基于集成神经网络模型的知情交易股票价格异象研究

The“Informed Trading Anomaly”in China’s A-Share Market—Research Based on Ensemble Neural Network

本文使用中国A股的日内高频交易数据,采用集成神经网络算法实现了针对知情交易行为的精准识别并证实A股市场中存在与股票知情交易程度相关的定价异象.研究发现:知情交易者的交易手法主要包括首尾盘操纵和策略化下单,具体表现为首尾盘时段内量价指标的异常变化和日内买卖价差、订单簿深度的短期突变,上述特征均可被本文建立的模型所捕捉.进一步研究发现,由于信息不对称所导致的流动性风险使得具有高知情交易倾向的股票需提供额外的风险补偿以吸引普通投资者的进入,基于本文计算的知情交易指数所构建的多空组合每月可获得1.38%的等权收益率.此外在市值规模较大、流动性较高、机构投资者和大股东持股比例较高的股票中组合收益更加显著.本文的研究对于完善金融市场监管、提升资本市场定价效率具有一定的启示意义.

Based on the high-frequency trading data of China’s A shares,we used ensemble learning algorithm to achieve accurate identification of informed trading activity,and found corresponding pricing anomaly in China’s stock market.We found that:The trading methods of insiders including manipulation at the beginning and end of the market and quick order placement during the trading hours,which is manifested in abnormal changes in volume-price indicators,bid-ask spreads and order book depth.Further research found that due to the liquidity risk caused by information asymmetry,stocks with high informed trading tendency need to provide additional risk compensation to attract ordinary investors to enter.The monthly long-short portfolio constructed based on our informed trading index can achieve significant excess returns.Besides this anomaly is more prevalent in large-sized stocks,high-liquidity stocks and stocks with high institutional ownership.Our paper is of great value for strengthening the financial market supervision and increasing the pricing efficiency of China’s A-share market.