The impact of the official statistics revision on the accuracy of the Russian macroeconomic indicators nowcasting models

Abstract

The paper presents the results of an accuracy analysis of nowcasting models for Russia's GDP and its components based on usage data for the period from the first quarter of 2014 to the third quarter of 2023. The novelty of the study lies in comparing the accuracy of various models — MIDAS, MFBVAR, DFM models, regularization models, and classical autoregression for order 1, evaluated on both the first and final versions of official statistics revisions, taking into account data availability within the examined quarter. The result of our study is a strategy for optimal selection of data and models for the most accurate first and final versions of nowcasting official statistics revisions for GDP and its components by use. Our paper also demonstrates that the nowcasting model for Russia's GDP, evaluated on the first revision, is more accurate regardless of data completeness within the quarter, despite multiple subsequent revisions of initial data by Rosstat. Furthermore, for GDP models, the forecasts of the first, most volatile, and imprecise revision turn out to be the most accurate.