Exact diffuse initialisation of states
Posted: Thu Jun 10, 2021 4:27 pm
Hello!
As I understand, EViews does not perform the exact diffuse initialisation of the initial state matrices, as outlined in Durbin and Koopman pg 126 - please correct me if I am wrong? Two questions:
1) The help page states: "EViews uses the authors’ recommendation that one first set and then adjust it for scale by multiplying by the largest diagonal element of the residual covariances."
- When estimating hyperparameters (via ML) is P(1|0) updated with each iteration using the procedure above until convergence?
2) Is anyone aware of an EViews subroutine / add-in / code example which does calculate the exact initialisation?
Thanks
Adam
As I understand, EViews does not perform the exact diffuse initialisation of the initial state matrices, as outlined in Durbin and Koopman pg 126 - please correct me if I am wrong? Two questions:
1) The help page states: "EViews uses the authors’ recommendation that one first set and then adjust it for scale by multiplying by the largest diagonal element of the residual covariances."
- When estimating hyperparameters (via ML) is P(1|0) updated with each iteration using the procedure above until convergence?
2) Is anyone aware of an EViews subroutine / add-in / code example which does calculate the exact initialisation?
Thanks
Adam