""" usdjpy_validation_v1.py ======================== Full validation of the best USDJPY config from the parameter sweep. Best config: z30 t2.0 tp0.50% sl0.50% h52h - Full Sharpe: 1.83 WR: 72.2% DD: -2.87% - OOS Sharpe: 2.40 WR: 73.7% N: 152 Tests: 1. Equity curve with buy & hold comparison 2. Yearly P&L heatmap 3. Monthly returns distribution 4. Drawdown chart 5. Walk-forward OOS validation 6. Win rate by z-score band 7. Direction analysis (LONG vs SHORT) 8. YCC regime split (pre/post 2016) Run from project root: python src/research/usdjpy_validation_v1.py """ import pandas as pd import numpy as np import matplotlib.pyplot as plt import matplotlib.gridspec as gridspec from pathlib import Path import sys import warnings warnings.filterwarnings('ignore') BASE_PATH = Path(__file__).resolve().parents[2] PROC_PATH = BASE_PATH / "data" / "processed" RATES_PATH = BASE_PATH / "data" / "raw" / "rates" OUT_PATH = BASE_PATH / "data" / "output" / "usdjpy" OUT_PATH.mkdir(parents=True, exist_ok=True) if str(BASE_PATH / "src") not in sys.path: sys.path.insert(0, str(BASE_PATH / "src")) # ── Locked best config ──────────────────────────────────────────────────────── Z_WINDOW = 30 THRESHOLD = 2.0 TP_PCT = 0.0050 SL_PCT = 0.0050 HOLD_HOURS = 52 FIB = 0.786 SPREAD_COST= 0.0001 NOTIONAL = 300_000 SESSION_S = 7 SESSION_E = 17 ZSCORE_EXIT= 1.5 WF_SPLIT = pd.Timestamp('2020-01-01') YCC_START = pd.Timestamp('2016-01-01') YCC_END = pd.Timestamp('2024-01-01') def load_rates(): us2y = pd.read_csv(RATES_PATH / "us2y.csv") dc = [c for c in us2y.columns if 'date' in c.lower()][0] vc = [c for c in us2y.columns if c != dc][0] us2y[dc] = pd.to_datetime(us2y[dc]) us2y = us2y.rename(columns={dc:'date', vc:'us2y'})[['date','us2y']].dropna() jp2y = pd.read_csv(RATES_PATH / "jp2y.csv") jp2y['date'] = pd.to_datetime(jp2y['date']) jp2y = jp2y[['date','jp2y']].dropna() rates = pd.merge(us2y, jp2y, on='date', how='outer').sort_values('date') rates = rates.set_index('date') full_idx = pd.date_range(rates.index.min(), rates.index.max(), freq='D') rates = rates.reindex(full_idx).ffill().dropna() rates['spread'] = rates['us2y'] - rates['jp2y'] s = rates['spread'] rates['zscore'] = (s - s.rolling(Z_WINDOW).mean()) / s.rolling(Z_WINDOW).std() return rates def load_prices(): fpath = next(BASE_PATH.rglob("USDJPY_15M*.csv"), None) df = pd.read_csv(fpath) df.columns = df.columns.str.strip().str.lower() dc = next(c for c in df.columns if 'date' in c or 'time' in c) df[dc] = pd.to_datetime(df[dc]) df = df.rename(columns={dc:'datetime'}).set_index('datetime').sort_index() return df def run_backtest(rates, m15): z_daily = rates['zscore'] full_idx_h = pd.date_range(z_daily.index.min(), z_daily.index.max(), freq='h') z_hourly = z_daily.reindex(full_idx_h).ffill() # Also load USDJPY 1H for buy & hold fpath_1h = next(BASE_PATH.rglob("USDJPY_1H*.csv"), None) bh_df = None if fpath_1h: bh_raw = pd.read_csv(fpath_1h) bh_raw.columns = bh_raw.columns.str.strip().str.lower() dc = next(c for c in bh_raw.columns if 'date' in c or 'time' in c) bh_raw[dc] = pd.to_datetime(bh_raw[dc]) bh_df = bh_raw.rename(columns={dc:'datetime'}).set_index('datetime').sort_index() trades = [] last_exit = None for dt in pd.date_range('2003-01-01','2026-04-01',freq='h'): if dt not in z_hourly.index: continue z = float(z_hourly[dt]) if abs(z) < THRESHOLD: continue h = (dt.hour + 2) % 24 if not (SESSION_S <= h <= SESSION_E): continue if last_exit is not None and dt <= last_exit: continue direction = 1 if z >= THRESHOLD else -1 bar_end = dt + pd.Timedelta(hours=1) - pd.Timedelta(minutes=1) bar_slice = m15.loc[dt:bar_end] if dt in m15.index else pd.DataFrame() if bar_slice.empty: continue bh = float(bar_slice['high'].max()) bl = float(bar_slice['low'].min()) bc = float(bar_slice['close'].iloc[-1]) if direction == 1: pull = bc - bl if pull <= 0.05: continue target = bc - FIB * pull else: pull = bh - bc if pull <= 0.05: continue target = bc + FIB * pull entry_window = m15.loc[ dt + pd.Timedelta(minutes=1):dt + pd.Timedelta(hours=6)] entry_price = entry_time = None for edt, ebar in entry_window.iterrows(): if direction==1 and float(ebar['low']) <= target: entry_price = target; entry_time = edt; break elif direction==-1 and float(ebar['high']) >= target: entry_price = target; entry_time = edt; break if entry_price is None: continue tp = entry_price*(1+TP_PCT) if direction==1 else entry_price*(1-TP_PCT) sl = entry_price*(1-SL_PCT) if direction==1 else entry_price*(1+SL_PCT) hold = m15.loc[ entry_time + pd.Timedelta(minutes=1): entry_time + pd.Timedelta(hours=HOLD_HOURS)] exit_price = None exit_reason = "hold_expiry" for hdt, hbar in hold.iterrows(): hr = hdt.replace(minute=0,second=0,microsecond=0) if hr in z_hourly.index: hz = float(z_hourly[hr]) if ((direction==1 and hz<=-ZSCORE_EXIT) or (direction==-1 and hz>=ZSCORE_EXIT)): exit_price = float(hbar['close']); exit_reason="z_exit"; break if direction==1: if float(hbar['high']) >= tp: exit_price=tp; exit_reason="tp"; break if float(hbar['low']) <= sl: exit_price=sl; exit_reason="stop"; break else: if float(hbar['low']) <= tp: exit_price=tp; exit_reason="tp"; break if float(hbar['high']) >= sl: exit_price=sl; exit_reason="stop"; break if exit_price is None: exit_price = float(hold['close'].iloc[-1]) if not hold.empty else entry_price pnl = round(((exit_price-entry_price)/entry_price*direction - SPREAD_COST)*NOTIONAL, 2) zscore_abs = abs(z) band = "2.0-2.5" if zscore_abs<2.5 else ("2.5-3.0" if zscore_abs<3.0 else "3.0+") trades.append({ 'entry_time': entry_time, 'direction': direction, 'zscore': round(z,4), 'zscore_abs': round(zscore_abs,4), 'zscore_band': band, 'exit_reason': exit_reason, 'entry_price': entry_price, 'exit_price': exit_price, 'pnl': pnl, 'year': entry_time.year, 'month': entry_time.month, 'regime': ('ycc' if YCC_START<=entry_time=YCC_END else 'pre_ycc')), }) last_exit = entry_time + pd.Timedelta(hours=HOLD_HOURS) return pd.DataFrame(trades), bh_df def make_charts(trades, rates, bh_df): print("Generating charts...") df = trades.copy().sort_values('entry_time').reset_index(drop=True) df['equity'] = 100_000 + df['pnl'].cumsum() df['peak'] = df['equity'].cummax() df['drawdown'] = (df['equity'] - df['peak']) / df['peak'] * 100 dates = pd.to_datetime(df['entry_time']) # 1. Equity curve fig, ax = plt.subplots(figsize=(14,5)) if bh_df is not None: try: bh = bh_df.loc[bh_df.index >= df['entry_time'].min()]['close'] ax.plot(bh.index, bh/bh.iloc[0]*100_000, color='grey', linewidth=1, alpha=0.5, label='Buy & Hold') except Exception: pass ax.plot(dates, df['equity'], color='#00ff88', linewidth=1.5, label='Strategy') ax.axvline(WF_SPLIT, color='gold', linewidth=1, linestyle='--', label='OOS split') ax.axhline(100_000, color='#666', linewidth=0.5) ax.set_title(f'USDJPY Equity Curve — z{Z_WINDOW} ±{THRESHOLD} TP{TP_PCT*100:.2f}% SL{SL_PCT*100:.2f}% H{HOLD_HOURS}h') ax.yaxis.set_major_formatter(plt.FuncFormatter(lambda x,_: f'${x:,.0f}')) ax.legend(); ax.grid(alpha=0.2) fig.tight_layout() fig.savefig(OUT_PATH / 'usdjpy_equity.png', dpi=150) print(' Saved: usdjpy_equity.png') plt.close(fig) # 2. Drawdown fig, ax = plt.subplots(figsize=(14,4)) ax.fill_between(dates, df['drawdown'], 0, color='#ff4444', alpha=0.6) ax.set_title(f'Drawdown | Max: {df["drawdown"].min():.2f}%') ax.set_ylabel('%'); ax.grid(alpha=0.2) fig.tight_layout() fig.savefig(OUT_PATH / 'usdjpy_drawdown.png', dpi=150) print(' Saved: usdjpy_drawdown.png') plt.close(fig) # 3. Yearly P&L fig, ax = plt.subplots(figsize=(14,5)) yearly = df.groupby('year')['pnl'].sum() colors = ['#00ff88' if v >= 0 else '#ff4444' for v in yearly.values] ax.bar(yearly.index, yearly.values/1000, color=colors, alpha=0.8) ax.axhline(0, color='#666', linewidth=0.5) ax.set_title('Yearly P&L ($k)') ax.set_ylabel('$k'); ax.grid(alpha=0.2, axis='y') fig.tight_layout() fig.savefig(OUT_PATH / 'usdjpy_yearly.png', dpi=150) print(' Saved: usdjpy_yearly.png') plt.close(fig) # 4. Monthly heatmap (pure matplotlib — no seaborn needed) hmap = df.pivot_table(values='pnl', index='year', columns='month', aggfunc='sum') / 1000 month_names = ['Jan','Feb','Mar','Apr','May','Jun', 'Jul','Aug','Sep','Oct','Nov','Dec'] hmap.columns = [month_names[m-1] for m in hmap.columns] fig, ax = plt.subplots(figsize=(16, max(6, len(hmap)*0.4+2))) vmax = max(abs(hmap.values[~np.isnan(hmap.values)]).max(), 1) im = ax.imshow(hmap.values, cmap='RdYlGn', aspect='auto', vmin=-vmax, vmax=vmax) ax.set_xticks(range(len(hmap.columns))) ax.set_xticklabels(hmap.columns, fontsize=9) ax.set_yticks(range(len(hmap.index))) ax.set_yticklabels(hmap.index, fontsize=9) for i in range(len(hmap.index)): for j in range(len(hmap.columns)): val = hmap.values[i, j] if not np.isnan(val): ax.text(j, i, f'{val:.1f}', ha='center', va='center', fontsize=8, color='black' if abs(val) < vmax*0.6 else 'white') plt.colorbar(im, ax=ax, label='P&L $k') ax.set_title('Monthly P&L Heatmap ($k)') fig.tight_layout() fig.savefig(OUT_PATH / 'usdjpy_heatmap.png', dpi=150) print(' Saved: usdjpy_heatmap.png') plt.close(fig) plt.close('all') def print_summary(trades, rates): df = trades.copy() years = (df['entry_time'].max() - df['entry_time'].min()).days/365.25 is_df = df[df['entry_time'] < WF_SPLIT] oos_df = df[df['entry_time'] >= WF_SPLIT] yrs_oos = (df['entry_time'].max() - WF_SPLIT).days/365.25 yrs_is = (WF_SPLIT - df['entry_time'].min()).days/365.25 def sh(d, yrs): if len(d)==0: return 0,0,0,0,0 pnl=d['pnl'].values; wr=(pnl>0).mean()*100 eq=100_000+np.cumsum(pnl); pk=np.maximum.accumulate(eq) dd=((eq-pk)/pk*100).min(); pyr=len(pnl)/yrs