Working paper: Suite of Latvia's GDP forecasting models
Abstract: We develop and assess a suite of statistical models for forecasting Latvia's GDP. Various univariate and multivariate econometric techniques are employed to obtain short-term GDP projections and to assess the performance of the models. We also compile information contained in the GDP components and obtain short-term GDP projections from a disaggregate perspective. We propose a novel approach assessing GDP from the production side in real time, which is subject to changes in NACE classification. Forecast accuracy of all individual statistical models is assessed recursively by out-of-sample forecasting procedure. We conclude that factor-based forecasts tend to dominate in the suite. Encouraging results are also obtained using disaggregate models of factor and bridge models, which could be considered as good alternatives to aggregate ones. Furthermore, combinations of the forecasts of the statistical models allow obtaining robust and accurate forecasts which lead to a reduction of forecast errors.
Keywords: out-of-sample forecasting, real-time estimation, forecast combination, disaggregate approach
JEL codes: C32, C51, C53