Nowcasting unemployment rate and new car sales in south-western Europe with Google Trends

This work presents a study describing the use of Internet search information to achieve improved nowcasting ability with simple autoregressive models, using data from four countries and two different application domains with social and economic significance: unemployment rate and car sales. The results we obtained differ by country/language and application area. In the case of unemployment, we find that Google Trends data lead to the improvement of nowcasts in three out of the four considered countries: Portugal, France and Italy. However, there are sometimes important differences in the predictive ability of these data when we consider different out-of-sample periods. For car sales, we find that, in some cases, the volume of search queries helps explaining the variance of the car sales data. However, we find little support for the hypothesis that search query data may improve predictions, and we present several possible reasons for these results. Taking all results into account, we conclude that, when Google Trends variables are significantly different from zero in-sample, they tend to lead to improvements in out-of-sample predictive ability. The results can have implications for nowcasting, by providing some indications regarding the advantage or not of the use of search data to improve simple models and indirectly by highlighting the sensitivity of the approach to the actual country-specific base, nowcasting period and search data