This blog and my forecasting models have nothing to do with high frequency trading.  This short post explains the situation.

A major theorem of the mathematical field of statistics says that high speed or high frequency trading can be profitable — at least in theory.

The Central Limit Theorem is a cornerstone of conventional statistics. If you take a sample of some sort of group, say, the height of third grade students in a particular school,  you tend to get survey results that look something like a bell curve.  The heights of the kids will vary from tall to short, but most of their heights will tend to fall somewhere near the middle.  The Central Limit Theorem doesn’t say anything about how that initial survey comes out.  (Maybe the third grade class you sampled is made up of kiddie basketball recruits mixed with tiny gymnasts.)

But, the Central Limit Theorem says if you run multiple samples of the same population, the median of those samples will form a very tight normal distribution (bell curve).  The variance, the total difference (squared) of things from the true average) should be much less for all the surveys taken together than for just a single survey.

In the context of high frequency trading this means that if you make a zillion trades, you can bet against trades that are being made by other people far from the current strike price. If someone is willing to make a statistically unusual (i.e. ‘stupid’) trade you can make money by betting against them.  Make enough of these trades and you should be able to make money.  This is what ‘algorithmic trading’ is all about.  It should work.

But.

High Frequency Trading  Only Works Until it Doesn’t
Long Term Capital Management was a firm that practiced highly leveraged,  high frequency probabilistic trading from 1993 to 1998.  It attracted roughly $120 billion in capital and was wildly successful.  In 1998 LTCM was betting in a very big way on a rebound in the falling Russian currency. But, suddenly Russia defaulted on its government debt. LTCM went bankrupt nearly instantly, and quickly the world economy was on the brink of economic collapse. Algorithmic trading works until it doesn’t work.

Low Hanging Fruit Was Eaten Years Ago
High frequency trading has been around for years. It is no big secret anymore. I attended a graduate seminar on it nearly a decade ago.

What has happened is that over time the opportunity for new players to profit has shrunk.  In the early stages ‘high frequency’ would mean trading with a gap of a few seconds.  Even a decade ago the frequency of trade response had shrunk to the range of nanoseconds.  Players fought to get their trading servers as near as possible to the stock market exchange computers.

An individual investor swimming in the pond of high frequency trading is like a minnow swimming with a school of sharks.