What is position correlation — and why are three 2% trades not three 2% risks?
This is the mistake that catches traders who have done everything right. They read the 1–2% rule, they use a position size calculator, every single order risks exactly 2% — and one red morning the account is down 6%, then 12%, because the “three trades” were one trade wearing three tickers. This page measures how bad that is, translates a correlated cluster into the single-trade risk it really carries, and shows what to do about it.
KEY TAKEAWAYS
- Three independent 2% positions lose all three stops 12.5% of the time. At a correlation of 0.8 they do it 34.7% of the time — the 6% losing round comes almost three times as often, with no change to the edge.
- Over 200 rounds, the chance of a 50% drawdown goes from 2.5% (independent) to 18.6% at 0.5 correlation, 35.6% at 0.8 and 60% at 1.0 — where three 2% trades are exactly one 6% trade.
- Translated into a single trade, three 2% positions at 0.8 correlation carry the drawdown risk of one trade risking about 4.8%. Three at 1.0 are one at 6%.
- Three positions at 0.8 correlation are worth about 1.15 independent bets, not three. Five are worth 1.19.
- The fix is a cluster budget: size the group as one trade. Three positions at 0.67% each brought ruin back to 0.02% at the same 0.8 correlation.
What is position correlation, in one paragraph?
Position correlation is the tendency of the trades you have open at the same time to win together and lose together. It runs from 0, where knowing that one stop was hit tells you nothing about the others, to 1, where the positions are effectively the same trade under different names. It matters because every rule of thumb about risk per trade — the 1–2% rule, risk of ruin tables, the Kelly criterion — silently assumes you are taking one bet at a time, and that the next bet does not know how the last one went.
Open three trades at once and you have made a second bet on top of the three you meant to make: a bet that they will not all go wrong together. In stocks across unrelated sectors that bet is often reasonable. In crypto, where the same news moves nearly every coin in the same direction on the same day, it is usually wrong, and the price of being wrong is that your carefully sized 2% is not the number your account experiences.
How often do correlated stops fire together?
Take a plain system: 50% of trades win, wins pay 1.2 times what losses cost, and every trade risks exactly 2% of the account. Now take three of those trades at the same time and ask one question: how many of the three stops get hit in the same round? We modelled the three outcomes with a shared factor — the same way a portfolio model treats “the market” — and turned the correlation dial from 0 to 1. Two million rounds per setting.
| Correlation between the three | 0 stops hit | 1 stop | 2 stops | All 3 stops (a −6% round) |
|---|---|---|---|---|
| 0.0 — independent | 12.5% | 37.4% | 37.5% | 12.5% |
| 0.3 | 19.7% | 30.2% | 30.3% | 19.8% |
| 0.5 | 25.0% | 25.0% | 25.1% | 25.0% |
| 0.7 | 31.0% | 19.0% | 19.0% | 31.0% |
| 0.8 | 34.6% | 15.3% | 15.4% | 34.7% |
| 0.9 | 39.2% | 10.8% | 10.8% | 39.2% |
| 1.0 — identical | 50.0% | 0% | 0% | 50.0% |
Read the last column. The win rate never moves: every row still wins exactly half of its trades. What moves is the shape of the results. Independent trades spread their losses out — most rounds lose one or two stops and the account absorbs it. Correlated trades stop doing that. At 0.8 the middle of the table empties and the outcomes pile up at the two ends: either everything works, or everything fails and the account takes the full −6% in one go. That −6% round arrived once in eight rounds when the trades were independent. At 0.8 it arrives once in three.
What does that do to your chance of ruin?
A lot, and the reason is compounding. Three small losses on three separate days are recoverable; one 6% loss followed by another two rounds later is how a 20% drawdown happens before the trader has noticed anything is wrong. Here is the same system run for 200 rounds — three trades per round, 600 trades in all — compounding on current equity. “Ruin” means a 50% drawdown from the running peak at any point, the definition used by our risk of ruin simulator. Two reference rows first, then the correlation dial.
| What you are running | Chance of a 50% drawdown | Median worst drawdown | Median ending account |
|---|---|---|---|
| One 2% trade at a time, 200 trades | 0.3% | 20.9% | 1.42× |
| One 2% trade at a time, 600 trades | 2.9% | 29.3% | 2.87× |
| Three 2% trades per round, correlation 0 | 2.5% | 27.8% | 2.87× |
| Three 2% trades, correlation 0.3 | 10.3% | 34.4% | 2.72× |
| Three 2% trades, correlation 0.5 | 18.6% | 38.4% | 2.61× |
| Three 2% trades, correlation 0.7 | 29.0% | 42.7% | 2.49× |
| Three 2% trades, correlation 0.8 | 35.6% | 45.1% | 2.41× |
| Three 2% trades, correlation 0.9 | 43.7% | 47.8% | 2.33× |
| Three 2% trades, correlation 1.0 | 60.0% | 53.6% | 2.15× |
| One 6% trade at a time, 200 trades | 60.2% | 53.6% | 2.15× |
Three things stand out. First, the independent row and the 600-single-trades row are the same to within simulation noise: three uncorrelated trades really are three trades. That is the case the rulebook assumes, and in that case the rulebook is right. Second, the bottom two rows are identical — three fully correlated 2% positions are, exactly and not approximately, a single position risking 6%. Third, and this is the part that should change how you trade: at a correlation of 0.8 the chance of losing half the account is 35.6%, against 0.3% for the one-trade-at-a-time trader over the same calendar period. That is more than a hundred times the ruin risk, produced by three orders that each looked like textbook 2% risk on the calculator.
