The 99.9 percent modelling quality badge
Posted: Sat Oct 31, 2026 9:09 pm
The 99.9 percent modelling quality badge
MetaTrader's strategy tester reports a modelling quality figure. Seeing 99.9 percent feels like a stamp of approval. It is nothing of the sort.
The figure describes how the testing engine built ticks from the underlying minute data. It says nothing about whether the underlying data is accurate, whether spreads were realistic, or whether the fills match real execution.
A strategy can show 99.9 percent quality and still be tested with a fixed spread, instant fills, and no slippage. Those three assumptions alone can turn a losing system into a winning backtest.
For scalping, realism matters more than any badge. That means variable spreads taken from real history, commission included, a reasonable slippage model, and a delay between signal and order.
Honest testing often shows that an apparently profitable scalper has a very small edge, or none, once costs are added properly.
One more point: the tester's modelling of ticks inside a bar is a simplification, not a recording. Two runs with identical settings can give identical results and still both be wrong about what happened inside each bar.
Some traders respond by testing on real ticks from a data vendor, and others by running a forward test on a small live account. Both cost effort, but they replace a comforting number with real evidence, and the evidence tends to be less flattering and far more useful.
What assumptions about spread and slippage have been used in backtests, and how did the results change when they were tightened?
MetaTrader's strategy tester reports a modelling quality figure. Seeing 99.9 percent feels like a stamp of approval. It is nothing of the sort.
The figure describes how the testing engine built ticks from the underlying minute data. It says nothing about whether the underlying data is accurate, whether spreads were realistic, or whether the fills match real execution.
A strategy can show 99.9 percent quality and still be tested with a fixed spread, instant fills, and no slippage. Those three assumptions alone can turn a losing system into a winning backtest.
For scalping, realism matters more than any badge. That means variable spreads taken from real history, commission included, a reasonable slippage model, and a delay between signal and order.
Honest testing often shows that an apparently profitable scalper has a very small edge, or none, once costs are added properly.
One more point: the tester's modelling of ticks inside a bar is a simplification, not a recording. Two runs with identical settings can give identical results and still both be wrong about what happened inside each bar.
Some traders respond by testing on real ticks from a data vendor, and others by running a forward test on a small live account. Both cost effort, but they replace a comforting number with real evidence, and the evidence tends to be less flattering and far more useful.
What assumptions about spread and slippage have been used in backtests, and how did the results change when they were tightened?