Taming the factor zoo in China’s equity market: A Bayesian approach

Abstract

This paper proposes an advanced Bayesian Model Averaging (BMA) framework to estimate the stochastic discount factor (SDF) in the Chinese stock market, addressing model uncertainty across 288 quadrillion factor combinations. By integrating the Moore–Penrose pseudoinverse and LDL decomposition, our methodology ensures sparsity, numerical stability, and robustness for high-dimensional, volatile datasets. We find that (i) the idiosyncratic volatility (STD) factor dominates with 60 percent posterior model probability, likely driven by retail investor herding and regulatory inefficiencies; (ii) the size factor (SMB) reflects distortions from state-owned enterprise (SOEs); (iii) the optimized BMA-SDF outperforms benchmark models in both in-sample and out-of-sample tests; (iv) no single model consistently excels across cross-sectional and time-series dimensions; and (v) the SDF relies on a dense set of observable factors. These findings highlight BMA’s efficacy in emerging markets and underscore the need for reforms to enhance transparency, reduce volatility, and optimize SOE performance.