The financing charge or credit applied to a position held past the broker's daily rollover, derived from the interest rate differential between the two currencies plus the broker's own markup.
also: rollover, overnight fee, financing cost, swap rate, carry
Updated · Reviewed
Holding a currency pair overnight means borrowing one currency and holding another, so interest changes hands. The broker settles that difference once a day, adds its own margin to it, and the result lands on the position as a credit or a charge.
Swap is the only trading cost that grows with time rather than with activity. A strategy that trades rarely and holds for days can pay more in financing than in spread, and it is the cost most often missing from the backtest that justified it.
The Windrose record published here, an AUD/CAD reverter that regularly holds for days to weeks. That record measures its financing rather than assuming it, which makes it the clearest illustration of what swap does to a slow strategy.
Nothing here depends on the strategy being right or wrong. It is the rent on holding the positions at all, and on this record the rent outran the drawdown.
Calculation 521 ÷ 1,445.84 ≈ 36% · 148 − 91 = 57 of swap on one position
Result Financing was 36% of the net it left behind, and larger than the record's worst drawdown
Swap is read against holding period, not against balance. Compute financing per trade first, then read it twice: as a share of the average win, which is the band below, and against the record's maximum drawdown, which is what tells you whether the cost is larger than the risk everyone already looks at.
| Range | What it means |
|---|---|
| Under 2% of the average win | Immaterial. Ignore it in strategy selection and keep it in the cost model. |
| 2–10% | A real cost worth checking against a second broker's table before committing. |
| 10–25% | Financing is now a design constraint. Holding period and broker choice both need to be justified. |
| Over 25%, or positive carry as the stated edge | The strategy is trading the financing rather than the market, and it lives or dies on a rate the broker sets. |
Fifteen of the 23 records published here state a swap total in currency; the remaining eight say only that the run booked swap, and leave the per-deal values in the downloadable ledger. Reading which of the two a listing does is the fastest way to see how closely a listing examined its own financing.
MT5 applies the swap values from the current symbol specification across the whole test period. A ten-year backtest therefore charges ten years of nights at the rate that happened to be in force the day it was run, which understates or overstates financing depending on where rates were.
Swap-free accounts replace financing with an administration fee, usually flat per lot per night after a grace period. For a strategy holding a week at a time this is frequently more expensive than the swap it replaced. The comparison is total cost per trade, not the presence of the word free.
One weekday settles three nights to cover the weekend. Five rollovers in a trading week therefore bill seven nights, so a position held across whole weeks accrues 40% more financing than a naive nights-held count suggests, and the excess lands on a fixed weekday that shows up as a step in the equity curve.
A stop bounds what the market can take. It does not bound what the calendar can take. One Windrose position held for three and a half months closed at −$148 against a nominal −$91 stop, because financing kept accruing under a position the stop had already priced. On any strategy that can hold for weeks, the real worst case per trade is the stop plus the swap the hold accumulates.
Four of the 23 records published here booked exactly zero financing, and in each case that is a property of the tested symbol rather than of the strategy. Stillwater's 833 deals all recorded zero swap because it only ever buys, and the USD/JPY history behind that run carries no long-side swap. Meanwhile 286 of its 416 trades stayed open across at least one night. A live account at ordinary rates pays every one of those nights.
Positive swap is set by an interest differential the broker marks up and revises. A strategy whose profitability depends on it is exposed to a rate decision and a broker policy at the same time, neither of which appears anywhere in its backtest.
| Across the 23 EAs published here | Figure |
|---|---|
| Records stating a swap total in currency | 15 |
| Of those, records whose total is exactly zero | 4 |
| Records that book swap without totalling it | 8 |
| Largest financing bill on one record | −$938.13 |
| Smallest non-zero bill | −$9.37 |
| Combined financing across the eleven non-zero records | about −$4,307 |
| Run manifests naming a per-night swap rate | 1 of 23 |
The spread between the top and bottom row of that range is the whole point. Ballast made $4,061.59 gross and paid $938.13 of it to financing, publishing $3,123.46 — roughly a quarter of the strategy’s output went to holding the positions. Nautical, on a custom symbol configured with near-zero EUR/JPY swap, paid $9.37 across 1,665 deal rows for holds with a three-day median. Same cost, same platform, two orders of magnitude apart, and neither number is knowable from the strategy logic.
The four zero rows are the trap. Stillwater recorded a swap of exactly zero on every one of its 833 deals, because it only ever buys and the USD/JPY history we tested on carries no long-side swap. Its positions still sat out the nights: 286 of its 416 trades stayed open across at least one, with a median hold near 51 hours and a longest of 44 days. A zero in that column describes the test symbol, not the strategy. Mistake it for a strategy property and you quietly delete a cost a live account will charge on every one of those nights.
The last row is the one that undermines reproducibility. Only one of the 23 manifests records a per-night rate at all; the rest defer to the per-deal values in the trade ledger, and for good reason. Re-running the Lattice Weave build on the same dates returned identical entries, identical exits and the same prices to the last digit. Only the total moved, and it moved by financing alone, because the broker had revised its stored swap history in the interval. Every other cost in a backtest follows from the data; this one follows from when you pressed start.