1
$\begingroup$

So I was trying to estimate the performance of a static hedge vs dynamic hedge in the electricity market and I came up with some weird findings. When I used the minimum variance hedge approach using regressed returns of spot to futures I was receiving an R-Square of around 0.05 (Hedge effectiveness also in this case) and a Hedge Ratio of 0.46, which seems extremely low, but when I regress the prices of spot to futures (both via OLS) I receive an R-Square of around 0.75 and a Hedge Ratio of 0.96, which makes a lot more sense to me since spot and futures are more or less moving at the same pattern. But if I remember correctly f.e. Hull (2012) sais to regress the returns.

My second question would be if it is correct to when using a Rolling-Beta approach (Hedge-Ratio = Beta) to calculate the hedge effectiveness, to estimate the series of Betas and then multiply it with the vector of the Futures prices. To get the hedge effectivness now I would calculate the vector of spot price - the beta vector multiplied with the vector of futures prices? I wasn't sure if this procedure is correct.

$\endgroup$
  • $\begingroup$ Probably better to split this into two actual questions $\endgroup$ – Ivan Apr 2 at 9:31
1
$\begingroup$

If you regress spots and futures prices you are likely to end up with a case of spurious correlations. Perhaps a cointegration analysis would be a better tool. This is because the time series may not be stationary. Returns are typically (more) stationary which is why regressing them is usually more sensible. But no guarantee there either.

That said, it is really not clear a priori what the correct answer is. Regression of returns may or may not be appropriate for the task at hand.

One specific point in any case is this: are you 100% sure your time series are synchronous ? If you sensibly expect two series to be highly correlated but they are in fact not based on collected data, then you may want to check that the sampling is done on the same basis. It is often the case with futures and spot that the “close” prices for example do not denote the same timestamp at all. So I would double check that in the first instance. If you can’t fix it, then look at longer horizon returns to smooth out the discrepancy (one hour lag over a day is a lot, but one hour over a week is less important).

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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