Knitting Multi-Annual High-Frequency Google Trends to Predict Inflation and Consumption

Abstract

We propose a regression-based algorithm that allows to construct arbitrarily many comparable, multi-annual, consistent time series on monthly, weekly, daily, hourly and minute-by-minute search volume indices based on the scattered data obtained from Google Trends. The accuracy of the algorithm is illustrated using old datasets from Google that have been used previously in the literature. We use our algorithm to construct an index of prices searched online (IPSO). Out-of-sample, the IPSO improves monthly inflation and consumption forecasts for the US and the Euro Area. In-sample it is contemporaneously correlated with US consumption, when controlling for seasonality, and Granger causes US inflation on a monthly frequency.