There has been a lot of talk in the press following the credit crisis, Bear Stearns and Soc Gen about risk management and its role in our industry. Last month saw articles by Alan Greenspan and Nout Wellink talking about the need for banks and regulators to be more "forward-looking"; advocating more use of simulation rather than relying on historical calculation. This sounds ideal but there are some intrinsic issues here.
Although it has taken some time (arguably changes should have happened
in 1998 following the collapse of LTCM) the financial world is rightly
discussing how the industry should be managing its risks. The problem
is however much deeper than changing focus from an historical to a more
"forward-looking" approach.
Currently most banks, hedge and institutional funds use VaR as the
benchmark analytic for assessing risk exposure. Herein lies the first
issue. VaR is based on a normal Gaussian distribution which does not
reflect the reality of how the financial markets actually work. You may
have heard of the phrase "fat-tails" - meaning that markets appear to
display kurtosis risks and that events that should occur one in a
million (or higher) years under that model actually happen all too
frequently.
The second issue is that all this is based upon
historical data - and again the markets have a nasty habit of deviating
from past behaviour. In the LTCM crisis even the models built by the
greatest minds in finance (Scholes and Merton) were not able to predict
the intricate correlation of disparate markets following the Russian
bond default.
Thirdly is the problem of technology. All of the above require
massive processing power that only the largest organisations can really
afford. Most institutions systems grow organically from the
departmental level - and thus weaving together all of that data to give
you an enterprise-wide view of your risk exposures is by no means an
easy thing.
So can we solve this? Yes I think so - but not in the near future.
If we are to replace VaR as the most widely used analytic, you
first have to find a replacement, have it accepted over time by
practitioners and build systems to actually do it. There is no
acceptance of alternatives at the moment - the most compelling one that
I have seen is Benoit Mandelbrot's Fractal Theory of the markets (a view shared by the maverick option trader and author Nassim Taleb). If we
are to move towards prediction then you also need an accepted way of
(mathematically) predicting the future market movements. Many banks
have been trying to incorporate neural networks into their
infrastructure in a bid to understand the biological complexity of the
markets - but we have not solved this yet nor I think are ever likely
to. Simulation is a great concept; knowing what and how to simulate
future events is the problem. Lastly more focus (or rather money) needs
to be put into Risk Management departments and systems that in some
areas have been neglected in recent years and the current debate will
surely ensure this happens. Once the good times start to roll again
however, you can be certain that this will again take a back seat.
So to answer the question of who should be doing this, yes all market participants will
need to change once we find the answers. But it is the banks (the
sell-side) who need it the most; the buy-side is by its very nature a
much more cautious beast. What is needed most to my mind is actually
agreement by all parties about what we need to do and where we need to
go to. My advice to you is to await the Corrigan report (due in July)
to see what the industry will do to face this challenge after the
recent credit crisis.
|
|
||||||||
Can banks be "forward-looking" in risk management
Comments
Re: Can banks be "forward-looking" in risk management
by
JBM
on Mon 05 May 2008 09:09 BST | Profile | Permanent Link
Sean,
An intermediate step on the way to risk nirvana ought perhaps be to combine VaR with scenario testing and some form of stress testing. For some investment managers this of itself is a considerable undertaking however one which is not without merit. The application of scenario testing goes some way to eliminating the time bias which may be inherent in VaR analysis (i.e. the time period over which the VaR inputs are gathered covers a historically low-vol period, skewing the output and leading to a misappreciation of risk).At the same time, stress testing where the variables are set by the risk committee to account for events which may be on the horizon can act as a further backstop. Clearly not a flawless proposal but one which is implementable in the short to medium term while we wait for the quants and statisticians to develop the next generation of risk measures and systems. Trackbacks
TrackBack URL: |
||||||||
|
|
||||||||

