1
$\begingroup$

The discrete time version of a Vasicek model is equivalent to an AR(1) model with opportunely chosen parameters, as showed in this paper: http://www.damianobrigo.it/toolboxweb.pdf.

Following this procedure, having fitted an AR(1) model I can recover the parameter of a Vasicek model, but what about the standard errors? Is there a way to obtain them from the AR(1) estimates as well?

Since I'm interested in the regime-switching version of these processes, I cannot use the usual MLE, and all the R or MATLAB packages have implementation just for the regime-switching AR model.

Improve this question
$\endgroup$
8
  • 1
    $\begingroup$ Are you familiar with the delta method and/or block bootstrapping? Those would be appropriate methodologies for obtaining the standard errors of the Vasicek model given the estimates and standard errors of the AR(1). $\endgroup$
    – Zachary Blumenfeld
    Commented Jul 19, 2015 at 4:27
  • $\begingroup$ I was trying the block bootstrap, but I have the same problem as stated in this question: stats.stackexchange.com/questions/160776/… . I didn't think about the delta method, I will try that. Thanks. $\endgroup$
    – Egodym
    Commented Jul 19, 2015 at 13:00
  • 1
    $\begingroup$ See nccur.lib.nccu.edu.tw/bitstream/140.119/35143/6/51007106.pdf and the references cited therein for a discussion on selecting block length. As with a lot of things in statistics there is no one right answer and block bootstrapping methods range from simple to complex (some argue that selecting the block length itself from another discrete distribution is most appropriate). $\endgroup$
    – Zachary Blumenfeld
    Commented Jul 19, 2015 at 22:18
  • $\begingroup$ Problem with delta method is how to get the variance-covariance matrix for my model (regime-switching autoregressive of order 1) coefficients. $\endgroup$
    – Egodym
    Commented Jul 20, 2015 at 15:15
  • $\begingroup$ You can still estimate the model with MLE, see this paper by JD Hamilton econweb.ucsd.edu/~jhamilto/palgrav1.pdf . So that will give you standard errors for your coefficients under the iid normality assumptions. Whether or not the iid normality assumptions are met is another story, which is why you might prefer block bootstrapping or going Bayesian. I don't know exactly what you mean. $\endgroup$
    – Zachary Blumenfeld
    Commented Jul 20, 2015 at 16:30

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.