On the Applicability of Dynamic Factor Models for Forecasting Real GDP Growth in Armenia

In this paper, we are trying to find out whether large-scale factor-augmented models can be successfully employed for forecasting real GDP growth rate in Armenia. We use Armenian data because as a developing country Armenia has experienced a relatively higher volatility of GDP growth rate in comparison to other countries. Based on our calculation using growth rate data from 40 countries, we argue that low-income countries have about 57% higher volatility of growth rates than high-income countries. Taking this into account, it is worth testing the forecasting performance of factor models on a country like Armenia to check the applicability of the advanced forecasting methods to economies with highly volatile growth rates. For this, we compare the forecasting performance of factor-augmented models such as FAAR, FAVAR and Bayesian FAVAR with their small-scale benchmark counterpart models like AR, VAR, Bayesian VAR and mixed-frequency VAR. Based on the ex-post out-of-sample recursive and rolling forecast evaluations and using RMSFE’s, we conclude that large-scale factor-augmented models outperform small-scale benchmark models when we apply these methods to forecasting real GDP growth. However, the differences in forecasts among the models are not statistically significant when we apply statistical test.