""" Is the Kalshi market wising up? Measure the Kalshi-vs-Pinnacle price GAP over calendar time, on ALL fixtures (not just settled bets) -- a far higher-n signal than win/loss P&L. For each fixture take the LAST pre-KO Kalshi snapshot (its "closing" line, the fairest comparison to Pinnacle's close). For the bettable home/away sides compute pin_fair = multiplicative de-vig of (pin_home,pin_draw,pin_away) [live default] kalshi_mid gap = kalshi_mid - pin_fair (signed; <0 means Kalshi cheap = +CLV) |gap| = mispricing magnitude CLV = (pin_fair - kalshi_mid)/pin_fair CONVERGENCE (market wising up) => |gap| shrinks and mean CLV -> 0 over time. CONFOUND: league mix shifts over the season (Apr = European majors, Jun = thin summer leagues w/ structurally bigger gaps), so a raw time trend can be pure mix. We therefore ALSO report the trend WITHIN each league that spans several weeks, which removes the mix. Edge-relevant view = longshot sides pin_fair in [0.05,0.40]. Read-only. """ import csv, glob, os, math from datetime import datetime, date from collections import defaultdict DIR = r"E:\soccer\live_data" LO, HI = 0.05, 0.40 # longshot pin_fair range where the strategy bets _FMTS = ['%Y-%m-%d %H:%M:%S', '%m/%d/%Y %H:%M', '%m/%d/%Y %H:%M:%S', '%Y-%m-%d %H:%M'] def _pts(s): for f in _FMTS: try: return datetime.strptime(s, f) except (ValueError, TypeError): pass return None def _f(x): try: return float(x) except (TypeError, ValueError): return None def fair_mult(o): r = [1.0/x for x in o]; s = sum(r); return [ri/s for ri in r] def isoweek(gd): try: d = datetime.strptime(gd, '%Y-%m-%d').date() except ValueError: return None y, w, _ = d.isocalendar(); return f"{y}-W{w:02d}" def load_sides(): """Return list of dicts: one per (fixture, side) at the closing pre-KO snapshot.""" fixtures = defaultdict(list) for path in glob.glob(os.path.join(DIR, "*.csv")): with open(path, newline='', encoding='utf-8') as fh: for r in csv.DictReader(fh): fixtures[(r.get('game_date'), r.get('league'), r.get('home'), r.get('away'))].append(r) out = [] for (gd, league, home, away), rows in fixtures.items(): if not gd: continue pre = [(_pts(r['timestamp']), r) for r in rows if (r.get('status') or '').upper() == 'PRE' and _pts(r['timestamp'])] if not pre: continue _, r = max(pre, key=lambda t: t[0]) # closing pre-KO snapshot ph, pd_, pa = _f(r.get('pin_home')), _f(r.get('pin_draw')), _f(r.get('pin_away')) if None in (ph, pd_, pa) or not (ph > 1 and pd_ > 1 and pa > 1): continue fair = fair_mult((ph, pd_, pa)) wk = isoweek(gd) for side, sidx in (('home', 0), ('away', 2)): mid = _f(r.get(f'kalshi_{side}_mid')) if mid is None or not (0 < mid < 1): continue pf = fair[sidx] out.append(dict(wk=wk, gd=gd, league=league, side=side, pf=pf, mid=mid, gap=mid - pf, agap=abs(mid - pf), clv=(pf - mid)/pf if pf else None, longshot=(LO <= pf < HI))) return out def summ(rows): n = len(rows) if not n: return None agap = sum(r['agap'] for r in rows)/n gap = sum(r['gap'] for r in rows)/n clvs = [r['clv'] for r in rows if r['clv'] is not None] clv = sum(clvs)/len(clvs) if