[R-SIG-Finance] Option valuation for arbitrary distribution using monte carlo simulation

[Brian G. Peterson]

On Thu, 2011-11-24 at 16:19 +0100, Joachim Breit wrote:
> Thank you for the interesting link.
> 
> I agree that you can use a (fit of a) stable distribution for sampling
> purposes. But please: Why not simply use the raw data and sample from 
> that? What could be a better starting point? Again, it is clear that
> you cannot use the raw return series as it comes off your data feed;
> there is a need for adjusting. But there cannot be a better fit to the
> raw data than the raw data itself... 

Of course you start with sampling from your data, but if all you do is
sample from the data, with no 'noise', then all you are doing is
smoothing out the prior observations.  This may lead you to a false
sense of security, and give you an incorrect idea of how likely a fat
tailed event is.  So by adding noise from a stable or other fat tailed
distribution (skewed Student's T is also popular), you are believing
that your prior data is both stationary and fully representative.

Regards,

  - Brian

-- 
Brian G. Peterson
http://braverock.com/brian/
Ph: 773-459-4973
IM: bgpbraverock