njrookie's random thoughts

A collection of my posts on investment, trading, and life...
個人資料
正文

Can we trust backtested results from optimised trading strategy?

(2014-10-30 08:00:57) 下一個

The answer is no due to data snooping bias.
 

Everyone wants holy grail, and believes it is out there to be found. If you start with 1000s of parameter permutations (say a 3 MA combination to determine entry, exit, and stop), optimizes over parameter space, and require 5% statistical signficance for viable strategy, then 5% strategies will seem to work well in sample, even with purely randomly generated data.
 
Inevitably those "successfully back-tested strategies" will fail miserably out of sample. This is called data snooping bias. In short, you can only massage a fixed path of data history so many times.
"optimized" strategies are usually not optimal. 

Sometime one has to resort to intuitive understanding of financial market/economy/human psychology to gain an edge. But that takes far more work and most people are not willing to make the commitement.
 
just my 2c
[ 打印 ]
閱讀 ()評論 (0)
評論
目前還沒有任何評論
登錄後才可評論.