高频选股因子周报（20250512- 20250516）：深度学习因子空头端失效，多头端强势，AI增强组合继续维持正收益-20250520

• The report includes high-frequency factors such as intraday skewness, downside volatility proportion, post-open buying intention proportion, post-open buying intention intensity, large order net buying proportion, large order net buying intensity, improved reversal factor, end-of-day transaction proportion, average single transaction outflow proportion, and large order-driven price increase factor[3][5][6] - Deep learning factors include GRU(50,2)+NN(10), residual attention LSTM(48,2)+NN(10), multi-granularity model with 5-day labels, and multi-granularity model with 10-day labels[3][5][6] - AI-enhanced portfolios are constructed based on deep learning factors, including CSI 500 AI-enhanced wide constraint portfolio, CSI 500 AI-enhanced strict constraint portfolio, CSI 1000 AI-enhanced wide constraint portfolio, and CSI 1000 AI-enhanced strict constraint portfolio[3][67][68] - The construction of high-frequency factors involves specific methodologies such as realized volatility decomposition and machine learning-based low-frequency application of high-frequency data[13][18][23] - Deep learning factors are built using models like GRU and LSTM combined with neural networks, and multi-granularity models are trained using bidirectional AGRU[57][60][62] - AI-enhanced portfolios optimize expected returns with constraints on stock weight, industry weight, market capitalization, and other financial metrics[67][68][69] - High-frequency factors show varying IC values, rank MAE, and multi-long-short returns across historical, 2025, and May data[6][7][9] - Deep learning factors demonstrate strong multi-long-short returns and excess returns, with GRU(50,2)+NN(10) achieving 17.68% multi-long-short return in 2025[3][9][57] - AI-enhanced portfolios deliver positive excess returns, with CSI 1000 AI-enhanced strict constraint portfolio achieving 10.89% excess return in 2025[3][10][88]