also: order execution model, broker execution type, dealing desk vs no dealing desk
An execution model is how a broker processes an order — by taking the other side of it as counterparty, or by routing it outward to external liquidity — and it decides who pays whom, how, and which EAs the account will tolerate.
When you press buy, either the broker sells to you itself, or it passes the order to someone else and keeps a fee for the trouble. That one choice decides how your costs appear, who gains when you lose, and which trading behaviour the broker will put up with.
Spread, commission, requotes, scalping bans, execution quality — every complaint a trader has about a broker traces back to where the order goes. Read the model first and the complaints become one explanation.
One standard lot of EUR/USD on the commission-free tier of each of six named brokers, and what the tester would have filled the same order at. The figures are the brokers' typical published values, not our measurements.
Source: the executionModel, baseSpreadEurUsd and commissionPerLot fields recorded for these six brokers. A $1-per-lot commission equals 0.1 pip on a standard lot of EUR/USD, so a stated commission converts into the same unit as the spread.
Calculation $3.5 per lot ÷ $10 per pip = 0.35 pip · $7 per lot ÷ $10 per pip = 0.7 pip (one standard lot of EUR/USD, per charge)
Result 0.7 to 1.6 pips across six commission-free tiers — one backtest, six different costs
Identify the model from how the broker gets paid and how orders fill, then match it to what the strategy needs rather than to whichever label sounds most advanced.
| Range | What it means |
|---|---|
| Raw quote plus stated commission, Market execution | Costs are explicit and the server never requotes. Suits frequent trading and small targets. |
| Variable spread with markup, Market execution | Routed outward, cost folded into the quote. Fine for targets measured in tens of pips. |
| Fixed spread, Instant execution | Counterparty model. Workable for slow strategies; read the holding-time terms before deploying. |
| Any model with a documented minimum holding time | A scalping EA cannot run here however good its record. |
Execution model is worth recording as a field for each broker you compare, so accounts line up on the same row. How the figures here were gathered is on the methodology page.
A counterparty broker under a serious regulator, with negative balance protection and published terms, is a legitimate arrangement and often the cheapest way to trade small size on slow strategies. The conflict of interest is real and is a reason to read the terms, not a reason to assume misconduct. All six brokers recorded here, the market maker included, list negative balance protection.
The Strategy Tester fills every order from the price series at the spread the history carries. Two accounts with completely different routing produce identical tester reports, which is why the record on the account itself is the only evidence about execution that exists.
Most retail brokers run different models on different account types, and often on different symbol groups within one account. The question is never what the broker is but what the specific account and symbol are.
Minimum holding times, scalping restrictions and the right to adjust trades executed on off-market prices live in the contract, not in the platform. An EA restricted by contract runs perfectly and then has its results reversed.
Deviation belongs to Instant execution, where it caps how far the price may move before the server requotes. On Market execution the server ignores the field, so an EA that relies on it has no slippage protection at all; a spread guard before the order and a check of POSITION_PRICE_OPEN after it are what remain.
Each listing names the history its record came from and, where the build has one, the spread guard it ships. Those fields are the whole execution footprint a tester report leaves behind.
| 33 published EA records, 6 brokers compared | Figure |
|---|---|
| Records naming Exness MT5 M1 history as the source | 30 of 33 |
| Records run on real ticks (Model 4) | 22 of 33 |
| Records run on ticks generated from M1 bars (Model 0) | 10 of 33 (1 record states no model) |
| Builds that ship a spread guard | 14 of 33, set at 20–60 points (40 on 6 of them) |
| Records carrying a tick-recorded average spread | 1 of 33 (19.14 points, GBPJPYm, in the tester) |
| Brokers by stated model | ECN 2 · STP 3 · market maker 1 |
| Commission-free tier EUR/USD spread across those brokers | 0.7 to 1.6 pips at $0 commission |
| Live spreads we have measured across brokers | 0 rows |
Thirty records share one venue’s history, so the costs inside them are one
broker’s costs; comparing the same lot across the six brokers above shows a
commission-free tier costing more than twice as much before any routing
question arises. The
spread guard is the only line that survives the move from tester to live
unchanged: it reads SymbolInfoInteger(_Symbol, SYMBOL_SPREAD) and skips the
signal when the live spread is wider than the fixed MaxSpread input the build
ships, so that input should be set from the account you fund, not from the backtest.
Read the cost the way this comparison does — the spread on the standard tier, the commission on the raw tier, and the ECN label only after both. The gap between the record and the live account is slippage plus whatever the model adds; a scalping build feels all of it, a daily-timeframe expert advisor almost none. The six accounts recorded here line up on the same fields.
