Predicting the COVID-19 pandemic in Canada and the US

We propose a time series model with the quartic trend function to make short-term forecasts of the COVID-19 confirmed cases in Canada and the U.S. Our one- to seven- days ahead out-of-sample forecast exercise demonstrates that the quartic trend model can produce very competitive short-term forecasts relative to the benchmark Susceptible, Infected, and Recovered (SIR) model. The bootstrap distance-based test of independence and the XGBoost algorithm reveals a strong link between the coronavirus case count and relevant Google Trends features (defined by search intensities of various keywords that the public entered in the Google internet search engine during this pandemic). Moreover, dynamic linear panel data models suggest a statistically significant relationship between the coronavirus case count and people's mobility trend provided by Google Mobility Reports (GMR) during the pandemic period.