Temporal dependence in multi-step density forecasting models

Kevin Dowd

Abstract

ABSTRACT

Temporal dependence in multi-step density forecasting models is examined. If the underlying random variable has a known dependence structure this suggests that suitably transformed observations from the density forecasting model have a unique (and often simple) dependence structure of their own. Knowledge of this structure enables the model validator to carry out more efficient tests that take account of, and also test, the dependence structure itself. The tests themselves are extremely simple and powerful. An illustrative empirical application of the proposed approach to the evaluation of the Bank of England’s “fan chart” forecasts of inflation uncertainty is also given.