Deep Learning and Bayesian Model in Empirical Asset Pricing

Abstract

This dissertation proposes one deep learning model and one Bayesian model for empirical asset pricing. The first model is a deep neural network with an application in the corporate bond market, and the second model is a Bayesian Mixture model with an application in the mutual fund market. The results are shown in two chapters.

In Chapter 1, we propose a parametric approach to estimate the tangency portfolio weights on high-dimensional individual assets by combining fundamental finance theory with deep learning techniques. The deep tangency portfolio combines the market factor and a deep long-short factor constructed using a large number of firm characteristics. We apply our approach to the corporate bond market. The deep factor acts as a market hedge and achieves a sizable market price of risk with an out-of-sample annualized Sharpe ratio of 1.79. The deep tangency portfolio outperforms those constructed from commonly used observable factors with an out-of-sample annualized Sharpe ratio of 2.29. We find evidence supporting the integration between the bond and equity markets.

In Chapter 2, we propose a hierarchical Bayesian model with a nonparametric mixture component that allows one to estimate the betas and, at the same time, cluster them. The hierarchical nature of the model allows one to borrow information across the securities, while preserving heterogeneity among the betas. The nonparametric mixture component of the model lets us perform clustering without assuming the knowledge of the number of clusters. Our model applies more generally to any panel data regression where the clustering of regression coefficients is desirable, as well as preserving their heterogeneity at the instance level. We derive an efficient Gibbs sampler for the model and demonstrate the performance gain achieved by our model through comprehensive simulation and empirical studies. In particular, we show various estimation improvements and competitive clustering results against known baselines such as K -means and Gaussian mixture models.

Date of Award	24 May 2024
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Guanhao Gavin FENG (Supervisor) & Zhixin ZHOU (Co-supervisor)