Theory: (From An SIR model with behavior)
Fact: (from Scott Gottlieb via Marginal Revolution)
I do not mean to toot my horn, as many other graphs from the model did not look like that. This particular graph did, and really offers a sad interpretation of what's going on. In the model that produced the graph, people and policymakers react to the current death rate in deciding how much risk to take by going out.
It is entirely individually rational for people to go out and party when very few around them are infected. Sadly, that means the disease collectively ramps up. Then it is individually rational for people to cut back, and the disease slows down. Cycles can result.
Public policy is supposed to get on top of these cycles, by stamping out disease when it is low, the same way you keep taking antibiotics even when you feel better. It is the policy that has failed rational expectations here, not people. (No, that does not mean lockdown business and print money so we all can stay home and order stuff that comes by magic from Amazon. Ambitious testing would have done the trick. Or at least containing the summer's wave of super spreading parties.)
Sherwin Rosen, Kevin Murphy and José Scheinkman have a beautiful JPE paper Cattle Cycles describing a related phenomenon. But our governments are supposed to be smarter than cows.