Expert Modeler (automatic forecasting model)
Tries exponential smoothing models (simple, Holt, Brown, damped trend, simple seasonal, Winters additive and multiplicative) and ARIMA models, and picks the one with the best normalized BIC (or BIC / AIC).
Menu: Analyze → Forecasting → Create Traditional Models (Expert Modeler)…
When should I use it?
- You need forecasts but are not sure which model fits the series.
- As a starting point before refining an ARIMA model by hand.
What are the assumptions?
- Equally spaced observations without gaps.
- The periodicity is set correctly (12 for monthly, 4 for quarterly, 1 for none).
How do I run it in StatLab?
- Choose the series and its periodicity.
- Keep All models, or restrict to exponential smoothing or ARIMA.
- Set the forecast horizon; tick Save to add the predictions.
How do I read the output?
| Output | Meaning |
|---|---|
| Model Description | The chosen model. |
| Model Fit / Model Statistics | Stationary R², R², RMSE, MAPE, MAE, normalized BIC and the Ljung-Box test of the residuals. |
| Models Considered | Every candidate with its criterion; ARIMA fits that sit at the invertibility boundary are listed but not eligible. |
| Forecast | Predictions (with limits for ARIMA models). |
Worked example
On the practice survey dataset (File → Sample dataset), this command — which the dialog's Paste button also produces — runs the procedure:
PROC tsExpert /v=score /period=12 /types=all /criterion=nbic /search=stepwise /maxP=2 /maxQ=2 /forecast=12 /save=NO.
Output tables: Model Description · Model Fit · Model Statistics · ARIMA Model Parameters · Models Considered · Forecast.
How would I report it in APA style?
The Expert Modeler selected ARIMA(0,0,0)(0,0,0) for Wellbeing score (normalized BIC = 4.177; stationary R² = .00, RMSE = 7.94), compared with 16 other candidate models.
Computed by StatLab's validated procedure code when this page was built. The example only illustrates the steps; choose variables that fit your own research question.
What are common mistakes?
- Treating the selected model as 'true' — check the residual Ljung-Box test and the plot.
- Using the wrong periodicity, which hides or invents seasonality.