Professor Alan Ker will be giving a seminar in the Economics department at Dalhousie University on Monday Sept 10.

Title: Borrowing Information for Like DGPs in Nonparametric Density Estimation with Applications.

Abstract: In this talk I consider three methods that borrow information from possibly similar DGPs without assuming the form of similarity or to what extent they may or may not be similar. The first method introduced by Racine and Li (Journal of Econometrics, 2004) essentially smooths across different DGPs. The second estimator introduced by Ker (Statistics and Probability Letters, 2016) reduces bias by incorporating data from like DGPs into an initial start estimate and then corrects that start using data from the DGP of interest. The third estimator introduced by Ker and Liu (Computational Statistics, 2017) uses Bayesian Model Averaging across the different DGPs rather than functional forms. These methods are considered in the application of estimating crop yield distributions and crop insurance premium rates. This application is notable because crop insurance is the cornerstone of domestic agricultural policy in many developed and developing countries and accurate rating of these contracts is paramount. For example, total liability for the U.S. crop insurance program in 2017 was \$106.1 billion with premiums of \$10.1 billion. To assess the efficacy of the three methodologies, we undertake two empirical simulations comparing this estimator with the current rating methodology used by the United States Department of Agriculture Risk Management Agency (USDA-RMA). Our results indicate that borrowing information across possibly similar DGPs decreases estimation error and enables statistically and economically significant rents to be captured.