Two curves, one chart, 270 million dollars apart
Capacity is the dollar figure at which a strategy’s own footprint removes its edge. Two backtests can look identical and differ by that figure alone.
· 7 min · capacity
4 / 4
| candidate | walk forward Sharpe | capacity |
|---|---|---|
| one | 0.81 | 40 million dollars |
| the other | 0.81 | 310 million dollars |
Put two candidates on one chart. Same walk forward Sharpe, 0.81. Same shape, same drawdowns, same start and end. If you were shown only the curves you couldn’t tell them apart, and most pitch decks show only the curves.
One of them carries 40 million dollars before its own trading eats a third of what it earns. The other carries 310 million. That is a 270 million dollar difference between two lines that overlap, and it is the difference between a candidate worth a meeting and one that is not. The number that separates them is capacity, and this note is about where it comes from and why it belongs in the first conversation rather than the fourth.
What capacity means, exactly
Capacity is the most capital a strategy can run before its own market impact removes the edge. Every strategy has one, because every trade moves the price against the trader, and the bigger the trade relative to what the market does that day, the further it moves. Scale up and the edge stays roughly fixed while the cost of collecting it rises. At some size the two cross.
The agent doesn’t wait for them to cross. It states capacity as the size at which impact exceeds a third of the edge. A third is a threshold, not a law, chosen so the number reported is a size you would actually run rather than the size at which the strategy stops working. Run where impact equals the whole edge and you earn nothing. Run where impact takes a third and you keep two thirds, and that is a business.
So when the candidate frame reads “310 million dollars before impact exceeds a third of the edge”, it means that at 310 million the strategy’s own footprint is consuming a third of what it makes. Past that, you are paying more and more of your edge to yourself.
Where the number comes from
Capacity isn’t a separate estimate bolted on at the end. It falls out of the same impact model that charged the fills at the costs gate, which is why the two numbers agree with each other.
That model rests on a measured result. Toth, Lemperiere, Deremble, de Lataillade, Kockelkoren and Bouchaud at Capital Fund Management, in Physical Review X in 2011, showed that the price impact of a large order grows with the square root of its size relative to the day’s volume. The square root is what makes capacity finite and computable, and it means a strategy’s capacity depends on two things it owns: how often it trades, and how liquid the names are.
Turnover first. A strategy that rebalances daily pushes its whole book through the market many more times a year than one that rebalances quarterly, so it pays impact many more times and hits the one third threshold at a far smaller size. That is why, on the simulator page, the 1 day horizon carries a capacity multiplier of 0.25 and the 63 day horizon carries 1.6. Same universe, six times the capacity, purely from trading less often.
Liquidity second. In 500 large caps a given dollar size is a small fraction of a large daily volume. In 3,000 all caps it is a bigger fraction of a smaller one. The universe sets the base: large cap in the simulator starts at 400 million dollars, all cap at 150 million.
Put the two together, run the impact model over the strategy’s actual fills, and solve for the size at which impact reaches a third of the edge. That is the number. It is in dollars, because dollars are what an allocator decides about.
| universe | capacity base ($M) | horizon | capacity multiplier |
|---|---|---|---|
| US large cap | 400 | 1 day | 0.25 |
| US all cap | 150 | 5 days | 0.50 |
| Global developed | 300 | 21 days | 1.00 |
| Futures | 600 | 63 days | 1.60 |
Why it goes in the first meeting
The habit in most research processes is to establish that a signal works, argue about it, and then, weeks later, ask the execution desk how much it could carry. That order wastes the weeks. If the answer is 40 million dollars and the fund needs 200 to make the allocation worth its risk budget, every hour of the argument went to a candidate that could never be sized.
The candidate frame prints three lines. Sharpe, with its in sample, walk forward and haircut figures. Capacity, in dollars. Decay watch, with a half life. Those are the three numbers a portfolio manager asks for second, after the curve. Printing them first means the second question is answered before the first one is asked.
In the canonical run, 14 specifications reached the capacity gate. Eight of them carried under 40 million dollars before impact exceeded a third of the edge. They had cleared point in time, cleared walk forward, and paid their costs. They were real, small, and not worth a meeting at a fund of any size. The log says so in one row, “killed 8, capacity: under 40 million dollars before impact”, and the desk moves on.
“Same walk forward Sharpe, 0.81. Same curve. One carries 40 million dollars, the other 310 million. Most pitch decks show only the curve.”
The companion number
Capacity tells you how much a strategy can carry today. It doesn’t tell you for how long. That is the third line, the decay watch, which estimates a half life for the signal, the time over which its predictive power falls to half, and rebuilds the estimate weekly from live data. The canonical candidate reads 19 months. A candidate with 310 million dollars of capacity and a 19 month half life is a different proposition from one with the same capacity and a 6 month half life, and both differ from one whose half life nobody is watching.
What to look for
- If you are handed a backtest without a capacity figure, ask for one. If you are handed a capacity figure, ask what impact model produced it, whether the same model charged the fills, and what threshold it uses. If the threshold is “where the strategy stops working”, the number is optimistic by a large factor. If the fills were charged a flat cost and capacity was estimated separately, the two numbers are not talking to each other.
- And if two candidates look identical on a chart, don’t choose between them on the chart.
Illustrative output. Backtested, modelled and simulated results are hypothetical. They do not represent actual trading, they carry the benefit of hindsight, and they may differ materially from any result achieved in the market. Past performance, real or simulated, does not indicate future results.
No output of the Services is a recommendation, an offer, a solicitation, or advice about any security, strategy or account. Every investment decision is made by the customer and the customer's own investment professionals.