Forecasting Performance of Hybrid ARIMA-ANN and VAR-ANN Models on Macroeconomic Data

The main aim of present paper is comparing the forecasting performance of ARIMA, VAR, Artificial Neural Network and Hybrid models on macroeconomic data. The main idea of Hybrid models consists in combination of classic econometrics models with Artificial Neural Network, which are very popular in scientific literature un recent times. It is assumed that hybrid models can make better forecasts than classic ARIMA and VAR themselves. Different Russian and US macroeconomic series were investigated. Twelve of them were used for testing ARIMA-ANN model and its components. Also short term interest rate, inflation rate, unemployment rate and GDP of both countries were used for testing VAR-ANN model and its components. Attained results allow to state that hybrid models can provide significant contribution to the field of macroeconomic data forecasting.