EcoSta 2022: Start Registration
View Submission - EcoSta2022
A0696
Title: A distributional perspective on autoencoder asset pricing models Authors:  Zhoufan Zhu - Shanghai University of Finance and Economics (China) [presenting]
Ke Zhu - University of Hong Kong (Hong Kong)
Dong Li - Tsinghua University (China)
Xuanling Yang - Tsinghua University (China)
Abstract: Quantitative trading and investment decision making are intricate financial tasks that rely on accurate asset selection. Despite advances in deep learning that have made significant progress in the complex and highly stochastic asset return prediction problem, the previous conditional asset pricing models face two major limitations. One is that they only focus on the mean of asset returns, and the other is that they ignore the potential randomness in latent factors. To get rid of these limitations, we consider a Dirac mixture model to represent the distribution over asset returns and employ the Variational Autoencoder to measure the randomness in latent factors. The key novelty is modeling the asset return distribution with a parametric model, which allows us to consider the expectation and variance(risk) of asset return simultaneously. Through simulations and an application to the US market spanning over sixty years of data, we show that our proposed method significantly outperforms previous relevant ones.