π Kelly Sizing in Practice β Day 51 Numbers
Wednesday lesson: Kelly Criterion sizing, applied to today's live board. WHY THIS MATTERS NOW Day 51 stats are in: β 172 wins, 83 losses, 16 pending β 67.5% win rate across 271 signals β ML alone: 69.6% (126W / 55L) β Crash Detection: 61.1% (44W / 28L) A real edge measured over 255 resolved markets is enough to size off. Sizing wrong is how good traders blow up. Kelly is the math that tells you how much to put on. THE FORMULA f* = (bp β q) / b β f* = fraction of bankroll β b = net odds (decimal payout β 1) β p = your estimated probability of winning β q = 1 β p APPLIED TO TODAY'S BOARD Take the WTI Crude Oil (LOW) $95 in May signal β YES priced at 39.5Β’. β NO costs 60.5Β’ and pays $1 if resolves NO β Net odds b = (1 / 0.605) β 1 = 0.653 β Crash Detection true NO probability estimated around 65%: p = 0.65, q = 0.35 β f* = (0.653 Γ 0.65 β 0.35) / 0.653 = 0.115 β 11.5% full Kelly Half-Kelly = 5.7%. Quarter-Kelly = 2.9%. WHY FRACTIONAL KELLY ON EVERY EDGEFINDER SIGNAL 1. The model's probability estimate is itself uncertain. Full Kelly assumes you know p exactly. You don't. 2. 2. Variance in a 67.5% win-rate strategy is real. Five-loss streaks happen. Half-Kelly keeps drawdowns survivable. 3. 3. Compounding works best when bankroll never gets cut in half. Fractional Kelly trades peak growth rate for resilience. RULES OF THUMB β Never exceed half-Kelly on a single signal β Cap total open exposure at 25-30% of bankroll across all pending positions β Reduce size on Crash signals (61.1%) vs ML signals (69.6%) β If a five-loss streak would put you under, you're sized too big The edge is real. Sizing keeps you in long enough to compound it. Learn more in The EdgeFinder Foundation course β Classroom tab above.