Look at the median drawdown column too, because it describes the ordinary run, not the disaster. The single-trade trader’s typical worst moment is a 21% drawdown. The 0.8-correlated trader’s typical worst moment is 45%, and one run in ten went past 64%. Nothing about the strategy changed. The trader simply stopped being one bet at a time.
What single-trade risk is a correlated cluster really?
The most useful way to hold this in your head is to translate a cluster into the single trade that carries the same drawdown risk. We ran the one-trade-at-a-time version at every risk setting and matched the ruin figures:
| Three positions at 2% each, correlation… | Chance of a 50% drawdown | Equivalent single-trade risk |
|---|---|---|
| 0.0 | 2.5% | about 2.4% (three separate trades) |
| 0.3 | 10.3% | about 3.3% |
| 0.5 | 18.6% | about 3.9% |
| 0.7 | 29.0% | about 4.5% |
| 0.8 | 35.6% | about 4.8% |
| 0.9 | 43.7% | about 5.2% |
| 1.0 | 60.0% | exactly 6% |
So the trader who would never risk 5% on a trade, and says so, is risking 4.8% every time they open three 2% positions in coins that move together at 0.8. Nobody sets out to do that. It happens because the calculator was asked the right question about each order and nobody asked it about the group.
How correlated are crypto positions, really?
High, most of the time, and it is worth checking rather than assuming. Correlation is usually measured between daily returns, and published matrices for the large coins sit well above the levels that matter on this page: one freely available correlation matrix describes the top-30 coins as typically 0.6–0.95 correlated to Bitcoin, with the pairs among the six largest running 0.7 to 0.95 in a risk-on market (Sharpe, correlation matrix guide). Coins that share a narrative — the same sector, the same chain, the same catalyst — cluster higher still.
Two cautions on turning that into a number for your own cluster. Daily-return correlation is not quite the same thing as the chance that your stops fire together, which is what the simulation above uses; but stops fire because prices move, so a return correlation of 0.8 is a fair first estimate of where you sit on the curve, and if anything it understates the problem, because correlations rise in sell-offs. The day you most need your positions to behave differently is the day they behave most alike. And correlations move: a pair at 0.9 over the last month may be 0.5 over the year. Check the 30-day figure before opening the cluster, not the long-run one that flatters it.
One honest simplification runs through the whole page: we treat every position as long. Three longs in three majors is the common case and the dangerous one. Section seven covers what changes when it is not.
How should you size a cluster of correlated trades?
Set the risk budget for the group, not the order. If your rule is 2% per trade, and three coins you want to buy move together, then the three of them together are the trade, and the three of them together get 2%. Here is what that does at the same 0.8 correlation, alongside the ways traders usually try to get around it:
| Sizing choice at correlation 0.8 | Total risk if all stops hit | Chance of a 50% drawdown | Median worst drawdown | Median ending account |
|---|---|---|---|---|
| Three positions at 2% each | 6% | 35.6% | 45.1% | 2.41× |
| Two positions at 2% each | 4% | 10.0% | 33.7% | 1.91× |
| Three positions at 1% each | 3% | 1.3% | 24.7% | 1.68× |
| Three positions at 0.67% each (cluster = 2%) | 2% | 0.02% | 17.0% | 1.44× |
| Five positions at 2% each | 10% | 84.2% | 63.5% | 3.28× |
| Reference: one 2% trade at a time, 200 trades | 2% | 0.3% | 20.9% | 1.42× |
The cluster-budget row lands almost exactly on top of the reference row: three positions sized as one trade behave like one trade, which is the whole point. The five-at-2% row is the one to remember. That is what “diversifying” into five alts looks like when they move together: an 84% chance of losing half the account, and the fattest median ending balance in the table — because it is simply the biggest bet. Bigger bets have bigger medians right up until the drawdown you cannot sit through, which is the same trap the Kelly page describes from the other side.
In practice the cluster budget is three steps. One: before opening a second position, ask whether it would be stopped out by the same move that stops the first. If yes, it belongs to the same cluster. Two: give the cluster your normal single-trade risk and divide it among the members — equally, or weighted towards the setup you like most. The position size calculator does the arithmetic once you hand it 0.67% instead of 2%. Three: if dividing the budget makes each position too small to be worth the fees, that is information: take the best one of the three at the full budget rather than three thin ones. Three tickers do not add edge; they add the illusion of it.
What about shorts, hedges and scaling in?
