Wednesday, October 16, 2013

High expectations kill your long-term returns

Let's assume, that you are a systematic trader, you search for trading strategies using some time window, one year, three year or maybe ten - as you like. Probably, in that window there are not only data artifacts  but even some real life factors, that impacted market prices and made price behavior somehow different from random walk. So it is possible that the strategy you finally are going to trade, is based not only on data artifacts but on some real market issues as well.

What can we say about those real life factors?

1. Existing of strong factors is less probable than week factors. Week factors, creating low-return opportunities are supposed to be more frequent, than strong factors with high-return opportunities.

So, the higher you strategy backtesting returns, the more chances are those returns are fake, data mining artifacts and there is no real life factor behind it.

2. Factor's life time is shorter for stronger factors. Low-return factors can hold for years, because index+3% is not what a crowd is looking for. Strong high-return factors, on the contrary, are quickly overcrowded, what quickly makes them adequately priced or even worse, they may start to perform negatively under the weight of the crowd. Long-term mean reverse, kind of.

So, the higher the return of the factor you base your trading strategy on, the higher is a probability, that there is no actual factor at all, or even if there is, it's going to dye soon or is already dead. And the higher a probability that after its death it will reverse its effect. Low-return factors can hold much longer and dye quietly, without catastrophic capital re-allocation from those who are late for dinner to those who got lucky. 

As a conclusion, I present you the best way to kill your long-term results: trade only those strategies, that showed highest backtesting profits. Fit the graal, trade, get surprised, fit next, trade, get surprised again and so on.

No comments:

Post a Comment