Forecast accuracy calculator
Compare actual values with forecasts using MAE, RMSE, signed bias, MAPE and WAPE. Inspect every error and download the full report.
Paste CSV with a header and 1 to 5000 actual and forecast pairs. Choose two distinct column names. Use dot decimals, no thousands separators, and at most six decimal places; values may be negative and must be within ±1000000000. Error is forecast minus actual, so positive bias means overforecasting. MAPE excludes zero actuals and reports its included and excluded counts. WAPE divides total absolute error by total absolute actual values. An undefined percentage is shown as Not defined. Percentage errors may be misleading for signed quantities or actuals near zero. Rows are equally weighted; this is not a forecasting model or a universal accuracy score. Complete JSON and CSV include every row, summary metric and assumption. Processing stays on your device. CSV accepts at most 100 columns and 100000 cells through the shared parser. Extra columns are ignored and not copied into the report. Calculations use floating-point arithmetic. A disposable local worker has a 10-second timeout. Up to 1,000,000 characters. Review the settings before running.
Your result
How to use Forecast accuracy calculator
- Paste CSV with actual and forecast columns. Match each pair to the same period or item.
- Select the CSV separator and the exact actual and forecast column names.
- Compare error metrics, inspect the row table and download the complete JSON or CSV.
Useful next steps
Measure error without hiding its definition
For each row, error equals forecast minus actual. MAE averages absolute errors. RMSE is the square root of the average squared error. Mean signed bias averages signed errors, so positive values indicate overforecasting. Errors use the same units as the supplied values.
Handle zero actuals explicitly
MAPE averages 100 times absolute error divided by absolute actual over nonzero actual rows only. Both the included and excluded counts are shown. WAPE is 100 times total absolute error divided by total absolute actual. A zero denominator gives an undefined percentage, not zero accuracy. Percentages can exceed 100.
Compare equivalent observations
Use the same periods, forecast horizons and units for each pair. The tool does not verify temporal alignment or create forecasts. Small actual values can dominate MAPE, and percentage interpretations may be inappropriate for signed quantities. Evaluate held-out forecasts and compare suitable baselines before judging a model.