Features Min Length

Dear Alphathon Support Team,

Our model may needs at least 50 days (for example)to train, I don’t know if your features data will cover at least 50 days data? I noticed you have given 1 year training, that means we have at least 1 year in your features data for training with allowing lookahead bias?

Alphien Dashboard

The strategy must comply with the following constraints:
  • Each weight must be in a range of [-100%; +100%]
  • The portfolio’s annualised volatility must lie between 8% and 12% for the entire 2007::2016 period.
  • First weights have to be generated before/on 2008-01-02 (1 year calibration/training. Rolling/expanding windows are allowed)
  • Payout has to be replicable and contain no lookahead bias
  • All time series data has to be called via the DataFeatures object - no data calls in your payout.
  • Whole notebook has to be ran under 15 mins (excluding the testFXCarrySubmission function). Training of models/processing of data inclusive.


I am not sure I fully understand your question here but look ahead bias or information leakage should be avoided (please check https://wiki.alphien.com/ALwiki/Preventing_look_ahead_bias_in_payout for more details).

50 days doesn’t seem to be a problem in any cases.