A0499
Title: Bayesian emulation for high-dimensional portfolio problem
Authors: Kaoru Irie - University of Tokyo (Japan) [presenting]
Abstract: The portfolio problem to select a small number of promising financial assets from all the assets available in the market is discussed. Based on the equivalence between the Bayesian expected utility optimization and the posterior analysis of synthetic statistical models, we ``mode'' the portfolio decision problem as the state space models, introducing the priors with different levels of shrinkage effects, such as Laplace, horseshoe and their combination, for the purpose of portfolio selection. This approach also allows for the use of computational methodology in statistics customized for the posterior mode, including the EM methods and other stochastic model search technique. The advantage of these decision models are exemplified by the sequential portfolio analysis of Nikkei 225 stocks, by alternatively computing the multiple step ahead predictions provided by the multivariate volatility models and making the portfolio decision.
