Friday, September 7, 2007

The quant menace: connecting the dots

When the number of computerized quant traders was relatively small, there was little need for them to interact much. But as the potential for huge profit crowds the investment field with quants, the potential for national and global economic catastrophe increases.

As one quant manager said in a Wall Street Journal story today: last month's sudden massive plunges in quant (as in high quantity trading fund) performance was a consequence of too many players. He of course is trying to fine-tune his fund's algorithms.

But the real trouble with computerized statistical arbitrage -- which relies on statistical analysis and the law of large numbers to favor profitability -- is not that it doesn't work, but that that eventually it will be such a powerful market force that it can't work as planned.

What happened in August seems to have been triggered by some unknown quant's short-selling. But short-selling is a key component of arbitrage pricing theory, a technique for taking advantage of statistical patterns in the market. That short-selling spree evidently triggered a round of forced repurchases, with massive losses.

As these arbitrage programs proliferate and their volumes grow, the statistical nature of the market is bound to change. That is, you get a "new force" that, like the market it mirrors, is highly non-linear. Non-linear processes have a strong tendency to become erratic, unpredictable and even mathematically chaotic.

Yes, some traders will improve their algorithms and squeeze out those who are less capable. But, the competitive process will mean that optimization of algorithms will reach a limit. In fact, it's a mathematical fact that an algorithm always has some maximum efficiency. You can't improve on it forever. Optimal algorithms may vary somewhat but they'll all have about the same bang for the buck.

So as computer arbitrage techniques race toward equilibrium, the quants face the likelihood of further sudden catastrophic losses at times that aren't terribly predictable. Of course, one could design an algorithm to monitor -- statistically and via espionage -- the subset of quant traders, but eventually this tactic will also zero out in value as others follow suit.

Consider several computerized players playing poker. Once optimal poker algorithms are achieved, the most probable outcome for any player, assuming each has unlimited funds, is to break even. If each player begins with a finite stake, then the player with the largest (using statistically meaningful differences) stake is most likely to eventually take all, with all the others going bankrupt. If each player has about the same finite stake, that means that there is an equal chance that any player will eventually bankrupt all the others.

This analogue may seem excessively simple. But these "forces" work the same, whether in poker or in a crowded APC field.

However, in the computerized poker games, a cascade effect isn't considered. But sudden, catastrophic APC movements can interact with each other and cascade into the general market, wreaking havoc.

The Bush administration's Wall Street watchers have failed to connect the dots, or, if they have, the intelligence has not reached higher officials. Yet the likelihood of one or more horrific economic shocks poses a far more terrifying threat to national security than anything that occurred on 9/11.

No comments: