Regional differences in modelling Covid-19 infections using Google Trends data : evidence from Hungary

Abstract

This study models the spread of Covid-19 based on internet search data at Hungary’s NUTS 2 regional level. Using a modified version of the search index composition method proposed by Li et al. (2020), adapted to Hungarian data, we explore the regional predictability of Covid-19 infections and examine its correlation with socio-economic variables. Our findings indicate significant regional variations in the effectiveness of internet search data for predicting coronavirus case numbers. We identify key Covid-19-related Google search terms relevant to the Hungarian context and assess their predictive power using ordinary least squares and Bayesian vector autoregression models, with the susceptibleinfectious–recovered model serving as a benchmark. Our findings suggest that internet search data can be a valuable tool for predicting the regional spread of Covid-19, though predictability varies. Forecasting infection numbers based on internet search data is more successful in regions with high health awareness influencing internet search behaviour. The case of Northern Hungary demonstrates that internet search-based forecasting can also be effective in areas where the population relies on the internet as an alternative source for health-related information. This study highlights the importance of regional analysis in pandemic forecasting, offering new insights for public health strategies and contributing to the literature on the applicability of internet search data.