a***m
发帖数: 74 | 1 Greetings!
I use fminsearch and fmincon in matlab to estimate parameters for a GARCH
model. The solution is not reliable: for example different initial values
result in different solutions, sometimes the optimization wouldn't be
reached. Because I need to estimate GARCH model for different sets of data,
it is impossible for me to debug or tweak solve for each dataset. I wonder
if there are better approach tothe problem. Any suggestions are welcome.
Many thanks! | W*****k
发帖数: 158 | 2 the econometrics toolbox of Matlab has a specific function to calibrate
GARCH
,
【在 a***m 的大作中提到】
: Greetings!
: I use fminsearch and fmincon in matlab to estimate parameters for a GARCH
: model. The solution is not reliable: for example different initial values
: result in different solutions, sometimes the optimization wouldn't be
: reached. Because I need to estimate GARCH model for different sets of data,
: it is impossible for me to debug or tweak solve for each dataset. I wonder
: if there are better approach tothe problem. Any suggestions are welcome.
: Many thanks!
| a***m
发帖数: 74 | 3 thanks, I am trying the GARCH toolbox now.
From the manual, it uses fmincon function too. | c**a
发帖数: 316 | 4 for non-convex problems, i dont think there is any method w/ guarantee that
optimum solution will be found.
【在 a***m 的大作中提到】
: thanks, I am trying the GARCH toolbox now.
: From the manual, it uses fmincon function too.
| a***m
发帖数: 74 | 5 I am confused by the results --
I tried the GARCH toolbox in Matlab. Compared to the results using
fminsearch, which I provided the likelihood function, the estimated
parameters are close for some cases, but very different for others, although
they both seem in reasonable ranges.
Could the differences be caused by the search methods themselves? | W*****k
发帖数: 158 | 6 rI guess it depends on the choice of starting points.
although
【在 a***m 的大作中提到】
: I am confused by the results --
: I tried the GARCH toolbox in Matlab. Compared to the results using
: fminsearch, which I provided the likelihood function, the estimated
: parameters are close for some cases, but very different for others, although
: they both seem in reasonable ranges.
: Could the differences be caused by the search methods themselves?
|
|
