Managing Supply and Demand for the Performing Arts in the Time of COVID

The COVID-19 pandemic forced performing arts organizations to shut down operations, and considerable uncertainty about future demand exists. We develop a data-driven framework and decision support tool to explain and predict demand for live, in-person arts performances in the time of COVID. Our granular panel dataset includes: (1) detailed household-level transaction panel data (2018-2021) data for 51 nonprofit performing arts organizations in the USA, (2) anonymized and aggregated mobile location data that captures movement flow patterns from household census tracts to the focal organizations, (3) health data that captures county-level COVID case rates and state-level vaccination rates, (4) census data (e.g., income, education, and age) matched to household tracts, and (5) organization location data with respect to competition and complements (hotels, restaurants, and bars). We develop model-free visualizations and a reduced-form econometric model to describe and estimate the impact of these factors, revealing generalizable insights that help us understand performing arts ticket purchase behavior. For example, our results suggest that lagging vaccination rates likely cost this industry $10M per month for every unrealized percentage point in vaccinations. Next, we implement a super learner-based demand prediction framework that provides better out-of-sample prediction than any of the base learners. We implement this framework in an R notebook that organizations can easily use to predict demand by augmenting our data with their organization specific data. Overall, our work contributes to the literature on demand estimation using machine learning methods and provides data-driven insights for managing COVID-era demand for the performing arts