An Empirical Study of Autoencoder Asset Pricing Models and the Impact of Arbitrage Constraints
Abstract: Following Gu et al. (2021), we implement a state-of-the-art machine learning asset pricing model, the conditional autoencoder, to capture the time-varying interactions between observable stock characteristics and factor loadings, while simultaneously extracting latent factors from stock returns. Unlike their no-arbitrage model setup, we explicitly compare the out-of-sample performance between arbitrage models and their no-arbitrage counterparts. Our results show that arbitrage models exhibit better predictive out-of-sample performance, which indicates that stock characteristics proxy not only for factor risk premia, but also compensation for mispricing.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)