I am modelling the dynamic conditional correlations of a couple of assets via DCC mgarch. I also have some exogenous variables that try to explain these correlations. Since my dependent variable is the correlation and this is of course bounded between -1 and 1. The exogenous variables are bounded between 0 and 100. How do I deal with this? I already used LS regression.
4
-
$\begingroup$ You could transform the response so that it's not negative and then fit a gamma or beta glm $\endgroup$– Robert LongCommented Jun 6, 2021 at 16:39
-
$\begingroup$ Thanks for your response sir, what is the rationale behind using a gamma or beta glm? $\endgroup$– AbderrahimCommented Jun 7, 2021 at 14:11
-
$\begingroup$ You're welcome. The beta and gamma distributions model positive values that are constrained within a range. $\endgroup$– Robert LongCommented Jun 7, 2021 at 14:37
-
$\begingroup$ Hi Robert, I have researched the beta glm and it indeed could be a suitable solution to my problem. However, my response is between -1 and 1 (correlation coefficient) and for beta regressions it has to be between (0, 1). What would be a suitable transformation for my response? $\endgroup$– AbderrahimCommented Jun 8, 2021 at 11:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Since you have a response variable that is bounded in $[-1,1]$, one approach is to transform this into $[0,1]$ and fit a beta generalised linear model. You can achieve such a transformation with:
$$f(y) = \frac{y+1}{2}$$
answered Jun 8, 2021 at 11:35