Learning in rent-seeking contests with payoff risk and foregone payoff information

We test whether deviations from Nash equilibrium in rent-seeking contests can be explained by the slow convergence of payoff-based learning. We identify and eliminate two noise sources that slow down learning: first, opponents are changing their actions across rounds; second, payoffs are probabilistic, which reduces the correlation between expected and realized payoffs. We find that average choices are not significantly different from the risk-neutral Nash equilibrium predictions only when both noise sources are eliminated by supplying foregone payoff information and removing payoff risk. Payoff-based learning can explain these results better than alternative theories. We propose a hybrid learning model that combines reinforcement and belief learning with risk, social and other preferences, and show that it fits data well, mostly because of reinforcement learning.