Suite of statistical models forecasting Latvian GDP

Abstract

We develop a suite of statistical models to forecast Latvian GDP. We employ various univariate and multivariate econometric techniques to obtain short-term GDP projections and to assess the performance of the models. We also comprise the information contained in components of GDP and obtain short-term GDP projections from disaggregated perspective. We run out-of-sample forecasting procedures to evaluate GDP projections and to assess forecasting accuracy of all individual statistical models. We conclude that factor and bridge models are among the best individually performing models in the suite. Forecasting accuracy obtained using disaggregated models of factor and bridge models is noteworthy and might be considered as a good alternative to aggregated ones. Furthermore, weighted combination of the forecasts of the statistical models allows obtaining robust and accurate forecasts which leads to a reduction of forecasted errors.