Strategy Capacity: When Trade Size Changes the Result

A backtest can look flawless and still be untradeable. I’ve watched rules with clean equity curves that assumed every signal filled at the closing print fall apart the moment real size had to go to work. That gap has a name most research readers skip past: strategy capacity, the amount of capital or trading activity a process can absorb before its own orders start changing the very result the test measured.

It sounds like an institutional problem, something only a large fund worries about. It shows up much earlier than that. Any time you read a study, a screen result, or a systematic rule and treat the reported return as the return you would have earned, you’re making a silent bet about capacity. Usually you’re making it without checking.

What strategy capacity actually measures

Capacity is a conditional property of a process, not a fixed number stamped on a strategy. It depends on what you trade, how large your order is relative to the activity already in the name, how often you turn the book over, how long you’re willing to wait for a fill, the spread you pay, the depth available, the volatility around your entries, and how many other participants are chasing the same signal. Change any one of those and the number moves.

Two misreads show up constantly. The first treats capacity as a liquidity label, as if a large, heavily traded stock automatically grants room for any strategy. Liquidity is one input among several. A high-turnover rule in a liquid name can exhaust its capacity faster than a patient rule in a thinner one. The second misread treats capacity as a forecast of profit. A rule can have all the room it needs to scale and still lose money, and a genuinely good rule can be capped at a size too small to matter. Capacity tells you how far a result travels before implementation erodes it. It doesn’t say a thing about whether the result was any good to begin with.

How a fill assumption quietly does the damage

Every backtest makes an assumption about how orders get filled, and most make the friendliest one on offer: that each historical signal was executed at the quoted price or the day’s close. At tiny size that assumption is close to harmless. As intended size grows, it stops describing anything that could actually happen.

Picture the order arriving at the book. It can consume the displayed liquidity sitting in the limit order book, wait for fresh orders to replenish the level, take only a partial fill, or push the price up before the whole position trades. The visible bid-ask spread is the part everyone sees, and it’s the smallest part. The larger costs hide in delay, in the price impact as your own buying lifts the offer, and in the fills you never get because the price left without you. When I read a backtest, the first thing I check is whether any of that was modelled, or whether the whole result rides on a closing-price fantasy.

There’s a lever people forget here: the execution horizon. You can shrink impact by feeding the order in slowly over hours or days, but patience isn’t free. The longer you take, the more the price can drift away from your signal, and the more of the move you were trying to catch has already happened by the time you’re filled. Trade fast and you pay impact. Trade slow and you pay drift and the fills that get away. Capacity sits inside that trade-off, and it’s why the same rule can look tradeable to a patient desk and impossible to a hurried one.

The same rule, tested at two sizes

Consider a neutral illustration, offered to show the mechanism rather than to suggest a size. Take a stock that trades about 500,000 shares a day near $40, so roughly $20 million changes hands in an average session. Run a historical breakout rule on it, and hold the rule fixed while you change only the money behind it.

At $50,000 of notional the order is about 1,250 shares, near a quarter of one percent of a day’s volume. The backtest’s assumption that you filled at the close is roughly true. You were a rounding error in the tape, so the recorded result and the realistic one aren’t far apart.

Now run the identical rule at $5,000,000 of notional. That’s 125,000 shares, about 25 percent of everything that trades in a day. You can’t be a rounding error at that size. Your order eats through the visible depth, waits for replenishment that may not arrive at the same price, and your own demand becomes part of the order flow that moves the quote before you’re done. Same rule, same signals, same dates. The realized result pulls away from the backtest as size climbs, and the strategy logic didn’t change. Only the relationship between your order and the market’s normal activity did. I keep the dollar-volume math in front of me while reading: 125,000 shares against 500,000 traded is a quarter of the session, and no rule survives being a quarter of the session for free.

Why turnover multiplies everything

A single entry pays the implementation cost once. A rule that rebalances every week pays a version of it every week, on both sides of every trade. High portfolio turnover takes whatever per-trade slippage the strategy carries and multiplies it by how often the rule trades, which is why two strategies with identical signal quality can have wildly different capacity. The patient one absorbs a small cost slowly. The frenetic one compounds that same small cost into a wall.

Short-horizon research is where this flatters hardest. A rapid rule can post a gorgeous gross return in a simulation that charges nothing for turnover, then hand almost all of it back once realistic costs and partial fills are applied at any serious size. The signal can be real and the strategy still fail, purely on the arithmetic of trading it too often at too large a size.

A big universe or a big return proves nothing

Two numbers get quoted as if they settle the capacity question, and neither one does. The first is the size of the investable universe. A rule that ran across three thousand names sounds like it has endless room, but capacity lives in each position, not in the count of tickers. If the rule keeps concentrating into the same forty small, crowded names every rebalance, the width of the universe is decoration. The second number is the headline historical return. A high figure tells you the rule worked on paper at whatever size the test assumed, and it doesn’t tell you how much capital that figure survives.

Crowding deserves its own line. When many participants chase the same signal, they compete for the same liquidity at the same moment, and the capacity each of them enjoys shrinks as the crowd grows. Livermore described this a century ago: his own buying moved the prices he wanted to buy, and the bigger he got, the harder it became to build or leave a position without paying for the privilege. The mechanics have new names now. The underlying constraint is the old one.

Questions to ask when you read the research

You don’t need a fund’s order book to read a capacity claim critically. You need a short list of questions and the willingness to downgrade a result when the answers are missing.

• What fill assumption did the test use? Closing price, midpoint, next open, or a modelled fill against real depth?
• What fraction of normal activity would the simulated order represent at the size being discussed? A result at 0.25 percent of daily volume and the same result at 25 percent are two different claims.
• Are costs and partial fills modelled, or is execution treated as free and complete?
• Does turnover change the burden? A rule that trades ten times as often needs ten times the cost discipline to hold up.

Every one of those questions is about realism. None of them tells you the result will happen again. Those are separate things, and confusing them is how good research gets misread in both directions: a fragile study gets trusted because the return was large, or a durable one gets dismissed because a single year disappointed.

Where capacity estimates run out of road

Now the honest limit. Capacity is genuinely hard to pin down from public end-of-day data, because the things that determine it are mostly invisible from the outside. Hidden liquidity never prints until it trades. Order-routing choices, the behaviour of other participants, and the degree of crowding around a signal rarely show up in a daily bar. Any capacity estimate built from close-only data is a rough sketch, and it’s a sketch drawn in conditions that refuse to hold still. When volatility rises, when market structure shifts, or when more capital piles into the same trade, the estimate you trusted last year can be wrong this year.

So treat capacity as a question you keep asking, not a certificate a strategy earns once. What it’s genuinely useful for is forcing the right follow-up whenever a backtest looks too clean: how much of this survives contact with real size? Read the research that way and you’ll be fooled less often by a beautiful curve that was never built to carry any weight.

Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.