Correlation and Portfolio Heat: Why Six Positions Can Be One

A trader holding six positions at 1% risk each will usually describe the risk as “1% per trade”, which is true of each ticket taken on its own and thoroughly misleading about the book as a whole, because 6% is what stands to be lost if everything goes wrong on the same afternoon.

Whether everything can go wrong on the same afternoon depends on correlation. Most retail risk management ignores it.

Portfolio heat

Portfolio heat is the sum of risk across all open positions. It is what you lose if every stop is hit at once.

Positions	Risk each	Total heat
3	1%	3%
6	1%	6%
10	1%	10%
6	2%	12%

Most professional frameworks cap heat somewhere between 5% and 6%, whatever the per-trade sizing might suggest. Reach the cap and you take no new positions until something closes. That constrains opportunity, which is rather the point of having it.

Why heat understates the real risk

That table assumes the positions are independent, and they rarely are.

Suppose the six are six semiconductor names. Six tickers, six stops, 6% of heat. In substance it is one bet on the semiconductor cycle, since poor guidance from any of them moves all six on the same day and a sector rotation moves all six harder than that.

So the effective risk is closer to 6% riding on a single outcome, accompanied by six sets of commission and six separate chances to be stopped out by something with no bearing on the thesis at all.

The reverse error exists too. A utility, a gold miner, a biotech and a regional bank answer to genuinely different drivers, and 4% spread across those four is materially less risky than 4% concentrated in one of them, although nothing on a position report will ever say so.

Count drivers before you count tickers

Sector is the obvious one. Factor comes next, and it is the one that catches people out, because growth against value, large against small cap and momentum against quality all cut across sector lines, which means four positions spread neatly across four different sectors can still add up to a single factor bet wearing four names. Then macro sensitivity, since rate-sensitive names such as utilities, REITs and long-duration growth move together irrespective of sector. And single inputs. An oil price, the dollar, one customer, one regulatory decision.

Then add a cap for each driver on top of the per-position one:

Limit	Cap
Risk per position	1%
Risk per sector	2%
Risk per factor or macro driver	3%
Total portfolio heat	6%

Under those rules the six-semiconductor portfolio cannot be built. You hold two, at 1% each, and go looking elsewhere for the rest.

The correlation problem nobody solves

Correlations move about. They also fail in one direction, converging towards one during broad liquidations.

In normal conditions a utility and a biotech have almost nothing to do with each other, which is the whole basis on which that pair gets described as diversification in the first place. In a genuine market-wide selloff both of them fall, because participants facing margin calls and redemptions sell whatever can be liquidated quickly and preference plays no part in the choice. Correlation measured on calm data understates risk in exactly the conditions where you needed it to hold, which is a defect no quantity of extra history will repair.

Sizing when positions relate

Two adjustments handle most of it.

The first is to treat correlated positions as fractions of one. Wanting semiconductor exposure across three names means sizing each at a third of a standalone position, which leaves the sector risk where it was and buys a little idiosyncratic spread along the way.

The second is to cut total heat when breadth narrows. When market internals show a narrowing advance, realised correlations are usually rising. Dropping the heat cap from 6% to 3% in those conditions costs some upside and removes a large share of the tail, and the upside it costs is mostly the part you would have given back in the event anyway.

Work the resulting per-position risk into share counts with the position size calculator. The arithmetic is unchanged. Only the risk percentage moves.

A worked comparison

Two traders. Both risk 1% per trade and both hold six positions.

Trader A holds six semiconductors. A sector downgrade arrives and all six gap through their stops. Realised loss: roughly 8% after slippage, out of a book its owner had believed was spread across six independent ideas, none of which turned out to be independent of the others.

Trader B holds one semiconductor, one utility, one energy, one healthcare, one industrial and one consumer staple, which is a duller list to read out and a completely different shape of risk. The same downgrade hits one position. Realised loss: about 1.3%.

Same rule, same position count, same per-trade risk. The difference is structural. None of it shows up on a per-trade risk report.

What this does to expectancy

Lower variance leaves expected return where it was. What it raises is the probability that you are still trading when the edge pays.

A strategy with +0.2R expectancy and uncontrolled correlation produces drawdowns that most people abandon before the edge has had any chance to compound, and abandoning it is a perfectly rational response to a curve that has started to look broken. Put driver limits on the same strategy and the curve becomes shallower and slower and survivable, and a survivable edge is the only kind that ever gets realised. See expectancy and sample size for why the number of trades you complete is the binding constraint.

Going further

Four companion pieces take the parts of this argument that deserve their own space. How to calculate correlation between two stocks covers the arithmetic, including the rolling version that shows how unstable a pair really is. Correlation vs beta separates the two measurements that get used interchangeably and explains which one belongs in a sizing decision. Why correlation goes to 1 in a crash deals with the convergence that makes calm-period diversification overstate itself. And the portfolio correlation checklist is the short version to run before opening a position.

Two further pieces deal with the questions that follow from counting drivers rather than tickers. How many stocks do you need to diversify? works through where the arithmetic of spreading risk stops paying, and why twenty holdings can be worth four. How to rebalance when correlations change covers the drift afterwards: positions opened as separate ideas become one bet without anything being bought, and the review that catches it before several of them fall together.