AI Questions

AI Machine Learning
1

Being new to the AI process, I have a few dumb questions:

  1. Why can’t a model built on one universe be backtested on a different universe. Seems like this would be a good robustness test.
  2. Why can’t a model built be backtested over the full model testing period? Again, seems like back testing with added filters or limiting the universe to even / odd would be a good testing process
1 Like
2
  1. You can, click on “Prediction Universe - No restrictions (advanced)”, but this is only really applicable to when you backtest with your predictor (AIFactor).
  2. You don’t want to back test on trained data, you would get unrealistic results. You can back test on your Holdout period with AIFactorValidation if you save your predictions. If you train with K-fold CV, you can back test the full period with AIFactorValidation.
1 Like