Everybody complains about increasing contagion, but can anybody do anything about it?

PARTICIPANTS

Joe Kolman, Moderator, editor, Derivatives Strategy Alexander Arapoglou, treasurer, Fuji Capital Markets Corp. Bluford Putnam, president, CDC Investment Management Corp. William Ferrell, president and chief investment officer, Ferrell Capital Management Farid Naib, CEO, FNX Mark Galant, president, FNX Ram Shivakumar, senior consultant, Arthur Andersen Lisa Polsky, managing director, Morgan Stanley Dean Witter

Joe Kolman: There's a general sense that the contagion effect in the financial markets is increasing. Are correlations in fact changing? If so, to what do we attribute this?

Alexander Arapoglou: One of the things we've seen in the last 15 years is changing patterns in asset ownership, which lead to more than historical correlation and contagion. When one person buys something from a broker-dealer, the head of the hedge fund sales desk may say, “Oh my goodness, what a great idea. Why don't I tell all my other clients about this wonderful idea?” It may not be exactly the same trade, but it's similar. So the idea tends to spread.

Initially, you may have had one set of correlations that was based on one group of investors. But now, five years later, you've got a completely different set of people holding that asset, all motivated by the same thing. As soon as that trade runs into trouble, everybody's going to go back to that same hedge fund sales desk and say, “Oh, please, take me out.” Now where will the firm's proprietary traders make markets? The bids are certainly going to be way back.

“The head of the hedge fund sales desk may say, ‘What a great idea. Why don't I tell all my other clients about this?'” —Alexander Arapoglou

Bluford Putnam: I would like to look at which kinds of portfolios were most negatively affected by the contagion phenomena and which were not. One can divide the portfolios into three camps. First, the long-only portfolios were obviously dead, since equity and credit markets declined and correlations between these exposures rose in the August–October 1998 crisis. In fact, as correlations and volatilities rose, the riskiness of the long-only portfolios increased dramatically.

The second group of portfolios is what I will call the yield/carry strategies. Although these are long/short portfolios, they behave just like long-only portfolios, because the short positions in U.S. Treasuries are there to finance cheaply a set of higher-yielding long positions and not to provide diversification. The hedge funds that were running yield/carry positions got killed as correlations and volatilities rose in their long positions.

The funds that did surprisingly well were the ones that were truly long/short and that had truly diversifying short positions. That is, their short positions were in securities and markets that were expected to perform less well in a capital gains sense. So when the correlations between exposures rose in these types of portfolios, there was not necessarily an increase in volatility, because the gains in the short positions offset the losses in the long positions.

The thing we learned is that you have to study the type of diversifying short positions you put in your portfolio much more carefully. If you are just short U.S. Treasuries to fund a yield/carry trade, then you are extremely vulnerable to the contagion effect, as well as the flight-to-quality effect that often occurs in a market crisis.

Contagion is a phenomenon that occurs mostly within one asset class. That is, all country equity markets went down together in August 1998. Currencies, however, such as the dollar, yen, euro-zone currencies, or sterling, did not move in lockstep. And with the flight to quality, G-7 government bonds rallied while credit-related fixed-income exposures were hit hard. So if you are taking advantage of trades across asset classes in a true long/short portfolio, you have a fighting chance to come out of a contagion problem looking pretty good.

William Ferrell: I'll say in response to Blu's comment that he's perfectly right. And that's the reason that, as a group, the established long/short managers have lower volatilities and better Sharpe ratios than traditional long-only managers. Establishing long/short positions is the best way to construct a portfolio, in part because market spikes and the increased correlations of markets that accompany such shocks do not put stress on the long/short portfolio. In extreme cases of market illiquidity, you don't have to cover until market liquidity improves.

One thing that hasn't been discussed so far is the value of stress tests. Value-at-risk can be modified quite a bit by using short-term and long-term lookbacks, by using Monte Carlo scenarios and by stress-testing the two basic components—the volatility of volatility and the volatility of correlation. The only way to be practical in managing risk in a large portfolio is to first assess the key drivers—that is, those elements of market risk that will, along with the expected correlations to other portfolio components, determine the changes in net asset value for the portfolio. Once the key drivers have been identified, the best way to set parameters on the range of future performance is to stress-test the foundations of the VAR analysis. Analyzing the impact of changes in the rate of volatility and changes in the degree of correlation to the rest of the portfolio mitigates the risk that historical data do not always reliably predict the future. After the key driver stress test, the portfolio manager can determine if action is necessary to protect the portfolio from market changes and use the analysis to identify the risks that need to be mitigated.

The question to ask is not what happens if all the correlations go to 1. The question is what happens if correlations change. You need to know how much of your portfolio risk is built on the fact that you're counting on correlations being constant.

We all know that emerging markets have low correlations to developed countries for about 20 minutes out of every year. You can't average these correlations. You have to analyze the range of correlations to understand fully how reliable—or, in the case of emerging markets, how unreliable—correlations can be.

