Online Early Stopping for Time-Varying Stock Return Prediction

Summary

This document studies neural network prediction when the relationship between inputs and returns changes over time. It proposes online early stopping, a training approach intended to let the model track an evolving function without requiring advance knowledge of how that function changes. The paper compares this method with existing approaches on monthly U.S. stock return prediction and reports better predictive performance.

The analysis also finds that the predictive value of familiar factors, including size and momentum, and of industry indicators varies over time. During market stress, industry information becomes more important relative to firm-level features. The document gives no specific performance metrics, model details, or stress-period definitions in the supplied text, so the reported superiority and changing feature importance cannot be independently assessed from this summary alone. The findings suggest that fixed training choices and stable interpretations of predictors may be unreliable when market relationships shift.

Key ideas

• Online early stopping is proposed to help neural networks adapt to changing return relationships.
• The approach is compared with existing methods for monthly U.S. stock return prediction and is reported to perform better.
• The predictive strength of size, momentum, and industry indicators changes over time.
• Industry indicators gain relative importance during market distress, while firm-level features lose importance.

Tags

# Time-varying neural network for stock return prediction


# Time-varying neural network for stock return prediction









We consider the problem of neural network training in a time-varying context. Machine learning algorithms have excelled in problems that do not change over time. However, problems encountered in financial markets are often time-varying. We propose the online early stopping algorithm and show that a neural network trained using this algorithm can track a function changing with unknown dynamics. We compare the proposed algorithm to current approaches on predicting monthly U.S. stock returns and show its superiority. We also show that prominent factors (such as the size and momentum effects) and industry indicators, exhibit time varying stock return predictiveness. We find that during market distress, industry indicators experience an increase in importance at the expense of firm level features. This indicates that industries play a role in explaining stock returns during periods of heightened risk.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.