| DATE: | Thursday, March 9, 2006 |
| TIME: | 1:15-2:15 PM |
| PLACE: | LN 2205 |
| SPEAKER: | Wolfgang Wefelmeyer, University of Cologne |
| TITLE: | Partially observed Markov chains |
We consider estimation of functionals in Markov chains when part of the observations are missing, either in a deterministic and periodic pattern, or randomly, with probabilities possibly depending on the last state. We show how best to use the information across the gaps. In certain situations, say when every second observation is missing, the distribution of the chain is not identifiable, except when we have sufficient information about its structure. We show in particular that for linear autoregression there are different ways of recovering the innovation distribution.