Abstract for this paper: Given the dramatic age variation in COVID death rates, we create a heterogeneous agent version of the Behavioral SI* contagion model of Keppo et al. (2020). Individuals randomly meet each others pairwise, unaware of their types. Inspired by auction theory, we compute the Bayes Nash equilibrium of the pairwise incomplete information games transpiring over time and across the population. This yields a simple new log-linear relationship between the case fatality rate (CFR) and COVID incidence: Everyone knows that everyone optimizes vigilance both for both the prevalence and their CFR.

We explain 2020 CDC incidence data for the USA north-east in terms of the CFR to age-specific COVID death data for Massachusetts. Our model is statistically significant: A 10% higher CFR reduces incidence by about 1%.