exc=pnl/100_000-0.04/pyr sharpe=exc.mean()/exc.std()*np.sqrt(pyr) if exc.std()>0 else 0 return round(sharpe,2),round(wr,1),round(dd,2),int(pnl.sum()),len(pnl) sf,wf,df2,tf,nf = sh(df, years) si,wi,di,ti,ni = sh(is_df, yrs_is) so,wo,do,to,no = sh(oos_df, yrs_oos) print(f"\n{'='*72}") print("VALIDATION SUMMARY — USDJPY z30 t2.0 TP0.50% SL0.50% H52h") print(f"{'='*72}") print(f"\n {'Period':<22}{'N':>6}{'N/yr':>6}{'WR%':>6}{'Sharpe':>8}" f"{'MaxDD%':>8}{'Total$':>12}") print(f" {'─'*65}") pyr_f = nf/years; pyr_i=ni/yrs_is; pyr_o=no/yrs_oos print(f" {'Full (2003-2026)':<22}{nf:>6}{pyr_f:>6.0f}{wf:>6.1f}" f"{sf:>8.2f}{df2:>8.2f}{tf:>12,.0f}") print(f" {'IS (2003-2020)':<22}{ni:>6}{pyr_i:>6.0f}{wi:>6.1f}" f"{si:>8.2f}{di:>8.2f}{ti:>12,.0f}") print(f" {'OOS (2020-2026)':<22}{no:>6}{pyr_o:>6.0f}{wo:>6.1f}" f"{so:>8.2f}{do:>8.2f}{to:>12,.0f}") # Compare with EURUSD print(f"\n COMPARISON WITH EURUSD MODEL:") print(f" {'Metric':<25}{'EURUSD':>12}{'USDJPY':>12}") print(f" {'─'*50}") print(f" {'Full Sharpe':<25}{'3.72':>12}{sf:>12.2f}") print(f" {'OOS Sharpe':<25}{'2.73':>12}{so:>12.2f}") print(f" {'Win Rate':<25}{'75.2%':>12}{wf:>11.1f}%") print(f" {'Max Drawdown':<25}{'-4.28%':>12}{df2:>11.2f}%") print(f" {'Trades/year':<25}{'182':>12}{pyr_f:>12.0f}") print(f" {'Threshold':<25}{'±2.75':>12}{f'±{THRESHOLD}':>12}") print(f" {'TP':<25}{'0.20%':>12}{f'{TP_PCT*100:.2f}%':>12}") # Direction split print(f"\n DIRECTION ANALYSIS:") for d, dlabel in [(1,'LONG (Short JPY)'),(-1,'SHORT (Long JPY)')]: sub = df[df['direction']==d] wr = (sub['pnl']>0).mean()*100 tot = sub['pnl'].sum() print(f" {dlabel:<22}: n={len(sub):>4} WR={wr:.1f}% Total=${tot:>10,.0f}") # Regime split print(f"\n BoJ REGIME ANALYSIS:") for reg, rlabel in [('pre_ycc','Pre-YCC (<2016)'), ('ycc','YCC Era (2016-2024)'), ('post_ycc','Post-YCC (2024+)')]: sub = df[df['regime']==reg] if len(sub)==0: continue wr = (sub['pnl']>0).mean()*100 tot = sub['pnl'].sum() print(f" {rlabel:<22}: n={len(sub):>4} WR={wr:.1f}% Total=${tot:>10,.0f}") # Exit reason breakdown print(f"\n EXIT BREAKDOWN:") for reason, n in df['exit_reason'].value_counts().items(): pct = n/len(df)*100 sub_wr = (df[df['exit_reason']==reason]['pnl']>0).mean()*100 print(f" {reason:<15}: {n:>4} ({pct:.1f}%) WR={sub_wr:.1f}%") # Z-score band breakdown print(f"\n Z-SCORE BAND BREAKDOWN:") for band in ['2.0-2.5','2.5-3.0','3.0+']: sub = df[df['zscore_band']==band] if len(sub)==0: continue wr = (sub['pnl']>0).mean()*100 avg = sub['pnl'].mean() print(f" z {band:<10}: n={len(sub):>4} WR={wr:.1f}% Avg=${avg:.0f}") print(f"\n Charts saved to: {OUT_PATH}") def main(): print("="*72) print("USDJPY VALIDATION v1") print(f"Config: z{Z_WINDOW} threshold=±{THRESHOLD} " f"TP={TP_PCT*100:.2f}% SL={SL_PCT*100:.2f}% hold={HOLD_HOURS}h") print("="*72) rates = load_rates() m15 = load_prices() print("\nRunning backtest...") trades, bh_df = run_backtest(rates, m15) print(f" {len(trades):,} trades generated") print_summary(trades, rates) make_charts(trades, rates, bh_df) # Save trades out = PROC_PATH / "usdjpy_trades_validated.csv" trades.to_csv(out, index=False) print(f"\n Trades saved: {out.name}") if __name__ == "__main__": main()