Theoretically Motivated Data Augmentation and Regularization for Portfolio Construction

TL;DR

This paper develops a theoretical framework for data augmentation in deep learning-based portfolio construction, showing that injecting noise proportional to previous returns improves model robustness.

Contribution

It introduces a novel theoretical understanding of data augmentation's role in finance and proposes a simple, effective noise injection method for portfolio models.

Findings

Injecting noise proportional to previous returns enhances model performance.

Theoretical justification for specific data augmentation techniques in finance.

Simple noise injection outperforms other augmentation methods.

Abstract

The task we consider is portfolio construction in a speculative market, a fundamental problem in modern finance. While various empirical works now exist to explore deep learning in finance, the theory side is almost non-existent. In this work, we focus on developing a theoretical framework for understanding the use of data augmentation for deep-learning-based approaches to quantitative finance. The proposed theory clarifies the role and necessity of data augmentation for finance; moreover, our theory implies that a simple algorithm of injecting a random noise of strength $\sqrt{|r_{t-1}|}$ to the observed return $r_{t}$ is better than not injecting any noise and a few other financially irrelevant data augmentation techniques.