Missing bars: the holes in the history that tests treat as nothing
Posted: Sun Nov 01, 2026 12:13 am
Downloaded history for one-minute data often has gaps. The platform simply skips minutes with no ticks. For a major pair in busy hours, there are none. For thin pairs or off hours, there are many.
A backtest treats the missing minute as if time jumped. Indicators computed over "the last 14 bars" now span a longer period than intended. A moving average covers more wall-clock time. Volatility measures shrink because no movement is recorded in the empty minutes.
Other holes appear around weekends and holidays, where the chart connects Friday's close to Monday's open without any visible indication of the pause.
More serious are data errors: a missing hour from a feed outage, a duplicated block, a bad import. These can hide inside long histories and distort results without any warning.
A simple check counts bars per day and flags days that fall short of expectations. Plotting the count over time shows where data quality varies.
Cleaning the data, or at least knowing where it's poor, avoids drawing conclusions from artifacts.
How was data quality verified before the last backtest, and did anything odd turn up?
A backtest treats the missing minute as if time jumped. Indicators computed over "the last 14 bars" now span a longer period than intended. A moving average covers more wall-clock time. Volatility measures shrink because no movement is recorded in the empty minutes.
Other holes appear around weekends and holidays, where the chart connects Friday's close to Monday's open without any visible indication of the pause.
More serious are data errors: a missing hour from a feed outage, a duplicated block, a bad import. These can hide inside long histories and distort results without any warning.
A simple check counts bars per day and flags days that fall short of expectations. Plotting the count over time shows where data quality varies.
Cleaning the data, or at least knowing where it's poor, avoids drawing conclusions from artifacts.
How was data quality verified before the last backtest, and did anything odd turn up?