A short against your longs has negative correlation with them, and negative correlation is the one kind that genuinely reduces cluster risk: the short is more likely to win on the day the longs lose. But it is not free insurance. A long BTC, long ETH, short SOL book is a bet on relative performance, and if the whole market rises the short loses while the longs win — the trades still resolve together, just with mixed signs. Treat a hedged pair as one position with its own stop, sized on the net exposure, rather than as three independent trades.
Scaling into one coin in three pieces is not three positions. It is one position at correlation 1.0, and its risk is the sum of the three pieces — the bottom row of the ruin table. Our scaling in and out lesson covers how to size the pieces so the total stays inside the single-trade budget.
Different timeframes do not rescue you either. A swing long on the daily chart and a scalp long on the five-minute chart in the same coin share the same shock: the sudden 8% drop hits both stops. Correlation is about the shock, not the chart you drew the setup on.
What position correlation is NOT
It is not a reason to trade one coin only. Diversification across positions that actually behave differently is one of the few free reductions in risk available. The point is to measure whether they behave differently instead of assuming three tickers mean three bets.
It is not the same as losing streaks. A losing streak is losses clustering in time, one after another; our risk of ruin page shows that alone can multiply ruin by 81. Position correlation is losses clustering at the same moment. A real account usually suffers both at once, which is why every calculator figure should be read as a floor.
It is not fixed. Correlation among crypto majors is high on average, higher in crashes and occasionally low during sector rotations. A cluster budget set from a 30-day reading is a reasonable working number, not a permanent property of the coins.
It is not an argument against the 1–2% rule. It is the missing half of it. The rule works exactly as advertised when the “trade” it is applied to is everything that would lose at the same time.
Where this reasoning breaks down
1. Correlation is modelled as constant. In the simulation it never changes; in markets it spikes exactly when losses arrive. Real cluster risk is therefore worse than the tables say at any given average correlation, which is a reason to size at the high end of the range you measure, not the middle.
2. Losses are assumed to cost exactly 1R. A move that stops three positions at once is often the kind of move that gaps through stops and fills late. Three simultaneous slipped stops are a larger loss than three separate ones, and this page does not model that.
3. When the advice is wrong: genuinely uncorrelated systems. If you run one strategy on a crypto pair and another on, say, a currency or a commodity, with results you have measured as close to independent, then three 2% positions across them really are three 2% risks, and cutting each to 0.67% throws away diversification that was doing its job. The test is the measurement, not the asset class label — and it has to be measured through a sell-off, not a calm month.
4. When the advice is wrong: the cluster is intentional. A trader who wants a 6% bet on “the market goes up” and expresses it as three 2% longs has not made an error, provided they know it is a 6% bet and their rules allow one. The error is only ever believing it is 2%.
5. Ruin here is a 50% drawdown, and 200 rounds is one horizon. Choose a 30% drawdown as your definition of ruin and every figure on this page gets worse. Choose a shorter horizon and they get better. The ranking of the sizing choices does not change under either.
Where should you go from here?
Before your next multi-position day, put the group through the position size calculator with the cluster budget instead of the per-trade one, and see how the order sizes change. Then read what risk of ruin is for the one-bet-at-a-time maths this page builds on, position sizing for how a percentage becomes an order, and avoiding the big loss for why the one 6% day matters more than the many 2% ones. If you keep a trading journal, add one column: which other positions were open when this one was stopped? After fifty trades that column will tell you your real correlation better than any matrix.
FAQ
What does position correlation mean in trading?
It is the degree to which your open positions win and lose at the same time. Correlation of 0 means the results are independent; 1 means they are effectively the same trade. In crypto, positions in the large coins commonly show daily-return correlations of 0.6 to 0.95, so three separate-looking trades often behave as one.
If I risk 2% on each of three trades, how much am I really risking?
Up to 6% if all three stops are hit, and the chance of that depends on correlation. With independent results it happens 12.5% of the time; at a correlation of 0.8 it happens 34.7% of the time. Measured by the chance of a 50% drawdown, three 2% positions at 0.8 correlation carry about the same risk as one position risking 4.8%.
How do I size correlated positions?
Give the whole cluster the risk you would normally give one trade, and divide it among the positions. If your rule is 2% per trade and three coins move together, size each at about 0.67% so the group risks 2% in total. In our simulation that brought the chance of a 50% drawdown from 35.6% back to 0.02%, matching a single 2% trade.
Does holding five different altcoins diversify my risk?
Usually not much. Five positions at an average correlation of 0.8 behave like about 1.2 independent bets, not five. At 2% risk each, that cluster hit a 50% drawdown in 84% of simulated 200-round runs, because it is simply a 10% bet on the market direction.
Is correlation risk the same as a losing streak?
No. A losing streak is losses arriving one after another in time, which our risk of ruin page shows can multiply ruin by 81 on its own. Position correlation is losses arriving at the same moment across open trades. Real accounts face both, which is why any calculator figure that assumes independent results should be treated as a best case.