Прогнозирование основных российских макроэкономических показателей с помощью TVP-модели с байесовским сжатием параметров

The paper examines the quality of forecasts of Russian GDP and its components (household consumption, investment, exports and imports) using a model with Bayesian shrinkage of time-varying parameters (TVP) based on hierarchical normal-gamma prior. Such models account for the possible nonlinearity of relationships and, at the same time, can deal with the overfitting problem. We find that, compared to simpler benchmarks, the Bayesian TVP model with exogenous predictors gives better forecasts for GDP at horizons of 2-4 quarters, and for investment – at horizons of 1-3 quarters. When predicting other components of GDP, Bayesian TVP models do not demonstrate systematic superiority over other models.