GENERATIVE ADVERSARIAL NETWORK (GAN) MODEL IN ASSET PRICING.

Abstract

An important topic in asset pricing is explaining the variation in expected returns of financial assets. No-arbitrage pricing theory suggests the idea of a pricing kernel that governs asset prices. However, it remains a challenge to estimate the asset-pricing kernel. The difficulties include (1) choosing the right factors, (2) estimating the pricing kernel’s functional form, and (3) selecting the right portfolio to estimate the kernel. Recently, Chen et al. (2021) proposed a Generative Adversarial Network (GAN) model that attempts to solve all three challenges in a single setup and claim to achieve the best performance compared to all existing models. This paper seeks to empirically validate Chen et al. (2021)’s research based on the United States stock data with the United Kingdom (UK) London Stock Exchange 1998 - 2017 data. This paper found that the GAN model outperformed the benchmark four-factor model in terms of Sharpe ratio.