clvs else float('nan') fpos = sum(1 for r in rows if r['clv'] is not None and r['clv'] >= 0.05)/n return dict(n=n, agap=agap*100, gap=gap*100, clv=clv*100, fpos=fpos*100) def corr(xs, ys): n = len(xs) if n < 3: return float('nan') mx, my = sum(xs)/n, sum(ys)/n cov = sum((x-mx)*(y-my) for x, y in zip(xs, ys)) sx = math.sqrt(sum((x-mx)**2 for x in xs)); sy = math.sqrt(sum((y-my)**2 for y in ys)) return cov/(sx*sy) if sx > 0 and sy > 0 else float('nan') def main(): sides = load_sides() ls = [r for r in sides if r['longshot']] print(f"closing-snapshot sides total={len(sides)}, longshot[{LO},{HI})={len(ls)}\n") # 1) raw weekly trend on longshot sides (CONFOUNDED by league mix) byw = defaultdict(list) for r in ls: byw[r['wk']].append(r) print("Weekly longshot Kalshi-vs-Pinnacle gap (RAW -- confounded by league mix):") print(f" {'week':9s} {'nSides':>6} {'|gap|pp':>8} {'gap_pp':>7} {'meanCLV%':>9} " f"{'%CLV>=5':>8} {'nLg':>4} topLeagues") weeks = sorted(byw) idx = [] for i, wk in enumerate(weeks): rows = byw[wk]; s = summ(rows) lgc = defaultdict(int) for r in rows: lgc[r['league']] += 1 top = ",".join(f"{l}:{c}" for l, c in sorted(lgc.items(), key=lambda kv: -kv[1])[:4]) print(f" {wk:9s} {s['n']:6d} {s['agap']:8.2f} {s['gap']:+7.2f} {s['clv']:+9.2f} " f"{s['fpos']:7.1f}% {len(lgc):4d} {top}") idx.append((i, s)) print(f"\n corr(week_index, |gap|) = {corr([i for i,_ in idx], [s['agap'] for _,s in idx]):+.2f}") print(f" corr(week_index, meanCLV) = {corr([i for i,_ in idx], [s['clv'] for _,s in idx]):+.2f}") print(" (positive |gap| corr => WIDENING; negative => converging. Mix-confounded.)") # 2) WITHIN-LEAGUE trend (removes mix): slope of weekly |gap| vs week, for # leagues present in >=3 distinct weeks with >=15 longshot sides total. print("\nWithin-league |gap| trend (mix-controlled): leagues spanning >=3 weeks:") print(f" {'lg':6s} {'nSides':>6} {'weeks':>5} {'|gap|first':>10} {'|gap|last':>10} " f"{'slope_pp/wk':>12} {'corr':>6}") bylg = defaultdict(list) for r in ls: bylg[r['league']].append(r) wk_to_i = {wk: i for i, wk in enumerate(weeks)} trends = [] for lg, rows in sorted(bylg.items()): wkrows = defaultdict(list) for r in rows: wkrows[r['wk']].append(r) if len(wkrows) < 3 or len(rows) < 15: continue pts = sorted((wk_to_i[wk], summ(rs)['agap']) for wk, rs in wkrows.items()) xs = [p[0] for p in pts]; ys = [p[1] for p in pts] # OLS slope n = len(xs); mx = sum(xs)/n; my = sum(ys)/n den = sum((x-mx)**2 for x in xs) slope = sum((x-mx)*(y-my) for x, y in zip(xs, ys))/den if den else float('nan') c = corr(xs, ys) trends.append(slope) print(f" {lg:6s} {len(rows):6d} {len(wkrows):5d} {ys[0]:10.2f} {ys[-1]:10.2f} " f"{slope:+12.2f} {c:+6.2f}") if trends: pos = sum(1 for s in trends if s > 0) print(f"\n within-league slopes: {len(trends)} leagues, {pos} widening / " f"{len(trends)-pos} converging, median slope = " f"{sorted(trends)[len(trends)//2]:+.2f} pp/week") print(" (consistent NEGATIVE within-league slopes = market wising up;" " mixed/near-zero = no convergence signal)") if __name__ == '__main__': main()