| """
|
| Multi-symbol screen of the RECOVERED live configuration.
|
|
|
| The parameters are not searched. They are the ones recovered from the surviving
|
| tester configs (z_*.ini / m*.ini) of the lost 2026-09-13 build:
|
|
|
| Period=H4 DipEntry=1 (z-score) DipZ=1.5 DipExitMA=20
|
| DipTrendMA=0 DipMaxBars=10 Direction=1 (long only) StopMode=3
|
|
|
| Testing a recovered configuration instead of fitting a new one is the single
|
| biggest defence against the curve-fit result this project keeps producing:
|
| there is no free parameter left to tune, so an OOS number means something.
|
|
|
| Every candidate is measured against a LONG-ONLY RANDOM ENTRY with matched
|
| holding period. On a drifting index that is the only honest control.
|
| """
|
|
|
| from __future__ import annotations
|
|
|
| import sys
|
| import numpy as np
|
|
|
| sys.path.insert(0, __file__.rsplit("\\", 1)[0] if "\\" in __file__ else ".")
|
|
|
| import backtest as bt # noqa: E402
|
|
|
| RISK = 0.01
|
| STOP_ATR = 3.0
|
| PROP_DD = 0.05 # 5% prop limit
|
|
|
|
|
| def zscore_entries(d, n=20, z_th=-1.5, trend_ma=0):
|
| c = d["c"]
|
| m = bt.sma(c, n)
|
| s = bt.rolling_std(c, n)
|
| z = (c - m) / np.where(s > 0, s, np.nan)
|
| e = z <= z_th
|
| if trend_ma and trend_ma > 0:
|
| e &= c > bt.sma(c, trend_ma)
|
| return np.nan_to_num(e, nan=0).astype(bool), bt.sma(c, n)
|
|
|
|
|
| def rsi_entries(d, n=2, th=10.0):
|
| return bt.rsi(d["c"], n) <= th, bt.sma(d["c"], 20)
|
|
|
|
|
| def breakout_entries(d, n=20):
|
| c, h = d["c"], d["h"]
|
| hh = np.full(len(c), np.nan)
|
| for i in range(n, len(c)):
|
| hh[i] = h[i - n:i].max()
|
| return (c > hh), bt.sma(c, 20)
|
|
|
|
|
| def random_entries(d, rate, seed=0):
|
| rng = np.random.default_rng(seed)
|
| return rng.random(len(d["c"])) < rate, bt.sma(d["c"], 20)
|
|
|
|
|
| def evaluate(d, entries, exit_ma, side=1, max_bars=10, lo=None, hi=None):
|
| ts = d["ts"]
|
| mask = np.ones(len(ts), bool)
|
| if lo is not None:
|
| mask &= ts >= np.datetime64(lo)
|
| if hi is not None:
|
| mask &= ts < np.datetime64(hi)
|
| e = entries & mask
|
| tr = bt.simulate(d, e, side=side, exit_ma=exit_ma, max_bars=max_bars, stop_atr=STOP_ATR)
|
| tr = bt.add_r(tr, d, stop_atr=STOP_ATR)
|
| sub = ts[mask]
|
| if len(sub) < 10 or not tr:
|
| return None
|
| return bt.metrics(tr, sub, risk_frac=RISK, stop_atr=STOP_ATR), tr
|
|
|
|
|
| def row(tag, m):
|
| if m is None:
|
| return f"{tag:<22} no trades"
|
| flags = []
|
| if m["per_month"] < 2.0:
|
| flags.append("cadence")
|
| if m["maxdd"] > PROP_DD:
|
| flags.append("maxDD")
|
| if not (m["ret_dd"] >= 2.0):
|
| flags.append("ret/DD")
|
| verdict = "PASS" if not flags else "fail:" + ",".join(flags)
|
| return (f"{tag:<22}{m['n']:>6}{m['per_month']:>7.1f}{m['exp_bp']:>9.1f}"
|
| f"{m['hit']:>7.2f}{m['total']:>9.1%}{m['maxdd']:>8.1%}"
|
| f"{m['ret_dd']:>8.2f} {verdict}")
|
|
|
|
|
| HDR = (f"{'candidate':<22}{'n':>6}{'/mo':>7}{'exp_bp':>9}{'hit':>7}"
|
| f"{'total':>9}{'maxDD':>8}{'ret/DD':>8} screen")
|
|
|
| SYMS = ["SP500", "NAS100", "US30", "DAX40", "EURUSD", "USDJPY", "XAUUSD"]
|
| SPLIT = "2024-01-01"
|
|
|
| if __name__ == "__main__":
|
| period = sys.argv[1] if len(sys.argv) > 1 else "PERIOD_H4"
|
| print(f"=== RECOVERED CONFIG, {period}, long only, z<=-1.5, exit SMA20, "
|
| f"max 10 bars, stop 3 ATR, risk {RISK:.0%} ===")
|
| print(f"=== IS: start..{SPLIT} OOS: {SPLIT}..end (prop limit {PROP_DD:.0%}) ===\n")
|
| for s in SYMS:
|
| try:
|
| d = bt.load(s, period)
|
| except Exception as ex: # noqa: BLE001
|
| print(f"{s}: load failed {ex}")
|
| continue
|
| e, xma = zscore_entries(d, 20, -1.5, 0)
|
| print(f"--- {s} ({d['ts'][0]} .. {d['ts'][-1]}, {len(d['c'])} bars, "
|
| f"median spread {np.median(d['cost'] / d['c']) * 1e4:.2f} bp) ---")
|
| print(HDR)
|
| for tag, lo, hi in (("dip-z IS", None, SPLIT), ("dip-z OOS", SPLIT, None),
|
| ("dip-z FULL", None, None)):
|
| r = evaluate(d, e, xma, 1, 10, lo, hi)
|
| print(row(tag, r[0] if r else None))
|
| # control: long-only random entry, matched trade count
|
| full = evaluate(d, e, xma, 1, 10, None, None)
|
| if full:
|
| rate = full[0]["n"] / len(d["c"])
|
| accs = []
|
| for sd in range(8):
|
| re_, rma = random_entries(d, rate, seed=sd)
|
| rr = evaluate(d, re_, rma, 1, 10, None, None)
|
| if rr:
|
| accs.append(rr[0])
|
| if accs:
|
| print(f"{'random-long control':<22}{np.mean([a['n'] for a in accs]):>6.0f}"
|
| f"{np.mean([a['per_month'] for a in accs]):>7.1f}"
|
| f"{np.mean([a['exp_bp'] for a in accs]):>9.1f}"
|
| f"{np.mean([a['hit'] for a in accs]):>7.2f}"
|
| f"{np.mean([a['total'] for a in accs]):>9.1%}"
|
| f"{np.mean([a['maxdd'] for a in accs]):>8.1%}"
|
| f"{np.nanmean([a['ret_dd'] for a in accs]):>8.2f} (8 seeds)")
|
| print(f"{' edge over control':<22}"
|
| f"{'':>6}{'':>7}{full[0]['exp_bp'] - np.mean([a['exp_bp'] for a in accs]):>+9.1f} bp")
|
| print()
|
ff46cfd57e