Cheating for the common good in a macroeconomic policy game

This paper presents a simple repeated-game model of interaction between an optimizing government and the private sector. Two polar cases are considered: (a) the private sector is represented by a single agent; and (b) there is a continuum of heterogenous atomistic private agents. In both cases, the government starts each repetition by making a non-binding announcement about its future actions. The players have complete and perfect information, with one exception: the private agents do not know whether or not the government will act as announced. Thus, each private agent i either behaves with probability π i as if it trusted the announcement, or plays with probability 1− π i as a Stackelberg leader. After observing the reaction of the private sector, the government implements the actual policy measures. Finally, the private agent(s) update π i as a function of the payoffs received. We show that, although the government's announcements are never respected, acting as if they were true leads to an outcome that simultaneously improves the situation of the government and of those players that trust it, compared to the standard equilibrium solutions. An overwhelming majority of private agents very rapidly learns to behave as if the announcements were true. The other agents also experience higher payoffs due to an herding effect. The fact that the announcements are always reneged is crucial for the solution to be pareto-improving. These results are in stark contrast to the conclusions usually presented in the related economic literature.