After the best forecasting model is selected from the candidate models, the selected model is fit to the full range of the data to obtain the most accurate model parameter estimates. If you excluded the holdout samplethe number of periods of the most recent data that should be excluded from the parameter estimation. The holdout sample can be used to evaluate the forecasting performance of a candidate model. in this step, you would be ignoring the most recent and influential observations. Most univariate forecasting models are weighted averages of the past data, with the most recent having the greatest weight. After the model is selected, excluding the holdout sample can result in poor forecasts. Holdout sample analysis is used only for forecasta numerical prediction of a future value for a specified time period for each unique combination of BY variable values model selection, not for forecasting.