A 99 percent modelling quality number does not mean the data is good
Posted: Sat Oct 31, 2026 6:37 am
MetaTrader reports modelling quality after a test, and traders see 99 percent and relax. The figure only says that tick-level modelling was applied, not that the source data is accurate.
If the underlying history is generated from minute bars, the ticks are artificial interpolations. They fill the gaps with plausible but invented paths. A score of 99 percent just means the generator ran across the whole period.
Real tick data from a reputable source differs substantially. It includes spread changes, gaps, and the ugly moments that interpolated data smooths away. Results often look worse, and that is the point.
Checking the data source before trusting a test is basic hygiene. Comparing a short test period with real tick data and looking at the trade-by-trade differences shows how much the generated data flatters the strategy.
Spread handling matters just as much. Many tests use a fixed spread, which ignores the expansion at rollover and news.
Which tick data source is used for backtests, and how did results compare with generated data?
If the underlying history is generated from minute bars, the ticks are artificial interpolations. They fill the gaps with plausible but invented paths. A score of 99 percent just means the generator ran across the whole period.
Real tick data from a reputable source differs substantially. It includes spread changes, gaps, and the ugly moments that interpolated data smooths away. Results often look worse, and that is the point.
Checking the data source before trusting a test is basic hygiene. Comparing a short test period with real tick data and looking at the trade-by-trade differences shows how much the generated data flatters the strategy.
Spread handling matters just as much. Many tests use a fixed spread, which ignores the expansion at rollover and news.
Which tick data source is used for backtests, and how did results compare with generated data?