Forecast-versus-Actual Plot Calculator

Data and calculator options

Use long-format data. Each row represents one model, series, and period.

Forecast data

Recommended columns: date, actual, forecast, lower, upper, model, series, horizon.

Upload CSV or TSV Delimiter

Column mapping

Date or period column

Actual column

Forecast column

Model column

Series column

Lower interval column

Upper interval column

Horizon column

Processing and validation

Missing-value method

Duplicate-period method

Value transformation

Moving-average window

Decimal places

Start period filter

End period filter

Series filter

Accuracy and tolerance

Model ranking metric

Absolute tolerance

Percentage tolerance

%

Interval label

Plot options

Plot style

Line width

Marker size

Chart height

Chart title

X-axis title

Y-axis title

Legend position

Y-axis scale

Y-axis minimum

Y-axis maximum

Date-label rotation

Show gridlines

Show interval bands

Show value labels

Highlight tolerance failures

Reverse Y-axis

Formula used

Error = Forecast − Actual

MAE = Σ|Forecast − Actual| ÷ n

RMSE = √[Σ(Forecast − Actual)² ÷ n]

sMAPE = 100 × mean(2|F − A| ÷ (|A| + |F|))

Bias = mean(Forecast − Actual)

Coverage = observations inside interval ÷ eligible observations × 100

How to use this calculator

Paste long-format data or upload a CSV file.
Confirm the column names in the mapping section.
Select processing, tolerance, metric, and plot options.
Press calculate to generate plots and accuracy tables.
Export the chart, results, cleaned data, or report.

Example data structure

date	actual	forecast	lower	upper	model	series	horizon
2026-01	120	118	108	128	Model A	Sales	1
2026-02	132	136	124	148	Model A	Sales	2
2026-01	120	122	111	133	Model B	Sales	1

Frequently asked questions

What does forecast versus actual mean?

It compares predicted values against observed outcomes. The plot reveals timing and size differences. Smaller gaps usually indicate stronger forecasting performance.

Which metric should rank models?

RMSE emphasizes large forecasting errors. MAE gives every absolute error equal weight. MASE supports comparisons across differently scaled series.

Why can MAPE become unreliable?

MAPE divides errors by actual values. Zero values make the calculation undefined. Near-zero values can create misleading percentages.

What does positive bias indicate?

Positive bias means forecasts are generally high. Negative bias means forecasts are generally low. Values near zero indicate balanced errors.

How are interval coverage results calculated?

Coverage counts actual values inside prediction bounds. It excludes rows without valid bounds. Higher coverage is not always narrower.

Can several models be compared?

Yes, use one row per model and period. Keep actual values consistent across models. Rankings summarize each model independently.

What is a tracking signal?

Tracking signal compares cumulative error with average absolute error. Large magnitudes can indicate persistent bias. Context determines a suitable alert threshold.

Can the calculator handle missing values?

Yes, several missing-value strategies are included. Interpolation works best for ordered continuous series. Dropping rows is safest for uncertain gaps.

Why inspect residual plots?

Residual patterns can reveal model weaknesses. Trends suggest missed structure or distribution shifts. Random scatter supports better model adequacy.