Position sizing: how big should each trade be?

The answer

Sizing 1 position is arithmetic and it is covered elsewhere. Sizing a portfolio of positions held at the same time is a second, separate calculation, and it is the one almost nobody does. Eight positions risking 1% each are not 8 risks if they move together. They are closer to 1 position risking 8%.

Why this costs you money

Assume you have done everything the single-position method asks. Each position risks 1%. Each stop sits where the reason for owning it stops being true. Each position value is capped. Nothing is wrong at the level of any individual trade.

You hold 8 of them. All long. Six are in financial companies or the index, because that is where your setup appeared this month.

Then a policy announcement lands, or an election result, or a global risk event. Everything long falls together on the same morning and your 8 stops trigger within 2 hours. You planned to risk 1%. You lost between 7% and 9% in a session, and the gaps made part of it worse.

The error was not in any trade. It was that nobody added them up.

This is the most common way a person who has read about risk management still loses a large amount quickly. The rule was applied where it is visible, on the order screen, and not where it is invisible, across the account.

How it works

Four ideas, and each one answers a question the single-position formula cannot.

1. Portfolio heat: the number nobody calculates

Heat is the sum of the money at risk across every open position, expressed as a percentage of the account. For each position: shares × (current price − stop price). Add them all. Divide by the account. That is your heat.

Set a maximum. For most retail accounts, 4% to 6% total heat is a workable ceiling. When heat is at the ceiling, you take no new position, however good the setup looks. If you want the new position, you must close or reduce an existing one first.

That last sentence is the whole mechanism. It forces every new trade to compete with the trades you already have, which is a comparison nobody makes voluntarily.

2. Correlation groups: why 8 positions can be 1 position

Two positions are correlated when they tend to move together. On the day that matters, correlated positions behave as a single position.

The practical method does not need a correlation coefficient. Assign each open position to a group, and cap the risk per group.

Sensible groups for a retail account:

• Direction. All long positions form one group. All short positions form another.
• Sector. Banking and financials. Technology services. Metals and mining. Consumer.
• Size band. Large capitalisation. Small and micro capitalisation. Small companies fall together in a risk event regardless of sector.
• Currency or country, if you hold positions in more than one market.
• Factor. Everything bought because it was rising. Everything bought because it was cheap. These 2 groups behave very differently in the same week.

Then the rule: no more than 2% of the account at risk in any one group. With a 6% total heat ceiling, that permits meaningful exposure to 3 groups and prevents 6 versions of the same bet.

A position often belongs to several groups. A small Indian private bank is long, financial, and mid-capitalisation. Count it in each group. The count is deliberately conservative, because on the day that matters it will behave like all 3.

3. Sizing when the stop distance varies

Different instruments need different stop distances, because they move different amounts per day. A stock that moves 1% a day and a stock that moves 4% a day cannot use the same stop distance without one of them being meaningless.

The instrument that makes this consistent is ATR, the Average True Range, which is the average daily movement of an instrument over a recent period, usually 14 sessions. Most charting software shows it.

Set every stop as the same multiple of ATR. For example, 2.5 × ATR below entry for every position. Then:

Shares = (account × 1%) ÷ (2.5 × ATR)

What this produces is worth stating plainly. The volatile instrument gets a wide stop and therefore a small position. The quiet instrument gets a tight stop and therefore a large position. The money at risk is identical in both. The position values are very different, and they should be.

This is the answer to the question "should I size down on a volatile stock?" You do not decide. The formula does it, because the volatility is already inside the stop distance.

One caution. A quiet instrument produces a large position value, which reopens the gap problem from cluster 16. The position value cap still applies on top.

4. Scaling into and out of a position

Scaling in means adding to a position after the first part is placed. It is useful and it is also the most common route to an accidentally large position.

Three rules make it safe.

• Never add to a losing position. Adding to a loser increases size exactly when the evidence says the idea is wrong. There is no version of this that works over many trades.
• Add only after the first part has moved in your favour by at least 1R, where 1R is the amount you risked on it. By then the stop on the first part can usually be moved to the entry price.
• Size every addition so that total open risk on the position stays within your per-position limit. If the first part now risks nothing, because its stop is at entry, the addition may risk the full 1%. If the first part still risks 0.6%, the addition may risk 0.4%.

The order matters. Move the stop first, then calculate what the addition may risk. Doing it in the other order is how a 1% position becomes a 3% position while feeling careful.

Scaling out means selling part of a winning position before the exit signal. It reduces the variation in your results and reduces your average result at the same time. You are selling part of the position that was working, which removes the largest wins from the distribution. In a method whose returns come from a small number of very large winners, scaling out can remove most of the return.

It is a legitimate choice with a stated cost. Use it if the reduced variation is what keeps you executing the plan. Whichever you choose, write it into the plan and apply it to every trade. Applying it only to the positions that make you nervous introduces a bias you cannot measure afterwards.

What it tells you, and what it does not

Portfolio heat tells you what a bad day costs. It does not tell you the worst case, because heat assumes every stop executes at its price. On the day when everything correlates, gaps are common. Treat heat as a floor for the loss, not a ceiling.

Correlation groups tell you where your concentration is. They do not tell you which correlations will appear. Relationships that were absent for 3 years arrive together in a crisis, when instruments that looked unrelated turn out to share a seller.

