[R-SIG-Finance] tawny: deriving

[Patrick Burns]

Rohan,

I'm a little confused about what you really
want, but perhaps 'var.shrink.eqcor' from
the 'BurStFin' package can help.

The package is not yet on CRAN (but it's
getting closer), get it from:

install.packages('BurStFin',
    repos="http://www.burns-stat.com/R")

(which doesn't work under 2.14.x because it
doesn't have a namespace).

'var.shrink.eqcor' returns the shrinkage that
it uses as an attribute.

To get estimates that are trying to mimic the
original Ledoit-Wolf code, you need to set the
weights to be all equal and the 'compatible'
argument to TRUE.

The shrinkage parameter is slightly different
from that estimated in 'tawny'.  I have a sample
size of one that suggests I get the value that
Ledoit and Wolf's Matlab code gets.  I couldn't
see where the difference comes from in the little
time I spent trying.

Pat

On 28/01/2012 05:42, Rohan Sadler wrote:
> Hi All,
>
> newbie question: I am interested in acquiring the value of Ledoit&  Wolf's
> (2003) optimal shrinkage estimate from the tawny package (or other). Having
> read the paper, and then had a look at tawny, then I can derive the
> shrinkage intensity (intensity.shrinkage, see below code). However, as the
> value of the intensity is larger than one then I wanted to confirm whether
> the L&W shrinkage parameter  is not tawny's intensity/T, where T is the
> length of the time series (i.e., rows of the returns matrix). That is,
> confirm whether tawny's shrinkage.intensity function returns what L&W
> termed as the 'optimal constant' kappa (p611), rather than L&W's optimal
> shrinkage intensity alpha.
>
>    sample.vcov<-cov.sample(returns.zoo)
>    F<- cov.prior.cc(sample.vcov)
>    k<- shrinkage.intensity(returns.zoo, F, sample.vcov)
>    shrunk.vcov<-cov.shrink(returns.zoo,prior.fun = cov.prior.cc)
>
> The output from Brian's site:
> http://nurometic.com/quantitative-finance/tawny/exploring-shrinkage-estimation
> doesn't
> seem
> to hold for the latest R/tawny version. It would be nice to acquire that
> information somehow, perhaps embedded in one of the shrunk objects, or at
> least one or two more lines of documentation in the associated help pages
> (apologies for being demanding!!). Packages fPortfolio and corpcor didn't
> seem to provide any of this information either for the newbie crowd.
>
>> S.hat<- cov.shrink(h)
> Got intensity k = 146.8918 and coefficient d = 0.9792788
>
>
> Regards
> Rohan Sadler
>

-- 
Patrick Burns
patrick at burns-stat.com
http://www.burns-stat.com
http://www.portfolioprobe.com/blog
twitter: @portfolioprobe