This paper considers the combination of forecasts using changing weigh
ts derived from switching regression models or from smooth transition
regression models. The regimes associated with the switches may not be
known to the forecaster and thus need to be estimated. Several approa
ches to this problem are considered. In two empirical examples, these
time-varying combining procedures produced smaller, in some cases subs
tantially smaller, out-of-sample squared forecast errors than those ob
tained using the simple linear combining model.