Forecasting the Birth Rate in Russia Based on Google Trends Search Query Statistics

This article proposes an approach to forecasting fertility using Google Trends search query statistics. Unlike existing studies on fertility forecasting, this paper introduces a method for aggregating thematic search queries using machine learning techniques, reducing the dimensionality of search queries through principal component analysis, and incorporating a lag structure of search queries ranging from 1 to 12 months depending on the search topic. The lag depth was determined based on the cross-correlation function, comparison of information criteria, and the statistical significance of the corresponding coefficients in the SARIMAX model. The study uses Rosstat data on the monthly number of registered births in Russia from January 2011 to December 2024. The results show that a SARIMAX-based fertility forecasting model that incorporates search query statistics and their lag structure improves forecast accuracy. The best one-year-ahead forecasting model includes the query group "Preparation for childbirth" and its lags, while for two- and three-year horizons the best performance is achieved using the model with "All queries." Forecast accuracy is evaluated using the MAPE and MAE metrics. To assess the robustness of the results, a cross-validation procedure with expanding and sliding windows was additionally applied.