Slippage in Trading: Why Good Strategies Lose Money
Slippage in trading is the gap between the price a strategy expected and the price it got. Most research treats that gap as a rounding error and subtracts a flat number for it somewhere near the end of the spreadsheet.
Measured against real fills, the gap is neither small nor steady. Across $1.7 trillion of live equity orders in 21 developed markets between 1998 and 2016, researchers at AQR found market impact averaging 9.97 basis points per order, and 18.95 basis points in small caps against 8.90 basis points in large caps. Those are one-way costs, achieved by a well-resourced institution that could choose when to trade.
A private account gets a different answer. In an experiment placing 85,000 matched market orders through five US brokers in 2022, the cost of a full round trip ranged from 0.07% at one broker to 0.46% at another, before commission.
QuantConnect’s LEAN and Backtrader both apply zero slippage until someone tells them otherwise, and Zipline applies a flat five basis points that does not grow with the size of the order.
- What is slippage in trading?
- How much slippage is normal?
- Where does slippage come from?
- Why do backtests look better than live trading?
- What can you do to reduce slippage?
- What happens to slippage when markets are stressed?
- Why does the same order cost more at some brokers?
- What should you ask a manager about slippage?
- Questions people ask about slippage
What is slippage in trading?
Slippage in trading is the difference between the price you expected when you sent an order and the price you were filled at. You click buy at $50.00 and the confirmation says $50.04. Those four cents are slippage, and they are gone whether or not the trade eventually works.
It happens on almost every order. On one trade it is invisible. The arithmetic is what makes it matter: four cents on a fifty-dollar share is eight basis points, you pay it going in and again coming out, and at twenty-four round trips a year that is roughly 3.8% of the position gone before the strategy has earned anything.
Slippage runs both ways. An order can fill better than expected, and a fairly routed account shows both directions over a few thousand orders. A book that only ever loses on the gap is worth a question about who is filling it and how.
How much slippage is normal?
Roughly 9 to 19 basis points one way on an institutional equity order, and 0.07% to 0.46% for a full retail round trip, depending on which broker fills it. Both ranges come from measured executions rather than estimates. Anyone quoting one number for slippage in trading has not measured it.
| Who is trading | What it costs | Where the number comes from |
|---|---|---|
| A large institution, one equity order | 0.10% buying, before selling back | Frazzini, Israel and Moskowitz: 11 million live orders |
| The same institution, in small caps | 0.19% buying, before selling back | Same study, small caps against 0.089% in large caps |
| A private account, bought and sold again | 0.07% to 0.46%, by broker | Schwarz and co-authors: 85,000 orders, 2022 |
| A strategy rebalancing once a year | under 0.1% a month | Novy-Marx and Velikov, 2016 |
| A strategy trading one to five times a year | 0.2% to 0.57% a month | Same study. Often more than half the gross return |
| A strategy trading more than five times a year | over 1% a month | Same study, for strategies built with no regard for cost |
Read the caveat with the institutional number. That is one well-resourced institution measuring its own patient execution, on trades it chose when to place, and the authors say plainly that they find costs an order of magnitude below what earlier academic models implied.
The monthly figures and the per-order figures are two different measurements. One prices a whole portfolio being rebuilt month after month. The other prices a single fill. A cost model needs both, and a pitch deck usually shows neither.
The frame that holds them together is implementation shortfall, which measures the whole distance from the decision to the position that ends up in the book:
- Market impact, the price your own order pushes away from you while it fills
- Delay, the price that moved between deciding and sending
- The trades that never got done, and the profit that was sitting in them
The spread is the visible part of that distance and the smaller one. A working cost model behind a quantitative trading strategy is built on all three.
Where does slippage come from?
From the distance between the market a model is looking at and the market an order meets. Five things open that distance. Which one dominates tells you where a strategy will stop working, and which part of its trade execution is worth examining first.
The bid-ask spread. Every order that crosses pays the difference between the buying price and the selling price. In a strategy holding for months this barely registers. In one holding for minutes it can be the entire result.
Market impact. A large order moves the price while it is still being filled. Size in dollars matters far less than size measured against what normally trades in that instrument at that hour.
Delay between signal and fill. The model decides at one price and the order completes at another, and the time in between belongs to routing, broker systems, approval steps, or a signal that decays faster than the order can be worked.
Liquidity that comes and goes. Depth, meaning the volume sitting on the order book ready to trade, is thinnest at the open, at the close, around scheduled releases, and in the sessions when a strategy most wants to move. A cost model calibrated on calm Tuesdays will understate all four.
