Title: A Bayesian transformation model for forecasting asset returns
Authors: Gelly Mitrodima - LSE (United Kingdom) [presenting]
Abstract: Jointly modelling a finite collection of quantiles over time is considered under a Bayesian nonparametric framework. To address the challenges of formal Bayesian quantile inference, we propose a flexible Bayesian transformation model. This allows the likelihood and the quantile function to be directly calculated, and define a novel stationary process which can be ``centred'' over a parametric model. The model is very general and its structure allows us to derive sufficient conditions for stationarity of some important sub-models. The application of the model to simulated and real data via Markov chain Monte Carlo (MCMC) methods shows that the model performs well for a range of data generating mechanisms.