Immigration enforcement affects millions who make daily economic decisions under uncertainty about state action. Using data from 2014 to 2018, we combine daily transaction data with arrest-level ICE records to study how households learn and respond to enforcement risk. Enforcement follows predictable weekday patterns, and communities have learned them: consumption is depressed not only on days when enforcement occurs, but also on the same weekday of other weeks—when no enforcement occurs—and rebounds in between, indicating that behavior tracks beliefs, not just realized enforcement. Instrumenting for beliefs using this learnable structure, we find that a 10 percentage point increase in perceived enforcement probability reduces Hispanic foreign-born consumption by about 5 percent. A structural model with Bayesian learning and pent-up demand reveals that roughly half of the immediate consumption decline is recovered through rebound; the rest is permanently foregone. Eliminating enforcement risk would increase Hispanic foreign-born consumption by 3.6 percent, but only 42 percent of these gains materialize within the first year—the remaining losses reflect learning frictions: even after enforcement ceases, beliefs must still adjust.