Journal of Public Economics 89(9-10), September 2005. PDF.

Standard models of tax evasion implicitly assume that evasion is either fully detected, or not detected at all. Empirically, this is not the case, casting into doubt the traditional rationales for interior evasion choices. I propose two alternative, dynamic explanations for interior tax evasion rates: First, fines increasing in the duration of an evasion spell, implying that the expected costs of evasion increase convexly with the time spent non-reporting, while the benefits increase linearly. Second, different vintages of income sources subject to aggregate risk and fixed costs when switched between evasion states. The dynamic approach yields a transparent representation of revenue losses and social costs due to tax evasion, novel findings on the effect of policy on tax evasion, and a tractable framework for the analysis of tax evasion dynamics.