Bayesian Density Forecasts for VIEWS

Abstract:

Dynamic analysis in VIEWS are important forecasts because the data are inherently serially correlated (in space, time and both). Here we consider dynamic forecasts as a baseline: above the density baselines proposed as part of VIEWS 2.0, the forecasts proposed here provide Bayesian density forecasts that allow for the evaluation of simple dynamics (autoregression and time trends) and for different distributional assumptions (e.g., Poisson, negative binomial, zero-inflated, Tweedie). The idea here is that the
forecasts proposed should be baseline in the sense that no-change or density forecasts account for the basic properties of the data. While modest, the idea is that the baseline bar for a “success” here can and should be higher.

Authors:

Patrick T. Brandt