Q2. how do the authors extend the concept of stochastic volatility?
the authors allow for the conditional variance of the measurement equation to depend on , thus extending the concept of stochastic volatility to allow for more general nonlinear patterns in the conditional variance.
Q3. What is the advantage of using a posterior simulator?
Since Bayesian methods for state space models are well-developed, the authors can use such methods and only add a block to an existing posterior simulator which characterizes the distance function.
Q4. What can be used to control dt?
in more general models dt can be a vector (e.g. it can have two components, one controlling breaks in coe¢ cients and the other in error variances).
Q5. How can the authors accommodate the state equation variances?
By allowing the state equation variances to depend on the distance between observations, the authors can accommodate a much wider variety of ways that their parameters can involve, including everything from abrupt change models (e.g. threshold autoregressive models or structural break models such as that of Bai and Perron, 1998) to those which allow gradual evolution of parameters (e.g. smooth transition autoregressive models or TVP models such as that of Primiceri, 2005).
Q6. How do the authors obtain posterior properties of any of the model parameters?
Using the methods of posterior simulation described above, with Xt = yt 1 and index de nition variable simply being the natural ordering (i.e. 1; 2; ::; T ), the authors can obtain posterior properties of any of the model parameters (or functions thereof).
Q7. What is the probability of the correct model with data ordered by yt 1?
In particular, the authors nd that there is a 94:3% probability that the (correct) model with data ordered by yt 1 is the correct one and only a 5:7% probability that the normal time ordering is appropriate.
Q8. What is the advantage of their approach?
The advantage of their approach is that the precise shape of the distance function would be estimated from the data and not imposed at the outset by choosing to estimate, e.g., a TAR or STAR model.
Q9. What is the relationship between the index variable and the measure of the eect?
Although it is true that very large positive oil price shocks (e.g. where the index is about 2) have the largest negative e¤ects on GDP growth, the relationship between the index variable and this measure of the e¤ect of an oil shock is highly non-monotonic.
Q10. What is the tting linear model?
The best tting linear model is the AR(2) and, hence, the authors consider extensions of this model and let Xt contain an intercept plus 2 lags of yt.
Q11. What are the different types of models used for the conditional mean and conditional variance?
using measures of real output, a wide variety of regime-switching and structural break models for the conditional mean and conditional variance have been used.