The honest planning assumption is that in a genuine risk event, every long position in your account is 1 position. Size so that this is survivable.

The decision rule

Before adding any new position, calculate 3 numbers and take the smallest answer.

1. The trade limit. Risk 1% of the account. `Shares = (account × 1%) ÷

(entry − stop)`.

1. The group limit. Would this take any correlation group above 2% of the

account? If yes, reduce the size or take a different trade.

1. The heat limit. Would this take total open risk above 6%? If yes, take

nothing until an existing position is closed or its stop is moved to entry.

And 1 rule for adding to a position: the stop moves first, then the addition is sized against what remains.

Try this now

Five minutes and a calculator. Most readers have never seen this number.

1. Open your positions screen and list every open position.
2. For each one, write 3 things: the quantity, the current price, and the stop price. If a position has no stop, write the price at which you would admit the idea had failed. Everybody has that number.
3. For each position, calculate quantity × (current price − stop price). That is the money at risk on that position.
4. Add all of them. Divide by your total account value, including cash. Multiply by 100. This is your portfolio heat.
5. Now group them. Mark each position with its sector, its size band, and long or short. Add the risk within each group separately.

What you should see. Two things, and the second is the one that matters.

Most readers find total heat between 6% and 20%. If yours is above 6%, a single correlated day costs more than you planned, and it does so without any of your individual rules being broken.

Then look at the groups. The common result is that 60% or more of the total risk sits in 1 group, usually the largest sector in the local index, and usually all in the same direction. That is not 8 positions. It is 1 position with 8 tickets, and you have been calculating the risk 8 separate times as though that made it smaller.

The fix takes 2 minutes. Reduce or close positions until no group exceeds 2% and the total is under 6%. You will find you have to remove 3 or 4 positions, and that they are mostly the ones you added because a setup appeared and there was cash available.

Three real cases

1. Melvin Capital Management, January 2021 (United States) — many positions, one exposure The fund held short positions in a number of heavily shorted American companies. When those shares rose sharply in January 2021, the positions moved against the fund together, because the thing they had in common was the reason for the position. The fund reported a very large loss for the month, received an outside capital injection, and announced its closure in May 2022. The individual shares were unrelated businesses. The positions were not unrelated, because all of them had been chosen by the same screen.

2. The May 2020 WTI crude oil contract, 20 April 2020 — the size was fixed and the price was not The expiring May 2020 West Texas Intermediate futures contract settled at -$37.63 on 20 April 2020, the first negative settlement in that contract. Holders of long positions owed money rather than owning something worth nothing. Interactive Brokers disclosed a loss of approximately $104 million arising from customers who held those positions. In China, retail investors in Bank of China's "Crude Oil Treasure" product, which tracked the same contract, faced losses exceeding their deposits. A position sized on the assumption that the worst case is a fall to zero is sized wrongly whenever zero is not the floor.

3. Anugrah Stock and Broking, 2020 (India) — correlated option positions in many client accounts The Mumbai broker ran options strategies across a large number of client accounts. Losses in 2020 exceeded the capital available, NSE declared the firm a defaulter, and SEBI and the exchanges took action. The relevant point for an individual is structural. When many accounts hold versions of the same position, a single market move produces a single loss event, however many account numbers it is spread across. The same is true of 8 positions inside 1 account.

The question that resolves it

A novice asks: how much should I put into this trade?

An expert asks: what does this trade add to what I already hold, and what is the total at risk if everything I own goes to its stop this week?

The first question can be answered without opening the positions screen. The second cannot, which is why it is almost never asked.

What would make this wrong

If correlations between positions in the same market were low and stable, group limits would be unnecessary and total heat would be close to the sum of independent risks. That claim is testable, and it fails in the direction that matters: correlations rise in falling markets, exactly when the assumption is being relied upon.

The honest limits.

Group limits reduce your return in ordinary conditions. If your method only produces signals in 1 sector, a 2% group cap will keep you out of trades that would have worked. That is a real cost, paid every ordinary month, to avoid a loss that arrives every few years.

Heat limits can be too tight for small accounts. A 6% ceiling with 1% per position allows 6 open positions. If your method needs 15, the per-position risk must fall to about 0.4%, which at a small account size produces positions too small to cover costs. The honest conclusion is that the account cannot run that method yet.

ATR sizing assumes recent volatility predicts near-term volatility. It usually does, which is why the method works. It fails at the moment volatility changes regime, which is the moment you needed it.

Scaling out is not always wrong. The claim here applies to methods where a few very large winners produce the return. In a method with a high win rate and small targets, scaling out changes very little.

Carry this

• Heat is the sum of money at risk across all open positions. Cap it at 6%.
• No more than 2% of the account at risk in any one correlation group.
• Set every stop at the same multiple of ATR. Equal risk, unequal position sizes, decided by the instrument rather than by you.
• Never add to a loser. Move the stop first, then size the addition.
• Scaling out lowers your results and lowers the variation in them. Choose knowingly and apply it to every trade.

Knowledge check

Related

• Risk management: surviving long enough to win
• How to scale: growing size without blowing up
• Risk management and position sizing (cluster 16)
• ← The trader's playbook