Forecast Error Results
Results appear here after analysis.
| Horizon | Model | Count | Mean Error | MAE | MSE | RMSE | MAPE | sMAPE | Bias | Std. Dev. | CI | Status |
|---|
Formula Used
Error: e(h) = Actual(h) − Forecast(h) MAE(h) = Σ|eᵢ(h)| / n(h) MSE(h) = Σeᵢ(h)² / n(h) RMSE(h) = √MSE(h) MAPE(h) = [Σ|eᵢ(h) / Actualᵢ(h)| / n(h)] × 100 sMAPE(h) = [Σ 2|eᵢ(h)| / (|Actualᵢ(h)| + |Forecastᵢ(h)|) / n(h)] × 100 Bias(h) = Σeᵢ(h) / n(h)
How to Use
Enter forecast horizons, actual values, and model predictions. Choose an error metric and grouping options. Then run the analysis to compare performance.
Use filters to isolate models, groups, or origins. Configure confidence bands and outlier handling carefully. Export the table or chart for reporting.
- Load the sample data or paste a CSV dataset.
- Select the main metric shown on the plot.
- Choose confidence, smoothing, threshold, and filtering options.
- Click Analyze Forecast Errors.
- Review summaries, diagnostics, charts, and horizon tables.
Example Data
| Horizon | Actual | Naive | Model A | Model B |
|---|---|---|---|---|
| 1 | 120 | 116 | 118 | 121 |
| 2 | 125 | 119 | 122 | 124 |
| 3 | 130 | 122 | 126 | 128 |
| 4 | 135 | 126 | 129 | 133 |
| 5 | 140 | 129 | 132 | 137 |
Interpretation Guide
A rising curve means forecasts become less accurate farther ahead. A flat curve indicates stable performance across horizons. Large bands show uncertain or inconsistent forecast errors.
Positive mean error indicates forecasts are generally too low. Negative mean error indicates forecasts are generally too high. Compare models using identical horizons and observations.
Frequently Asked Questions
What is an error-by-horizon plot?
It shows forecast error for each prediction distance. The horizontal axis contains forecast horizons. The vertical axis contains a selected error metric.
Why does forecast error often increase?
Longer forecasts depend on more uncertain future conditions. Small assumptions can compound over multiple steps. Wider horizons therefore often produce larger errors.
Which error metric should I use?
MAE is easy to interpret in original units. RMSE penalizes large mistakes more strongly. Percentage metrics help compare series with different scales.
Can multiple models be compared?
Yes, add one column for each model. Every model appears as a separate plot series. The table also reports horizon-level performance.
What do confidence bands show?
They summarize uncertainty around the aggregated error estimate. Narrow bands indicate more stable observed errors. Wide bands indicate variability or limited observations.
How are zero actual values handled?
MAPE excludes observations with zero actual values. sMAPE remains available in many zero-value situations. Warnings identify any excluded percentage calculations.
Should outliers be removed?
Remove outliers only with a defensible reason. Extreme errors may reveal genuine model weaknesses. Compare filtered and unfiltered results before deciding.
What is forecast bias?
Bias is the average signed forecast error. Positive bias means forecasts are generally too low. Negative bias means forecasts are generally too high.
Can the chart be exported?
Yes, download PNG or SVG chart files. Export the results table as CSV. Print the page to create a PDF report.