Slippage is two distributions reported as one number
Positive and negative slippage have different causes and different beneficiaries. Averaging them produces a figure near zero that is true, useless, and hides a systematic transfer.
Average slippage is reported as a single number close to zero, and that number is constructed so that it always will be. Positive and negative slippage are different phenomena with different causes, and averaging them together is how a systematic transfer gets recorded as noise.
Two distributions wearing one label
Slippage is the difference between the price shown when an order was submitted and the price it filled at. Positive slippage benefits the client, negative benefits whoever took the other side. A venue that passes on adverse movement but retains favourable movement produces a distribution with a long negative tail and a truncated positive one, and a mean of almost exactly nothing.
mean = ( Σ positive × P(pos) ) + ( Σ negative × P(neg) )Symmetry is the test, not magnitude. A venue that slips you both ways equally is behaving like a market. One that only slips you one way is behaving like a counterparty.
Where asymmetry legitimately comes from
Not all of it is extraction. Distinguishing the mechanical causes from the behavioural ones is the difference between a productive provider conversation and an accusatory one.
| Cause | Signature | Legitimate? |
|---|---|---|
| Latency in the quote path | Symmetric, scales with round-trip time | Yes, fixable |
| Genuine thin liquidity | Clusters at session boundaries and rollovers | Yes |
| Order size above top of book | Scales with notional, not with time | Yes |
| Price improvement withheld | Positive side truncated, negative side full | No |
| Selective last look | Asymmetry concentrated in fast markets only | No |
The rollover and news-window confound
Aggregate slippage measured across a whole month is dominated by a handful of minutes. Rollover, the first seconds after a scheduled release, and the illiquid window around session close will produce genuine asymmetry at any venue. Excluding those windows and measuring the remainder separately is the only way to see whether the baseline is clean.
- 01Timestamp at submission, not at fillClient-side, not venue-reported
- 02Split positive and negativeNever report the mean alone
- 03Exclude scheduled windowsMeasure them as their own population
- 04Normalise by order sizeSo large orders do not dominate
- 05Compare venues on shapeRatio of tails, not difference of means
| Venue A | Venue B | |
|---|---|---|
| Reported mean slippage | +0.01 | −0.02 |
| Positive fills, share | 46% | 19% |
| Negative fills, share | 44% | 71% |
| Tail ratio, neg to pos | 1.07 | 4.83 |
| Cost per $1M, adjusted | $31 | $88 |
Illustrative figures, chosen to show the mechanism. Your own numbers will differ.
Why clients notice before you do
Clients do not compute distributions, but they do remember. Asymmetric slippage produces a specific and consistent complaint: “I never get the price I see, but only when it moves against me.” That complaint is usually treated as a support issue and closed with an explanation of market conditions.
- A rise in execution complaints concentrated in one instrument group, not across the board.
- Complaints clustered in the same few minutes each day, which points at a window rather than a venue.
- Clients who reduce size rather than leave, which is invisible in churn reporting.
Joining execution telemetry to the support ticket that mentions it is the step nobody takes, because the two live in unrelated systems. It is also the fastest available route to knowing whether a venue problem is real.