High-Dimensional Macroeconomic Forecasting: A Partial-Correlation Based Panel Vector Autoregressive Model Estimation Method

The Vector Autoregressive Model (VAR) has been successful in forecasting macroeconomic time series. With the growing inter-dependencies between economies, econometricians have developed the Panel-VAR model to account for the international linkages between macroeconomic variables. However, the number of explanatory variables of a Panel-VAR model often greatly exceeds the number of observations, which inhibits the use of ordinary estimation methods. Utilized the persistent-pattern robust inference construction method introduced by [Müller and Watson (2018)], this paper developed an original partial-correlation based estimation method for Panel-VAR model. Together with other existing methods, the method is used to forecast European economic variables. Results of out-of-sample forecasting show that the estimation method provides a parsimonious yet competitive estimation method for PVAR models.