Backtest results do not match live trading

Causes, most likely first

1. The test did not model spread and commission realistically

Many tests use a fixed spread that never widens. Live spread expands at rollover, at session transitions and around news — exactly when signals cluster. On short-target gold strategies this alone can move a profitable test to a losing account.

2. Slippage was not modelled at all

Replay fills at the modelled price. Reality does not. Every entry and exit is slightly worse live, and the effect compounds across hundreds of trades. See slippage.

3. The test used a fixed lot and the account uses percentage risk

This one is rarely listed and it is large. At a fixed lot every trade carries equal weight, so the result measures whether the rules win more than they lose. Under percentage risk the weights differ and a larger-sized loser can outweigh several smaller winners. In our own data one month reads +$10.18 at a flat 0.01 lot and −$0.17 at $1,000 with real percentage sizing — same trades, same order, opposite sign. Check which sizing produced any figure you are comparing against, and never divide fixed-lot dollar totals by an account balance to get a percentage. See how the same strategy produced two different results.

4. The strategy was overfitted to the test period

If parameters were tuned until the backtest looked good, the result describes that period and nothing else. The check is an out-of-sample period the strategy has never seen — and it must only be used once.

5. Intrabar sequence was assumed favourably

On bars where both the stop and the target were touched, the tester must assume which came first. Optimistic assumptions inflate results substantially, and the effect is largest on tight stops. See tick data.

6. The live account took trades the test never allowed

We found this in our own results — several strategies firing in the same direction within minutes, which replay treated as one position and live treated as several. Three of ten trades in one month were affected, carrying roughly half the printed profit.

Why this happens

Some gap between test and live is unavoidable and expected. A gap large enough to flip the sign of the result means one of the assumptions above is doing the work, and it is worth identifying which rather than concluding the strategy is broken.

The most reliable structural defence is to test using the same code that trades. Our replay engine runs real broker bars through the live signal path with the real performance tracker driving selection, so there is no separate backtest implementation that could quietly diverge.

Being able to name the specific gaps is more valuable than a smaller gap. A vendor who can tell you exactly where their backtest overstates reality has measured it; one who reports no difference has not looked.