Why 99.9 percent modelling quality can still be fiction

[Amanhasnoname]

The MT4 strategy tester shows a modelling quality figure. Seeing 99.9 percent reassures a lot of people. It shouldn't.

That number describes how much of the test period was built from real tick data versus interpolated from M1 bars. A high value means the data is detailed. It doesn't mean the data matches what a live broker would have delivered.

The ticks may be generated from a third-party source, with different spreads, different gaps and no slippage. The test often assumes a fixed spread, instant fills and no rejections.

Add the fact that history data differs between brokers, and the simulation is only an approximation of one possible reality.

Better practice is to treat the result as an upper bound. Then stress it: add extra spread, add random slippage, delay fills, and see whether the edge survives.

If a strategy collapses when costs rise by a fraction of a pip, it won't survive live trading.

Another trap is that history data from the broker often contains small holes filled with synthetic ticks. A test that passes through those holes can generate trades at prices nobody ever saw, and the modelling quality number never mentions it.

What stress tests do you run before trusting a backtest?