Dynamic model averaging

Rather than committing to a single model specification, dynamic model averaging (DMA) maintains a universe of candidate models and weights them according to their recent predictive performance.

Note

This page is a stub. A worked treatment of DMA in DLMAX will follow.

The weights adapt over time, so the blend can favour different structures as the character of the data changes — a simpler model in calm periods, a richer one when structure emerges.