Dear all,
I am trying to fit various GARCH models to my data. Unfortunately, when I add variables to the variance equation the constant of the variance equation becomes negative. I will give you an example of my problem.
The data in the model is as follows:
- 522 observations of weekly Bitcoin log returns (~10 year period)
- 522 observation of the log first-difference of the Google Search Volume Index (SVI) as a proxy for investor attention
As can be seen in the screenshot attached below, the constant is negative, thereby violating the non-negativity constraint. In addition, a negative variance does not make sense intuitively.
When I estimate, e.g., a AR(1)-GARCH(1,1) or when I include additional variables in the mean equation, the constant in the variance equation remains positive. However, only when I add additional variables to the variance equation, negative coefficients appear.
This same problem shows up when I estimate a similar model on the daily interval and even when I include other variables instead of the Google SVI in the variance equation.
On the one hand, I doubt whether this might not simply be due to a bad model specification for the data (i.e., for the BTC log returns). On the other hand, the coefficients are significant.
I have also tried (as suggested by my econometrics book Introductory Econometrics for Finance [Chris Brooks, 3rd]) to impose a constraint on the constant in the variance equation:
Unfortunately, when I add this constraint to the variance equation, I obtain the following:
Can one of you help me with these issues?
Kind regards,
Hessel Bouwman
I am trying to fit various GARCH models to my data. Unfortunately, when I add variables to the variance equation the constant of the variance equation becomes negative. I will give you an example of my problem.
The data in the model is as follows:
- 522 observations of weekly Bitcoin log returns (~10 year period)
- 522 observation of the log first-difference of the Google Search Volume Index (SVI) as a proxy for investor attention
As can be seen in the screenshot attached below, the constant is negative, thereby violating the non-negativity constraint. In addition, a negative variance does not make sense intuitively.
When I estimate, e.g., a AR(1)-GARCH(1,1) or when I include additional variables in the mean equation, the constant in the variance equation remains positive. However, only when I add additional variables to the variance equation, negative coefficients appear.
This same problem shows up when I estimate a similar model on the daily interval and even when I include other variables instead of the Google SVI in the variance equation.
On the one hand, I doubt whether this might not simply be due to a bad model specification for the data (i.e., for the BTC log returns). On the other hand, the coefficients are significant.
I have also tried (as suggested by my econometrics book Introductory Econometrics for Finance [Chris Brooks, 3rd]) to impose a constraint on the constant in the variance equation:
Code:
constraint 1 _b[HET:_cons] >= 0
Code:
Constraints invalid: not possible with test
Kind regards,
Hessel Bouwman
