I propose a self-tuning logit model of adaptive repetitive choice which addresses currently presented issue of high heterogeneity in the logit parameter estimates. Model modifies the strength of agents' reactions to payoffs using the largest observed payoff difference as a scaling factor. I apply the model to the data from two laboratory experiments, which were design to study adaptive choices, and show that self-tuning model captures the main features of the observed behavior. The Model Confidence Set approach suggests that the proposed novel model best fits the data in comparison with different specifications of the logit models presented in the literature. Similar values of a single parameter estimated across all sessions highlight the external validity of the model. Example of model applications is provided.
Алексей Чернулич (выпуск 2014), PhD студент Технологического института Сиднея (University of Technology Sydney).