlated effects across markets reasonably well and, moreover, every new crisis adds to the list of likely events for future testing. But the real advantage is that it gets the senior management’s attention.

Unfortunately—and partly because market participants learn from past mistakes—new crises rarely tend to be exactly like the old ones, and the story goes on and on. Like military generals, risk managers seem to be continually preparing to fight the last war. The moral for the risk manager: don’t cry wolf too often.

Exit strategy

Now that value-at-risk is so popular, it is worth remembering an important parameter that rarely gets discussed: the time horizon over which VAR is computed. This is supposed to reflect a time period that allows for an orderly liquidation of the position in question. Typical choices are a week or 10 days. While that time frame may be appropriate for highly liquid instruments such as G-7 currencies, it is highly questionable for emerging and other markets where liquidity can dry up quickly, and orderly liquidation becomes an oxymoron. Perhaps one day is a better choice in some cases. Rather than worry about the profit-or-loss consequences of a sudden widening of spreads by 100 basis points, perhaps a better strategy is to have contingency plans ready for gradual unwinding or to hedge the positions as spreads widen by, say, 20 basis points, 40 basis points and so on. Readers may recognize this as a crude portfolio insurance strategy. Unfortunately, as is thought to have happened in the stock market crash of 1987, such strategies don’t work if everyone is stampeding for the same exit at the same time. The LTCM bailout was intended precisely to avoid this problem.

Finally, no amount of stress testing can save a firm from the most scary and dangerous event of all—management failure. Most serious losses in firms can be traced to deliberate and conscious policy choices made at the highest levels to enhance the firm’s risk profiles in certain markets. One does not need a degree in astrophysics to understand the potential risks in these strategies. As they say in Turkey, “if you can’t take the heat, get out of the bath.” In the financial world, this often happens only after a blood bath.

Disclaimer: The views expressed in this article are those of the author’s and do not necessarily reflect those of Bankers Trust or its subsidiaries.

Nagarajan can be reached at shankar.nagarajan@bankerstrust.com.

Was this information valuable?
Subscribe to Derivatives Strategy by clicking here!