Risk Professional - December 2009

INduStRy cOmmENtARy

Imperfect, but Still Very Useful

The solution invented 20 years ago, and the bedrock of risk management since, is value-at-risk (VaR). You identify the range of circumstances in which your models work, test it rigorously, and then make contingency plans to survive beyond those circumstances. Those plans do not count on the ability to trade — or prices being rational when you can trade — or having good information. The result is 99% of the time you let complex models make optimal decisions, but at a clearly-defined 1% of the time, you switch to simpler, more robust models, bring in human judgment and change the goal from optimum risk/reward ratio to simple survival. (Just think of a commercial airliner. Most of the time, an autopilot flies the airplane, optimizing speed, comfort and fuel consumption — but a human pilot must take control, with computerized assistance, for takeoff, landing and emergencies. Would a computer have known it was better to land a plane in the Hudson than in Central Park? And even if a programmer had anticipated this situation and programmed the correct response, I’m confident there are some equally likely scenarios that are missing from the autopilot.)

VaR is not perfect, but it works better than anything else. Each crisis teaches us more about the conditions under which our models work and what kind of contingency plans are helpful. Among the surprises of the last two years to a serious risk manager are the length of time the economy can operate without liquidity or trust in banks, the extraordinary unwillingness of governments to allow popular regulators or politically favored institutions to fail (or even to have to admit error), the shallowness of faith in the market when it delivers unpopular news and the depth of the faith in the US dollar.

There’s one area where quants have an edge over legislators (who think we need more legislation), regulators (who think we need more regulation) and financial executives (who think we need more profits): we remember. We embed our lessons in equations and computer code.

That’s the major flaw in the case of people who want to eliminate quantitative risk management. However bad it is, it gets better — the bigger the disaster, the bigger the improvement. Would anyone say the same about legislation, regulation or general management?

Humans: More Likely to Fail

Psychologist Dietrich Dorner once performed an experiment in which people were asked to maintain the temperature in a simulated meat freezer for 15 minutes. The instructions were

simple: when the temperature is too high, turn up the cooling; when it is too low, turn it down. Simple mechanical devices have been doing this flawlessly for 150 years, but none of Dorner’s subjects succeeded on their first attempts.

If the temperature starts out high, they turn up the cooling too much, and they subsequently don’t reverse until the temperature reaches the desired point, meaning they overshoot. So they overreact again, in a cycle that always leads to disaster. Even with extended explanations and lots of practice, few people can keep the meat frozen. Think about that before you ask humans to run the financial system without computer models.

Quantitative risk models, like refrigerator thermostats, re-act smoothly and efficiently. When they work, they steer a firm between the Scylla of inadequate risk — when the employees and shareholders bolt and the firm is bought out at a knock down price by a competitor — and the Charybdis of excessive risk — when you blow up.

Everyone remembers the firms that failed in the crisis, but there’s a longer list of firms that failed shareholders and employees in less spectacular fashion over the preceding five years. Chuck Prince made himself a symbol of Wall Street irresponsibility when he told the Financial Times, “As long as the music is playing, you’ve got to get up and dance.” Reckless as that sounds, critics should know that if you only dance to safe music, you’re not going to please your partners, and you will soon have no opportunities to dance at all.

Steering this narrow course is a difficult job, and quants fail a lot. But there would be a lot more failures without us — if firms were managed by qualitative feel alone. The financial world is far too complex and technical for that. It would be nice if the seas were safer, but occasional financial turmoil seems to be the price of a dynamic economy. So we build the best models we can, and update them constantly — but we still insist on lifeboats for when the models fail.

There will always be a few dreamers who think better models are all we need, as well as critics who want to abolish models. The consensus of professional opinion among practitioners is we need to keep improving both our models and our readiness to survive without them.

Aaron Brown is the risk manager at AQR Capital Management and the author of A World of Chance (Cambridge University Press, 2008) and The Poker Face of Wall Street (Wiley, 2006). He serves on Risk Professional’s editorial board and can be reached at Aaron.Brown@aqr.com.