How the order was designed. Market or limit, aggressive or passive, all at once or sliced across an hour, one broker or several. This is the only source on the list that is entirely within your control.
| Source | What it looks like on a fill report | What it tells you about the strategy |
|---|---|---|
| Bid-ask spread | A small, steady loss on every order | Whether the holding period is long enough to pay for the trading |
| Market impact | Cost that grows as the order grows | How much money the strategy can take before it stops working |
| Delay between signal and fill | Fill price sitting away from the decision price | Whether the signal survives the time it takes to trade it |
| Liquidity stress | Costs spiking in a handful of sessions | What happens on the days the strategy is tested for real |
| Order design | Wide dispersion between similar orders | Whether execution is run deliberately or left to a default |
Why do backtests look better than live trading?
Because the software fills every order instantly, in full, at a historical price nobody had to compete for. Real orders queue, fill partly, or miss. Most of the difference between those two worlds is a settings problem rather than a market problem.
Start with what your engine assumes when you have not told it anything. QuantConnect’s LEAN ships with a brokerage model that applies the NullSlippageModel, which sets slippage on every order to zero. Backtrader’s two slippage parameters both default to 0.0.
Zipline is the exception and charges equities a flat five basis points, capping any single fill at 10% of the bar’s volume. Better than nothing, and still a fiction: the rate a small order pays and the rate a very large one pays are identical.
QuantConnect’s own reconciliation guidance says the quiet part out loud. Market impact is not modelled at all, so fills on large orders come out better in a backtest than they do in live trading.
Fill logic can also be wrong in ways that flatter you after you have set a number. LEAN issue #4545, filed in June 2020 and later closed as a duplicate, described stop-market orders on gap bars filling at the bar’s closing price when price had opened below the stop, and filling at the stop price with no cost at all when the bar gapped straight through it.
The first case inflated the cost of a stop that price walked through. The second handed the strategy a gap for free. Neither shows up as an error, and the point is the class of mistake rather than that one ticket, which is why the fill model in whatever engine you run today is worth reading before the results are.
The equity curve comes out too high either way. Three broken strategies then look identical from the outside: the one that never survived an honest out-of-sample test, the one fitted to noise in the first place, and the one that failed only on trading costs.
| Engine | Default slippage on equities | Does the rate grow with order size? |
|---|---|---|
| QuantConnect LEAN | Zero (NullSlippageModel) | No, and market impact is not modelled at all |
| Backtrader | Zero (slip_perc and slip_fixed both 0.0) | No |
| Zipline | 5 basis points, fixed | No, though a 10% volume cap limits how much fills |
| A real fill | Varies by instrument, size, hour and session | Yes, and faster than linearly once the order is large |
What can you do to reduce slippage?
Trade less, size against the instrument rather than your account, choose the moment, and measure what you are paying. None of it requires better technology. Four changes do most of the work in a retail or small systematic book.
Stop paying for speed you do not need. A market order buys certainty of execution with an unknown price. A limit order fixes the price and accepts that some orders will go unfilled. Both sides cost something, and the missed trades hurt most when the signal was right, so the comparison to run is your fill costs against the profit on the orders that never filled.
Size against the instrument’s volume rather than against your account. A position that looks modest next to your capital can still be large next to what the instrument trades in a day, and that second ratio is what sets your market impact. It is also the ceiling on how much money the approach can ever take, which is the subject of strategy capacity.
Avoid the minutes everyone else is trading. A rebalance that has to happen in the closing auction is paying the widest spread of the day, every time, by design.
Log every order. Expected price, fill price, difference in basis points, order type, instrument, time of day. Two hundred rows is enough to show whether the assumption in your backtest was optimistic. Then put your measured numbers back into the engine, split by instrument if you trade more than one, and run the strategy again.
What happens to slippage when markets are stressed?
It rises fastest in exactly the sessions a strategy most needs to trade, and it rises in the deepest markets as well as the thin ones. The instruments a portfolio runs toward in a crisis are not exempt, which is the part most cost models leave out.
US Treasuries are the deepest government bond market in the world. In April 2025 that market widened, thinned and became expensive to move within a week.
- Bid-ask spreads in the Treasury market widened markedly after the tariff announcement of 2 April 2025
- The volume resting on the order book fell to its lowest level since March 2023, so the same order pushed the price further
- Price impact, the amount the price moves per $100 million traded, rose abruptly on 2 April and peaked around 9 April
- The ten-year yield rose from about 3.9% to about 4.5% between the evening of 6 April and midnight two days later
- All three measures recovered after the tariff postponement, and the authors conclude liquidity moved in line with volatility rather than breaking down
Two earlier episodes make the same point at higher intensity. In March 2020 the selling pressure ran across every advanced sovereign bond market and caused what a March 2022 Federal Reserve Bank of New York staff report calls a deterioration in market functioning that led to central bank interventions, concentrated disproportionately in US Treasuries.
