Forecasting inflation with dynamic factor model – the case of Poland

Abstract

The purpose of the article is to evaluate the forecasting performance of dynamic factor models in forecasting inflation in the Polish economy. The factor models are based on the assumption that the behavior of most macroeconomic variables can be well described by several unobservable factors, which are often interpreted as the driving factors in the economy. Such models are very often successfully used for forecasting. Employing several factors instead of a large number of explanatory variables may increase the number of degrees of freedom with the same information content. In the article we compare forecast accuracy of dynamic factor models with the forecast accuracy of three competitive models: univariate autoregressive model, VAR model and the model with leading indicator from the business survey. We have used 92 monthly time series from the Polish and world economy to conduct the out-of-sample real time forecasts of inflation (consumer price index). The results are encouraging. The dynamic factor model outperforms other models for both 1-step ahead and 3-step ahead forecast. The advantage of factor models is more straightforward for 1-month than for 3-month horizon.