[R-SIG-Finance] DEoptim MSGARCH

[Bastian Offermann]

Am 2/21/2014 3:16 PM, schrieb Brian G. Peterson:
> On 02/21/2014 02:06 PM, Bastian Offermann wrote:
>> Dear all,
>> there is a JSS paper on DEoptim including a Markov Switching GARCH
>> example . Since the optimizer can only handle box constraints, how would
>> the standard covariance stationarity condition alpha+beta < 1 in a
>> simple GARCH(1,1) be included here?
>>
>> My guess is sth like
>>
>> foo = function(x) {
>>
>> ...
>>
>> loglikelihood = ...
>>
>> if(alpha+beta >= 1) loglikelihood = loglikelihood + penalty
>>
>> }
>
>
> I'm not entirely sure your question makes sense.
>
> DEoptim in that case was used to calculate a probability of one Markov 
> state over another.
>
> There are lots of ways to calculate regime switching GARCH. None of 
> the ones I can think of assume covariance stationarity.  These are 
> different regimes, so different parameters apply to the two (or more) 
> regimes.
>
My question is more general: How would one take linear inequality 
constraints into account using DEoptim since the DEoptim function only 
allows box constraints? Couldnt find an optimizer so far that can handle 
box and linear inequality constraints.
Thanks!