A0856
Title: Recursive non-parametric predictive for a discrete regression model
Authors: Lorenzo Cappello - Bocconi University (Italy) [presenting]
Stephen Walker - University of Texas at Austin (United States)
Abstract: A recursive algorithm is proposed to estimate a finite set of conditional distributions. The procedure is fully nonparametric and has a Bayesian interpretation. Indeed the recursive updates follow a Bayesian update to a certain extent. We prove weak convergence of the distribution estimates, and demonstrate numerical accuracy via simulations. A novel fixed-point argument is used to prove the result, along with a novel assumption on the user-supplied sequence of weights that govern the iterations. The estimate is very fast and requires limited computing power; being also parallelizable. We show that it is competitive with both frequentist and Bayesian nonparametric models.
