Paper trade every number and rule on this page before you risk live capital on it. Nothing here is investment advice, a tip or a return promise. EliteAlgo is a proprietary trading desk; this is analytical writing about how sizing maths behaves.

Search for the Kelly criterion and you get coin flips, casino betting and a few blog posts about stocks. Almost none of it deals with what a NIFTY option seller actually faces: a high win rate, a payoff that is lopsided against you, and a loss distribution with a fat tail. That is exactly where Kelly looks most attractive and is most dangerous.

Here is the short version. The formula gives you the bet size that maximises long-run growth if your inputs are exact. Yours won't be. A small error in win rate, or one bad tail day, turns "optimal" into "ruin". So option sellers use a fraction of Kelly, or a fixed-loss rule, and then let lot size and margin cap the answer anyway. The rest of this page shows why, with a simulation we ran for this article.

Advertisement EliteAlgo advertisement

The Kelly formula in plain English

John Kelly Jr. published the idea at Bell Labs in 1956, in a paper on information rate. Gamblers and fund managers picked it up later. The Wikipedia entry on the Kelly criterion gives a fair history and the general form.

For a simple bet it reads:

f* = p − q / b

  • f* is the fraction of capital to put at risk on one trade.
  • p is the probability of a win, and q is 1 − p.
  • b is how much you win for every 1 you risk.

Take a coin that pays 1 for 1 and lands your way 60% of the time. Then f* = 0.60 − 0.40 / 1 = 0.20. Risk 20% of your bankroll each flip. Bet less and you grow slower. Bet more and you also grow slower, and past a point (roughly double Kelly) the long-run growth turns negative.

That last sentence matters most. The growth curve is not a plateau you can safely overshoot. It is a hill, and the far side drops off faster than the near side. Over-betting costs more than under-betting by the same amount.

What the formula assumes, and what short options break

Kelly was built for bets with a few clean properties. Short premium breaks most of them.

1. Known, stable odds

A coin has a win rate you can state in advance. An option strategy has a win rate you estimate from a sample of trades, and the sample comes from one market regime. A strategy that won 85% of its trades in a calm year can win far fewer when volatility changes. The formula has no box for "I'm not sure about p".

2. A fixed payoff

Kelly's b is one number. A short strangle has many outcomes: a full premium win, a partial win, a small stop-loss, a big stop-loss, a gap through your stop. Averaging them into one b hides the part that kills you.

3. Independent bets

Coin flips don't cluster. Option losses do. A volatility spike hurts every short position you hold in the same week, and your NIFTY and SENSEX trades are not two separate bets. They are the same bet on the same market, so sizing each one at full Kelly over-bets in total.

4. Continuous sizing

Kelly says "risk 11.3% of capital". Exchanges sell you whole lots with a margin requirement. You cannot buy 0.37 of a lot. We come back to this in the lot size section, because it often decides the answer before the maths does.

5. Losses you can price

This is the big one. The regulator has published studies showing that most individual F&O traders lose money, and fat-tail losses are part of that picture. Find the research on the SEBI site at sebi.gov.in. A short option has capped gains and uncapped, fat-tailed losses. A win rate of 85% says nothing about the size of the other 15%.

A useful academic view comes from a Georgia Tech talk on Kelly's criterion for option investment. It shows how the criterion changes once the payoff is no longer two-valued. The direction of the change is the same as ours: the answer shrinks.

An illustrative example: the number Kelly gives, and why it is too big

This example is made up to show the mechanics. The numbers are not a record of any real strategy, and they are not a recommendation.

Suppose a rule-based short-premium strategy, measured over 60 trades, shows:

  • Win rate p = 85%, so q = 15%.
  • Average win = 0.25 units, where 1 unit is the average loss.

That is a payoff of 4 to 1 against you, which is typical for premium selling: small wins, bigger losses. Then b = 0.25 and:

f* = 0.85 − 0.15 / 0.25 = 0.85 − 0.60 = 0.25

Kelly says risk 25% of capital on the average loss of every trade. A trader who sees that number should already be nervous. And they should be, because we then tested what happens when the inputs are slightly off.

How we ran the test

Our methodology was a Monte Carlo simulation. We built a simple script in Python, ran 20,000 random paths of 250 trades each (about a year of daily trading), and recorded the median final equity, the median worst drawdown, and how often the path fell more than 50% from a peak. We tested the same edge at full Kelly (25%), half Kelly (12.5%), quarter Kelly (about 6%) and a flat 2% risk. We then repeated the run under three assumptions about reality.

