Same trades, different fates: position sizing decides who survives

Most trading discussion is about entries: the indicator, the setup, the timing. Far less is said about a number many traders choose almost without thinking, which is how much to put on each trade. Yet that one number can turn a good strategy into steady growth, into gut-wrenching swings, or into an empty account. And it does so with exactly the same trades.

The experiment: one strategy, many sizing rules

We used a hypothetical strategy with a genuine, modest edge. It wins 42% of its trades; its winners make 1.8R and its losers lose 1R, and every trade costs 0.05R in brokerage, taxes and slippage. ("R" is the amount risked on a trade: the distance from entry to stop, times the position size.) That works out to an expectancy of +0.126R per trade after costs. In plain terms: over the long run, every ₹100 put at risk earns about ₹12.60.

We generated 10,000 sequences of 500 trades from it, roughly three years of active trading each. Every sizing rule below was applied to the same 10,000 sequences, trade for trade, so luck is identical across rows. Only the sizing changes. Each rule risks a percentage of the account as it stands at the time, so positions grow after gains and shrink after losses.

A few terms before the table. Results are shown as a multiple of the starting account (2× means it doubled). "Wiped out" means the account fell below 10% of where it started. A drawdown is a fall from the account's highest point to a later low: if ₹10 lakh grows to ₹15 lakh and then drops to ₹12 lakh, that is a 20% drawdown. The median is the middle result: half the traders did better, half did worse.

The strategy is profitable in every row, and its trades are identical in every row. Yet the outcomes range from steady growth with mild drawdowns to nearly nine traders in ten wiped out.

Lesson one: beyond a point, more size means less money

From 0.5% to 1% to 2% to 5% risk, the median result grows: 1.35×, 1.79×, 2.91×, 7.21×. Among the sizes we tested it is highest at 6.9% (8.41×), the growth-maximising size, and then it turns. At 10% the median falls back to 5.45×, and at 20% the typical account is destroyed, on a winning strategy.

The reason is that losses and gains do not compound symmetrically. A loss of 50% needs a gain of 100% just to get back to where you started. The bigger each bet, the deeper the holes that ordinary losing streaks dig, and the harder it becomes for the edge to climb out:

The size that maximises long-run growth is known as the Kelly fraction, after John L. Kelly Jr., a Bell Labs researcher who described it in a 1956 paper on information and betting. For this strategy it is about 6.9% per trade. Above it, risking more lowers long-run growth while deepening the drawdowns. At about twice the Kelly fraction (roughly 14% here) the typical trader's growth falls to zero, and beyond that the typical account shrinks towards nothing, even though every trade still has a positive expectancy.

Lesson two: the growth-maximising size is one nobody can live with

If 6.9% risk produced a median 8.4× in three years, why would anyone risk less? Look across that row. 99.8% of traders at that size saw their account halve from a peak at some point, and the typical worst drawdown was 81%. Even 5% came with a typical 67% drawdown.

Picture what that means. Your ₹10 lakh account grows to ₹25 lakh, then falls to ₹5 lakh, with no mistake on your part: just the strategy's ordinary losing streaks at that size. Few traders have the discipline to keep trading a system through that. Most cut size near the bottom, or switch strategy, or stop altogether, which locks in the loss and forfeits the strategy's future edge. A size that is optimal on paper is worthless if you cannot hold on to it.

Lesson three: sizing by mood costs you either way

The last two rows are two familiar ways of letting emotions set the size.

The "hot hand" trader starts at 1% and adds 1% for every consecutive win, up to 5%, because a winning streak feels like being in sync with the market. In this simulation, by construction, a streak tells you nothing about the next trade. (If you believe your own trades do run in streaks, test it rather than assume it.) The fair comparison is a fixed size equal to the hot hand's average risk, which was 1.70%. Same sequences, same average risk:

The fixed size did better on every measure. Sizing up on confidence bought a little extra risk and gave up a little return. The cost is small here, but it is all downside.

