Andreas Pick, Erasmus University, Rotterdam
"Optimal forecasts from Markov switching models and the effect of uncertain break dates"
We study the eect of uncertain break dates on forecasts based on optimal weighting of observations. We focus on Markov switching models where break dates correspond to switches between states.
It emerges that the forecasting performance increases drastically when the construction of the optimal weights takes uncertainty around the
time of breaks into account. Analytic expressions for the weights are provided both under exact knowledge of the break points and when the break points are uncertain.
In the latter case, forecasting improvements are substantial even in large samples. The performance of the optimal weights is shown through simulations and an application to
forecasting U.S. GNP over 30 years. In the application we find that using the optimal weights leads to a signicant reduction of the MSFE.