2. Slippage: the price you saw vs the one you got

Costs 2: What you pay and cannot see

The spread is a cost you can work out before you trade: you look at the book, you see the distance between bid and ask and you already know what crossing is going to cost you. Slippage is the cost you only find out about afterwards, and it is the one that ruins the most backtests.

What slippage actually is

Slippage is the difference between the reference price you had when you took the decision and the average price at which you were eventually filled. Nothing more. The serious argument is over which price that reference should be, because depending on what you pick the number comes out flattering or realistic.

The honest benchmark is the arrival price: the mid at the instant you decided to trade. Any other one tends to hide something. And what separates that price from the one you end up with is three distinct mechanisms, best kept apart:

• Delay. Between your system generating the signal and the order leaving, traveling and reaching the market, the price has moved. In a slow strategy we are talking about cents; in a fast one, that delay is the entire strategy.
• Crossing the spread. The business of the previous module. It is the predictable part of slippage.
• Sweeping the book. If your order is larger than the size available at the best price, the remainder fills at worse levels, each one further from the mid. Your average price gets worse without the market having done anything of its own accord.

Slippage is not zero-mean noise

Here is the mental error that costs the most. A great many people assume slippage is random: sometimes it goes your way, sometimes against you, and in the long run it evens out. It does not even out, and there are structural reasons why.

• You trade when everybody trades. Signals fire on news, on the open, on macro releases and on breaks of key levels, which is to say at the moments when the spread is widest and the book thinnest.
• Adverse selection. When the price runs away from you it is because there is flow pushing in that direction; when it comes toward you, it is usually because somebody informed is selling precisely what you are buying.
• You chase moves. A trend-following strategy buys, by definition, whatever is going up: the mechanics of the signal itself guarantee that the price runs away from you while you execute.
• Stop orders fill at the worst possible moment. A stop order guarantees no price: once triggered it turns into a market order, and it triggers precisely when the market is moving fast against you.

The upshot is that real average slippage carries a negative sign almost always. It is not a measurement error: it is a consequence of when and why you decide to trade.

How much it takes: the sum you have to do

Slippage is subtracted in absolute terms, not proportional ones. That detail changes everything. Suppose slippage of 8 basis points a side, a perfectly reasonable figure in mid-sized securities with normal-sized orders: 16 basis points per complete trade.

• High-turnover strategy. Gross edge of 25 basis points per complete trade. Take away 16 and 9 are left. You have lost 64% of the return, and that is before commissions and before the spread you already paid.
• Low-turnover strategy. You buy and hold for two years, with a gross gain of 40%, that is, 4,000 basis points. Take away the same 16 and 3,984 are left. You have lost 0.4% of the return.

The same cost, exactly the same, destroys one strategy and is invisible in the other. The lesson is not that slippage is large or small, but that the only thing that matters is the ratio between the edge per trade and the cost per trade. A strategy with a big edge per trade can afford mediocre execution; one that lives by scraping a few basis points needs excellent execution or it does not exist.

A second calculation, this one about backtests. Picture a system validated on the assumption of execution at the closing price, with 150 complete trades a year. In real life each side fills 4 basis points worse than that theoretical close. That is 8 basis points per complete trade, 150 times over: 1,200 basis points, twelve percentage points a year that the backtest invented. A system that promised 15% delivers 3%. Not one signal has failed: the simulation has.

The case of the stop that did not protect you

A concrete example earns its place here, because this one generates a great many arguments with the broker that lead nowhere. You hold a share bought at 55 euros and a sell stop at 50. Dreadful quarterly results come out after the close and the next day the security opens at 46.

Your stop triggers and you are filled around 46, not at 50. The slippage is 4 euros, an extra 8% on top of what you believed was your maximum loss. There is nothing to complain about: not a single trade crossed between 50 and 46, quite simply because no price existed in that stretch. A stop order defines when your order is sent, not at what price it is filled. Anyone who confuses the two has miscalculated their maximum risk, which is a far graver error than paying a few basis points too many.

Measuring it: the benchmark decides the answer

To finish, the commonest measurement trap of all. Execution is routinely judged against the volume-weighted average price of the session or of the execution window. It sounds objective and it has a serious problem: if your order is an appreciable part of the volume, you are the one forming that average, so you are comparing yourself with yourself. Beating that average is entirely compatible with having executed disastrously as measured from the price that was there when you decided.

Robert Kissell, one of the authors who has worked most on the measurement of execution costs, has been insisting on this point for years: the benchmark has to be the decision price, because it is the only one your own activity does not contaminate. If you are not measuring slippage against that price, you are not measuring it: you are dressing it up.