Beschreibung
When dealing with Bayesian Model Averaging in large model spaces, the selection of 'reasonable' models is a difficult task. So-called dilution priors have gained importance and can be used to adjust for similar models. This is especially relevant for model spaces that contain interacted covariates. In such a setting models which only include interacted variables can catch the effect of the omitted parent variable. To compensate for such models the literature proposes the use of dilution priors according to the strong or weak heredity principle. Accordingly a strong heredity prior would exclude models with interaction but no parent variable completely from the model space (assign them a zero model probability). This very informative prior has been criticised as being too strict in recent literature. The author proposes a weak heredity prior which penalizes such uncommon models, but does not exclude them from the model space entirely. In the paper both strong and weak heredity priors are compared using simulated data and several benchmark models. Furthermore the predictive performance of the weak heredity prior is evaluated.| Zeitraum | 17 Dez. 2011 → 19 Dez. 2011 |
|---|---|
| Ereignistitel | Computational and Financial Econometrics |
| Veranstaltungstyp | Keine Angaben |
| Bekanntheitsgrad | International |