Quantitative Alpha Mining on CSI 1000 2026-01 to 2026-05 (Holdout)
1.737Per-alpha Sharpe Ratio (Median)GP (gplearn / AlphaGen ref.)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GP (gplearn / AlphaGen ref.)Selection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 1.737 | 0.0366 | 1.737 | 0.294 | 0.0366 | 0.733 | |
| AgoraSelection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 1.06 | 0.0479 | 1.87 | 0.484 | 0.0894 | 0.285 | |
| Single LLM, iterativeSelection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 0.676 | 0.0375 | 0.468 | 0.123 | 0.0382 | 0.697 | |
| Alpha101Selection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 0.541 | 0.0053 | 0.821 | 0.177 | 0.0186 | 0.503 | |
| Random searchSelection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 0.526 | 0.0368 | 0.683 | 0.168 | 0.072 | 0.321 | |
| AlphaGen-PPOseed=42, Selection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 0.377 | 0.0212 | 1.334 | 0.354 | 0.05 | 0.673 | |
| Frozen libraries (ablation)Selection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 0.227 | 0.0219 | 0.379 | 0.054 | 0.0468 | 0.067 | |
| Single LLM, one-shotSelection rule=train-segment IC, Backtest protocol=10 deciles, after-cost, Rebalance frequency=5-day, Alpha count=Top-302026.06 | 0.086 | 0.0247 | 0.657 | 0.204 | 0.0516 | 0.527 |