Multi-term momentum effect and driving mechanisms based on machine learning

Given that investors engage in non-synchronous trading activities, the paper integrates prior monthly returns with different lags and employs the Support Vector Regression (SVR) model to develop a Multi-term Momentum indicator (Multi-MOM). By constructing the winner-minus-loser portfolios, the paper reveals the existence of a multi-term momentum effect in the Chinese stock market. Through the spanning regression, the paper finds that the traditional momentum and other machine-learning-based momentum LS returns cannot fully explain the SVR-based momentum LS returns. To further explore the underlying drivers of this anomaly, the paper decomposes the multi-term momentum effect using a range of factors, including firm characteristics, information uncertainty, trend, lottery, and others. The findings reveal that firm characteristics, lottery, and information uncertainty are the key factors driving the multi-term momentum effect, accounting for approximately 16.4%, 11.7%, and 7% of the effect, respectively. Moreover, the paper examines the drivers of multi-term momentum under different market conditions and concludes that firm characteristics, lottery, and information uncertainty remain the primary drivers of multi-term momentum regardless of market conditions. Additionally, considering the varying momentum performance across different levels of Economic Policy Uncertainty (EPU), the paper investigates the drivers of multi-term momentum under different EPU conditions. The results indicate that firm characteristics, lottery, and information uncertainty consistently exhibit significant explanatory power for the multi-term momentum effect at different EPU levels.