On 5 August 2024 a crowded trade borrowing cheaply in yen to buy higher-yielding assets went into reverse. The BIS recorded in September 2024 that the VIX briefly passed 60 in pre-market trading before closing under 40, and that Japan’s TOPIX banks index took the worst single-day loss of its forty-year history.
The August 2024 detail worth keeping sounds the least dramatic. The BIS found liquidity conditions deteriorated while still allowing trading to continue. Costs rose sharply and the market kept working, which is the ordinary case and the one a cost model should be built around.
Any strategy sold on the promise that it trades only liquid instruments should produce its own fills from a week like that one. Not the month’s average. The week.
Why does the same order cost more at some brokers?
Because where an order is routed and who fills it changes the price you receive, and the gap between the best and worst outcome is wider than most cost assumptions allow for. The experiment that measured this found one account paying more than six times what another paid.
Schwarz, Barber, Huang, Jorion and Odean opened six accounts at five US brokers and pushed 85,000 matched market orders through them in 2022, publishing the result in the Journal of Finance in 2025. Which broker you use is a cost decision of the same size as which order type you use.
Regulators have penalised the extreme version of this in retail currency dealing. The CFTC’s 2011 order against FXCM found the firm kept $8.26 million of favourable price moves belonging to more than 57,000 customer accounts, passing the unfavourable moves through and retaining the favourable ones.
Two years later the CFTC found FXDirectDealer rejecting orders once price moved more than two pips in the customer’s favour while filling them when it moved two pips against, across more than 24,900 accounts. Both cases are old, both were retail currency rather than equities, and both describe an incentive that has not gone anywhere.
Disclosure has been catching up. The SEC adopted amendments to Rule 605 on 6 March 2024 that widen execution-quality reporting to far more brokers and add statistics for fractional shares, odd lots and orders larger than a round lot, with average time to execution measured in milliseconds or finer.
The compliance date was extended to 1 August 2026. Reports in that format answer this question with a broker’s own numbers instead of its marketing.
What should you ask a manager about slippage?
Ask how the cost was estimated, what it is measured against, what happens as assets grow, and how the live number compares with the modelled one. Four questions, each answerable with a document rather than a description. A manager who has controlled slippage produces all four inside a day.
Each of those four should arrive as an artefact. Five of them, taken together, are what a complete answer looks like.
- A cost assumption that varies by instrument, liquidity, order size and holding period, rather than one number applied to everything the strategy trades
- Decision price, the arrival price when the order reached the market, and the final fill price kept as three separate figures, so market movement and execution quality are not blended into one
- A capacity estimate that links assets to expected cost, showing where the cost curve bends rather than asserting the strategy scales
- Realised against modelled cost tracked over time, with an explanation for any month the two separated
- Fills from a single named stressed week, rather than the quarterly average that contains it
One flat cost assumption applied across every instrument and every regime is the single most common thing we find in the material we read. It is sometimes convenience. More often it is the only assumption under which the backtest still looks good, and it is the first thing worth pulling on.
There is one question we have learned to ask early, because the answer is hard to rehearse. Can the manager name a strategy they chose not to launch, or cut back after launch, because the trading cost made the economics work out badly?
A real example comes with an instrument, a size and a date. The alternative answer, that every good idea can be run at scale, tells you the question has never come up.
Execution oversight belongs in the same conversation. FINRA’s 2026 Annual Regulatory Oversight Report identifies as an effective practice conducting the required regular and rigorous execution-quality reviews quarterly or more frequently, with some firms doing it monthly. A manager who cannot say when their own review last happened has answered the question anyway.
Most strategies we look at cannot show their fills from a bad week
Those five artefacts are not a high bar. A cost model with more than one number in it, three prices kept apart, a capacity estimate, a realised-against-modelled history, and one stressed week. Very few of the strategies that reach us produce all five without being asked twice, and slippage in trading is where most of them stop.
Backtest slippage realism, rejected-order handling and trade reconciliation are three of the five criteria scored under execution quality in the framework behind every profile we publish.
A systematic gold strategy profiled in The Algo & Quant Review enters with limit orders only and a hard cap on the slippage it will accept, cancelling the orders it cannot fill rather than chasing them. Its cost assumptions were built from tick data on the venue it trades. It still scores 3 out of 5 on rejected-order handling, and that number is published next to the good ones.