Sunday, September 23, 2007

The Myth of Noncorrelation

[This is a modified version of an article I wrote that appeared in the September issue of Institutional Investor].

With the collapse of the U.S. subprime market and the aftershocks that have been felt in credit and equity markets, there has been a lot of talk about fat tails, 20 standard deviation moves and 100-year event. We seem to hear such descriptions fairly frequently, which suggests that maybe all the talk isn’t really about 100-year events after all. Maybe it is more a reflection of investors’ market views than it is of market reality.

No market veteran should be surprised to see periods when securities prices move violently. The recent rise in credit spreads is nothing compared to what happened in 1998 leading up to and following the collapse of hedge fund Long-Term Capital Management or, for that matter, during the junk bond crisis earlier that decade, when spreads quadrupled.

What catches many investors off guard and leads them to make the “100 year” sort of comment is not the behavior of individual markets, but the concurrent big and unexpected moves among markets. It’s the surprising linkages that suddenly appear between markets that should not have much to do with one other and the failed linkages between those that should march in tandem. That is, investors are not as dumbfounded when volatility skyrockets as when correlations go awry. This may be because investors depend on correlation for hedging and diversifying. And nothing hurts more than to think you are well hedged and then to discover you are not hedged at all.

Surprising Market Linkages

Correlations between markets, however, can shift wildly and in unanticipated ways — and usually at the worst possible time, when there is a crisis with volatility that is out of hand. To see this, think back on some of the unexpected correlations that have haunted us in earlier market crises:

  • The 1987 stock market crash. During the crash, Wall Street junk bond trading desks that had been using Treasury bonds as a hedge were surprised to find that their junk bonds tanked while Treasuries strengthened. They had the double whammy of losing on the junk bond inventory and on the hedge as well. The reason for this is easy to see in retrospect: Investors started to look at junk bonds more as stock-like risk than as interest rate vehicles while Treasuries became a safe haven during the flight to quality and so were bid up.
  • The 1997 Asian crisis. The financial crisis that started in July 1997 with the collapse of the Thai baht sank equity markets across Asia and ended up enveloping Brazil as well. Emerging-markets fund managers who thought they had diversified portfolios — and might have inched up their risk accordingly — found themselves losing on all fronts. The reason was not that these markets had suddenly become economically linked with Brazil, but rather that the banks that were in the middle of the crisis, and that were being forced to reduce leverage, could not do so effectively in the illiquid Asian markets, so they sold off other assets, including sizable holdings in Brazil.
  • The fall of Long-Term Capital Management in 1998. When the LTCM crisis hit, volatility shot up everywhere, as would be expected. Everywhere, that is, but Germany. There, the implied volatility dropped to near historical lows. Not coincidentally, it was in Germany that LTCM and others had sizable long volatility bets; as they closed out of those positions, the derivatives they held dropped in price, and the implied volatility thus dropped as well. Chalk one up for the adage that markets move to inflict the most pain.

And now we get to the crazy markets of August 2007. Stresses in a minor part of the mortgage market — so minor that Federal Reserve Board chairman Ben Bernanke testified before Congress in March that the impact of the problem had been “moderate” — break out not only to affect other mortgages but also to widen credit spreads worldwide. And from there, subprime somehow links to the equity markets. Stock market volatility doubles, the major indexes tumble by 10 percent and, most improbable of all, a host of quantitative equity hedge funds — which use computer models to try scrupulously to be market neutral — are hit by a “100 year” event.

When we see this sort of thing happening, our not very helpful reaction is to shake our heads as if we are looking over a fender bender and point the finger at statistical anomalies like fat tails, 100-year events, black swans, or whatever. This doesn’t add much to the discourse or to our ultimate understanding. It is just more sophisticated ways of saying we just lost a lot of money and were caught by surprise. Instead of simply stating the obvious, that big and unanticipated events occur, we need to try to understand the source of these surprising events. I believe that the unexpected shifts in correlation are caused by the same elements I point to in my book as the major cause of market crises: complexity and tight coupling.

Complexity

Complexity means that an event can propagate in nonlinear and unanticipated ways. An example of a complex system from the realm of engineering is the operation of a nuclear power plant, where a minor event like a clogged pressure-release valve (as occurred at Three Mile Island) or a shift in the combination of steam production and fuel temperature (as at Chernobyl) can cascade into a meltdown.

