Communication with Detectable Deceit

We analyze a model of communication with an extreme conflict of interest in which deceit is detectable with positive probability. The sender observes (possibly imprecise) information about his type, and can send any message to the receiver at no cost. If he lies---either by claiming more precise information or a false type---his message ends up incoherent with positive probability. In any informative equilibrium, the types of the sender close to the mean of the type distribution are revealed, while the extreme types of the sender pool. Moreover, the sender reveals less precise information when lying. There is a trade-off between facing a skilled liar who observes the state precisely and a poor liar who has only imprecise information. The receiver is more likely to prefer a skilled liar when the distribution of the type is more concentrated around the mean.