Driving mechanisms of a new SVR-based trend anomaly in China’s A-share market

Abstract

This paper constructs a multi-horizon trend signal using support vector regression trained on short-, medium-, and long-term moving averages to capture asynchronous trading behaviours of heterogeneous investors. Long-short portfolios sorted on the trend signal yield significant risk-adjusted excess returns, and such outperformance persists across alternative formation windows, sub-samples and market states. To uncover its economic foundations, we employ a covariance-based decomposition framework to attribute cross-sectional variation in the anomaly to firm fundamentals, information uncertainty, profitability, momentum, lottery characteristics, and other features. Our analysis reveals that information uncertainty and momentum constitute the primary drivers of the trend premium, accounting for approximately 25% and 24% of its cross-sectional return variation, respectively. These results retain robust economic and statistical significance after excluding small-cap stocks, low-price equities, and turbulent market periods. More importantly, we identify strong state dependence among the anomaly’s underlying mechanisms: momentum exerts greater explanatory power during bull markets and high-sentiment regimes, while information uncertainty dominates in bearish environments with weak investor sentiment.