Nowcasting GRP in the Russian Economy Using Quantile Econometric Models

The paper aims to develop a model for probabilistic nowcasting of the Russian regions’ GRP growth rate, as the official statistiÑ s on this indicator are published with a long delay. Taking into account uncertainty and risk management problems, probabilistic nowcasting becomes especially relevant. However, this topic is poorly developed in the national researches considering regional forecasting. Linear and quantile regression, as well as quantile neural networks of various specifications are used in this paper as modeling tools. The models are estimated with the help of regional panel data and further compared in terms of interval and point nowcasts’ accuracy. Pooled additive quantile neural network turns out to be a promising model as it provides the most valid picture of the GRP growth slowing risks and is generally more robust in building point nowcasts. But when modeling the right-hand side of the distribution, that is, the scenario of extreme GRP growth, pooled linear regression is preferred. In turn, models with fixed individual effects, on average, give unsatisfactory results, but they are optimal for some regions. The constructed models can be used by policymakers to monitor recession risks and may help to to take prompt anti-crisis economic policy measures