Predicting the unemployment rate: Analyzing statistics on search engine query

Abstract

Unemployment is one of the key macroeconomic indicators that play a significant role in the state economic policy. There is a significant variety of methods for forecasting the situation on the labor market and the unemployment rate. We show the evolution of relevant approaches, including traditional methods (labor balance model, structural changes in regional labor markets, multifactorial labor market models such as ARIMA, TAR, ARNN) and modern ones. The development and the possibility of applying innovative approaches to forecasting unemployment is associated with digitalization and the development of Internet technologies, which provide researchers with new analysis tools. When compared to traditional methods, Google search query data and other internet activity data result in better forecasts, helping to solve the problem of lagging data provided by the official statistics, and adding relevant information to analyze and predict unemployment. Based on the experience of using the statistics of search queries in predicting and nowcasting the unemployment rate, the authors developed several models for predicting the unemployment rate in Russia. The research findings show that adding multiple query variables to the autoregressive model is able to improve the predictive accuracy of the model. Superiority of the hybrid model over the autoregressive variation is due to its ability to respond to future labor market shocks.