### MATLAB >> GARCH estimation

I am an Italian student facing my thesis about Dynamic Conditional
Correlation. I use Matlab 7 and I would like to estimate different
univariate GARCH model and then with the standardized residuals for
each of the univariate series to model the correlation. For this
purpose I need to have the scores of the log-likelihood function of
the univariate GARCH models. How can I obtain them?
Where can I go to change the settings and obtain the scores?
Have you some suggestions for my problem?

### MATLAB >> GARCH estimation

Ciao Francesco
have you developed by yourself the GARCH models or are you using The
Financial toolbox and/or the GARCH toolbox under MATLAB?

Let me know
-Francesca

### MATLAB >> GARCH estimation

Hello Citta,

The best results I have obtained for this problem type if with TOMLAB
/KNITRO using MAD for the first derivatives (if not available) and a
BHHH approximation of the Hessian. The best would be if you had
point precision.

An interior point solver is usually needed since you have
singularities (or extreme nonlinearity) close to the bounds. An SQP
solver would bounce into these more easily.

If you are using the UCSD toolbox (Kevin's) I can send you some

Best wishes, Marcus
Tomlab Optimization Inc.
< http://tomopt.com >

### MATLAB >> GARCH estimation

Hi, all

The econometrics toolbox estimates GARCH models however I could find anywhere details about the estimation, like:

1. The solver (and algorithm) used (I am guessing fmincon)
2. The initial value (the value in Sigmas(1)) for the conditional variance equation
3. The lower and upper bounds constraints used in the estimation procedure

Can anyboby from mathworks provide these details ?

### MATLAB >> GARCH estimation

Manthos,

This is all dicusssed in the documentation. Please see the section entitled "Estimation" in the Uer's Guide:

http://www.mathworks.com/access/helpdesk/help/toolbox/econ/f1-44824.html

-Rick

```Hi All,

I recently bought an intel Quad core cpu and I am looking for a way to make the cpu work at full capacity for doing GARCH calculation in a loop 1000 times.
So simplifying my program, I do something like the following

For n=1:1000
A=prices(n:n+1000,1);
spec = garchset(mpla mpla);
[coeff,errors,LLF,innovations,sigmas] = garchfit(spec, A);

end
Instead of using 25% of the cpu I want to use 90% (or more).

Any help will be greatly appreciated.

BR

Chris
```

```Hello everyone
I followed the GARCH toolbox documentation in Matlab using the data shipped with it. I then applied the pre-estimation analysis to a different dataset, and do not see significant correlation in the standard deviation. The dataset are daily peak wholesale power prices at a particular hub. What should be my next step? Here are the results:

>> [H,pValue,Stat,CriticalValue] = ...
lbqtest(retdata-mean(retdata),[10 15 20]',0.05);
>> [H  pValue  Stat  CriticalValue]

ans =

0    0.3408   11.2175   18.3070
0    0.3852   15.9523   24.9958
0    0.1514   26.4511   31.4104

>> [H,pValue,Stat,CriticalValue] = ...
lbqtest((retdata-mean(retdata)).^2,[10 15 20]',0.05);
>> [H,pValue,Stat,CriticalValue] = ...

>> [H  pValue  Stat  CriticalValue]

ans =

0    0.9056    4.7759   18.3070
0    0.7469   11.0802   24.9958
0    0.8290   14.0286   31.4104

>> [H,pValue,Stat,CriticalValue] = ...
archtest(retdata-mean(retdata),[10 15 20]',0.05);
[H  pValue  Stat  CriticalValue]

ans =

0    0.1910   13.6216   18.3070
0    0.3451   16.5722   24.9958
0    0.7166   16.0000   31.4104
```

```Hi,

I want to set up a bivariate GARCH model. Since there's a standard
GARCH toolbox in matlab, I think I might make it by adjusting the
univariate garchfit file.  Do you think it is feasible?

Also, I'm reading through the garchfit model right now. In the
optimization part, fmincon.m calls the garchllfn.m function
(log-likelihood function for normal distribution), then garchllfn.m
calls a function named as garchllf.

But I couldn't find a function named as  garchllf.m, instead I found a
mex-file named as garchllf, which can't be edited.

Anybody know how I can read and modify this file.

Thanks.

Best,
Norah

```

```Hello everybody!
Do you know how to get the variance-covariance matrix by
ML-estimation of the GARCH(1,1)-process?
I've used garchfit, but this function gives as result in Summary only
the dimension of the var-covar matrix. For my purpose (confidence
interval for Value at Risk based on GARCH(1,1)) i need this matrix
numerically.
Thanks.
```