""" Monte Carlo on pure Away CLV, OOS-only seasons (2122-2526). No outcome selection -- just Away bets where B365A > Pinnacle fair by >5%. Seasons 1920+2021 treated as the "identification" period; 2122 onward is OOS. """ import pandas as pd import numpy as np import matplotlib matplotlib.use('Agg') import matplotlib.pyplot as plt from matplotlib.ticker import FuncFormatter np.random.seed(42) N_SIMS = 10_000 START = 1_000.0 KELLY_MULTS = [0.10, 0.25, 0.50] EDGE_THRESH = 0.05 OOS_SEASONS = ['2122','2223','2324','2425','2526'] PLOT_OUT = r'E:\soccer\mc_oos_away_clv.png' seasons = ['1920','2021','2122','2223','2324','2425','2526'] league_names = { 'E0': 'EPL', 'E1': 'Championship', 'E2': 'League One', 'E3': 'League Two', 'EC': 'Conference', 'SP1': 'La Liga', 'SP2': 'Segunda', 'I1': 'Serie A', 'I2': 'Serie B', 'D1': 'Bundesliga', 'D2': '2.Bundesliga', 'F1': 'Ligue 1', 'F2': 'Ligue 2', 'N1': 'Eredivisie', 'B1': 'Belgian Pro League', 'P1': 'Primeira Liga', 'T1': 'Super Lig', 'G1': 'Greek SL', 'SC0': 'Scottish Prem', 'SC1': 'Scottish Champ', 'SC2': 'Scottish L1', 'SC3': 'Scottish L2', } def remove_margin(h, d, a): over = 1/h + 1/d + 1/a return 1/h/over, 1/d/over, 1/a/over # ── Load data ────────────────────────────────────────────────────────────────── print("Loading data...") all_dfs = [] for code, name in league_names.items(): for s in seasons: path = f'E:/soccer/league_data/{code}_{s}.csv' try: d = pd.read_csv(path, encoding='utf-8-sig') if len(d) < 10 or 'FTR' not in d.columns or 'B365A' not in d.columns: continue d = d.dropna(subset=['FTR','B365A','B365H','B365D','PSCH','PSCD','PSCA']) if len(d) < 10: continue d['Season'] = s d['LeagueCode'] = code d['LeagueName'] = name all_dfs.append(d) except: pass df = pd.concat(all_dfs, ignore_index=True) df['Date'] = pd.to_datetime(df['Date'], dayfirst=True, errors='coerce') df = df.dropna(subset=['Date']).sort_values('Date').reset_index(drop=True) print(f" {len(df):,} matches across {df['LeagueName'].nunique()} leagues") # ── Compute fair probs and edge ──────────────────────────────────────────────── tH, tD, tA = zip(*df.apply(lambda r: remove_margin(r['PSCH'],r['PSCD'],r['PSCA']), axis=1)) df = df.copy() df['p_A'] = pd.Series(tA, index=df.index) df['PSC_fair_A'] = 1 / df['p_A'] df['edge_A'] = (df['B365A'] - df['PSC_fair_A']) / df['PSC_fair_A'] # ── Filter to OOS seasons, Away CLV > 5% ────────────────────────────────────── sub = df[(df['Season'].isin(OOS_SEASONS)) & (df['edge_A'] > EDGE_THRESH)].copy() # Build bet list with Kelly fractions rows = [] for _, r in sub.iterrows(): b = r['B365A'] - 1 p = r['p_A'] kelly_f = max((p * b - (1 - p)) / b, 0) if b > 0 else 0 if kelly_f > 0: rows.append({ 'Date': r['Date'], 'Season': r['Season'], 'League': r['LeagueName'], 'Odds': r['B365A'], 'Kelly': kelly_f, 'p_win': p, 'Won': r['FTR'] == 'A', }) bets = pd.DataFrame(rows).sort_values('Date').reset_index(drop=True) print(f" {len(bets):,} OOS Away CLV bets ({bets['Date'].min().date()} to {bets['Date'].max().date()})") print(f" Overall win rate: {bets['Won'].mean()*100:.1f}%") print(f" Avg