Trying to replicate the results of testing shown in the results tab for this factor, which has a target of future 50 day relative returns. As shown below, the training period for the test was from October 3, 2003- Oct. 3, 2020, with a 26 week gap; the test period was April 7, 2021-April 10, 2024.
I ran a screen backtest with buy rule of Rank>95 using a ranking system based on a predictor trained over the same period as the test model.
I have done this with a few different models and cannot even come close to replicating the results shown on the AI Factor page.
2 Likes
I raised this issue here:
Use case for lift charts: When predictions of the actual returns are made available it will be nice to know how much bias there is in the predictions. Especially, if predictions of returns are added to estimates of transactional costs, by the member, to determine whenever to buy, hold or sell an asset.. Plus, why not add bias estimates to the variance estimates as part of a basic understanding of a model just for completeness, if nothing else?
Wow. I don't even have to add a feature request for…
We haven't received an answer that reconciles. Chime in over there on what you think.
Thank you!
1 Like
Do you have slippage cost in the screen?
