[R] Hurst Exponent Estimation

[Hans W Borchers]

Spencer Graves <spencer.graves <at> structuremonitoring.com> writes:
> 
>   Have you tried something like the following:
> 
> library(sos)
> H <- ???Hurst
> summary(H)
> H
> 
>        This identified 50 links in 15 packages, and displayed the 
> results in a table in a web browser with links to the best match in the 
> best package first.  The "installPackages" and "writeFindFn2xls" 
> functions are designed to produce an Excel file with a summary page that 
> can help you identify the package that seems to be most actively 
> maintained among the relevant packages, as explained in a vignette.
> 
>        Hope this helps.
>        Spencer Graves


Dear Spencer,

sometimes I am a bit worried about answers that simply redirect
the requestor to searching the Internet or R sites. It fuels the
impression that experience is not so much asked for nowadays.

Anyway, I looked into the 'fractal' package that comes up in your
search. The same applies here that I said before. For none of the
test series the result of computing H was nearly as accurate as
what 'pengFit' returned.

I really would be interested to hear from others about experiences
with financial times series.

Best, Hans Werner


> On 7/22/2010 12:25 AM, Hans W Borchers wrote:
> > Lorenzo Isella<lorenzo.isella<at>  gmail.com>  writes:
> >> Dear All,
> >> I am a novice when it comes to time-series analysis and at the moment I
> >> am actually interested in calculating the Hurst exponent of a time
> >> series.
> >
> > Some time ago I tested some of the classical chaotic time series
> > (such as the logistic map and others, no financial time series) with
> > available functions in R and Matlab. In my experience, Peng's method
> > (realized in R as fArma::pengFit) works reasonably reliable and is
> > more accurate than most others on these series.
> >
> > Unfortunately, the available R and Matlab implementations of the same
> > method -- and refering back to the same literature article -- can give
> > quite different results, with varying success for both sides.
> >
> > AFAIK, in TISEAN there is no function estimating the Hurst exponent.
> >
> > Regards
> > Hans Werner
> >