The New Program Trading

There's a Wired article up today about making money through semi-automated and even fully-automated stock trading becoming more popular among certain individual traders and small funds.

The article is all very interesting, and I even know a few people who are doing this (they claim to be making money, though of course I can't verify those claims--and properly calculating stock trading returns is harder than it seems at first glance). But if this is really becoming as big as the article claims, I wonder what effect it is having on the stock market as a whole?

Program trading got a bad rap in the late 1980's, when it was blamed for the 1987 stock market crash. The problem in 1987 was not just program trading: it was the fact that all the programs were following the same strategy (called "portfolio insurance"), and furthermore, that the portfolio insurance strategy creates a positive feedback loop. That is to say, portfolio insurance programs buy when the market is going up and sell when the market is going down, reinforcing whatever trend the market is taking. In October 1987, the market took enough of a dip that the feedback loop got out of control, and the market crashed.

[Aside 1: It was called "portfolio insurance" because the buying and selling pattern was designed to mimic the behavior of a put option, selling the entire portfolio when the price drops to a certain point. The idea was that the program could minimize risk by selling off stocks when the market drops, just like buying an actual put option limits how much money you can lose if the stock price goes down. The fatal flaw was that the people who invented the strategy assumed that the market is always liquid; that is, that you can always find a buyer or a seller at or close to the current stock price. But when the market crashed, the liquidity went away, and buyers of stocks were hard to find at any price.]

[Aside 2: Feedback loops come in two flavors: "positive" and "negative." Positive feedback loops reinforce whatever the underlying trend is, and make systems unstable because they lead to runaway conditions. Negative feedback loops oppose the underlying trend, and make systems more stable because they oppose change. An example of a stock trading strategy which creates a negative feedback loop would be "Buy Low, Sell High."]

Portfolio Insurance was largely discredited after the crash of 1987, when it not only failed to perform as advertised (it didn't prevent anyone from losing money when they needed the protection most), but it actually made things worse.

After 1987 program trading never went away, but there were some important changes. First, the markets imposed "circuit breakers" which cut off all program trading under certain conditions (mainly when the market drops a certain amount. Nobody seems to be too worried about runaway feedback loops pushing the market up). More important, the programs started following different strategies. No longer did you have the situation where so much money was programmed to follow the same trading strategy. So the risk of runaway feedback loops was mitigated somewhat.

But all program trading buys and sells based on market trends, and so basically boils down to a feedback loop. Current programs are much more complicated than the old portfolio insurance, but it can still be described as nonlinear feedback loops with variable delays. What's more, when taken in aggregate (i.e. adding up the effect of everyone's program trading strategy), it still boils down to a complex feedback loop.

The people doing the program trading tend to be very secretive about the exact models they're using, since they don't want everyone copying their trading strategies (which, ironically, probably means that many of them are following the same trading strategies, since there are a handful of obvious things most people will think of without realizing that everyone else is thinking of the same thing). But from what I've been able to glean through conversations and reading online, the vast majority of the current crop of programs are trend-following. That is, they attempt to identify a trend, then pile on.

Trend following always creates a positive feedback loop (almost by definition). So even though there is a diversity of trading programs out there today, they're mostly pushing the market in the direction of more volatility.

[Aside 3: These program traders are constantly refining their models, since as people catch on to a given method it tends to become less effective. But if you're in a market of trend followers, how do you make money? By being the first to spot the trend and jump on before everyone else. So I suspect what is really going on is that everyone is tweaking their programs to make them more and more sensitive, perhaps without even realizing that's what they're doing. Anecdotally, I have a hunch that a lot of people don't really understand their computer models. They just trust the computer and the data.]

By the way, trend following is not inherently a bad strategy. Trend following works because there are natural buyers and sellers in the market (that is, people who trade to build or liquidate a portfolio position, rather than day trading to catch a short term trend). If there's someone trying to accumulate a million shares of Google, for example, then that buyer is creating his own little mini-trend. All the program traders are hitching a ride, and essentially forcing the buyer to either pay more or stop buying. It is effectively a mechanism which seeks out the highest price the natural buyer is willing to pay.

The really interesting question is whether there's any way to make money by exploiting all these positive feedback loops and trend following behavior. I suspect there is, though it might take a fair amount of capital to make it work.

The strategy would basically be this:

Step 1: Create a trend.

Step 2: Let the programs amplify the trend.

Step 3: Bail out before the natural buyers or sellers bring the stock back to reality.

Step 1 is simple. All you need to do to create a trend is buy or sell enough stock to start moving the price. Depending on the stock, this might just take a few hundred shares, or it might take millions. Multiple small trades may be more effective than one big trade (I don't know for sure), though the brokers often split trades up into multiple transactions at slightly different prices.

Step 2 is also simple. If you've chosen a good stock (that is, one with lots of trend-following traders, but few natural buyers or sellers), then it will happen all by itself. The key to making this strategy work is achieving positive amplification of your trend. Sort of like in a laser, you want your initial trade to induce other trades, which will induce other trades. You need the stock to move enough, and with enough volume, that you can execute step 3 at a profit.

Step 3 is the tricky step. Part of the reason it is tricky is that the market makers (the human ones, at least) are wise to the behavior of the trend-followers, and they'll also be trading against the trend as it matures (knowing that all those day traders will go the other way in a short while). In addition, natural buyers and sellers (which are the inherent price-stabilizing mechanism of the stock market) will see the trend as an opportunity and trade against it. So there's an element of timing and especially not being greedy.

Probably the hardest part of this strategy is choosing the right stock. You want one you can move (enough to create a trend) with the capital you have available. In addition, you want one with a lot of trend-followers (so boring industrial stocks are probably out), and you need to find a time when the stock is ripe for trend amplification. Good candidate stocks will have a lot of inter-day price volatility but not much movement over days or weeks, high trading volume compared to the number of shares available to trade, and probably a certain amount of "buzz" or sexiness.

You also need to keep in mind that this is a zero-sum game (actually, negative-sum game when you consider commissions). By creating trends for the trend followers to amplify, then cashing out, you are literally taking money from the trend followers' collective pockets. Every dollar of profit you make is a dollar of loss for the program traders, since at the end of the day everyone has cashed out. This will, over time, drive the program traders out of whatever stocks you're trading and/or lead them to refine their models (probably making them less aggressive). So after some time, you will have to try something else.

Previous
Previous

Getting to Know my Stove

Next
Next

Wikipedia Horror Story