What could have happened here?! I ran a normal AI-factor test, but I'm getting an extremely unrealistic return.
This is due to your feature FutureRel%Chg_D(65, GetSeries("SPY")). Dont use any 'future' functions as a feature. I tried adding that as a feature and then adding some basic valuation features and I get 800% annual returns.
If you get crazy results like this, you could create a predictor using the linear model and train it. Then go to the Importance page. It will show you that the FutureRel%Chg_D feature is the only 'important' one. That is a good indication that something is wrong.
1 Like