Fraud Power Laws

We develop a mechanism explaining the emergence of heavy tails in the distribution of detected frauds. In our model, a manager manipulates earnings for a private benefit and updates her belief about monitoring quality over time, perceiving the monitor to be increasingly less effective as frauds remain undetected. The learning process leads to a slippery slope, whereby fraud size increases over time. Using misstatements data, we show that the empirical distribution of detected frauds is consistent with a power law and provide support for the slippery slope and learning channel. Surprisingly, a higher detection intensity, because it allows the manager to identify more quickly an ineffective monitor, will eventually increase frauds. Our analysis provides an economic argument against systematically dismissing extreme frauds as rare events, as we demonstrate that such frauds constitute a relatively large fraction of all frauds. As a policy implication, we show that, while fraud cannot be prevented altogether, extreme frauds can be averted by tolerating small ones and imposing steep punishments (i.e., increasing at a greater rate than the crime committed)