Open to read. This in-depth guide is part of the member library — a subscription unlocks all guides plus the AI trade journal.
Why this matters
Most traders know WHAT to trade but not HOW MUCH to risk. The Kelly Criterion is a mathematically optimal formula for position sizing that maximizes long-term growth while minimizing the risk of ruin. Used by hedge funds, professional poker players, and systematic traders worldwide, it answers the single most important question in trading: given your edge, how much should you bet?
Section 1: What Is the Kelly Criterion?
In 1956, John L. Kelly Jr., a physicist at Bell Labs, published a paper on information theory that solved a fundamental problem: if you have an edge in a repeated game, what fraction of your bankroll should you bet each time to maximize long-term growth? His answer became known as the Kelly Criterion, and it has been adopted by everyone from casino card counters to Renaissance Technologies.
The core insight is that betting too little leaves money on the table, but betting too much risks catastrophic drawdowns or ruin. The Kelly Criterion finds the exact sweet spot — the bet size that maximizes the geometric growth rate of your capital over many repetitions.
The Formula
f* = (bp - q) / b. Where f* = fraction to bet, b = odds (reward/risk ratio), p = win probability, q = loss probability (1 - p). This gives you the optimal fraction of capital to risk.
Maximizes Growth Rate
Kelly does not maximize expected value of any single bet. It maximizes the expected logarithm of wealth — which is equivalent to maximizing long-term compound growth. This distinction is critical.
Full Kelly Is Aggressive
Full Kelly often produces uncomfortably large bet sizes and drawdowns of 50-80%. Most practitioners use Half Kelly or Quarter Kelly to reduce volatility while retaining most of the growth benefit.
Requires Known Edge
Kelly works only when you have a genuine statistical edge (positive expectancy). If you do not know your win rate and average reward/risk ratio from historical data, Kelly cannot be applied meaningfully.
Dynamic Sizing
Kelly sizes are proportional to your current capital. When you win, you bet more. When you lose, you bet less. This naturally protects against ruin — you can never lose 100% because each loss reduces the next bet.
Edge / Odds = Kelly
A simpler way to think about it: Kelly fraction = Edge / Odds. Your edge is (probability x reward) - (probability x loss). Divide by the odds. This gives the same answer as the formal formula.
Section 2: Breaking Down the Formula
The Kelly Criterion formula is: f* = (bp - q) / b
Let us define each variable with a trading example. Suppose you have backtested a Nifty options strategy and found: you win 55% of trades (p = 0.55), you lose 45% (q = 0.45), and when you win, you make 1.5x what you risk (b = 1.5, i.e., your reward-to-risk ratio).
Worked Example: Nifty Options Strategy
Given Values
p (win rate) = 0.55
q (loss rate) = 0.45
b (reward/risk) = 1.5
Kelly Calculation
f* = (1.5 x 0.55 - 0.45) / 1.5
f* = (0.825 - 0.45) / 1.5
f* = 0.25 (25%)
Result: Kelly says you should risk 25% of your trading capital on each trade. If you have Rs 10,00,000, that is Rs 2,50,000 per trade. This is FULL Kelly — and for most traders, far too aggressive.
Understanding What "Risk" Means Here
The Kelly fraction (f*) represents the amount you are willing to lose on this trade, not your position size. If you are risking 25% of Rs 10L = Rs 2.5L, and your stop loss is 5% below entry, your position size would be Rs 2.5L / 0.05 = Rs 50L (with leverage). This is why Full Kelly often implies leverage and is extremely aggressive.
For Indian F&O traders, think of it this way: if you sell a Nifty put with a maximum loss of Rs 2,50,000 (your risk amount per Kelly), and your strategy has a 55% win rate with 1.5:1 reward, Kelly says this is the optimal risk to take per trade. But one bad streak of 5 losses would wipe out most of your account.
Section 3: Full Kelly vs Fractional Kelly
In theory, Full Kelly maximizes long-term growth. In practice, it produces gut-wrenching drawdowns. Simulations show that a Full Kelly bettor will experience drawdowns of 50% or more with near-certainty over a few hundred trades. Most humans cannot psychologically withstand watching their Rs 10L account drop to Rs 5L, even if the math says it will recover.
