RVRAE: A Dynamic Factor Model Based on Variational Recurrent Autoencoder for Stock Returns Prediction
Yilun Wang, Shengjie Guo
TL;DR
RVRAE tackles predicting cross-sectional stock returns in noisy, high-dimensional markets by framing factors as latent variables in a dynamic latent factor model. The method combines a factor network that uses a Variational Recurrent Autoencoder to infer $f_t$ from $r_t$ and a beta network based on LSTM to map firm characteristics $x_t$ to loadings $eta_t$, trained under a joint ELBO objective that aligns the posterior $q_{ ext{ } }(z_t|r_t)$ with the prior $p_{ heta}(z_t)$. The model predicts returns with historical data while restricting future data to influence factor identification, and provides probabilistic risk estimates from latent space, achieving superior out-of-sample $R^{2}$ and Sharpe ratios relative to IPCA, CA, FactorVAE, ALSTM, and Transformer baselines. Overall, RVRAE advances asset pricing by delivering a dynamic, robust, and interpretable method for latent factor extraction and risk modeling in volatile stock markets.
Abstract
In recent years, the dynamic factor model has emerged as a dominant tool in economics and finance, particularly for investment strategies. This model offers improved handling of complex, nonlinear, and noisy market conditions compared to traditional static factor models. The advancement of machine learning, especially in dealing with nonlinear data, has further enhanced asset pricing methodologies. This paper introduces a groundbreaking dynamic factor model named RVRAE. This model is a probabilistic approach that addresses the temporal dependencies and noise in market data. RVRAE ingeniously combines the principles of dynamic factor modeling with the variational recurrent autoencoder (VRAE) from deep learning. A key feature of RVRAE is its use of a prior-posterior learning method. This method fine-tunes the model's learning process by seeking an optimal posterior factor model informed by future data. Notably, RVRAE is adept at risk modeling in volatile stock markets, estimating variances from latent space distributions while also predicting returns. Our empirical tests with real stock market data underscore RVRAE's superior performance compared to various established baseline methods.