How Markets Move

Market theory is founded upon the idea that prices are based on information. For example, if company X comes out with a nasty press release revealing earnings are going to fall, standard economics tells us that the stock price should adjust rapidly to factor in this new information.

Earlier this week, I wrote about how the actual speed of this movement seems to depend on both the hard numerical news in the release as well as the softer information conveyed in the text, like sentiment and confidence. Hard news is believed to be easily digestible while soft information can take time (days, weeks, months) to work its way into prices. Any deviations from fundamentals are then explained away as noise trades.

But a body of recent research says that the role of information in prices may be at best secondary. That view comes from physicist and hedge funder Jean-Philippe Bouchaud, who like another French-speaking market heretic, seems intent on overturning, or at least drastically updating, some economic beliefs.

I wrote about Bouchaud's essay in the journal Nature yesterday on the need for a "scientific revolution" in economics, and last night I spent some time digging into his latest piece of research which reviews recent work on the microstructure of markets.

Bouchaud's points out that for even the most liquid stock, only a tiny amount (0.1 to 1 percent) of its total market capitalization is traded per day. So that means if any big investor wants to buy or sell a large quantity of shares, it'll take time to find buyers and sellers for each stock. Bouchaud's main contention, then, is that market movements are primarily driven by supply and demand.

Here's an illustration:

Consider again the case of a typical US large cap stock, say Apple, which (as of Nov. 2007) had a daily turnover of around 8B$. There are on average 6 transactions per second, and on the order of 100 events per second affecting the order book. These are extremely small time scales compared to the typical time for public news events, in which a hot stock like Apple might be mentioned by name every few hours during a period of fast information arrival. Perhaps surprisingly, the number of large jumps in price is much higher. For example, if we define a jump as a one minute return exceeding three standard deviations, there are the order of ten such jumps per day, reflecting the very heavy tailed distribution of high frequency returns (Joulin et al. [2008]). More often than not such jumps occur in the absence of any identified news. It is obviously a particularly important question to understand the origin and the mechanisms leading to these jumps. The difference between the frequency of news and the frequency of jumps already suggests that something else must be at work, such as fluctuations in liquidity, that may have little or nothing to do with external news entering the market.

And that "something else" is the practice of order-splitting:

in which large institutional funds split their trading orders into many small pieces. Because of the heavy tails in trading size, there are long periods where buying pressure dominates, and long periods where selling pressure dominates.

The market only slowly and with some difficulty "digests" these swings in supply and demand.

The upshot is that at any point in time, prices are very unlikely to reveal all available information, so prices are almost always "perfectly unpredictable, even on very short time scales."

If true, Bouchaud's assertions can help explain why there has never been conclusive evidence that technical analysis does not work, or why markets have gyrated wildly since the collapse of Lehman.