Abstract Book of the 12th International Conference on Business, Management and Economics
Year: 2026
[PDF]
Dissecting The Equity Risk Premium in The Chinese Stock Market Using Machine Learning and Post-HOC Inspection Methods
Dr. Aya Nasreddine, Xuankai Zhao, Wen-Qiang Li
ABSTRACT:
The equity premium is a requisite element for several operations and appraisals in the finance field. Many econometric challenges are faced by researchers to estimate this variable and catch the factors that may drive it. Besides, considering the profusion of trading signals, anomalies and return-driving firm characteristics documented in the literature, it is essential to define which are those that influence the most the equity risk premium over a specific market. To overcome the shortcomings of classical linear models in estimating the cross-section of stock returns, our work involves machine learning (ML) techniques that can handle non-linear relationships that may exist between asset risk premia and numerous stock-level characteristics inspired by the financial literature. We implement and compare the results of ordinary least squares, Ridge, LASSO, LightGBM, XGBoost, random forest, and multiplayer perceptron models, respectively. However, as the complexity of ML techniques often makes them appear like “black boxes”, we dive into interpretability techniques to better understand the underlying process of ML models and the results that emanate from them. Ergo, we compute permutation importance, partial difference plots as well as the Shapley Additive exPlanation (SHAP) and the Shapely Additive Global importancE (SAGE) for each tested model. This empirical analysis focuses on the case of the Chinese stock market, which has been prone to several evolutions. By scrutinizing the equity risk premium, past returns seem to have a crucial influence in shaping the latter, followed by, to a lesser extent, liquidity and risk variables, respectively.
Keywords: Asset pricing; Factors; Interpretability; Machine learning; Momentum