[R-SIG-Finance] Portfolio Value at Risk - A conceptual problem

[Brian G. Peterson]

On 10/20/2010 08:24 AM, Arun.stat wrote:
>
> Dear Amy, assuming you have sufficient amount of historical data, you should
> calculate return collectively, because calculating return separately will
> destroy the correlation pattern implicit within the historical quotes
> therefore you may not address properly the diversification effect in your
> VaR figure.
>
> Instead of directly valuing portfolio (and then portfolio return) based on
> historical quotes I would prefer to find out the historical realized return
> (may be logarithmic or percentage) and construct a hypothetical price
> distribution (by taking antilog or something like) for next day (assuming
> VaR horizon is 1 day) and then find the possible distribution of portfolio
> values for next day. This always makes sense as price series is never
> stationary/stable but returns are. Here the fact is to find answer like,
> previously on some date I got that return therefore what if same return gets
> realized from 1 day holding. This inference you can never make had you taken
> raw price because 1 year back price may be 50pt lower than current level,
> hence in a single day you can never achieve that level (ofcourse assuming
> tomorrow another layman brother would not fall on your head!) of price but
> with return you can.

I prefer to have all the individual returns so that I can also do 
component risk contribution.  If you simply use univariate portfolio 
returns, you lose the interdependence between the assets, and can't find 
out the *contribution* to the overall portfolio risk, only the result. 
Component risk measures sum to the univariate portfolio risk, so you get 
that too.

# e.g.
######

require(PerformanceAnalytics)
data(edhec)
VaR(edhec,portfolio_method="component")
ES(edhec,portfolio_method="component")

######

Weights, of course, may also be applied.

Cheers,

    - Brian

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