Forecasting a Nonstationary Time Series Using a Mixture of Stationary and Nonstationary Factors as Predictors

We develop a method for constructing prediction intervals nonstationary variable, such as GDP. The uses Factor Augmented Regression (FAR) model. predictors in the model include small number of factors generated to extract most information set panel data on large macroeconomic variables that are considered be potential predictors. novelty this article is it provides and justification mixture stationary FAR model; we refer mixture-FAR method. This important because typically data, example FRED-QD, likely contain variables. In our simulation study, observed proposed performed better than its competitor requires all nonstationary; MSE was at least 33% lower mixture-FAR. Using FRED-QD United States, evaluated aforementioned methods forecasting variables, GDP Industrial Production. competitors.