Growth Curves: Kelly Fractions vs Fixed Sizing
Full Kelly
Max growth, 50%+ drawdowns
Half Kelly
75% growth, 25% drawdowns
Quarter Kelly
50% growth, 12% drawdowns
Over-Kelly
Negative growth, eventual ruin
Half Kelly: The Practical Choice
Half Kelly means using exactly half of the Full Kelly fraction. In our example, Full Kelly was 25%, so Half Kelly = 12.5%. You risk Rs 1,25,000 per trade on a Rs 10L account. Half Kelly gives you approximately 75% of Full Kelly's long-term growth rate but with significantly smaller drawdowns. Most professional traders and fund managers use Half Kelly or lower.
The mathematical beauty of Half Kelly is that it retains the core benefit (proportional sizing based on edge) while dramatically reducing the emotional and financial stress of large drawdowns. A 25% drawdown is recoverable. A 60% drawdown (common with Full Kelly) is psychologically devastating and takes 150% gains just to recover.
Quarter Kelly: The Conservative Choice
Quarter Kelly (6.25% in our example) is for traders who prioritize capital preservation above growth. You get roughly 50% of Full Kelly's growth rate but with drawdowns rarely exceeding 10-15%. This is ideal for traders with smaller accounts who cannot afford deep drawdowns, or for strategies where the estimated win rate and reward ratio have uncertainty.
Over-Kelly: The Path to Ruin
Betting MORE than Full Kelly is mathematically guaranteed to produce negative long-term growth. If Full Kelly says 25% and you bet 30%, you will grow slower than someone betting 20%. Bet 50%, and you will eventually go to zero. This is not intuition — it is mathematical proof. Over-Kelly turns a winning strategy into a losing one purely through position sizing.
Section 4: Kelly Criterion for Indian F&O Trading
Let us apply Kelly to three realistic Indian F&O trading strategies. These numbers come from typical backtested results — your actual statistics may differ.
| Strategy | Win Rate | R:R | Full Kelly | Half Kelly | Rs Risk/Trade (10L) |
|---|---|---|---|---|---|
| Iron Condor (weekly) | 70% | 0.6:1 | 21.7% | 10.8% | Rs 1,08,000 |
| Breakout Trading (daily) | 40% | 2.5:1 | 16.0% | 8.0% | Rs 80,000 |
| Swing Momentum | 50% | 2:1 | 25.0% | 12.5% | Rs 1,25,000 |
Example: Iron Condor on Bank Nifty
You sell a weekly Bank Nifty Iron Condor. Your backtested stats: 70% win rate, average win = Rs 12,000, average loss = Rs 20,000. The reward-to-risk ratio b = 12,000/20,000 = 0.6. Kelly = (0.6 x 0.70 - 0.30) / 0.6 = (0.42 - 0.30) / 0.6 = 0.20 = 20%. Half Kelly = 10%. On a Rs 10L account, risk Rs 1,00,000 per Iron Condor (the maximum loss on the spread).
Since Iron Condor max loss is defined by the spread width (e.g., 500 point spread x lot size), you size your trade so that the maximum loss on the position equals Rs 1,00,000. If one lot has a max loss of Rs 25,000, you can do 4 lots. This is how Kelly translates into actual lot sizes in F&O.
Example: Breakout Trading on Nifty
You trade Nifty breakouts using options. Win rate: 40%, but winners are 2.5x the size of losers. Kelly = (2.5 x 0.40 - 0.60) / 2.5 = (1.0 - 0.6) / 2.5 = 0.16 = 16%. Half Kelly = 8%. On Rs 10L, risk Rs 80,000 per breakout trade.
Notice that this strategy has a lower win rate but higher reward ratio — and Kelly still gives a meaningful bet size because the edge exists. If your system had a 30% win rate with the same 2.5:1 reward, Kelly would give only 2% — telling you the edge is razor-thin and position sizes should be minimal.
