2
$\begingroup$

Let $r_{s, t}$ and $r_{f, t}$ be the return rates of the spot and futures of a commodity at time $t$. The hedging ratio based on variance minimization is calculated by finding the minimum of the variance of the combined returns:

$$\beta_t = \rho_{sf, t} \frac{\sigma_{s, t}}{\sigma_{f, t}},$$

where $\rho_{sf, t}$ is the time-varying correlation and $\sigma_{s, t}$ and $\sigma_{f, t}$ are the corresponding time-varying volatility of the spot and future returns respectively. From the several papers that I went through, the volatility measures are calculated using different GARCH-type models and the filtered standardized residuals are used to estimate various Copula models. My question is how the estimated static Copula models are then transformed to the time-varying correlation that is later used to calculate the hedge ratio? I would really be thankful for any kind of guidance.

$\endgroup$
0

1 Answer 1

4
$\begingroup$

Using a static copula model implies $\rho_{s,f,t}\equiv\rho_{s,f}$. In such case fitting a copula model to obtain $\rho_{s,f}$ is an overkill, since it can be estimated very simply by the empirical correlation of the two standardized residual series from the two GARCH models. Of course, a availability of the joint distribution via a copula-GARCH model facilitates all kinds of interesting calculations, so the model may well be worth fitting, just not for estimating $\rho_{s,f}$ alone.

If you want time-varying correlation, you may be tempted to consider using BEKK-GARCH or DCC-GARCH models; this is what many authors do. However, the models seem to be seriously flawed – except for the case of diagonal BEKK-GARCH; see Caporin & McAleer (2013), McAleer (2019a), McAleer (2019b), Allen & McAleer (2018). Other alternatives are time-varying copula GARCH and GO-GARCH, among other, though I am not sure how sound they are theoretically. In any case, the latter two as well as DCC-GARCH are available in the rmgarch package in R should you decide to try them out.

References:

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.