HSE Economists Use Search Queries to Forecast Birth Rates

© iStock

Researchers from the HSE Faculty of Economic Sciences have shown that the accuracy of birth rate forecasts for Russia can be improved by almost 50% by incorporating the dynamics of online search queries related to pregnancy and childbirth into forecasting models. In the best-performing models, the forecasting error fell from 4.6% to 3.2%. The findings have been published in Populations and Economics.

Increasing the birth rate and supporting families remain among the government's key priorities. Reliable birth rate forecasts help estimate future demand for nurseries and schools, plan social infrastructure, anticipate labour market trends, and project long-term public expenditure.

HSE Faculty of Economic Sciences researchers Lilia Rodionova and Elena Kopnova, together with doctoral researchers Nikita Rodionov and Svetlana Kamelendinova, used internet search queries as one of the predictors of birth rates. According to the authors, analysing online user behaviour, particularly digital search activity, can provide a valuable source of data and serve as a meaningful predictor of demographic processes, including fertility.

The researchers analysed monthly data from Rosstat (the Federal State Statistics Service) on the number of births in Russia between 2011 and 2024. These figures were compared with Google Trends data, which measures relative changes in search interest over time. Google Trends assigns a value of 100 to the period with the highest search volume for a given query and measures all other values relative to that peak. For the study, the researchers compiled a corpus of 56 search terms, which machine learning methods grouped into four thematic categories: pregnancy planning, pregnancy, preparation for childbirth, and general-purpose queries. The analysis was based on a SARIMA model, which captures birth dynamics while accounting for seasonal patterns.

Using a one-year forecasting horizon, the standard model produced an average forecasting error of 4.62%, equivalent to approximately 4,600 births per 100,000 births in absolute terms.

Adding search query data reduced the forecasting error from 4.62% to 3.2%.

Lilia Rodionova

‘The most informative category of search queries was “Preparation for Childbirth.” Searches such as “maternity hospital” or “hospital bag” are typically made by women who already know they are pregnant and are actively preparing for delivery. This makes them a clear and reliable predictor for short-term birth rate forecasting,’ explained Lilia Rodionova, Associate Professor at the HSE Faculty of Economic Sciences.

The strongest effect emerged after accounting for time lags between search activity and births. According to the authors, users generally search for information about pregnancy well in advance, whereas queries such as ‘hospital bag’ or ‘breathing during labour’ are typically made shortly before admission to hospital. Such searches therefore indicate that childbirth is imminent. The ‘Pregnancy Planning’ category entered the model with an average lag of 7.4 months, while the ‘Preparation for Childbirth’ category had a lag of around six months.

For longer forecasting horizons, the best results were achieved by combining all four categories of search queries together with their respective time lags. With a two-year forecasting horizon, the error declined to 2.7%, and with a three-year horizon it fell further to 2.6%.

‘The model was tested using data up to December 2024, covering both the COVID-19 pandemic and the onset of geopolitical instability. Its high forecasting accuracy during the validation period, which included these years of crisis, demonstrates the model's strong potential,’ said Lilia Rodionova.