Title: Simulating and Estimating Dark Rates in Collusion
Abstract: Dark rate estimation of (illegal) conduct is inherently estimating population sizes of unknown populations. We reversely engineer prediction of the collusive population in an economy from a non-randomly sampled set of observed cases. If conducted successfully, our method applies in various situations where the population size is in doubt. We start by simulating collusive behaviour as well as properties of law enforcement mechanisms to show that conventional methods estimating the population of undetected cartels do not provide true population sizes. Second, we aim to identify suspected sample selection biases in the detection of collusive industries; (i) industries being prone to collusion and (ii) industries being prone to detection. Incorporating strategic firm and competition policy agency behaviour (i.e. the fact that various cartels happen to be detected, dissolve, form again, and get detected again (repeat offenders), and external shocks like the introduction of leniency (e.g. 1978 in the US, 1997 in the EU)), we build a framework to estimate cartel activity over time. On the simulated as well as real data (EC DGCompetition) we test our new framework in terms of predictive power as we understand the data generating process.
