¿Can Google Trends improve predictions of consumption? evidence from a bayesian structural time series model of ecommerce transactions in Mexico

Abstract

In recent years, electronic commerce has emerged as a significant economic driver, making the surveillance of ecommerce markets crucial for various stakeholders including business owners, investors, and policymakers. A large portion of online consumers use search engines like Google to look for specific products, brands, or marketplaces. In this sense, it is relevant to ask the following question. Can data from Google search queries reveal insights into the ecommerce sector in Mexico? Data from ecommerce transactions is obtained from Banxico and data from search queries is available in GoogleTrends.com. The econometric framework present in this study follows the Bayesian Structural Time Series (BSTS) methodology, which combines time series and regression analysis. Results prove that the inclusion of contemporary information from Google Trends in the BSTS framework does reflect a lower cumulative absolute error than simple Structural Time Series models. Particularly, three clusters of significant Google Trends queries are identified: 1) online marketplaces that belong grocery stores, 2) two-sided marketplaces and 3) online apparel stores.