Abstract

This paper examines the quality of nowcasts and forecasts for Russian GDP and its components (in constant and current prices) using a mixed-frequency Bayesian vector autoregression model (MFBVAR) which is currently one of the most advanced time series forecasting models. It enables use of quarterly and monthly frequency data within a single monthly frequency VAR model in a statespace form while taking into account the intra-quarter dynamics of monthly indicators; this approach improves forecasting accuracy when new monthly data is published. The MFBVAR model’s resistance to the jagged edge problem is especially important for real-time forecasting, and it can incorporate a large number of predictors because of its Bayesian estimation with a Minnesota-type prior distribution. The paper sets up three experiments with differing availability of monthly data in order to test pseudo out-of-sample nowcasting and forecasting. The MFBVAR model exhibits statistically significant outperformance compared to a naïve benchmark, as well as to ARIMA and quarterly BVAR models, in nowcasting and forecasting a few steps ahead for GDP, consumption and foreign trade variables. The test sample is also quite representative and covers two crisis periods, specifically 2015 and 2020. In both crises, the model accurately estimates the scale of the recession and recovery of economic activity. Nevertheless, there was no significant improvement in the quality of forecasts when new available monthly data was introduced.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call