[R-SIG-Finance] performance attribution

[aamm]

Ben,

‘Performance attribution’ means different things to different people. There
are two different questions addressed under this term. 
One seeks an explanation of performance in terms of controllable investment
decisions. (“Interaction” is not a well-formed investment decision and,
thus, any well-formed investment-decision-attribution analysis will not have
‘Interaction” as one of its explanatory variables.) Decision attribution is
normally associated with the name Brinson and the most general and robust
implementation I know of is from Opturo. Opturo’s approach allows for
balanced funds, since it allows the user to input their customized decision
structure (employing, in any given order and structure, any parameter
selection, such as duration, convexity, cap, and bucket allocation, such as
asset, sector and/or industry, and currency decisions) rather than forcing
the analysis into predetermined and, thus, inappropriate evaluations. So it
can also address both equity and fixed alone and the decisions appropriate
to each. It also provides the necessary correlate of risk attribution since,
as modern portfolio theory insists, performance analysis without parallel
risk analysis can only be dangerously misleading. Finally, it is the only
system that correctly includes trades in its evaluation of the weighs and
returns that decision attribution requires. The common daily trade-inclusive
performance calculations employed by other systems are all variations on
Deitz calculations, which are all deeply flawed and can easily be shown to
produce absurd results. These absurdities show that even when the results do
not appear absurd to the intuition, they are always unreliable.
The other kind of performance attribution seeks an explanation of
performance in terms of uncontrollable market factors. I presume this is
part of what you are seeking since you use the word “exposure.” All these
models are difficult to significantly tinker with. Thus, their issue-level
exposures are usually provided as a coherent unit by providers such as
Barra, Northfield and APT. 
Specifically in regards to these market factor models, the other question
you should consider is whether you are interested in using them ex ante or
ex post. If you are considering using them ex post then be sure that the
weights that you employ are properly trade-adjusted. You do not want to be
attributing exposure to a factor that you traded out of. Again, Opturo is
the only package that I believe gets these trade-inclusive weights correct.
Others, it they adjust for trades at all, employ some version of Dietz
calculations, which are always flawed. Opturo allows you to take whatever
factor exposure models you employ and join them with its properly
trade-adjusted weights.
Most dangerous are models that conflate the investment-decision and the
market-factor approaches. Their mixed methodologies do not provide answers
to any clearly formulated economic questions. So you end up with numbers
that have names that sound informative but all these actual values are not
correct answers to the questions that their names imply or to any coherent
question at all.


--
View this message in context: http://r.789695.n4.nabble.com/performance-attribution-tp3519023p3545594.html
Sent from the Rmetrics mailing list archive at Nabble.com.