Metrics

What is Kelly Criterion?

Reviewed by Karthikeyan Anbarasan (Founder, TradeLyser) · Updated 19 Oct 2026 · Editorial policy

DefinitionThe Kelly criterion is a formula for the fraction of capital to risk per trade that would maximise long-run growth if your win rate and payoff were known exactly. With a 45% win rate and average wins 1.5 times average losses, full Kelly is 8.3%, about ₹41,700 at risk per trade on ₹5,00,000, an aggressive size given that both inputs are estimates.
Formula
f* = W − (1 − W) ÷ R, where W = win rate and R = average win ÷ average loss
Example
0.45 − 0.55 ÷ 1.5 = 0.45 − 0.367 = 0.083, or 8.3% of capital at risk per trade
0%40%80%0.54.0Reward : risk (average win ÷ average loss)1:1 → 50%2:1 → 33%3:1 → 25%ProfitableLosing money
Figure 1. Kelly is zero exactly on the breakeven line: 50% at 1:1, 33% at 2:1 and 25% at 3:1 reward to risk. Above the line Kelly is positive and grows with the distance from it; below the line it is negative.

What the formula assumes

Kelly treats each trade like a bet with fixed odds: you either lose the amount risked or win R times it. Real trades have varied outcomes, slippage and gaps, and the win rate and payoff you feed in are estimates from past trades. The answer is only as good as those estimates.

The result is a fraction of capital to lose if the stop is hit, not a fraction to put into the position. With ₹5,00,000 and f* = 8.3%, the formula points to about ₹41,700 of risk per trade, which then has to be turned into shares or lots using the stop distance.

Small input errors, large swings in the answer

Measured win rate 45%
8.3%
If the true rate is 40%
0%
If the true rate is 50%
16.7%
Half Kelly at 45%
4.2%

All four use average wins 1.5 times average losses. A five-point error in win rate, well inside the range you would expect after 100 trades, moves full Kelly from 8.3% to zero or to 16.7%. Full Kelly on an edge that is smaller than it looks is over-betting.

Why many use a fraction

For this reason, traders who use Kelly at all often use a fraction of it, such as half or a quarter. In the standard model for this example, half Kelly keeps about 75% of full Kelly’s long-run growth rate with much smaller swings, while twice Kelly brings growth close to zero.

Above full Kelly
Lower long-run growth than full Kelly, and deeper drawdowns.
Full Kelly
Highest growth only if the inputs are exact. Very large swings.
Half Kelly
About 75% of full Kelly’s growth rate in the example, with smaller drawdowns.

Worked example: Bank Nifty futures on ₹10,00,000

Illustrative. A Bank Nifty futures setup has 200 trades in the journal, net of charges, with a 50% win rate and average wins 1.4 times average losses. Full Kelly is 0.50 − 0.50 ÷ 1.4 = 0.50 − 0.357 = 0.143, or 14.3%. Each lot of 30 with a 200-point stop risks ₹6,000.

Fraction of ₹10,00,000Risk per tradeLots at ₹6,000 each, rounded downActual risk
Full Kelly, 14.3%₹1,42,85723₹1,38,000
Half Kelly, 7.1%₹71,42911₹66,000
Quarter Kelly, 3.6%₹35,7145₹30,000
Fixed 1% for comparison₹10,0001₹6,000

Even a quarter of Kelly here risks 3% of the account on one trade, above the 1% to 2% rule of thumb. That is a common reason Kelly is read as an upper limit on size rather than a target. With a 50% win rate, a run of eight losses is quite possible over a few hundred trades, and at full Kelly it would cut the account by about 70% (0.857 to the power 8 is 0.29).

Using Kelly with a trading journal

Kelly criterion

Works out the fraction to risk from your measured win rate and payoff. The answer changes whenever those inputs change.
Input: your measured edge

Fixed fractional

Risks a fraction you choose in advance, such as 1%, whatever the measured edge. Simpler and steadier.
Input: your chosen fraction

Position sizing

Turns either answer into shares or lots using the stop distance and the lot size.
Input: rupee risk and stop
  1. Compute win rate and average win ÷ average loss per setup, from net P&L, not for the whole account.
  2. Wait for at least 100 trades on the setup, and recompute every 50 trades or each quarter rather than after every trade.
  3. Run the formula on a win rate a few points below the measured one as well. If that version is near zero, the edge is too thin to size up on.
  4. Log the fraction you actually risked on each trade next to the Kelly figure, so the review shows whether you sized above or below it.
Common mistake
Feeding in a backtest win rate or the planned reward-to-risk instead of realised results. Kelly needs the average win and loss you actually got, after slippage and charges. At a 45% win rate, planned 2R targets give 17.5%, but if winners actually average 1.3R the answer falls to 2.7%.
For F&O traders in India
Any Kelly figure still has to fit whole lots. Illustratively, ₹10,000 of risk with a 100-point Nifty futures stop allows 1 lot, since each lot of 65 risks ₹6,500. Check the current NSE circular for lot sizes. This page explains the formula; it is not a recommendation of any position size.

Reference guide

ContextValueReading
Kelly zero or negativeNo measured edgeOn these numbers the formula says not to risk anything on the setup
Inputs from fewer than 100 tradesUnreliable resultA five-point error in win rate can halve the answer or turn it to zero
Risking more than full KellyOver-bettingLong-run growth falls and drawdowns deepen; at twice Kelly growth is close to zero

Questions

What is the Kelly criterion formula for trading?

f* = W − (1 − W) ÷ R, where W is your win rate and R is average win ÷ average loss. The result is the fraction of capital to risk per trade. With W = 0.45 and R = 1.5 it is 8.3%.

Why do traders use half Kelly?

Because win rate and payoff are estimates, and full Kelly on an overestimated edge is over-betting. In the standard model, half Kelly keeps about 75% of the growth rate with much smaller drawdowns.

What does a negative Kelly mean?

Your measured win rate is below breakeven for your payoff, so the strategy shows no edge on those numbers. The formula says not to risk anything on it.

Is Kelly the same as position sizing?

No. Kelly is one way to choose how much to risk per trade. Position sizing then turns that rupee risk into shares or lots using the stop distance.

How often should I recalculate Kelly?

When the inputs have changed meaningfully, for example every 50 trades or each quarter, and always from at least 100 trades on the same setup. Recalculating after every trade makes the size swing with short-term luck.

Does the Kelly criterion work for options trading?

The formula needs a known amount at risk and an average payoff, so it fits defined-risk trades such as bought options or spreads, where the most you can lose is known. For short options with open-ended losses the amount at risk is not fixed, and the simple formula does not apply cleanly.

Related terms