Enhanced Quantile Portfolio for Multifactor Model with Deep Learning

Abstract

Stock return predictability is an important research theme as it reflects our economic and social organization, and significant efforts are made to explain the dynamism therein. Although machine learning methods are increasingly popular and effective in stock return prediction in the cross-section, still most of the previous studies rely on a simple quantile portfolio. In this paper, we apply deep learning for stock return prediction in the cross-section and propose a more sophisticated portfolio construction framework called Enhanced Quantile Portfolios. These portfolios are inspired by Pure Quantile Portfolio that overcome the main drawbacks of simple quantile portfolios based on a single sort. The formulation of Enhanced Quantile Portfolio is a quadratic programming problem that considers the trade-off between portfolio alpha and stock diversification, while maintaining the characteristics of a simple quantile portfolio. The experimental comparison shows that the proposed approach outperforms a simple quantile portfolio.