The potential of using textual data to estimate inflation expectations in Russia

Abstract

The paper proposes a methodology for measuring inflation expectations of firms in Russia based on the analysis of textual data from major media outlets. Over 2.3 million news publications from 13 Russian media were collected from the social network VKontakte for the period from January 2015 to October 2024. To identify publications specifically devoted to inflation, rather than merely containing relevant keywords, Latent Dirichlet Allocation (LDA)—a probabilistic topic modeling method—is applied for the first time in the Russian literature. The constructed indicator of inflation expectation intensity exhibits a 78% correlation with the Bank of Russia's measure of firms' price expectations, which is substantially higher than the correlation with household expectations (48%). Using LASSO regression, it is established that the largest contribution to explaining the dynamics of firms' expectations comes from three business-oriented media outlets—Kommersant, Vedomosti, and Expert—which jointly account for 75% of the variance in firms' price expectations. The results confirm that media textual data can serve as an effective complement to survey-based methods of measuring inflation expectations.