Matsushima, Hitoshi, Tomomi Tanaka and Tomohisa Toyama

We examine repeated prisoners' dilemma with imperfect private monitoring and random termination where the termination probability is low. We run laboratory experiments and show subjects retaliate more severely when monitoring is more accurate. This experimental result contradicts the prediction of standard game theory. Instead of assuming full rationality and pure self-interest, we introduce naiveté and social preferences, i.e., reciprocal concerns, and develop a model that is consistent with, and uniquely predicts, the observed behavior in the experiments. Our behavioral model suggests there is a trade-off between naiveté and reciprocity. When people are concerned about reciprocity, they tend to make fewer random choices.