Group 1 - The article discusses the classification of market states into "trend" and "range" based on historical data, emphasizing the importance of recognizing these states for investment strategies [1][4] - In a trending market, momentum strategies like "buy high, sell higher" yield greater returns, while in a ranging market, mean-reversion strategies perform better [1][4] - A two-phase algorithm is developed to label historical trends and ranges in the Shanghai Composite Index, enhancing the accuracy of market state identification [2][3] Group 2 - The backtesting period is set from January 2020 to August 2025, reflecting a shift in market behavior post-2020, with increased frequency of state changes [7] - A feature variable system is constructed to identify market states, focusing on price, volume, and volatility, rather than traditional indicators [8][15] - The model training shows that all six feature indicators have an accuracy above 50%, with the volume feature achieving the highest accuracy of 63.48% [22][23] Group 3 - The decision tree model outperforms other models in predicting market states, achieving an accuracy of 80.10% in the test set [36][39] - The strategy based on the decision tree model yields a total return of 77.20%, significantly outperforming the benchmark [63] - The research highlights the potential of combining strategic signals for long-term market state identification with tactical signals for short-term changes to enhance strategy performance [64]