The RATS Software Forum

Dear Listers,

I am trying to run the oil_stock_msvar_irf2.RPF program I have got always the same following error:

## SR3. Tried to Use Series Number 45844208, Only 5 Are Available
The Error Occurred At Location 531, Line 72 of MSSYSREGRESSION

Would you like please to help me with this.

Kind regards,
N.K

That's not the data file that the program uses.

Thank you for your reply but I used the excel file and I got the same error message.

1. It has nothing to do with the data file. However, I couldn't run a program which references a data file that I don't have.
2. Use a newer version of @MSSYSREGRESSION.
3. A MS model is completely inappropriate to your data. There are definitely multiple regimes, but there's nothing "Markov" about them---there's a 30 year period in the middle of the data where the oil prices are controlled.

Many Thanks Tom.

It is not my data basis, just I tried to replicate the estimation of a published paper (Energy Economics 49 (2015) 317–327) with their data basis and their program.

N.K

I have an e-mail in to the authors with some questions about what they did. At any rate, you need a newer version of MSSYSREGRESSION.

Dear Tom,

I am so grateful.

Did you mean the last version of MSSYSREGRESSION the attached file.

Kind regards and many thanks,
N.K

Yes.

A cleaned up version of the program is now posted at https://estima.com/forum/viewtopic.php?f=8&t=2536. As in the paper, this finds two regimes, one where the price of oil doesn't change, one where it does. As I said above, it's not clear that a "Markov" assumption is reasonable for that, since for about 30 years, the price of (U.S.) oil was controlled by law.

Thanks a lot, Tom.

Dear Tom thank you for your valuable efforts.

I have a question:

Why the error:

## SR3. Tried to Use Series Number 70201344, Only 9 Are Available
The Error Occurred At Location 531, Line 72 of MSSYSREGRESSION

repeated after the following part of program

@mssysregression(states=2,switch=ch)
# dlsp dloil
# dlsp{1 to lag} dloil{1 to lag} constant ect{1}

Thanks,
N.K

You're using an outdated version of the program. The whole MS suite of procedures has changed.

Dear Tom,

How can test for the presence of Markov assumption in time series?

Thanks,
N.K

There really isn't a way to do that. Tests for n+1 vs n regimes are very complicated and generally not worth the effort---people typically use information criteria to compare those, though that requires that you can actually fit an n+1 regime model, which is often quite a task if n regimes are adequate as the n+1 regime model isn't identified. And I'm not sure I've ever seen a test for Markov vs some other form of switching.

Many thanks Tom.

Actually, I just found this paper that developed a new test for the Markov property in time series. A Gauss code to implement
this test is available from the authors.

https://hong.economics.cornell.edu/pape ... Series.pdf

Thanks a lot Tom,
N.K