The failure of financial institutions to manage risk and the role this has
played in creating today's economic crises has created awareness of the need
for a new risk-management approach. This may well become the new Holy Grail
of financial economics. There are two essential constituents: first,
developing a more diverse mindset regarding financial risk; and second,
applying more powerful techniques to capture financial reality than risk
managers have used in the past.
A new method for risk management does not mean that all elements of the
present system must be replaced or that all the discredited financial
instruments they spawned should be discontinued. Discrimination is needed:
Many existing techniques have validity, but only in certain market
conditions; and innovative financial products can contribute to wealth
creation, but only if they are used judiciously.
A new approach must start by understanding the reasons for the failure of
risk-management systems. Why did risk models grossly understate risks? Why
did managements fail to understand this? And why did regulators fail to
provide effective checks and balances against the banks' risk-management
mistakes?
Financial risks were understated partly because the world economy enjoyed a
long period of moderate growth and low financial volatility and this “great
moderation” was wrongly assumed to be a permanent benefit of prudent central
banking.
But there was a more fundamental error. Risk-management models used in banks
were generally based on the simplified assumption that markets fluctuated
randomly following a “normal” statistical distribution. This implied very
low probabilities of extreme losses, ignoring financial history. Even
complex models used to price more esoteric products such as credit
derivatives were based on the same weak foundations — for example, using
simulations based on random numbers, which in turn assumed a normal
distribution.
Many of these simplistic models grew out of Modern Portfolio Theory (MPT),
which was developed half a century ago and proved highly seductive due to
its simplicity and ease of computation. While MPT was useful in stable
markets, it failed in extreme market conditions. More advanced techniques
were developed to anticipate successions of extreme events, but these were
largely ignored. Even hedge funds and investment banks that used
sophisticated techniques for their trading strategies still relied on
“normal” distributions for their risk management reports. This produced
over-optimistic risk-return profiles and may have contributed to the
excessive leverage employed.
Risk managers proved as inadequate as their models for reasons familiar from
behavioural economics: emphasis was placed on precise-sounding numbers from
quantitative models, while non-quantitative judgments based on experience
were ignored. And although risk models produced useful numbers, managements
often failed to interpret these objectively, instead applying wishful
thinking. For example, extreme events such as the 1987 stock market crash,
or the Long-Term Capital Management collapse in 1998, which clearly
invalidated the assumptions of conventional risk models — but these were
either disregarded or rationalised later as “exogenous shocks”.
Finally, there was the failure of external regulators and rating agencies to
monitor the banks' risk-controls. Despite deficiencies in bank management,
some of the problems could have been averted with stronger checks and
balances from the regulatory bodies charged to assess risk. But rating
agencies undervalued risk when assessing new products. And regulators
ignored the dangers of macroeconomic contagion, as well as failing to apply
any independent checks to the risk models used by banks.
One driving force behind these policy errors was competition between
governments to attract financial businesses with “light-touch” regulations.
Another was excessive faith among regulators in investment bankers' skills.
Risk-management methods developed long ago by investment banks, for example
the normally distributed Value at Risk (VAR) models, were treated by
regulators as a “gold standard” — long after sophisticated financial
institutions had recognised their limitations and moved on.
What reforms in risk management are needed to overcome these limitations? The
key is to combine quantitative statistical methods with qualitative, or
judgmental, assessments — and then to use a multi-prong strategy, involving
a wide variety of methodologically independent approaches, to reach
comprehensive conclusions about financial risk. When several approaches
highlight similar risks, these can be flagged. Because financial uncertainty
evolves in many unexpected ways, no single model will work all the time and
new approaches will be required.
A good guiding principle is a “none and all” mindset. This means that no
approach is ever used exclusively and all approaches are respected for what
they may contribute. A simple macroeconomic example might be the ratio of
debt to GDP. The growth of this ratio to historic highs in many countries
was not on its own a conclusive reason to tighten regulation, but simply
ignoring this warning signal violated the rule that “all” models should be
considered in judging financial risks. Such qualitative judgments may create
“soft” risk management boundaries, but nonetheless are important in
detecting the irrational and subjective elements in market behaviour. To
avoid future disasters, regulators and managers must use such “soft”
qualitative approaches alongside “hard” statistical tools.
Equally urgently, the financial world must embrace new mathematical techniques
that capture market reality better than standard models.
One such approach, developed at Imperial College by Nicos Christofides, uses
techniques from control engineering, including neural networks and dynamic
programming to produce mathematically complex but highly intuitive results.
This approach is visualised through a State Transition Graph (STG) — a path
of possible future price movements with probabilities assigned to each
outcome. An STG can be developed for any combination of traded assets and
based on long periods of historical data, including several market
dislocations. STGs produce probability distributions of future price moves,
which, in contrast to standard models, do not assume continuous or
“normally” distributed market movements.
By mixing continuous market behaviour with discontinuous market disruptions,
STG models can produce realistic probabilities of large “shocks”. They allow
illiquidity risks to be considered and produce natural clusters of volatile
periods. They reflect precisely the high-risk conditions that can arise in
turbulent markets, but which standard models ignore. The success of STG
models has been demonstrated by their performance in the credit crunch, as
well as in retrospective testing on the 1987 stock market, with the
statistical analysis restricted to data before each event.
The STG is only one of several innovative approaches developed in recent years
by academics and sophisticated financial institutions, but largely
disregarded by bank managements and regulators. Instead, traditional
risk-management models have continued to be employed, even though they are
known not to work. To avoid a recurrence of the present crisis, this
narrow-minded attitude must change. Risk management must be reformed not by
seeking a new “right” answer to replace discredited VAR models, but by
creating a broader mindset — what we call the “none and all” approach.
No risk model should be relied on completely and all evidence should be taken
into account. Without such an inclusive approach, even the most powerful
models will fail us, at least at certain times. But with a more
comprehensive understanding of risk, involving both advanced quantitative
models and “soft” qualitative judgments, the chances of avoiding future
financial catastrophes could be much improved.
- Shahid Chaudhri and Richard Griffiths are founding partners of
Innovation4Now, a risk-management consultancy
