scripts/backtest_c15.py

#!/usr/bin/env python3
"""C15: Trend quality features — trend_efficiency_5d (directional purity of 5d move)
+ return_accel_5d (momentum acceleration). Pure OHLCV, no external data dependencies.
Source: arxiv.org/html/2512.12924v1 — "trending vs choppy behavior" as top feature class.
Pass gate: dir_accuracy > 43.5% AND up_precision > 54%.
"""
import sys
from pathlib import Path
ROOT = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(ROOT))
try:
    from dotenv import load_dotenv; load_dotenv(ROOT / ".env")
except ImportError:
    pass

import warnings; warnings.filterwarnings("ignore")
import json

sys.path.insert(0, str(ROOT / "scripts"))
from improvement_harness import BASELINE_FEATURES, run_comparison

STOCKS = ["2330", "0050", "2317", "2454", "2881"]

CURRENT_FEATURES = [f for f in BASELINE_FEATURES if f not in {
    "macd_cross_up", "macd_cross_down", "price_volume_div",
    "foreign_net_vol_ratio", "trust_net_vol_ratio", "dealer_net_vol_ratio",
    "institutional_net_vol_ratio", "institutional_5d_net_vol_ratio",
    "institutional_20d_zscore", "foreign_trust_alignment", "institutional_streak",
}]

C15_FEATURES = CURRENT_FEATURES + [
    "trend_efficiency_5d",  # |return_5d| / (vol_20d * sqrt(5)) — clean trend vs noise
    "return_accel_5d",      # return_5d.diff(5) — momentum acceleration
]

PASS_DIR_ACC = 43.5
PASS_UP_PREC = 54.0


def main():
    print(f"C15: baseline={len(CURRENT_FEATURES)} → candidate={len(C15_FEATURES)} features")
    print(f"New: trend_efficiency_5d, return_accel_5d\n")

    cmp = run_comparison(
        feature_sets={"baseline": CURRENT_FEATURES, "c15_trend": C15_FEATURES},
        label_mode="triple_barrier",
        stocks=STOCKS,
        output_path=None,
        pass_criterion={"dir_accuracy": PASS_DIR_ACC},
        verbose=True,
    )

    b = cmp["aggregate"].get("baseline", {})
    c = cmp["aggregate"].get("c15_trend", {})

    dir_delta  = c.get("dir_accuracy", 0) - b.get("dir_accuracy", 0)
    prec_delta = c.get("up_precision", 0) - b.get("up_precision", 0)

    passed = (
        c.get("dir_accuracy", 0) >= PASS_DIR_ACC and
        c.get("up_precision",  0) >= PASS_UP_PREC and
        dir_delta >= -0.5 and prec_delta >= -0.5
    )

    print("\n── Results ──────────────────────────────────────")
    print(f"  baseline  dir_acc={b.get('dir_accuracy')}%  up_prec={b.get('up_precision')}%")
    print(f"  c15_trend dir_acc={c.get('dir_accuracy')}%  up_prec={c.get('up_precision')}%")
    print(f"  Δ dir_acc={dir_delta:+.1f}pp  Δ up_prec={prec_delta:+.1f}pp")
    print(f"  Pass gate: dir>{PASS_DIR_ACC}% AND up_prec>{PASS_UP_PREC}%")
    print(f"  Result: {'PASSED' if passed else 'FAILED'}")

    result = {
        "experiment": "C15_trend_quality",
        "new_features": ["trend_efficiency_5d", "return_accel_5d"],
        "aggregate": cmp["aggregate"],
        "results": cmp["results"],
        "passed": passed,
    }
    Path("docs").mkdir(exist_ok=True)
    with open("docs/c15_result.json", "w") as f:
        json.dump(result, f, indent=2)
    print(f"  Saved → docs/c15_result.json")

    if passed:
        print("\n  ACTION: Add to FEATURE_COLUMNS in models/predictor.py:")
        for f in ["trend_efficiency_5d", "return_accel_5d"]:
            print(f"    \"{f}\",")


if __name__ == "__main__":
    main()