An Information Bottleneck Asset Pricing Model

TL;DR

This paper introduces an information bottleneck neural network model for asset pricing that effectively filters out noise and redundant information, improving the robustness and accuracy of financial predictions.

Contribution

It proposes a novel information bottleneck approach that constrains mutual information to enhance neural network performance in financial asset pricing.

Findings

Reduces overfitting by filtering noise in financial data

Improves asset pricing accuracy with compressed representations

Balances information retention and noise elimination

Abstract

Deep neural networks (DNNs) have garnered significant attention in financial asset pricing, due to their strong capacity for modeling complex nonlinear relationships within financial data. However, sophisticated models are prone to over-fitting to the noise information in financial data, resulting in inferior performance. To address this issue, we propose an information bottleneck asset pricing model that compresses data with low signal-to-noise ratios to eliminate redundant information and retain the critical information for asset pricing. Our model imposes constraints of mutual information during the nonlinear mapping process. Specifically, we progressively reduce the mutual information between the input data and the compressed representation while increasing the mutual information between the compressed representation and the output prediction. The design ensures that irrelevant information, which is essentially the noise in the data, is forgotten during the modeling of financial nonlinear relationships without affecting the final asset pricing. By leveraging the constraints of the Information bottleneck, our model not only harnesses the nonlinear modeling capabilities of deep networks to capture the intricate relationships within financial data but also ensures that noise information is filtered out during the information compression process.