Section 5: Why Fractional Kelly Is Essential
Reason 1: Edge Estimation Error
Kelly assumes you know your exact win rate and reward ratio. In reality, these are estimates from backtesting or limited live trading data. If your true win rate is 50% but you estimated 55%, you are over-betting relative to your actual edge. Fractional Kelly provides a buffer against estimation errors. Half Kelly with a 10% estimation error is far better than Full Kelly with perfect estimates.
Reason 2: Non-Independent Trades
Kelly assumes each trade is independent. In reality, market regime changes can cause correlated losses. If the market crashes, your last 10 trades might all lose — violating the independence assumption. Fractional Kelly reduces exposure to these correlated-loss scenarios.
Reason 3: Psychological Limits
Full Kelly drawdowns regularly exceed 50%. After a 50% drawdown, you need a 100% gain to recover. Even if the math says you will recover, your psychology might break first. Traders who blow up often have positive-expectancy strategies that they abandoned during a deep drawdown. Fractional Kelly keeps drawdowns manageable enough that you actually stick with the strategy.
Edward Thorp's Rule: Ed Thorp, the mathematician who invented card counting and ran one of the first quant hedge funds, never used more than Half Kelly. He called Full Kelly "aggressive to the point of recklessness." If the man who proved Kelly works in practice recommends Half Kelly, that should tell you something.
Section 6: Common Kelly Mistakes to Avoid
Using Full Kelly
Full Kelly is theoretically optimal but practically dangerous. Drawdowns of 50-80% are normal with Full Kelly. Use Half Kelly or Quarter Kelly in live trading. Always.
Applying Kelly Without Data
You need at least 50-100 trades of data to estimate your win rate and reward ratio reliably. Using Kelly with 10 trades of data is just guessing with a fancy formula. Garbage in, garbage out.
Ignoring Regime Changes
Your backtest from a bull market does not apply to a bear market. Kelly calculations should be updated as your strategy performance evolves. Recalculate quarterly at minimum.
Using Kelly Across Multiple Strategies
If you run 3 strategies simultaneously, the Kelly for each should account for correlation between them. Running 3 Full Kelly strategies is effectively 3x over-Kelly. Divide your capital allocation.
Confusing Position Size with Risk
Kelly gives you the fraction of capital to RISK (lose if stopped out), not your total position size. A 10% Kelly risk with a 2% stop loss means your position size is 5x your capital — that requires leverage and careful execution.
Not Adjusting for Taxes and Costs
In India, F&O profits face 30% tax (short-term) plus STT, brokerage, and GST. Kelly should be calculated on net returns (after all costs), not gross returns. Costs reduce your edge, which reduces Kelly sizing.
Section 7: Practice Exercises
Calculate Your Kelly Fraction
- 01. Pull your last 50+ trades from your trading journal or broker statement
- 02. Calculate your win rate (number of winners / total trades)
- 03. Calculate your average winning trade and average losing trade
- 04. Compute b = average win / average loss (your reward-to-risk ratio)
- 05. Plug into Kelly: f* = (bp - q) / b
- 06. Take Half Kelly: f_practical = f* / 2
- 07. Compare with your current position sizing — are you over-betting or under-betting?
- 08. Run a simulation: apply Kelly sizing retroactively to your last 50 trades and compare the equity curve with your actual results
"Sizing your bets correctly is more important than which bets you make. A mediocre strategy with perfect sizing beats a perfect strategy with mediocre sizing — every time."
Key Takeaway
The Kelly Criterion is the only mathematically proven method for optimal position sizing. But "optimal" in theory and "practical" in real life are different things. Use Half Kelly or Quarter Kelly. Calculate it from real trading data, not backtested perfection. Update it as your strategy evolves. And remember: Kelly tells you the maximum you should risk. You can always risk less, but you should never risk more. Position sizing is the boring skill that separates traders who survive from traders who blow up.
Your progress
0 read in Risk Management
Ready to apply this?
Apply risk management to your actual portfolio. ArthaLearn calculates your position sizes, R:R ratios, and max drawdown automatically.
Free forever for trade logging. AI features start at ₹599/month.