The issue of assessing liquidity ties in closely with what other market participants are going to do. In any portfolio, you know there are certain instruments that are going to become less liquid than others if there's a market spike or shock. And you can almost count on the fact that market participants are going to sell the liquid instruments first. So you want to figure out what kind of relationship there is between your own portfolio and that of the portfolio that's widely held in the marketplace, whose liquidation might have a negative impact.

Farid Naib: I'm wondering how contagion has been encapsulated in pricing. If there is increased contagion in the world, you'd expect there to be more external shocks to those markets. And if there are more external shocks, you probably end up with a more leptokurtic distribution than a normal distribution. I've also started thinking that the volatility smile might be a little bit steeper, with out-of-the-money options priced a little higher than at-the-money options were 10 or 15 years ago. But I have no idea if that's true.

Mark Galant: There's certainly a leptokurtic effect for countries that are in play. I remember that before the European monetary union crisis in 1992, both the skew and the steepness of the leptokurtic distribution increased dramatically. So this might be useful when discussing what might happen, before any damage has occurred. What I want to know is what the volatility surface looked like before any problems broke out in Asia in 1997. I don't know.

“A portfolio denominated in currencies that experience a crash will generally have a return distribution that is more peaked than the normal distribution. As a consequence, VAR estimates will grossly understate potential losses.” —Ram Shivakumar

Arapoglou: One problem with trying to predict the future is that this Latin American crisis is quite different from previous ones. If everything is attributable to changing patterns in asset ownership, once somebody has been burned once, he or she is less likely to get into the same trade a second time. So even though you may have contagion, the degree of increase in correlation is not likely to be as great. How much not as great? Who knows.

I also think it's going to be difficult to choose exactly the right scenario. Everybody looks at historical scenarios. What happens if we have a crash like 1987? What happens if we have an EMU crisis? But the fact is, it's difficult for lightning to strike twice in the same place.

Ram Shivakumar: Many trading organizations and corporations are still using what may be called first-generation models for computing portfolio risk. These models are predicated on two key assumptions—that portfolio returns are normally distributed, and that the mean and standard deviation of portfolio returns and the correlation between securities in the portfolio are stable.

These assumptions are questionable even during “tranquil” periods when one is evaluating portfolios with exposure to currency risk. During periods of turmoil, the assumptions are palpably false, and, as a consequence, the risk measurement models are completely unreliable.

Consider what crashes imply for the measurement of VAR if one is using a first-generation risk-measurement model. A portfolio whose elements are denominated in currencies that experience a crash will generally have a return distribution that is more peaked than the normal distribution—and with fatter tails than the normal distribution. As a consequence, VAR estimates from the standard risk measurement model will grossly understate potential losses, while sensitivity analyses that assume a one-standard-deviation change will overstate potential losses.

Crashes have important implications for regulatory capital. If a bank were to use this simple model to determine its regulatory capital, it would be woefully unprepared to deal with crashes. On the other hand, the required capital to comply with regulatory requirements would be extremely large if one were to factor in crashes.

Putnam: I think the regulatory authorities are going to wake up to the fact that the risk methodologies they have been pushing have some flaws. The VAR methodologies they have advocated have tended to be based on risk measurement calculations using high-frequency daily data, for example. Daily data underestimate correlations, particularly in global portfolios, because of the time-zone effects. That is, something could happen in the United States after Europe and Japan are closed. The markets will catch up the next day, but in a daily VAR system it is close of business prices that matter. Thus, the markets will appear less correlated than they actually are. Time-zone effects are minimized in weekly or monthly data. If you are using hundred-day historical data for your VAR calculations, you are going to have correlations that look lower than they truly are, and your portfolio will appear less risky than it really is. That is why you have to stress-test with a scenario that implies a higher correlation among your exposures to see how your portfolio reacts in a contagion-type crisis.

“You don't stress-test everything. You only stress-test those elements of the portfolio that make the portfolio value move up and down the most.” —William Ferrell

I also get the impression that some people think contagion should not happen. They think, If Thailand goes down, why should Argentina be affected?

I want to point out that investment managers are in the business of allocating capital. If the market takes the price way down on one asset, in my mind the market has also taken the risk way down. I do not care what your historical VAR says. My forward-looking view is that when the price of a security is cut in half, there is probably less risk in that security than there was the week before and more potential for future capital gains.

Now the security with the new and lower price looks more attractive to me relative to the rest of the market, since I am assuming other securities still have higher prices, lower expected returns and higher risks. Even if the fundamentals are totally different, and even if the causes are totally different, it is time for me to adjust my portfolio to consider the new relative risk/return profile in the market. Whenever one asset's risk/return profile changes, my portfolio changes.

