The RATS Software Forum

Dear Tom,

I estimate the Dueker (1997)'s paper using my RATs 8.2. It works fine, but when it comes to the last part of the program to estimate the model with fixed means, I cannot get the value of p1 and p2. In fact, p1 and p2 are both zero. Could you help me with this? Thanks a lot for your help in advance!

Which of the five programs are you talking about?

Dear Tom,

Sorry, the file I am using is SWGARCH_NF.rpf

That one (of the twenty or so combinations in the paper) apparently isn't converging given the guess values. That looks like it's the only model that uses a fixed mean and switches the variance only and I can't seem to find any way to make that have anything other than a corner solution where the lower variance state has 0 retention probability. You can see what happens to that as you reduce the .1 in the below to .01 and .001.

nonlin(parmset=nomeans) mu(2)=mu(1) p(1,1)=.1

So basically, it collapses to a one-regime model.

Dear Tom,

Thanks for your reply. But I have another question when reading the output: How to intepret the P(1,1) and P(1,2)? Does P(i,j) suggest the probability of switching from j to i? Then how to calculate the value of P(2,1) and P(2,2)?

zy761 wrote:Dear Tom,

Thanks for your reply. But I have another question when reading the output: How to intepret the P(1,1) and P(1,2)? Does P(i,j) suggest the probability of switching from j to i? Then how to calculate the value of P(2,1) and P(2,2)?

Subtract from 1

Dear Tom,
I estimated Dueker(1997) SWARCH-Df. The value of NUV(2) coefficient is turned out to be 36690.719780 with t-test value of 358.63824. The coefficient value is very large and is different from the value in Dueker(1997) paper. Could you please help? What went wrong in the estimation? I attach my output file. The last table has the output.
Looking forward for your reply,
Prashant

The big coefficient simply means that that branch is conditionally Normal.

So far as I can tell, Dueker's original results for the DF model are wrong---if you feed in his estimates, the log likelihood is much lower than he shows, and it's not even a local mode.

Dear Tom,
Thanks..so you mean nothing wrong with the results. I just write it is conditionally normal in my analysis if I get the big coefficient values!! right?
Looking forward for your reply,
Prashant

Yes. That would be the interpretation. Whether a model with means and degrees of freedom switching, but with the variance multiplier fixed makes any sense is far from clear.

Dear Tom,
THANKS FOR THE reply. I used the same programme-SWGARCH_DF for my data set with little change to my requirement. The message I got in the output at last is
"## MAT15. Subscripts Too Large or Non-Positive Error was evaluating entry 712" I could not understand the reason why it happened. Could you please help me out and offer solution for it? I attach the programme, output and dataset.
Looking forward for your reply,
Prashant

Because you chopped the top 16 lines of data off with TOP option on the DATA instruction:

open data nifty.xls
data(format=xls,org=columns,top=17,left=2,nolabels) 1 726 niftychange

Rule #1. Check your data.
Rule #2. Don't forget rule #1.

Dear Tom,
Thanks for your reply. I checked the data,made changes and got the result. Everything is fine, But in the out put the value of centered and uncentered R square were zero. also the in the same table does not contain the value of Y{1}. is it possible? I have attached the output file.
Prashant

You're looking at the output for a preliminary regression being used to get guess values. It has nothing at all to do with your final results.

Perfect, Tom. I am impressed by the forum.
I wanted to learn the interpretations of the output. I would like to know interpretation of output like what is the interpretation of values of mean, C, A, B and also MU, MSG etc..becuase when I try to compare them with Dueker results, I could not relate C, A, B etc of RATS and the Dueker. the interpretations in the paper also not clear. Could you please help or suggest me from where do I learn it? I also bought SWiTCHIng course but interpretation is still an issue for me.
Looking forward for your reply,
Prashant