For financial markets, complexity is spelled d-e-r-i-v-a-t-i-v-e-s. Many derivatives have nonlinear payoffs, so that a small move in the market might lead to a small move in the price of the derivative in one instance and to a much larger move in the price in another. Many derivatives also lead to unexpected and sometimes unnatural linkages between instruments and markets. Thanks to collateralized debt obligations, this is what is at the root of the first leg of the contagion we observed from the subprime market. Subprimes were included in various CDOs, as were other types of mortgages and corporate bonds. Like a kid who brings his cold to a birthday party, the sickly subprime mortgages mingled with these other instruments.

The result can be unexpected higher correlation. Investors that have to reduce their derivatives exposure or hedge their exposure by taking positions in the underlying bonds will look at them as part of a CDO. It doesn’t matter if one of the underlying bonds is issued by a AA-rated energy company and another by a BB financial; the bonds in a given package will move in lockstep. And although subprime happens to be the culprit this time around, any one of the markets involved in the CDO packaging could have started things off.

Tight Coupling

Tight coupling is a term I have borrowed from systems engineering. A tightly coupled process progresses from one stage to the next with no opportunity to intervene. If things are moving out of control, you can’t pull an emergency lever and stop the process while a committee convenes to analyze the situation. Examples of tightly coupled processes include a space shuttle launch, a nuclear power plant moving toward criticality and even something as prosaic as bread baking.

In financial markets tight coupling comes from the feedback between mechanistic trading, price changes and subsequent trading based on the price changes. The mechanistic trading can result from a computer-based program or contractual requirements to reduce leverage when things turn bad.

In the ’87 crash tight coupling arose from the computer-based trading of those running portfolio insurance programs. On Monday, October 19, in response to a nearly 10 percent drop in the U.S. market the previous week, these programs triggered a flood of trades to sell futures to increase the hedge. As those trades hit the market, prices dropped, feeding back to the computers, which ordered yet more rounds of trading.

More commonly, tight coupling comes from leverage. When things start to go badly for a highly leveraged fund and its collateral drops to the point that it no longer has enough assets to meet margin calls, its manager has to start selling assets. This drops prices, so the collateral declines further, forcing yet more sales. The resulting downward cycle is exactly what we saw with the demise of LTCM.

And it gets worse. Just like complexity, the tight coupling born of leverage can lead to surprising linkages between markets. High leverage in one market can end up devastating another, unrelated, perfectly healthy market. This happens when a market under stress becomes illiquid and fund managers must look to other markets: If you can’t sell what you want to sell, you sell what you can. This puts pressure on markets that have nothing to do with the original problem, other than that they happened to be home to securities held by a fund in trouble. Now other highly leveraged funds with similar exposure in these markets are forced to sell, and the cycle continues. This may be how the subprime mess expanded beyond mortgages and credit markets to end up stressing quantitative equity hedge funds, funds that had nothing to do with subprime mortgages.

All of this means that investors cannot put too much stock in correlations. If you depend on diversification or hedges to keep risks under control, then when it matters most it may not work.

4 comments:

  1. A corollary to the myth of non-correlation is that there are now so many leveraged players that no market is safely isolated. Take muni bonds -- no foreign holders, dominated by individuals and retail mutual funds that are agents for individuals: yet munis still took a hit (short-term, to be sure) in August. It appears that some hedge funds looking for action were taking advantage of the fact that the muni yield curve had not inverted. Nowhere to hide.

    ReplyDelete
  2. Is seems that pursuit of hedging though pairing of non-correlated assets assures an effective arbitrage of this non-correlation to much lower levels.

    So is it then any consequent surprise that previously non-correlated assets are now far more correlated than historical data?

    ReplyDelete
  3. The overall Systemic Risk of sudden asset illiquidity, resulting in too many unknowns, e.g., from non-normal return distributions, 100+ yr events, and resultant feared domino of defaults, indeed proves the fallacy of non-correlations.

    This begs the question if the 21st century electronic interdependence amongst all debits & credits, quantified down to risk/reward lognormal probability tables, is akin to the dice table? And, therefore, should same be regionally, politically, and electronically be decoupled?

    An aside -- 35 years of Nixon's fiat dollar vs. 3,500 years of gold as the backing for the most accepted mediums of exchange? . . . Hum.

    ReplyDelete
  4. Yup, I was 100% correct . . . where's MY Nobel Prize in Economics?

    ReplyDelete