Title: Using GLMMs to Predict the Effects of COVID-19 on Airline Passenger Throughput

Abstract:

COVID-19 has had a major impact on the U.S. economy and resulted in unprecedented changes in consumer behavior and mobility patterns. Many companies and government agencies face challenges predicting the future amidst so much uncertainty surrounding the virus. Traditional methods for time series prediction work well to forecast seasonality and trend components of a time series when the future state of the time series is expected to mimic past behavior. Sudden events that change the underlying behavior patterns render these models ineffective. To account for this, we use a Generalized Linear Mixed Effects Model (GLMM) to explain changes in airline passenger throughput subject to differences in how policy restrictions have been implemented and followed throughout the United States.

The method also supports scenario simulations where we can explore how different social distancing interventions that may be implemented in the future would affect the number of people traveling on a daily basis. We use Apple mobility trends to understand how far off from baseline the driving conditions are for a region, UMD Mobility data to understand the proportion of people that are staying at home, and B2C data derived from cell phones to understand consumer visits and spending habits at the store level. With this family of models, we were able to predict national air passenger volume within a 50% confidence level more than one month into the future during early stages of the pandemic when uncertainty was at its peak. Airports and airlines could use these forecasts to inform workplace staffing plans and could also adapt their prices to meet expectations of how travel patterns are expected to change in the near future.