Results, from our own data set of simulated paths (starting equity = 1.00):

ScenarioSizing (fraction at risk)Median end equityMedian worst drawdownPaths with a drawdown over 50%
A. Inputs exactly right, fixed lossFull Kelly, 25%9.6780%99%
A. Inputs exactly right, fixed lossHalf Kelly, 12.5%5.0249%46%
A. Inputs exactly right, fixed lossQuarter Kelly, 6.25%2.5027%2%
A. Inputs exactly right, fixed lossFlat 2%1.379%0%
B. True win rate 80%, not 85%Full Kelly, 25%0.1097%100%
B. True win rate 80%, not 85%Half Kelly, 12.5%0.5975%93%
B. True win rate 80%, not 85%Quarter Kelly, 6.25%0.8845%39%
B. True win rate 80%, not 85%Flat 2%0.9916%0%
C. Win rate 85%, but fat-tailed lossesFull Kelly, 25%1.0697%100%
C. Win rate 85%, but fat-tailed lossesHalf Kelly, 12.5%3.6564%82%
C. Win rate 85%, but fat-tailed lossesQuarter Kelly, 6.25%2.3336%15%
C. Win rate 85%, but fat-tailed lossesFlat 2%1.3512%0%

In scenario C, 85% of losing trades lose 0.6 units and the other 15% lose 2.6 units, scaled so the average loss is still exactly 1 unit. The average did not change. Only the shape did.

What the table says

Look at scenario A first. Full Kelly has the biggest median result (9.67), and nearly every path went through a drawdown of more than half the account. The median worst drawdown was 80%. Nobody sits through an 80% drawdown on rules alone. Real traders switch the system off, change settings in panic, or run out of margin. The growth Kelly promises is only collected by someone who never flinches.

Scenario B is the one that should worry you. We changed the true win rate from 85% to 80%. That is five points, and a 60-trade sample cannot tell those two apart. At 80% the edge is exactly zero (0.80 × 0.25 = 0.20, the same as the 0.20 expected loss). Full Kelly then ended at 0.10 in the median path: a 90% loss. A zero-edge strategy bet at 25% is a machine for destroying capital.

Scenario C keeps the win rate and the average loss but adds a tail. Full Kelly's median end equity falls from 9.67 to 1.06, and nearly every simulated path (over 99%) suffered a drawdown above 50% at some point. Half Kelly was still ugly, with 82% of paths past a 50% drawdown.

Flat 2% risk produced a median drawdown of 9% to 16% in every scenario and no path in our run fell more than 50%, and it gave up most of the headline growth. That is the trade you are making: a little less on paper, a lot less chance of being shut out.

We should say what this does not show. It is a toy. Real trades have slippage, costs, overlapping legs and regime shifts that no 30-line script captures. Treat it as a picture of how fast the cliff arrives, not as a forecast.

How wrong can your win rate be? Less than a coin flip can tell you

The simulation changed p from 85% to 80% and the strategy fell apart. Is a five-point miss a far-fetched case? It is not. Take the 60-trade sample from our example. The standard error of a win rate is the square root of p × q / n, which here is the square root of 0.85 × 0.15 / 60, or about 4.6 points. A rough 95% range is therefore about 76% to 94%. Your "85%" strategy is compatible with a true win rate of 77%, where the edge is negative, and Kelly would then tell you to bet nothing.

To narrow that range to plus or minus two points you would need roughly 300 trades from the same regime. For a weekly-expiry seller trading one setup, that is years. Meanwhile the market changes. This is the practical reason to be modest about p, and why a fraction of Kelly is a floor on humility, not a fancy extra.

One habit helps: before sizing, compute the break-even win rate for your payoff. With wins of 0.25 and losses of 1, break-even is 1 / 1.25 = 80%. If your measured rate is 85% with a plausible range down to 76%, you are living very close to the line, and full Kelly sits right on top of it.

From a risk percentage to a number of lots

Here is the mechanical step that most Kelly articles skip. The numbers below are placeholders to show the arithmetic, not a suggested size.

  1. Pick the capital the strategy may use. Say ₹10,00,000.
  2. Pick the loss you accept per trade. Say 1%, so ₹10,000.
  3. Find the worst loss per lot if the stop is hit. That is the stop distance in points times the units per lot (look up the current figure on the exchange circular), plus an allowance for slippage. If the stop is 40 points and the allowance 5 points, you use 45 points.
  4. Divide: lots = ₹10,000 ÷ (45 × units per lot). Round down.

If the answer is below 1, the strategy is too big for the account at that risk level. The honest options are a tighter or defined-risk structure, a smaller risk percentage with a conscious choice, or more capital. Rounding up to "just one lot" quietly turns a 1% rule into something else, and that gap is where accounts get hurt.

Also check the margin side. The same trade needs SPAN plus exposure margin blocked for as long as it is open, and that figure can jump in a volatile week. A good plan leaves unused margin as a buffer, so a spike does not force you to close positions at the worst price.

Advertisement EliteAlgo advertisement

Fractional Kelly and fixed-fractional sizing

Fractional Kelly

Half Kelly means betting half of f*. Quarter Kelly means a quarter. It is the standard patch for the uncertainty problem. In our table, going from full to half Kelly cut the median worst drawdown in scenario A from 80% to 49%, while keeping about half of the median growth. You give up a lot less than half the return to cut the pain by a lot more.

A common reading of the maths is that betting fraction c of Kelly gets you close to c × (2 − c) of the maximum growth rate. At c = 0.5 that is 75% of the growth for 50% of the bet. The same logic is why traders who trust their estimates least go lowest.

