A Machine Learning Approach to Constructing Weekly GDP Tracker Using Google Trends

Abstract

[EN] The outbreak of the COVID-19 pandemic further highlighted the limitation of existing traditional indicators as policy formulation, particularly during crisis periods, demands timely and granular data. We construct the first weekly GDP tracker in the Philippines using topic- and category- based Google Trends search volumes with the aid of machine learning models. We find that our weekly GDP Tracker is a useful high-frequency tool in nowcasting economic activity, especially during periods of extreme economic duress as the trends and developments in the actual GDP growth are well-captured by the model. Our weekly Tracker was able to capture about 96 percent of the slumpobserved in actual GDP growth in Q2 2020, reflecting the tracker’s overall good performance and the potential of the use of Google Trends. The top three Google Trends searches in predicting GDPgrowth using the SHAP interpretability tool are “unemployment”, “subsidy”, and “investment”. We also showed that the machine learning-based GDP tracker outperforms the traditional autoregression models under study in terms of lower root mean square error (RMSE) for both train and test datasets. Thus, pending the availability of quarterly national accounts, our weekly GDP tracker can serve as useful complementary surveillance tool for monitoring economic activity.