Generating short-term forecasts of the Lithuanian GDP using factor models

This paper focuses on short-term Lithuanian GDP forecasting using a large monthly dataset. The forecasting accuracy of various factor model specifications is assessed using the out-of-sample forecasting exercise. It is argued that factor extraction by using a simple principal components method might lead to a loss of important information related GDP forecasting, therefore, other methods should be also considered. Performance of several factor models, which relate the factor extraction step to GDP forecasting, was tested. The effect of using weighted principal components model, with weights depending on variables’ absolute correlation with GDP, was explored in greater detail. Although factor models performed better than naive benchmark forecast for GDP nowcasting and 1 quarter ahead forecasting, we were unable to set up the ranking among different factor model specifications. We also find that a small scale factor model with 5 variables (which could be regarded as the most important monthly variables for GDP nowcasting) is able to nowcast GDP better than models with a full data set of 52 variables, which might indicate that for the case of the Lithuanian economy, a smaller scale factor models may be more suitable.