Friday, February 10, 2006

Thinking About Historical Analyses

I'll take a break today from the historical analyses to highlight several issues when using historical data to define possible trading edges:

1) All such analyses provide hypotheses only, not conclusions. The historical analyses are not "right" if they are replayed in the upcoming markets; nor are they "wrong" if the historical odds are not manifested on the next occasion. The historical studies merely indicate what has occurred in the past and whether such occurrences have been skewed or not. Our job is to use this information as background and see if the obtained patterns are indeed playing themselves out in the market. As I mentioned in a recent article for The Traders Journal in Singapore, Pasteur's observation that "Chance favors the prepared mind" captures the intent of historical study. It is meant as preparation--not as a fixed opinion to get locked into.

2) Disconfirmed hypotheses also provide information. Excellent trade ideas can be derived once you identify that a market is behaving contrary to its historical norms.

3) The utility and validity of historical analyses depends not only on having clean data and a promising set of indicators to study, but also on selecting proper lookback periods. The patterns that were common during the 2000-2002 bear market are not patterns we've seen in the past 2-3 years. Nor will the patterns of 2008 replicate those of the recent past. Recently, we've seen that patterns that were strong early in the current bull market (2003-4) are no longer robust. This means that we must weight recent data more strongly and be open to the possibility that we are undergoing a market transition. Indeed, shifts of trending and volatility patterns are perhaps the best evidence we have that market cycles are shifting.

4) The utility and validity of historical analyses also depends upon lookforward periods. Forecasting directional tendencies 1-3 days in the future requires a different set of predictors than forecasting 1-3 months or years hence. Of particular importance is what I call the optimum predictive horizon: the time frame that is most reliably forecast with a given set of predictors. This horizon, which almost always is not intraday, can help one define a directional bias within which shorter-term trading can occur.

5) Historical analyses are best employed in combination, with different sets of predictors and timeframes. Occasions in which multiple analyses converge to yield similar directional biases provide very high confidence trade ideas. Occasions when multiple analyses generate no meaningful directional biases help alert us to markets where our edge might be slim to none. Much of total P/L comes from trading known edges aggressively and staying away from risk when there is no edge--something any good poker player knows.

I will follow up with further thoughts on the Trader Performance blog of my personal site. Thanks for the many kind comments and suggestions I've received regarding the TraderFeed blog.