A VAR model for short-term forecasting of the Finnish macroeconomy

Abstract

The study examined short-term forecasting of the Finnish macroeconomy using unrestricted vector autoregressive models (VAR models). The aim is to make use of the constructed VAR models in the research institute’s short-term forecasting practice in the autumn and spring. The study was prefaced by an examination of the history of economic forecasting in Finland and of the sources of uncertainty associated with forecasting. In addition, the key features of different types of forecasting models were examined, and the structure and properties of VAR and VARX models were explored in depth. Quarterly data from 1990 onwards were collected for the estimation of the VAR models.

The models were subject to three selection criteria: the accuracy of out-of-sample forecasts as measured by the root mean squared error (RMSE) or the sum of squared forecast errors of annual growth rates; the macroeconomic consistency of the models as assessed using impulse responses and Granger causality tests; and the results of sensitivity analysis. The models selected on the basis of these criteria shared the characteristics of a small number of lags and estimation beginning from the earliest possible period. All selected models contained the following three variables: Finland’s adjusted GDP, the European Union’s GDP, and the ratio of nominal compensation per employee in Finland and in the EU15. In addition, at most two of the following variables were included: private investment and labour costs per employee. Transforming the models from VAR models to VARX models by setting the ratio of nominal compensation per employee in Finland and the EU15 as exogenous further improved the forecasting accuracy of the models. Forecasts for 2017 and 2018 were produced using the estimated models. Forecasts were made starting from the periods 2017Q1, 2017Q2, and 2017Q4. In forecasts starting from 2017Q4, the average estimate for the growth rate of Finland’s gross domestic product is 3.1 percent in 2017 and 1.8 percent in 2018. (AI translation)

• ISSN: 1795-1801 (Online)
• ISBN: 978-952-209-178-9 (Online)
• Publication in PDF-format