Does Training on Future Data Pay? Look-Ahead Bias in Forecasting with Pretrained Models
Updated
Updated · arxiv.org · Sep 17
Does Training on Future Data Pay? Look-Ahead Bias in Forecasting with Pretrained Models
1 articles · Updated · arxiv.org · Sep 17
Summary
A new study has found that using pretrained financial models trained on future data can introduce look-ahead bias in historical backtests.
Researchers showed that forecasts from models trained on post-forecast data often reduce predictive accuracy and portfolio value compared to origin-aligned models.
The findings highlight the importance of aligning training and forecast dates to avoid overstating investment strategy performance in empirical finance.