Bayesian Forecasting Using Stochastic Simulation

In this article, we present a general framework to construct forecasts using simulation. This framework allows us to incorporate available data into a forecasting model in order to assess parameter uncertainty through a posterior distribution, which is then used to estimate a point forecast in the form of a conditional (given the data) expectation. The uncertainty on the point forecast is assessed through the estimation of a conditional variance and a prediction interval. We discuss how to construct asymptotic confidence intervals to assess the estimation error for the estimators obtained using simulation. We illustrate how this approach is consistent with Bayesian forecasting by presenting two examples, and experimental results that confirm our analytical results are discussed.