Nowcasting and forecasting Russian GDP and its components using quantile models

The paper examines the quality of probabilistic nowcasts and short-term forecasts of the Russian GDP and its components in constant prices (consumption, investment, exports and imports) based on the standard quantile regression model and its shrinkage modifications, aimed at reducing the risk of overfitting (averages of quantile forecasts, partial quantile regression, regressions with regularization, Bayesian quantile regression). We find that quantile models with predictors are superior to autoregressive and OLS models in terms of CRPS (Continuous Ranked Probability Score) metrics in nowcasting exercises for investment and consumption. When forecasting 1–4 quarters ahead, shrinkage models yield the most accurate forecasts of GDP and consumption distributions at all horizons. For investment and imports, shrinkage methods turn out to be the best performers at three forecast horizons out of four. There is no single shrinkage model, which would provide the best probabilistic forecasts of macroeconomic variables much more often than others.