But notice what scenario B did to quarter Kelly: a median end of 0.88 and a 45% median drawdown. A fraction of a wrong number is still a wrong number. Fractional Kelly shrinks the damage; it doesn't fix bad inputs.

Fixed-fractional and fixed-loss rules

Most systematic desks skip the formula and set a flat rule: risk a fixed percentage of capital on the worst realistic loss of each trade, and cap total open risk. For option sellers "worst realistic loss" is not the premium. It is the stop-loss distance, or for a defined-risk spread, the width minus credit.

The advantage is that it needs no estimate of p. It only needs a number you control: how much you let a position lose before it is closed. We cover the rule family in more depth in position sizing vs technique in option trading.

Lot size and margin: the hard limits Kelly ignores

Before the formula ever gets a vote, three practical limits decide how big you can trade.

Lot size is a step, not a dial

You trade in whole lots. On a small account, even one lot of a NIFTY strangle may already risk more than quarter Kelly allows. Then the Kelly answer is not "trade 0.4 lots". It is "you are already over-sized with one lot". Lot sizes are revised by the exchange from time to time, so check the current figure on the exchange circular or your broker's contract page rather than a blog post. We explain how to turn a lot size into a risk figure in our guide to NIFTY lot size and position sizing.

Margin caps the upside of sizing

For short options, your broker blocks SPAN plus exposure margin. The amount changes with volatility and with the exchange's rules. That means Kelly's "bet more" is not even available when the account is fully margined, and, worse, margin goes up exactly when a spike hits and your open trades are losing. A sizing plan has to survive that squeeze. Our note on SPAN and exposure margin for a NIFTY strangle walks through how the requirement is built.

Correlation across your own positions

If you run a NIFTY strangle and a SENSEX strangle on the same expiry week, a single gap can hit both. Sizing each at its own Kelly fraction double counts. Count them as one position, or size the pair so the combined worst case sits inside your cap.

Costs eat the edge Kelly depends on

Brokerage, STT and slippage all lower b and p in practice. Tax and charge rates change, so use the figures in force from the exchange or your broker. The point for sizing: a thin edge on paper can be zero after costs, and Kelly at zero edge is zero. In scenario B above, that is what a 5-point slip did.

Sizing rules a systematic desk uses instead

These are patterns, not instructions. Paper trade any of them first, and set your own numbers.

  1. Cap the loss per trade. Pick a percentage of capital you accept losing if a position hits its stop. Keep it small enough that a run of ten stops is survivable.
  2. Cap total open risk. Add up the worst-case loss across everything open. Put a ceiling on the sum, not on each trade.
  3. Size down after a regime change. If realised volatility moves up sharply, your old win rate no longer applies. Reduce size until the new regime has enough trades to measure.
  4. Use a drawdown brake. Define a drawdown level that halves size automatically. It stops you from deciding in the middle of a bad week.
  5. Tie size to the tail, not the average. Backtest the worst days for the strategy and make sure that day, at your size, leaves the account usable.
  6. Re-estimate on a schedule. Recompute win rate and payoff every month on a rolling window, and treat a big change as a signal to resize, not to argue with.

If you want rules like these run by software without second-guessing at 10:15 in the morning, that is the gap algorithmic execution fills. It brings its own risks, which we list honestly in the disadvantages of algo trading. And if you'd rather talk through a managed setup, see our managed algo trading service.

On method: the figures in this article come from our own simulation, not from a live account, and the formula and history come from the sources linked in the text. If you spot an error, tell us and we will correct it.

Frequently Asked Questions

What is the Kelly criterion?

It's a formula for the share of capital to risk on a bet that has a known edge. For a simple bet, f* = p − q / b, where p is the win probability and b is the payoff per unit risked. It maximises long-run growth only when p and b are known exactly.

Can I use the Kelly criterion for options trading?

You can compute it, but treat the output as a ceiling, not a target. Short options have fat-tailed losses and an uncertain win rate, which both push the safe size well below the formula's answer. Most sellers use a fraction of Kelly or a fixed-loss rule.

What is half Kelly?

Half Kelly means risking half of the fraction the formula gives. It keeps a large part of the growth while cutting drawdowns sharply. In our illustrative run, it cut the median worst drawdown from 80% to 49% in the clean-inputs case.

Why is full Kelly risky?

Because it assumes your inputs are exact and your losses are well-behaved. A five-point error in win rate took our simulated edge to zero and a full Kelly account to a 90% loss. Full Kelly also produces drawdowns most people cannot hold through.

How does lot size affect position sizing in NIFTY options?

You trade whole lots, so size moves in steps. If one lot already risks more than your cap, the answer is to trade a defined-risk structure or wait for a larger account, not to round up. Check the current lot size on the exchange circular or your broker page.

Is Kelly better than a fixed risk per trade?

Not for option sellers, in our view. Fixed risk needs no win-rate estimate and is easy to audit. Kelly can be a useful sanity check on whether your edge justifies any size at all. If the formula returns zero or a negative number, don't trade the setup.

Reminder: nothing on this page is investment advice. Paper trade first, and use your own risk limits.