scripts/backtest_c30.py

#!/usr/bin/env python3
"""C30: Sector peer return features.

Hypothesis: Stocks within the same sector exhibit lead-lag relationships.
TSMC often leads UMC by 1 trading day; Fubon leads Cathay and Mega.
Adding yesterday's average peer return as a feature lets the model capture
this intra-sector momentum signal.

New features (non-leaking — all use shift(1)):
  peer_ret_1d  — prior-day mean return of same-sector peers
  peer_ret_5d  — 5-day return of peers ending yesterday

All 12 stocks are loaded to compute peer returns even for the 5-stock eval set.
Stocks with no peers (2412 telecom) get 0.0 fill.
"""
import sys
from pathlib import Path

ROOT = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(ROOT))

import warnings; warnings.filterwarnings("ignore")
import json
import argparse
import numpy as np
import pandas as pd

from scripts.improvement_harness import (
    fetch_df, build_triple_barrier_labels, walk_forward, compute_metrics,
    CURRENT_FEATURES, DEFAULT_STOCKS, EXTENDED_STOCKS,
    PASS_DIR_ACC, PASS_UP_PREC,
)
from models.predictor import _build_features

SECTOR_MAP = {
    "semis":       ["2330", "2454", "2303"],
    "electronics": ["2317", "2382", "2308"],
    "financials":  ["2881", "2882", "2886"],
    "etfs":        ["0050", "0056"],
    "telecom":     ["2412"],
}
STOCK_SECTOR = {s: sec for sec, members in SECTOR_MAP.items() for s in members}

C30_FEATURES = ["peer_ret_1d", "peer_ret_5d"]
NEW_FEATURES  = CURRENT_FEATURES + C30_FEATURES

ALL_STOCKS = [s for members in SECTOR_MAP.values() for s in members]


def _build_peer_close(stock_dfs: dict[str, pd.DataFrame]) -> dict[str, pd.Series]:
    """Return {stock_no: close Series indexed by YYYY-MM-DD string}."""
    out = {}
    for s, df in stock_dfs.items():
        if df is not None and not df.empty and "date" in df.columns:
            out[s] = df.set_index("date")["close"].astype(float)
    return out


def _add_peer_features(feat: pd.DataFrame, df: pd.DataFrame,
                       stock_no: str, peer_close: dict[str, pd.Series]) -> pd.DataFrame:
    sector  = STOCK_SECTOR.get(stock_no, "")
    peers   = [p for p in SECTOR_MAP.get(sector, []) if p != stock_no and p in peer_close]

    if not peers or "date" not in df.columns:
        feat["peer_ret_1d"] = 0.0
        feat["peer_ret_5d"] = 0.0
        return feat

    date_col = df["date"].astype(str).reset_index(drop=True)

    r1_cols, r5_cols = [], []
    for p in peers:
        s = peer_close[p]
        r1 = s.pct_change(1).shift(1)   # prior-day 1d return — non-leaking
        r5 = s.pct_change(5).shift(1)   # 5d return ending yesterday — non-leaking
        r1_cols.append(date_col.map(r1.to_dict()).astype(float))
        r5_cols.append(date_col.map(r5.to_dict()).astype(float))

    feat = feat.reset_index(drop=True)
    feat["peer_ret_1d"] = pd.concat(r1_cols, axis=1).mean(axis=1).ffill().bfill().fillna(0.0).values
    feat["peer_ret_5d"] = pd.concat(r5_cols, axis=1).mean(axis=1).ffill().bfill().fillna(0.0).values
    return feat


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--extended", action="store_true")
    args = parser.parse_args()

    stocks = EXTENDED_STOCKS if args.extended else DEFAULT_STOCKS
    tag    = "12-stock" if args.extended else "5-stock"
    suffix = "_12stock" if args.extended else ""

    print(f"\n=== C30: Sector peer return features [{tag}] ===")
    print(f"  New features: {C30_FEATURES}\n")

    # Load all 12 stocks for peer close prices
    print("  Loading all stock data for peer returns...")
    all_dfs: dict[str, pd.DataFrame] = {}
    for s in ALL_STOCKS:
        df = fetch_df(s)
        if df is not None and not df.empty:
            all_dfs[s] = df
    peer_close = _build_peer_close(all_dfs)
    print(f"  Loaded {len(peer_close)} stocks\n")

    def _avg(results, key):
        vals = [m[key] for m in results if m and not np.isnan(m.get(key, float("nan")))]
        return round(float(np.mean(vals)), 1) if vals else float("nan")

    per_stock = {}
    base_results, c30_results = [], []

    for stock_no in stocks:
        print(f"  {stock_no}...", end=" ", flush=True)
        df = all_dfs.get(stock_no)
        if df is None or df.empty:
            print("no data"); continue

        feat   = _build_features(df)
        feat   = _add_peer_features(feat, df, stock_no, peer_close)
        close  = (df.set_index("date")["close"] if "date" in df.columns else df["close"]).values
        labels = build_triple_barrier_labels(close)

        sector = STOCK_SECTOR.get(stock_no, "?")
        peers  = [p for p in SECTOR_MAP.get(sector, []) if p != stock_no and p in peer_close]

        m_base = walk_forward(feat, labels, CURRENT_FEATURES)
        m_c30  = walk_forward(feat, labels, NEW_FEATURES)

        per_stock[stock_no] = {"baseline": m_base, "c30": m_c30, "peers": peers}
        base_results.append(m_base)
        c30_results.append(m_c30)

        b_dir = m_base.get("dir_accuracy", float("nan"))
        b_up  = m_base.get("up_precision", float("nan"))
        c_dir = m_c30.get("dir_accuracy",  float("nan"))
        c_up  = m_c30.get("up_precision",  float("nan"))
        print(f"peers={peers}  base dir={b_dir}% ↑prec={b_up}%  →  c30 dir={c_dir}% ↑prec={c_up}%")

    base_agg = {"dir_accuracy": _avg(base_results, "dir_accuracy"),
                "up_precision": _avg(base_results, "up_precision")}
    c30_agg  = {"dir_accuracy": _avg(c30_results,  "dir_accuracy"),
                "up_precision": _avg(c30_results,  "up_precision")}

    passed = c30_agg["dir_accuracy"] >= PASS_DIR_ACC and c30_agg["up_precision"] >= PASS_UP_PREC

    print(f"\n  Baseline avg: dir={base_agg['dir_accuracy']}%  ↑prec={base_agg['up_precision']}%")
    print(f"  C30 avg:      dir={c30_agg['dir_accuracy']}%   ↑prec={c30_agg['up_precision']}%")
    print(f"  Gate (dir≥{PASS_DIR_ACC}% AND ↑prec≥{PASS_UP_PREC}%): {'PASS ✓' if passed else 'FAIL ✗'}")

    result = {
        "experiment":   "C30",
        "description":  "Sector peer return features (peer_ret_1d, peer_ret_5d)",
        "new_features": C30_FEATURES,
        "stocks":       stocks,
        "aggregate":    {"baseline": base_agg, "c30": c30_agg},
        "passed":       passed,
        "pass_gate":    {"dir_accuracy": PASS_DIR_ACC, "up_precision": PASS_UP_PREC},
        "per_stock":    per_stock,
    }

    out = ROOT / f"docs/c30_result{suffix}.json"
    out.parent.mkdir(exist_ok=True)
    out.write_text(json.dumps(result, indent=2))
    print(f"\n  Saved: {out}")
    return 0 if passed else 1


if __name__ == "__main__":
    sys.exit(main())