Kelly Criterion — Kitna Bet Karna Chahiye?

John Kelly Jr. ne 1956 mein Bell Labs mein yeh formula develop kiya tha. Formula hai: f = (bp minus q) divided by b. Yahan f hai fraction of capital to bet, b hai odds received (net), p hai probability of winning, aur q hai probability of losing (1 minus p).

Simple example: Ek coin flip game. Agar heads aaye toh 2:1 odds milte hain (bet 1000, win 2000). Probability of win = 50 percent. f = (2 multiplied by 0.5 minus 0.5) divided by 2 = (1 minus 0.5) divided by 2 = 0.5 divided by 2 = 0.25. Toh Kelly kehta hai apne total capital ka 25 percent bet karo.

Stock trading mein Kelly apply karna thoda zyada complex hai kyunki outcomes binary nahi hote. Win percentage aur average win/loss ratio estimate karne hote hain jo khud uncertain hote hain.

Full Kelly bet karna theoretically optimal hai long term wealth maximization ke liye. Lekin practically bahut problematic hai. Kyunki? Full Kelly maximum drawdowns produce karta hai jo emotionally unbearable hote hain.

Simulation karo. 100 trades. Each trade pe full Kelly. Aapko periods milenge jab portfolio 30 to 50 percent neeche chala jaata hai even though you are following optimal strategy. Bahut log aise drawdowns pe strategy abandon kar dete hain, exactly worst time pe.

Practitioners generally Half Kelly ya Quarter Kelly use karte hain. Half Kelly mein expected geometric return thoda kam hota hai full Kelly se, lekin drawdowns dramatically reduced hote hain. Quarter Kelly aur bhi conservative. Majority of professional traders aur quant funds 25 to 50 percent of Kelly use karte hain.

Kelly ki sabse badi practical problem hai ki yeh assume karta hai ki aapko apni edge accurately pata hai. Lekin real trading mein win rate aur average profit/loss estimate hi hote hain, exact nahi.

Agar aap apni edge overestimate karte ho toh Kelly aapko over-bet karwayega. Consistent overestimation of edge with full Kelly sizing is a recipe for ruin. Yeh ek aur reason hai kyun fractional Kelly preferred hai.

Ek practical approach: Pehle backtested results se apni estimated edge nikalo. Phir usse conservative estimate maano, maybe 50 percent discount karo. Phir us conservative estimate pe Half Kelly apply karo. Yeh approach significant cushion deta hai estimation errors ke against. Overall shayad aap 25 percent of 'true' Kelly pe operate karoge.