Computerized trading using algorithms to sift through massive amounts of data and pick stocks to buy and short has its good and its bad sides. A recent profile of quant trader Cliff Asness, who built a successful such model, made me see a couple of the good points more clearly.
Asness and his partners were among the first to build a stock portfolio--and now a very successful business--by using computer models to combine two simple concepts: buying undervalued stocks (a strategy known as value investing) and betting against overvalued ones (which are called "momentum" stocks, referring to the tendency of securities that are rising in price to keep going up for a time, even when they're overvalued). Using a variety of metrics, the AQR models spit out the names of hundreds and hundreds of stocks that are undervalued (which the firm buys and holds) and hundreds more stocks that are over-valued (which they short, or bet will fall).
Asness explained the differences between quants and quals this way: "A qual digs very deeply into potential investments, but he can only do that with so many stocks, so he needs to have a relatively high level of conviction that he is right, since he's going to hold a pretty concentrated portfolio, say 10 or 20 stocks ... A qual needs to be careful about not making mistakes--one bad mistake in a 10-stock portfolio can get ugly!" He continued: "A quant, on the other hand, has the ability to study thousands of stocks at once, and thus can hold much more broadly diversified portfolios. Because quants hold so many stocks, ones that are even slightly misvalued may still make sense ... If you can find 500 stocks to bet on where each has a 51 percent chance of beating the market, then through diversification, the odds of your overall portfolio start to look pretty good."
This could actually be quite efficient. There’s a host of studies showing that human judgment is poor at synthesizing and weighting a large number of different types of evidence, and that simple, statistical models can outperform humans on tasks such as predicting recidivism, making clinical judgments (psychiatry and medicine), predicting divorce, predicting future academic success, etc. (for an entrypoint to this literature, see here for a blogpost I found that has some good quotes from J.D. Trout and Michael Bishop).
I guess the point is that algorithmic trading can be good or bad depending on the algorithm – and that the danger it brings is more if the ecology of trading algorithms active in a market is of a kind that could create cascading ripples destabilizing the market: One set of algorithms lowering the price of a set of stocks, triggering another set of algorithms to sell these stocks to avoid loss, triggering another set of… and so on. The lightning-fast feedback cycles set up by a changing ecology of (proprietary and secret) algorithms, increasing and decreasing in weight and influence depending on past results in the market, is difficult to predict. Which the article briefly touches on:
Sometimes, though, the quants get too clever for their own good, with potentially devastating effects. Such a moment occurred in the second week of August 2007, when a wave of selling by a group of quant funds using the same trading strategies led to terrible losses, as the firms all tried to sell the same stocks at the same time. As Andrew Lo, a professor at MIT's Sloan School of Management, observed in a September 2007 paper on the event, an "apparent demand for liquidity" that week "caused a fire sale liquidation." Patterson estimated that AQR lost $500 million in a single day, and close to $1 billion in the four-day rout before the markets steadied and started to recover on August 10.
Asness […] told the New York Post that he blamed the sudden losses not on AQR's computer models but on "a strategy getting too crowded ... and then suffering when too many try to get out the same door" at the same time. He told me he finds the argument that quants are "black boxes" of dangerously opaque trading strategies annoying and wrong. "We don't think of ourselves as 'black box,' " he said. "It is a great irony to us that even though a quant can, if willing, fully describe his investment process, it's often called 'black box,' even as the fundamental investor, who can never accurately describe his process, is not tagged with that label. A friend of ours, who is both a quant and fundamental investor, thinks quant is more accurately called 'glass box.' We think that's pretty accurate."
Seems like an interesting thing to study – maybe along the lines suggested in this paper by Brian Arthur, and it also seems related to evolutionary game theory (where strategies increase and decrease depending on their relative payoff in the current environment).