About "curve-fitting"

Discussion in 'Trading' started by c.chugani, Sep 4, 2011.

  1. Isn't any backtested strategy that produces positive results based on a certain time series of past data, curve fitted? How does one avoid this?

    Whatever entry/exit parameters you use to trade - however "loose" and open ended you make them - (values for your momentum indicators, channels, volatility, etc. etc.) need certain values of input that will perform well in the past.

    Aren't all backtested strategies then, subject to hindsight bias? How does one determine that the past will repeat itself, in order for the selected trading parameters to continue to remain profitable in the future?

    Moreover, wouldn't other traders pick up on the repetition of the pattern (provided it continued in the future) and erode this so-called edge?

  2. Visaria


  3. No and Why should it be avoided?

    "Well" has many different meanings for different people. Some look at Sharpe, others at pf, others at win rate, and so on.

    Of course, almost everything in life is based on hindsight bias. Why do you think this is bad?

    One goes to college and takes several courses on statistics and probability theory. Then uses the knowledge to run tests on patterns.

    Yep, this is why trading is a difficult job. You have to be ahead of the crowd.