High-Dimensional Spatiotemporal Dependencies and Behavioral Heterogeneity Co-Modeling: LSTM-Multi-Factor Penetrative Analysis and Risk Premium Decomposition of the Reversal Effect in China's A-Share Market

China’s A-share market, with its unique retail-dominated investor base and policy-driven ecosystem, serves as a compelling laboratory for testing market anomalies. This study investigates the efficacy and mechanisms of momentum and reversal effects in this emerging-market context, uncovering how behavioral forces dynamically shape price patterns under structural inefficiencies. Building on monthly data spanning January 2014 to January 2024, we systematically evaluate strategy performance through winner-loser portfolio construction across formation periods (1, 3, 6, 12 months), employing long-short tactics, risk-neutralizing techniques (CAPM and Fama-French three-factor models), and an LSTM deep learning framework. Key findings reveal reversal strategies’ striking dominance over momentum approaches mainly during an economic recession, with win rates consistently hovering at 50%-60% in both economic expansion and recession and peaking with an average of 61.95% in 3-month formation windows—evidence of retail-driven overreaction followed by swift price correction. Our novel LSTM model, integrating implied volatility (IV), turnover, market capitalization, and technical indicators, achieved 81.3% prediction accuracy in out-of-sample tests, with the relative strength index (RSI) emerging as the most influential predictor. By bridging multifactor risk adjustment with nonlinear machine learning, this research marks a breakthrough in quant methodology, offering practitioners a robust toolkit to navigate China’s sentiment-fueled volatility while advancing behavioral theory in nonlinear markets. This study also redefines emerging-market asset pricing paradigms through theoretical and technological breakthroughs. Theoretically, it pioneers a dynamic pricing model that captures the interplay between institutional frictions and behavioral biases, particularly during policy-sensitive windows and sentiment extremes—a critical advance beyond traditional frameworks’ inability to decode the “policy-sentiment feedback loop.” Practically, the novel LSTM-multifactor framework offers modular applications: its real-time tracking of RSI and implied volatility (IV) thresholds could power next-gen robo-advisors, while its sector rotation analytics might optimize multi-asset funds’ tactical pivots. For regulators, the research unveils a stealth weapon: by mapping cointegration between reversal anomalies and manipulative trading patterns, it equips watchdogs with precision tools to flag suspicious activity—a leap forward for China’s ambitions in penetrative, AI-augmented market surveillance.