Cost sensitivity: how much slippage can this edge survive?
Every backtest result is really a bet: this edge is worth more than what it costs to trade it. Cost sensitivity tells you where that bet stops paying — the break-even slippage, the point where costs cancel the entire result.
It's on every finished run that stored its trades. Open the run and scroll to Cost sensitivity.
Why break-even beats the P&L figure
"Net $59,366" answers a question nobody can act on, because it's true only for one cost assumption you typed in earlier. "Breaks even at 3.8 ticks per side" is a claim you can actually check against reality: is your broker filling you inside 3.8 ticks on this contract, at the times this strategy trades?
Two results with identical P&L can be completely different bets:
| | Net P&L | Breaks even at | |---|---|---| | Strategy A | $50,000 | 12 ticks/side | | Strategy B | $50,000 | 1.2 ticks/side |
A is a real edge with room to spare. B is a rounding error away from worthless — one bad fill per trade and it's gone. The P&L column can't tell them apart. The break-even column can.
As a rough guide: if break-even is under ~1 tick per side, treat the result as noise. Under 2, treat it as fragile. Well above your realistic fill quality, you have something worth testing further. These are starting points, not rules — a strategy trading one contract in a liquid session at 09:35 faces different fills from one trading ten contracts at 03:00.
How to read the panel
The line is net P&L against slippage. It's straight, because slippage is a flat charge per contract per side — every extra tick costs the same dollars regardless of whether the trade won. The orange marker is where it crosses zero.
The slider moves the assumption. The stat cards below it — net, profit factor, average trade, max drawdown — update instantly to that cost level.
Nothing is re-backtested. The fills are fixed; only what they cost changes. That's why the slider is instant, and it's also the point: it isolates one variable cleanly.
The part this can't tell you
Costs hit a strategy twice, and only one of them is measured here.
- Directly — every trade costs more. That's what the slider shows, and it's exact.
- Through selection — costs are part of the score the optimizer maximizes, so raising them changes which settings win each search window, which changes which trades you get at all.
The second effect needs a real re-run to measure, and it can be large enough to move the answer in a surprising direction: a re-run at higher slippage can produce a better headline number, because the optimizer, now penalised for churn, picks calmer settings that trade less.
That's a genuinely useful thing to know about a strategy — but it is not "slippage sensitivity", and confusing the two is one of the easier ways to talk yourself into a bad system. If you want the selection effect, re-run the walk-forward with the new cost setting in Settings and compare. If you want to know whether the edge survives worse fills, use the slider.
Portfolios: members don't share a break-even
Each member trades its own instrument, and instruments have different tick values — a tick of GC is $10, a tick of ES is $12.50, a tick of MES is $1.25. So the same tick of slippage costs each member a different amount, and the member table shows each one's own break-even.
This is worth reading before you trust a portfolio. A combination can look healthy while one member is already underwater at realistic costs and the other is carrying it.
When it isn't available
- "This run stored no trade list" — the study didn't keep its trades. Runs that store them: optimize, walk-forward, robustness, perturbation, certify and portfolio walk-forward.
- "Trades aren't tagged with an instrument" — a portfolio run made before per-trade tagging was added. Its trades were merged into one stream with no record of which member each came from, so they can't be priced at the right tick value. Re-run it.
- Commission fixed, slippage only — an older run stored its net P&L but not the cost that produced it, so the two can't be separated back out. Slippage can still be varied relative to what was applied. Re-run it to vary both.
Refusing is deliberate in each case. A cost curve built on a guessed instrument would look exactly as convincing as a correct one.