High-Frequency Trading in the Chinese Stock Market: Predictability, Profitability, and Regulation

We study the predictability and economic value of ultra-high-frequency stock re turns in the Chinese A-share market, extending Aït-Sahalia et al. (2022) to a market with ±10% price limits, dominant retail participation, and T+1 short-selling con straints. Using Level-2 order book and transaction data for 347 CSI 300 stocks from February 2021 to May 2023, we apply penalized linear models and tree-based methods to forecast returns at 5- to 60-second horizons. The median out-of-sample R2 reaches 15.5% at the 5-second horizon, approximately 50% higher than compa rable U.S. estimates. Proximity to price limits is the single most powerful predictor, and a difference-in-differences analysis links predictability causally to limits to ar bitrage. Two trading strategies confirm economic significance: a VWAP execution optimization saving 3.41 basis points per day and an intraday market strategy gen erating 15.38% annualized returns net of costs. Profitability is robust across fee levels, frequencies, and liquidity conditions, with break-even costs of approximately 27 basis points.