wilmott.com

Hello everyone;I was checking out the bloomberg terminals and I found out about the Hurst Exponent; I googled the subject and I found this interesting website:http://www.bearcave.com/misl/misl_tech/ ... /hurst/Now... I like the subject, but the estimation looks rather complicated, I'm not saying that it's impossible, but, what if I already have the ACF and PACF or any of the statistical tests for let's say farima's exponent, then I shouldn't bother about this number... right? Is there any particular advantage about this number? and if so, do you know of an easier way to calculate it?

I looked at the Hurst exponent for a while. The problem is unless you have a HUGE data set, estimates of the HE tend to have very large confidence bands. Try an MC simulation, estimating HEs on random walks and you will see what I mean.

A long time ago, I looked to Hurst coeff determination, and it was- and it is still, I presume - just bullshit: at that time, the number of past data used for computung the coeff was max 250, which is far to small to give any significative result (see also answer from JamesH83).

I have programmed Hurst methods like R/S and ADF in VBA, and I built a CI using bootstrap way. Like James said, it is difficult to get confidence band narrow due to limited data length. So it is hard to find something have significant long memory though the Hurst exponent is greater than 0.5. And I still not sure what I should believe. One thing I found interesting is that I tried three or four Hurst programs and I cannot find a pair which give the same results using the same data. cheers,Gary

"One thing I found interesting is that I tried three or four Hurst programs and I cannot find a pair which give the same results using the same data. "What programs / data are you using?