Pinnacle fair prob: {bets['p_win'].mean()*100:.1f}%") # Flat stake summary flat_pl = [(r['Odds']-1) if r['Won'] else -1 for _, r in bets.iterrows()] print(f" Flat ROI: {sum(flat_pl)/len(flat_pl)*100:+.2f}%") print(f"\n By season (flat stake):") for s in OOS_SEASONS: sg = bets[bets['Season'] == s] if len(sg) == 0: continue fp = [(r['Odds']-1) if r['Won'] else -1 for _, r in sg.iterrows()] print(f" {s} n={len(sg):>4} WR={sg['Won'].mean()*100:.1f}% ROI={sum(fp)/len(fp)*100:+.1f}%") # ── Deterministic Kelly compounding ─────────────────────────────────────────── print(f"\n{'='*65}") print("DETERMINISTIC KELLY COMPOUNDING (OOS only)") print(f"{'='*65}") print(f" {'Kelly mult':<16} {'Final':>12} {'x':>7} {'MaxDD':>7}") print(" " + "-"*50) for km in KELLY_MULTS + [1.0]: bk = START; peak = START; max_dd = 0.0 for _, bet in bets.iterrows(): f = min(bet['Kelly'] * km, 0.25) stake = bk * f bk += stake * (bet['Odds']-1) if bet['Won'] else -stake peak = max(peak, bk) max_dd = max(max_dd, (peak - bk) / peak) print(f" {km:.2f}x Kelly {bk:>12,.2f} {bk/START:>7.1f}x {max_dd*100:>6.1f}%") # ── Monte Carlo ─────────────────────────────────────────────────────────────── print(f"\n{'='*65}") print(f"MONTE CARLO ({N_SIMS:,} simulations, OOS bets only)") print(f"{'='*65}") N = len(bets) odds = bets['Odds'].values p_win = bets['p_win'].values kelly = bets['Kelly'].values mc_results = {} N_PLOT = 500 for km in KELLY_MULTS: stakes = np.minimum(kelly * km, 0.25) terminal = np.zeros(N_SIMS) max_dds = np.zeros(N_SIMS) paths = np.zeros((N_PLOT, N + 1)) paths[:, 0] = START for sim in range(N_SIMS): outcomes = np.random.random(N) < p_win bk = START; peak = START; max_dd = 0.0 last_i = N - 1 for i in range(N): s = bk * stakes[i] bk += s * (odds[i] - 1) if outcomes[i] else -s if bk <= 0: bk = 0; last_i = i; break peak = max(peak, bk) max_dd = max(max_dd, (peak - bk) / peak) if sim < N_PLOT: paths[sim, i+1] = bk terminal[sim] = bk max_dds[sim] = max_dd if sim < N_PLOT and bk <= 0: paths[sim, last_i+1:] = 0 mc_results[km] = {'terminal': terminal, 'max_dds': max_dds, 'paths': paths} print(f"\n {km:.2f}x Kelly:") print(f" Median: {np.median(terminal)/START:>8.1f}x (£{np.median(terminal):>10,.0f})") print(f" P10: {np.percentile(terminal,10)/START:>8.1f}x") print(f" P25: {np.percentile(terminal,25)/START:>8.1f}x") print(f" P75: {np.percentile(terminal,75)/START:>8.1f}x") print(f" P90: {np.percentile(terminal,90)/START:>8.1f}x") print(f" P(profit): {(terminal>START).mean()*100:>7.1f}%") print(f" P(> 5x): {(terminal>START*5).mean()*100:>7.1f}%") print(f" P(> 10x): {(terminal>START*10).mean()*100:>7.1f}%") print(f" P(ruin <50%): {(terminal7.1f}%") print(f" Median MaxDD: {np.median(max_dds)*100:>7.1f}%") print(f" 95th pct DD: {np.percentile(max_dds,95)*100:>7.1f}%") # ── Summary table ───────────────────────────────────────────────────────────── print(f"\n{'='*70}") print(f" {'Metric':<22}", end="") for km in KELLY_MULTS: print(f" {km:.2f}x Kelly", end="") print() print(" " + "-"*60) rows_table = [ ("Median", lambda t: