What portfolio heat is

Four positions, one account 

A position sizing calculation answers a question about one trade. It takes the distance from entry to stop, takes the fraction of equity that is to be at risk, and returns a size. Run four times, on four different instruments, it returns four sizes, each one internally correct. What it never returns is a statement about the account, because at no point in that arithmetic does the account appear as a whole. The four calculations do not know about each other.

Portfolio heat is the missing statement. It is the sum, across every position currently open, of the money each position would lose if it closed at its stop. One number, expressed either as an amount or as a fraction of equity, describing the whole account rather than any trade inside it. The word heat is the term trading literature settled on for how much of an account is exposed at once, and it describes the arithmetic well enough: heat accumulates as positions are added, and it comes off as they are closed.

The calculation is deliberately unclever. Each open position contributes the distance from its current stop to the price at which it would close, multiplied by its size, converted into the account currency. Those contributions are added. Nothing is weighted, nothing is discounted for the likelihood of a stop being reached, and no allowance is made for the possibility that the positions are not all wrong at once. That last omission is the point rather than an oversight: heat measures the arithmetic of everything going wrong together, which is the only case in which the sum matters.

Reading the block from the bottom up is the useful direction. The final two rows are the same account under two symmetrical outcomes, and neither is more likely than the other on the strength of this arithmetic alone. The row above them is the only one that describes the account rather than a trade, and it is the row that no individual sizing calculation produced.

Why a per trade limit does not bound the total 

A fraction of equity per trade fixes one of the two terms in the sum. The other term is the number of positions open at once, and nothing in the sizing arithmetic constrains it at all. A fraction that looks small on a single trade becomes an unremarkable total across three positions and a substantial one across twelve, without any single decision along the way having looked reckless. Each new position was sized correctly. The total was never calculated.

This is why heat is treated as a separate quantity rather than an implication of the per trade fraction. The per trade fraction is chosen deliberately. The total, in an account where nobody has written it down, is chosen by accident: it is whatever number of ideas looked interesting in the same week. Traders who track heat are tracking the term that was previously being set by circumstance.

Heat is also a snapshot rather than a fixed property of a strategy. It rises the moment a position is opened, falls when one is closed, and changes whenever a stop is relocated, because the distance from price to stop is one of its inputs. Two accounts running the identical method can carry very different heat at the same moment purely because of what happens to be open.

Positions that share a driver 

The sum treats every position as a separate event. Markets frequently do not. Several currency pairs quoted against the same currency move together when that currency moves, and a set of positions taken on the same side of it is a single bet on that currency wearing several names. Shares in one sector respond to the same input. An index and its largest constituents overlap by construction. Two commodities in the same complex share a supply story. In each case the positions are separate contracts with separate tickets, and the thing that would trigger their stops is one event.

Correlation is the statistic that describes this. It compares how two series have moved relative to their own averages over a stated window and returns a coefficient between minus one and plus one, where plus one describes two series that have moved in lockstep, minus one describes exact opposition, and zero describes no linear relationship over that window. Three properties of the statistic matter more than the number itself. It is backward looking, so it describes a window that has already happened. It is unstable, so a pair of instruments can read one way over a month and differently over a year. And correlations between risk assets have historically tended to rise during periods of market stress, which is the same period in which several stops are most likely to be reached together.

The two middle rows contain the same money and describe different situations. Three independent positions of one fraction each would need three separate things to go wrong. One position of three times that fraction needs one. The sum is blind to the difference, which is the specific limitation of heat as a measure: it counts money at risk correctly and counts events incorrectly whenever the positions are not independent.

Heat and the margin level 

Two different aggregates run over the same set of open positions, and the margin module taught the other one. Used margin is the total held against everything open, and margin level divides equity by used margin and expresses the result as a percentage. Heat is not that number and does not replace it. Used margin describes what is currently held. Heat describes what would be lost if every stop were reached. Both rise as positions are added, for unrelated reasons, and an account can be comfortable on one while stretched on the other.

The two meet through equity. Adverse moves reduce equity as they accrue, while used margin against open positions does not fall with them, so margin level falls as heat is realised. That mechanism operates on the full contract value rather than on the margin held against it, which is why a loss can exhaust the margin posted and is not limited to the amount deposited, and a favourable move is measured on exactly the same basis and to exactly the same degree. On a YAL account, open positions are closed automatically once the margin level reaches 50%, a level set by the counterparty rather than by the trader, and it is arithmetic on equity rather than a judgement about any position.

The block shows heat and margin level moving together but at different speeds, because they have different denominators. Heat is measured against equity, so it is a direct statement about how much of the account is committed. Margin level is measured against used margin, so it responds to the size of the exposure as well as to the loss. An account carrying modest heat on very large exposure and an account carrying substantial heat on small exposure are different situations, and neither figure describes both.

Where practitioners disagree 

Trading literature discusses caps on total heat, and there is no settled figure and no agreement that a fixed cap is the right instrument. One tradition treats a ceiling on the sum as the natural extension of a ceiling on each trade, on the grounds that a rule which binds only at the level of the individual trade leaves the aggregate unmanaged. Another argues that a fixed number is arbitrary once correlation is admitted, since the same total means different things depending on what is open, and that a cap therefore encourages a false sense of precision. This page does not put forward a figure for any reader, and no figure exists that removes the risk of loss.

The basis of the calculation is contested as well. Counting stop risk, as above, describes what would be lost if every stop functioned. Counting notional exposure instead describes how much of the market the account is facing, which is a different quantity and a much larger one. Some desks track both and treat neither as sufficient alone. There is also disagreement about how correlated positions should enter the sum: whether to count a correlated group once at its largest member, to apply a haircut, or to count everything in full on the reasoning that a correlation estimated from a past window is too unstable to lean on. Each convention is defensible and each produces a different number from the same set of positions.

What the number does not cover 

Every heat calculation rests on an assumption stated plainly in each worked block above: that a stop closes a position at the level specified. It is an assumption rather than a fact. A market that gaps over a weekend, an announcement, or a period of very thin liquidity can produce a fill at a worse level, and the loss recorded is then larger than the one the sum anticipated. Positions that share a driver are exposed to this together, because the event that reaches several stops at once is also the event most likely to move price through them.