So contagion, at least in my definition, is built into the way financial markets work, which is to consider efficiently the opportunity costs of the positions one does not have. As markets have become more efficient, however, contagion happens faster and is more obvious now, but contagion is part of the natural asset allocation process and it is a characteristic of efficient markets, not a symptom of market deficiencies.

Lisa Polsky: Correlations increase in a market panic as a result of the rational behavior of market participants. If you have a position in a market that has dislocated, in which liquidity is extremely low and bid/offer spreads have therefore widened dramatically, you ask yourself, “Would I rather pay this bid/offer spread, or would I rather take the tracking risk in hedging against something correlated? When people make these trade-offs, when they decide that the cost is so large that they'd rather take on the tracking or basis risk by hedging with the correlated asset, then their actions make everything move together.

“If you are just short U.S. Treasuries to fund a yield/carry trade, then you are extremely vulnerable to the contagion effect, as well as the flight-to-quality effect that often occurs in a market crisis.” —Bluford Putnam

You're always balancing costs, risks, returns and opportunities. When these relationships get dramatically out of line, market participants bring them back into line by their actions, which implicitly say, “This cost is too high relative to the risks, or this risk is too high relative to the return.”

Shocking correlations is an extremely important thing to do, because we all know that markets are more correlated when there's a dislocation. The scenarios that will get you in big trouble will generally be those in which markets are abnormally highly correlated, because it's generally the alignment of two or more risk events simultaneously that spells disaster. Most firms calculate the profit-and-loss impact of abnormally large moves on each risk silo. But many then make the mistake of assuming that they are well-diversified because they are aggregating their stress tests using normal correlation assumptions. When you run market stress tests, you need to look at the results in the context of stressed correlations, not normal correlations. You can look back to other time periods when markets have dislocated and look at the correlation matrix in those stressed time periods to give you an idea of what could happen.

Ferrell: That is really the beauty of the stress test. The historical data are important, but if you stress-test, you cover the flaws in empirical data. By using a different set of data, based on a hypothesis about what the future might bring that is different from the past, you learn how the risk of changing volatilities and correlations affect the performance of your portfolio.

Take a look, for example, at AAA corporates and U.S. Treasuries, which have traditionally had a high correlation. All of a sudden, because of credit pressure, they may have a low correlation. Spread risk can be analyzed and quantified just as easily as volatility risk. If your assumption is that the correlation of a portfolio is no less than 0.9, and it goes to 0.8, 0.7 or 0.6, or becomes negative, you can measure how much money the change in correlation will cost you.

Conversely, if you set up a portfolio with a group of assets that have low correlations, you know that the risk compression—that is, the difference between the gross amount of risk in the portfolio and the net amount of risk in the portfolio—is highly dependent on the zero correlation you're assuming for the asset classes. But if that low correlation moves to +1 or -1, then the portfolio's risk characteristics change completely. That's why stress testing for correlation is just as important as any stress testing you do on volatility.

Kolman: Are there limits you can give as far as what one shouldn't do or what one should watch out for? Theoretically, if you stressed correlations enough, you'd get a number that would be worthless or that would prevent you from trading at all.

Ferrell: You don't stress-test everything. You only stress-test those elements of the portfolio that can move to produce the biggest change in asset value—the key drivers. Those are the components of the portfolio that make portfolio value move up and down the most. If you can manage the risk profile of the key drivers, you can achieve a target volatility for the performance of your portfolio.

Kolman: Have we reached some sort of conclusion about modeling liquidity? I think there's a consensus that liquidity has an extremely important effect on contagion, but that it's hard to nail things down in a model. Are there some other techniques we can use?

Polsky: I tend to start at the micro and work my way up to the macro, because risk management at the aggregate level is the sum of all the little risks. First of all, you have to understand where you have large, illiquid, concentrated positions and what it costs to unwind them. Unwind costs are a function of the size of your position relative to the normal size of trades in the market. The bid/offer spread you observe on a screen is the unwind cost for normal market-size transactions. Ask yourself, How would the spread widen as I increase the size of my position? This will help you measure the liquidity risk in your portfolio.

When you look at your portfolio, you should be able to say, “Here are my major exposures, here are my risk concentrations and here are the potential unwind costs of these positions.” You can then measure the liquidity-adjusted risk in your portfolio.

On most trading desks, you can get a fairly good feel not only for your own concentration risk but also for how much is out there in the market and how you will be affected if everybody has to liquidate simultaneously.

This is clearly a critical issue in assessing extreme market moves and credit risk, because these are situations in which everybody will be trying to get through the same door at the same time. The key issue in understanding your contagion risk is evaluating your profit-and-loss risk in situations where the time to unwind is abnormally long. A one-day VAR doesn't help you here—you need to run multiday unwind scenarios, evaluating the number of days to unwind your position market by market, as well as by size of position within a market in which you have major concentrations. If you think the whole market has the same position and you're evaluating contagion risk, then you need to increase your assumption about the number of days to unwind commensurately with your assumption of market illiquidity.

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