Before leaving measurement, note that an aspect of it is reviewing the outcome.
The point of serious specific risk is that you cannot measure it. The type that is
really a problem – the rare, unforeseen event – has probably not happened before
and so we have no glimmering of an idea what its probability is.
Financial risk
Indicators of financial risk such as gearing or interest cover or the Z-score (see
below) can be measured but it is a mistake to imagine that, in most cases, this
results in a percentage probability of a particular risk. Rather, however sophisticated
the analysis, the best outcome of measuring is likely to be just an indication of high
or low risk. The fallacies of accurate risk assessment have been written about by
Nassim Taleb, as discussed earlier, and are worth considering in more detail.
The problem with assessing risk
Perhaps the biggest problem is that many approaches to risk, particularly those
that use financial instruments, are built on an assumption that the frequency of
occurrence of events follows a normal distribution (also described as a Gaussian
distribution). This is the familiar bell-shaped graph, of the type shown in Figure
6.1, and is used because there are well-established mathematical techniques for
calculation using this sort of distribution of probabilities.
If the outcome of an investment decision is not checked against the origi-
nal proposal then we have no way of learning and correcting errors for the
future. If it is clear that a business area tends to overstate income from invest-
ments or understate the cost of investment then it is important either to find why
this is and adjust the techniques that are used or else to put through a central
adjustment to their figures to account for the risk of error.
Figure 6.1 The normal distribution curve
The application of this assumption ranges from the calculation of variations
in stockmarket prices to the likelihood of severe weather events and its use
is perfectly understandable. When I studied engineering we often made such
assumptions, because without simplifying complex physical systems they were
impossible to analyse and calculate so we approximated. The engineer then
tests the results of their calculations in a variety of circumstances and, having
proven that they work in stated conditions, uses the method in future problems of
that type.
The issue with applying this approximation technique in business is that there
is a wealth of evidence to show that the probability distribution of many of these
events is not Gaussian. In particular, we find that extreme events occur far more
often than the Gaussian distribution predicts. Looking at Figure 6.1, it shows that
the bell shape almost touches the ‘base’ within a short range of outcomes. Where
it seems to do so, the probability of occurrence of whatever we are measuring
is very low indeed. Unfortunately, in the worlds of economics and business that
is misleading.
Consider Figure 6.2. This has the same probability distribution as above
except that there is a greater chance of a negative outcome and then, suddenly,
there is quite a high probability of a particularly negative outcome. Suppose this
is a representation of the performance outcomes for an internet business. Most of
the probability is well represented by a normal probability distribution, except that
there is, say, a 15% chance of a catastrophic software collapse that will lead to
a disastrous outcome. In such circumstances it is not particularly helpful to think
only about the risks that are described by the normal probability distribution and
to ignore this fairly low but very present risk.
0–1–2–3–4 1 2 3

Get Criticial Finance Issues: Financial Times Briefing now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.