INduStRy cOmmENtARy
A MODEL DEFENSE
Though quantitative risk models have come under attack
during the crisis, they are still very effective —
particularly when supplemented by human judgment.
ometimes I feel that the only way a risk
manager gets in the paper is to claim
he or she knew the problem all
along, but was ignored because everyone else was blinded by greed.
My reaction is to wonder why, if
the firm was headed for disaster and no one
listened to you, you didn’t quit.
Instead of risk managers, we hear from
two kinds of non-practitioner experts on
the subject.
The first criticizes risk management
models for leaving out some important
factors: human behavior, parameter
uncertainty, long-term data, changes
in markets, fat tails, bubbles, fractal
dimensions or whatever else the
critic likes to study. The second attacks the very idea of quantitative
risk models.
I don’t know of a prominent
person in either camp who
has changed opinions
significantly over the
last 15 years, which I
find suspicious. You
have to be awfully
smart to have zero
surprises over those
volatile times, and if
you’re that smart, you
should be awfully rich.
Stupid as we quant risk
S
managers are, we do learn.
The first set of critics dream of the ubermodel
that never fails. It just has to be bigger, more com-
plex, incorporate more dimensions. It’s the mod-
ern form of the communist dream, running the
economy with math instead of people.
Everyone entrusted to manage the risk
of real businesses gave up the ubermodel
dream after the crash of 1987, the first di-
saster quants were well placed to observe.
The surprise was not that you could get a
12 standard deviation move in one mar-
ket; quants knew all about fat tails. The
surprise was that once things moved 3
standard deviations, all bets were off in
all markets, not just the affected one.
You couldn’t count on trading, on ar-
bitrage relationships being enforced
or even on good information.
Financial models use mar-
ket prices as inputs, as-
sume certain economic
relationships and output
suggested trades. If you
can’t get good prices
and the relationships
don’t hold and you
can’t trade anyway,
your models are as
useful as a laptop com-
puter to an ocean pas-
senger washed overboard
in a storm.
