Why EA Results Differ Between Brokers

An Expert Advisor is just code: feed it the same prices, the same clock and the same account conditions and it will always make the same decisions. Run it with identical settings on live accounts at two brokers, though, and after a month the two histories disagree — different entries, different exits, sometimes a different trade count entirely. The code did not change; every input did.

This is a separate question from why live results lag the Strategy Tester, which the backtest-vs-live guide covers. Here both accounts are live, so every divergence must come from something the brokers supply differently: the server clock, the quotes, the costs and the symbol rules. Seven mechanisms do most of the work — and the least obvious one does the most.

Every candle in MetaTrader is cut by the broker’s server clock. A broker running GMT+3 starts each new H1, H4 and daily bar at a different instant than a broker running GMT+0 — the same stream of ticks, sliced at different points. Where those offsets come from is explained in the server time guide; what matters here is the consequence. Most EAs are bar-based: they evaluate once per new bar, read the previous bar’s open, high, low and close, and compute indicators from series of bar closes. Shift the boundaries by three hours and all of those values change — a 20-period average of H4 closes becomes an average of different numbers, and yesterday’s high can belong to a different day.

Time filters compound the effect. An input like “trade 09:00–17:00” runs in server time, so the same setting trades different UTC hours — and therefore different liquidity — at each broker. The two accounts can even drift in and out of alignment twice a year, as daylight-saving transitions move one server’s offset and not the other’s.

Spread differences obviously change costs, but for an EA they also change behaviour. Many EAs carry a maximum-spread filter — for instance, skip the entry if the spread exceeds 12 points. At a broker averaging 0.8 pips during the EA’s active hours, nearly every signal passes; at one averaging 1.3 pips and widening past 2 around rollover, a meaningful share of entries is silently skipped. The trade counts diverge before a single fill differs.

Stops levelsare stricter still. Each broker sets a minimum distance, in points, at which stop loss, take profit and pending orders may be placed or modified. With a stops level of 0, a 3-pip trailing stop tightens freely; with a stops level of 40 points, the same modification request is rejected every time it comes within 4 pips of the market. The EA’s exit logic effectively runs on one account and fails on the other — identical code, different broker rule.

Commission and swap are set per broker per symbol, and unlike a one-off misfire they apply to every trade and every held night. Individually trivial, they accumulate into a steady gap between two accounts trading the same signals.

Neither line decides anything on its own — the broker charging the higher commission may quote the tighter spread. Only the all-in cost per round trip, with every component converted to the same unit, is comparable across accounts.

How an order becomes a fill depends on the broker’s execution model — the dealing-desk and STP/ECN mechanics are mapped out in the broker execution guide. For an EA, the practical difference is the failure mode: market execution converts a moved price into slippage, while instant execution can answer with a requote that the EA must handle or lose the entry. Retry logic that gives up after two attempts produces missed trades on one account that simply never existed on the other.

Liquidity is also hour-specific. An EA active in the thin hour around rollover might average 0.2 pips of slippage at one broker and 0.6 at another, while the same comparison at the London open could lean the other way. Slippage is a property of broker, symbol and hour together — exactly the combination an EA’s schedule concentrates.

Each broker publishes its own specification per symbol, and an EA inherits every line of it:

Underneath the specification sits the feed itself. Each broker aggregates quotes from its own liquidity providers, so tick density and the exact micro-highs and lows differ. A buy stop at 1.27555 fills at a broker whose feed prints a high of 1.27562 and never triggers where the move tops out at 1.27549. Tick-level logic — tick counters, very tight breakouts — sees those differences amplified, because it consumes the feed at its rawest.

All numbers above are illustrative, and the direction could just as easily reverse — Broker B could quote the tighter spread while Broker A enforces the awkward stops level. The point is structural: with this many independent inputs, matching results would be the surprise. How visible the divergence is scales with the EA’s average profit per trade — a strategy netting 2 pips per trade feels every mechanism, one netting 30 shrugs most of them off.

A fair comparison controls everything the brokers do not, so that whatever remains is the genuine broker effect:

• Same window, same build: compare identical calendar periods with the same EA version and inputs, and note each server's GMT offset for the period.
• Keep a forward record per broker rather than extrapolating one account's results to the other.
• Give the EA a unique magic number on each account so its trades can be isolated from anything else running there.
• Compare trade lists before totals: entries present on one account and missing on the other point to candle boundaries, spread filters or rejections; matching entries with different outcomes point to costs and fills.
• Normalise costs into one figure — spread, commission, swap and slippage in pips per round trip — before judging anything else.
• Scale expectations to the edge: the smaller the average profit per trade, the larger the divergence that counts as normal.

Run that way, the comparison stops being an argument about which account is “right” and becomes attribution. Reviewing your own synced MetaTrader accounts side by side — the EA’s trades isolated by magic number on each — turns “it behaves differently over there” into a short list of measured causes. No broker ranking follows from that, and none is intended; what follows is an accurate picture of what your EA actually does in each place it runs.