Invited Paper —Learning Models: An Assessment of Progress, Challenges, and New Developments

This present survey has three main objectives. The first is to summarize the existing literature on learning models. We will discuss the different types of models that have been estimated, and assess the progress that has been made – both empirically and methodologically. The second objective is to identify the main issues and challenges that confront future progress in this area. Our third objective is unusual for a conventional survey in that we propose new solutions to some of these challenges, and give an extended example of how to apply them. In particular, one problem in learning models with forward-looking agents is they are still difficult to estimate – requiring the solution of a dynamic programming (DP) problem. Here, we extend a method developed by Geweke and Keane (2000) to make it possible to estimate these models without solving the DP problem. Another key issue is whether learning or inventories provide a better explanation of choice dynamics. This issue has not been addressed, because it has never been computational feasible to include inventories in learning models. By using the Geweke and Keane (2000) method we can, for the first time, estimate a model with both learning and inventories, and shed light on the role of each. Our example focuses on demand for diapers.