config.json
| 1 | { |
| 2 | "model_type": "v4_regime_threshold", |
| 3 | "base_model": "LightGBM regressor (MSE, 47 features)", |
| 4 | "objective": "predict next-day log return per stock", |
| 5 | "feature_columns": [ |
| 6 | "logret_1d", |
| 7 | "ret_2d", |
| 8 | "ret_3d", |
| 9 | "ret_5d", |
| 10 | "ret_10d", |
| 11 | "ret_20d", |
| 12 | "ret_60d", |
| 13 | "ret_120d", |
| 14 | "ret_252d", |
| 15 | "vol_5d", |
| 16 | "vol_20d", |
| 17 | "vol_60d", |
| 18 | "dd_20d", |
| 19 | "dd_60d", |
| 20 | "dd_252d", |
| 21 | "mom_12_1", |
| 22 | "px_over_ma_20", |
| 23 | "px_over_ma_50", |
| 24 | "px_over_ma_200", |
| 25 | "logv", |
| 26 | "logv_over_ma_5", |
| 27 | "logv_over_ma_20", |
| 28 | "logv_over_ma_60", |
| 29 | "dollar_vol_20", |
| 30 | "range_pct", |
| 31 | "body_pct", |
| 32 | "gap_pct", |
| 33 | "up_frac_10", |
| 34 | "up_frac_20", |
| 35 | "ret_20d_chg", |
| 36 | "vwap_gap", |
| 37 | "ret_z60", |
| 38 | "days_since_earn", |
| 39 | "days_until_earn", |
| 40 | "near_earn_window", |
| 41 | "eps_est_known", |
| 42 | "dow", |
| 43 | "spy_ret_20", |
| 44 | "spy_vol_20", |
| 45 | "spy_dd_252", |
| 46 | "spy_over_ma_200", |
| 47 | "xs_ret_5d", |
| 48 | "xs_ret_20d", |
| 49 | "xs_vol_20d", |
| 50 | "xs_mom_12_1", |
| 51 | "xs_px_over_ma_50", |
| 52 | "xs_dd_60d" |
| 53 | ], |
| 54 | "transaction_cost_bps": 5.0, |
| 55 | "regime_rule": { |
| 56 | "trigger": "SPY above 200-day moving average", |
| 57 | "bull_threshold": -0.003, |
| 58 | "bear_threshold": "tuned per split on last 10% of training", |
| 59 | "description": "In bull regime, stay long unless prediction < bull_threshold. In bear regime, long only if pred > bear_threshold." |
| 60 | }, |
| 61 | "training": { |
| 62 | "base_params": { |
| 63 | "objective": "regression", |
| 64 | "metric": "l2", |
| 65 | "learning_rate": 0.03, |
| 66 | "num_leaves": 63, |
| 67 | "min_data_in_leaf": 200, |
| 68 | "feature_fraction": 0.8, |
| 69 | "bagging_fraction": 0.8, |
| 70 | "bagging_freq": 5, |
| 71 | "lambda_l2": 1.0 |
| 72 | }, |
| 73 | "num_boost_round": 2000, |
| 74 | "early_stopping_rounds": 50, |
| 75 | "val_frac": 0.1, |
| 76 | "val_selection": "chronological last 10% of train dates" |
| 77 | }, |
| 78 | "decision": "long if position=1 else cash (no shorts, no leverage)", |
| 79 | "splits": { |
| 80 | "2010": { |
| 81 | "train_start": "2005-01-01", |
| 82 | "train_end": "2009-12-31", |
| 83 | "test_start": "2010-01-01", |
| 84 | "test_end": "2010-12-31", |
| 85 | "bear_threshold": -0.002, |
| 86 | "best_iter": 1 |
| 87 | }, |
| 88 | "2018": { |
| 89 | "train_start": "2005-01-01", |
| 90 | "train_end": "2017-12-31", |
| 91 | "test_start": "2018-01-01", |
| 92 | "test_end": "2018-12-31", |
| 93 | "bear_threshold": -0.002, |
| 94 | "best_iter": 90 |
| 95 | }, |
| 96 | "2020": { |
| 97 | "train_start": "2005-01-01", |
| 98 | "train_end": "2019-12-31", |
| 99 | "test_start": "2020-01-01", |
| 100 | "test_end": "2020-12-31", |
| 101 | "bear_threshold": -0.00175, |
| 102 | "best_iter": 280 |
| 103 | }, |
| 104 | "2022": { |
| 105 | "train_start": "2005-01-01", |
| 106 | "train_end": "2021-12-31", |
| 107 | "test_start": "2022-01-01", |
| 108 | "test_end": "2022-12-31", |
| 109 | "bear_threshold": -0.002, |
| 110 | "best_iter": 1 |
| 111 | }, |
| 112 | "live": { |
| 113 | "train_start": "2005-01-01", |
| 114 | "train_end": "2025-04-23", |
| 115 | "test_start": "2025-04-24", |
| 116 | "test_end": "2026-04-23", |
| 117 | "bear_threshold": -0.00125, |
| 118 | "best_iter": 120 |
| 119 | } |
| 120 | } |
| 121 | } |