flat or discontinuous region encountered

[anastasia]

~~
Dear members,
I am trying to run a DCC-GARCH model on STATA 12 which looks like:
mgarch dcc (uk = L.uk L.us , arch(1) garch(1)) (singapore= L.singapore L.us , arch(1) garch(1)) ( france= L.france L.dlus , arch(1) garch(1))
and it works fine.
All the variables are stock return time series.
But when I add Germany and have
mgarch dcc (uk = L.uk L.us , arch(1) garch(1)) (singapore= L.singapore L.us , arch(1) garch(1)) ( france= L.france L.dlus , arch(1) garch(1)) (germany= L.germany L.us , arch(1) garch(1))
STATA gives me an error : “could not calculate numerical derivatives -- flat or discontinuous region encountered”
It would be of a great help if somebody knows what goes wrong here?
The time series for Germany do not have gaps or comma and visually look similar to other variables.
Thank you

[beyza oktay]

hi anastasia, did you solve your problem, because this is exactly what i have faced with stata... if you could solve it, maybe you can help me

[Joao Santos Silva]

Dear Both,

I never worked with this particular model, but convergence in Stata is sensitive to the scale of the variables. In my experience, you should scale your data so that the objective function takes values, say between 10 and 1000.

Of course, you can also help Stata by providing good starting values.

Best wishes,

Joao

[beyza oktay]

actually there was no problem when i did it before... I dont know why the program changed its mind

[Joao Santos Silva]

Beyza, Stata rarely changes its mind... Are you using the same computer? Did you update Stata? Are you using the same data? Did you sort the data or are you using any command that does a sort?

[beyza oktay]

i am trying to conduct an analysis about the spillover effects of US financial crisis on selected economies by using daily stock returns. My dataset consists all these indices numbers calculated as daily returns.

I used the command mgarch dcc ( re_us=L. re_us)( re_sp= L. re_sp L. re_us), arch(1) garch(1) one by one for all countries. re_us is the return for us and for example re_sp is the return for spain. but when i run this stata gives the error telling,

could not calculate numerical derivatives -- discontinuous region with missing values encountered
could not calculate numerical derivatives -- discontinuous region with missing values encountered
r(430);


This is my story... i did the same before it worked.. again it works well for all the countries except one :/

[beyza oktay]

One more thing after this aforementioned command. I try to find conditional variances for all the dates in the timeseries. I am using the command as " predict condvar*, variance"... When i run this, For some countries it gives 50, for some 100 and for other country, some other result from a whole of nearly3000 observation... I also dont know why is this happen and why it doesnt give the whole variances but instead stop at some point... (by the way, it doesnt give any error message )

[beyza oktay]

i solved everything by re arranging my data... it solved all the problems... thanks for interest

[Joao Santos Silva]

Thanks for letting us know. Yes, Stata is sensitive to these things.

Joao

[Farah Naz]

Mr. Beyza Oktay,
Would you please share how you had resolved the problem while running GARCH because I am facing the similar problem now.
I am unable to figure out why I am receiving message "could not calculate numerical derivatives
flat or discontinuous region encountered"

[Parul Gupta]

beyza oktay Could you please share your solution?

[spencer keene]

Does anyone know what beyza oktay means by "i solved everything by re arranging my data"? I am also having difficulties.

I get "flat or discontinuous regions" and "Hessian has become unstable or asymmetric" but there doesn't seem to be a pattern of commands and errors that let me know what the underlying problem is...

[Beri Parfait]

Does anyone know what beyza oktay means by "i solved everything by re arranging my data"? I am also having difficulties.

I get "flat or discontinuous regions" and "Hessian has become unstable or asymmetric" but there doesn't seem to be a pattern of commands and errors that let me know what the underlying problem is...

I am facing this same problem and don't know how to proceed, unfortunately.