Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Conference Presentation |
| Scope | Discipline-based scholarship |
| Title | Multivariate Markov chain approximations |
| Organization Unit | |
| Authors |
|
| Presentation Type | paper |
| Item Subtype | Original Work |
| Refereed | Yes |
| Status | Published electronically before print/final form (Epub ahead of print) |
| Event Title | ERCIM |
| Event Type | conference |
| Event Location | London |
| Event Start Date | December 17 - 2011 |
| Event End Date | December 19 - 2011 |
| Abstract Text | In order to solve equilibrium models numerically, it is necessary to discretise vector autoregressive processes (VAR) into a ?nite number of states. The univariate case for such Markov chain approximations is well studied; however, the multivariate case has been scarcely addressed in the literature. We argue that the common approach to apply multivariate Gaussian quadrature has dif?culties for small numbers of states. To address this weakness, two alternatives are proposed: moment matching and direct bin estimation. Moment matching constructs the discrete Markov chain, such that the ?rst moments are exactly the same as in the VAR. Direct bin estimation leaves out the ?rst step of estimating the VAR and estimates the discrete Markov chain directly, by forming bins and then estimating transition probabilities with maximum likelihood. Finally, moment matching and bin estimation are compared to four different implementations of Tauchen’s approach in a standard Lucas asset pricing framework. |
| Export |
BibTeX
|