Capturing the timing of crisis evolution : a machine learning and directional wavelet coherence approach to isolating event-specific uncertainty using Google searches with an application to COVID-19

Abstract

The phases of a crisis are critical to understanding its evolution. We construct an economic agent-determined machine learning-based Google search index that associates search terms with uncertainty to isolate COVID-19-related uncertainty from overall uncertainty. Subsequently, we apply directional wavelet analysis that discriminates between positive and negative associations to study the evolving impact of the COVID-19 pandemic on financial market uncertainty and financial markets. Our approach permits us to delineate crisis phases with high precision according to information type. The analysis that follows suggests that policy responses impacted uncertainty and that the novelty of the COVID-19 outbreak had a significant impact on global stock markets. Regression analysis, wavelet entropy and partial wavelet coherence confirm the informational content of our uncertainty index. The approach presented in this study is applied to the COVID-19 crisis but is generalisable beyond the pandemic and can assist in decision-making during times of economic and financial market turmoil and should be of interest to policymakers, researchers and econometricians.