Detection without deterrence: long-run effects of tax audit on firm behavior

We exploit a program of randomized audits covering the entire population of VAT filers from Pakistan to study how much evasion audit detects and how much evasion it deters by changing behavior. We document substantial evasion at the baseline: almost one-third of firms engage in tax evasion, and conditional on some evasion, the average evasion rate exceeds 40 percent. We find remarkable heterogeneity in evasion by firm size with the evaded amount exceeding the reported liability in the bottom three quartiles but is merely 7 percent in the top. Despite detecting substantial liabilities, audit does not deter tax evasion. Examining more than ten outcomes, we detect no effect of audit on proximate or distant behavior. We offer an explanation of the detection-without-deterrence result and discuss its optimal policy implications.