SWHY(SZ:000166) Shenwan Hongyuan Securities·2025-09-22 07:11Group 1 - The report emphasizes the importance of identifying the current market state as either "trend" or "range," which aids in timing and stock selection strategies [4][8][83] - A two-phase, layered diagnostic algorithm is employed to define the index state, utilizing a "zig-zag" algorithm combined with breakpoint correction to distinguish historical performance [4][10][11] - Six feature variables are constructed from price, volume, and volatility dimensions, and machine learning models such as logistic regression and decision trees are used for state prediction, achieving over 80% accuracy in future market state predictions after smoothing [4][31][32] Group 2 - A dynamic position management strategy is designed based on model predictions, switching between "momentum" logic in trend states and "mean-reversion" logic in range states, with weekly adjustments to balance risk and return [4][55][83] - The decision tree model-driven strategy yielded a total return of 77.26% and a Sharpe ratio of 1.12 during the backtesting period from 2020 to 2025, significantly outperforming the buy-and-hold benchmark return of 14.68% [4][55][78] - The report concludes that the decision tree model's strategy not only outperforms the benchmark but also demonstrates lower volatility and maximum drawdown, achieving a Sharpe ratio of 1.12, indicating robust performance [4][83]