f"{np.median(t)/START:>10.1f}x"), ("P10", lambda t: f"{np.percentile(t,10)/START:>10.1f}x"), ("P25", lambda t: f"{np.percentile(t,25)/START:>10.1f}x"), ("P75", lambda t: f"{np.percentile(t,75)/START:>10.1f}x"), ("P90", lambda t: f"{np.percentile(t,90)/START:>10.1f}x"), ("P(profit)", lambda t: f"{(t>START).mean()*100:>9.1f}%"), ("P(> 5x)", lambda t: f"{(t>START*5).mean()*100:>9.1f}%"), ("P(> 10x)", lambda t: f"{(t>START*10).mean()*100:>9.1f}%"), ("P(ruin)", lambda t: f"{(t9.1f}%"), ("Med MaxDD", lambda t: f"{np.median(mc_results[km]['max_dds'])*100:>8.1f}%"), ] for label, fn in rows_table: print(f" {label:<22}", end="") for km in KELLY_MULTS: t = mc_results[km]['terminal'] if label == "Med MaxDD": print(f" {np.median(mc_results[km]['max_dds'])*100:>9.1f}%", end="") else: print(f" {fn(t)}", end="") print() # ── Chart ───────────────────────────────────────────────────────────────────── colors = {0.10: 'steelblue', 0.25: 'seagreen', 0.50: 'darkorange'} fig, axes = plt.subplots(1, 3, figsize=(18, 6)) fig.suptitle( f'Monte Carlo: Pure Away CLV (>5% edge vs Pinnacle) — OOS Seasons 2122-2526\n' f'{N_SIMS:,} simulations, {N:,} bets, {bets["Date"].min().date()} to {bets["Date"].max().date()}\n' f'All leagues, no outcome selection. Outcomes sampled from Pinnacle fair probabilities.', fontsize=10, fontweight='bold' ) ax = axes[0] for km in KELLY_MULTS: t = mc_results[km]['terminal'] cap = np.percentile(t, 95) ax.hist(np.minimum(t, cap), bins=100, alpha=0.45, color=colors[km], label=f'{km:.2f}x Kelly median={np.median(t)/START:.1f}x', edgecolor='none') ax.axvline(np.median(t), color=colors[km], linewidth=2, linestyle='--') ax.axvline(START, color='gray', linewidth=1, linestyle=':', label='Start £1,000') ax.set_xlabel('Terminal bankroll (£)') ax.set_ylabel('Frequency') ax.set_title('Terminal distribution (capped at 95th pct)') ax.legend(fontsize=8) ax.xaxis.set_major_formatter(FuncFormatter(lambda x, _: f'£{x:,.0f}')) ax = axes[1] x = np.arange(N + 1) for km in KELLY_MULTS: pths = mc_results[km]['paths'] p5 = np.percentile(pths, 5, axis=0) p50 = np.percentile(pths, 50, axis=0) p95 = np.percentile(pths, 95, axis=0) ax.fill_between(x, p5, p95, alpha=0.12, color=colors[km]) ax.plot(x, p50, color=colors[km], linewidth=2, label=f'{km:.2f}x Kelly median') ax.axhline(START, color='gray', linewidth=0.8, linestyle=':') ax.set_xlabel('Bet number') ax.set_ylabel('Bankroll (£)') ax.set_title('Median path +/- 5th-95th percentile') ax.legend(fontsize=8) ax.set_yscale('log') ax.yaxis.set_major_formatter(FuncFormatter(lambda x, _: f'£{x:,.0f}')) ax = axes[2] for km in KELLY_MULTS: dd = mc_results[km]['max_dds'] * 100 ax.hist(dd, bins=80, alpha=0.45, color=colors[km], label=f'{km:.2f}x Kelly median={np.median(dd):.0f}%', edgecolor='none') ax.axvline(np.median(dd), color=colors[km], linewidth=2, linestyle='--') ax.set_xlabel('Max drawdown (%)') ax.set_ylabel('Frequency') ax.set_title('Max drawdown distribution') ax.legend(fontsize=8) plt.tight_layout() plt.savefig(PLOT_OUT, dpi=150, bbox_inches='tight') print(f"\nChart saved to {PLOT_OUT}") print("Done.")