"Win it back" doubles the stake after every loss, so that the next win recovers everything. It is a centuries-old gambling system, the martingale, and it feels rational because it usually works: most losing streaks are short. But over hundreds of trades, long ones are almost certain to come. In our simulation, 94% of sequences contained eight or more losses in a row and 60% contained ten or more, and after five straight losses this trader is already risking 32% of the account on the next trade. Half of these traders were wiped out, and 55% ended below where they started, trading a strategy with a positive edge.

How sizing rules are usually built

Good position sizing is not complicated. It is a calculation done the same way every time, before the trade, from three numbers: the account size, the percentage you are willing to risk, and the distance to your stop.

1. Risk per trade is fixed in advance. In this simulation, at 0.5% to 2% per trade the typical (median) worst drawdown was between about a tenth and a third of the account, though at 2% some 7% of traders still saw their account halve from a peak. The right range depends on a strategy's losing streaks and how many positions are open at once, and a backtest shows both.
2. The size comes from the stop, not the other way round. First decide where the trade is wrong, then calculate how much to buy so that being wrong costs the chosen risk.
3. Size from current equity, or know the trade-off. Sizing from the account as it stands shrinks positions after losses and grows them after gains. As a check we also ran a fixed-rupee rule, always risking 2% of the starting account. It earned less (2.26× against 2.91× at a recalculated 2%) and, because its risk becomes a smaller share of a growing account, it had shallower drawdowns (median 23.7% against 33.3%). But its risk becomes a bigger share of a shrinking one, and 0.2% of those traders were wiped out, against none at a recalculated 2%. Fixed rupees are gentler while you are winning and harsher while you are losing.
4. Lot sizes are respected. In futures and options, if even one lot risks more than the budget at a sensible stop, the trade is too big for the account, and the disciplined choice is to skip it rather than widen the risk.
5. The day is capped, not just the trade. A daily loss limit, set as a number of R, stops a bad day from turning into a bad month.
6. The size is checked against the backtest's drawdowns, with a margin. Live results are often worse than backtests, because of costs, slippage and overfitting, so a common check is to take the worst drawdown the strategy produced in testing and ask honestly whether you could sit through about twice that.

Automation can apply a sizing rule identically on every trade. A system calculates the position from the stop and current equity on every trade, never sizes up because it feels lucky, never doubles down because it feels behind, and enforces the daily limit even on the days you would rather it didn't. It will apply a bad sizing rule just as faithfully as a good one, so the rule has to be tested first. For the losing streaks that test every sizing rule, see our losing-streak study.

Methodology

Sequences

10,000, shared across all rules

Trades per sequence

500 (about 3 years, assumed)

Strategy

42% wins at +1.8R, 58% losses at −1R

Costs

0.05R per trade (net +1.75R / −1.05R)

Expectancy

+0.126R per trade

Sizing

fraction of current equity, unless stated

Hot hand

1% + 1% per consecutive win, max 5%

Win it back

1% × 2 per consecutive loss, max 32%

"Wiped out"

equity below 10% of start at any point

RNG

mulberry32, seed 20261017

This tests one hypothetical strategy; other win rates and payoffs have different growth-maximising sizes. The script also runs two comparison rules quoted in the text: a fixed 1.70% (the hot hand's average risk per trade) and a fixed-rupee rule risking 2% of the starting account. "Fell 50% from a peak" means a drawdown of 50% or more at any point. The Kelly fraction is computed net of costs, numerically, by maximising expected log growth. Trade outcomes are independent, so a streak carries no information about the next trade. Losses stop exactly at the stop (no gaps), costs are a fixed share of R (real Indian charges such as STT and exchange fees scale with trade value, while flat per-order brokerage does not, so tight stops cost more in R), and there are no margin, leverage or lot-size limits; in real markets, gaps, slippage and margin make large positions worse than shown, and several open positions at once multiply the risk. Every number comes from sim-sizing.js (plain Node.js: node sim-sizing.js).