Core Viewpoint - The article discusses the development and performance of a timing strategy based on a neural network model, specifically the AGRU model, which has shown promising results in stock selection and market timing [1][4][75]. Group 1: Timing Strategy Performance - The timing strategy using the CSI All Share Index as the investment target achieved an annualized return of 15.86%, a Sharpe ratio of 1.18, and a maximum drawdown of -12.66% [10][11][12][75]. - The strategy's performance was tested over different rebalancing periods, revealing that weekly and monthly rebalancing negatively impacted the strategy's stability and returns [17][18][31][75]. - The strategy's information coefficient (IC) was calculated at 15.81%, indicating a strong predictive ability for the timing signals [15][75]. Group 2: Daily Rebalancing ETF Rotation Strategy - A daily rebalancing ETF rotation strategy, assuming a portfolio of 10 ETFs, yielded an annualized excess return of 17.13% with an information ratio of 1.25 and a maximum excess drawdown of -14.38% [2][49][75]. - When combining the timing signals with the cross-sectional signals, the strategy's annualized return improved to 31.04% with a Sharpe ratio of 1.86 and a maximum drawdown of -10.08% [50][75]. Group 3: Comparison of Timing and Cross-Sectional Signals - The correlation between timing and cross-sectional signals was found to be low at -1.96%, suggesting that they provide different predictive advantages [76]. - Timing signals were more accurate in predicting afternoon and overnight returns, while cross-sectional signals performed better in predicting overnight and late trading session returns [76][68][72]. Group 4: Sensitivity to Rebalancing Prices - The strategy's performance was tested under different execution prices, showing that using the opening price resulted in the best performance, but the strategy remained robust even when using prices from the first hour of trading [33][37][54][55]. - The annualized return using the opening price was 15.86%, while using a 60-minute TWAP execution resulted in a return of 13.15% [33][37]. Group 5: Performance Across Different Indices - The timing strategy demonstrated consistent performance across various broad indices, indicating that the model is not overfitted to specific stocks [39][40][75]. - For example, the annualized return for the CSI 300 timing strategy was 13.43%, while the CSI 1000 strategy achieved a return of 17.51% [40][75].