Forecasting unemployment in the Republic of Belarus using Google Trends data

The problem of unemployment forecasting in the Republic of Belarus keeps its relevance against the background of the dynamically changing economic environment. Conventional forecasting methods based on official statistics often fail to account for real-time changes in the labour market and thus lose their accuracy. At the same time, search query data has proven to be effective as leading indicators in other countries, but its application in Belarus remains unexplored. The paper aims to assess the potential of using search query data to improve the accuracy of unemployment rate forecasting in the Republic of Belarus. Methodologically, the study rests on the theoretical propositions of macroeconomics and SARIMA, VAR, and SARIMAX models. The methods include time series decomposition, the Dickey–Fuller stationarity test, differencing, data standardisation, and the Granger causality test. The evidence is unemployment data of the National Statistical Committee of the Republic of Belarus for 2015–2024 and Google search query data related to job searches. The study found that the SARIMAX model incorporating search query data outperforms classical models of unemployment forecasting by demonstrating a minimum of errors. According to this model, the predicted values of unemployment rate reflect a downward trend indicating the sustained improvement in the labour market of the Republic of Belarus. The findings emphasise the importance of combining traditional data with digital metrics to increase the forecast accuracy as well as open up prospects for further research into the use of internet data for socioeconomic analysis, including the development of more advanced unemployment rate forecasting models.