Risk Management Technology in Financial Services

Elsevier UK Jobcode:RTF Chapter:CH13-H8304 22-3-2007 5:12p.m. Page:255 Trimsize:165×234MM

Fonts used:Ocean Sans & Sabon Margins:Top:42pt Gutter:54pt Font Size:10/12 Text Width:30p6 Depth:47 Lines

Technology for time management, high frequency financial data and high impact events 255

Critics say, not without reason, that as experience is gained with the analysis

of high frequency and low frequency risk events, we will find out that the normal

distribution hypothesis does not hold. Rather, the distributions will be leptokyrtotic,

or platokyrtotic, the latter being encountered very frequently in fuzzy engineering

(see Chapter 9). A characteristic of leptokyrtotic distributions is the fat tail in their

right leg, which indicates that certain events repeat themselves more frequently than

expected.

In the loss distribution we have just examined, with LF/HI risk events are at the right

tail. This has a certain similitude with the 1996 Market Risk Amendment’s one-tailed

distribution. With value at risk (VAR), for example, the regulatory 99 per cent level

of confidence represents maximum amount of losses leaving out of risk measurements

1 per cent of events – which is too coarse.

Quite often, our ability to handle LF/HI items and their pattern statistically is

limited by lack of data, hence the need for data analysis over several decades in

conjunction with extreme value theory. We are not there yet, because the content of

our databases is not what it should be. Alternatively, we can explore high frequency

financial by looking for ‘anomalies’, which in reality may be outliers.

As work along the HF/LI and LF/HI lines starts gaining momentum, the databases

get richer, and experience on the analytics of risk events accumulates, we will be

confronted with a different type of challenge. The more astute analysts will want to

know why the worst losses are due to risks whose behavior is non-linear in terms of:



Size,



Frequency, and



Severity.

These will become very interesting studies giving the financial institutions which

undertake them a competitive edge. For the time being, however, the priority is to

establish a firm basis for data collection, including appropriate emphasis on HFFD;

and for frequency-and-impact studies which allow analysing real life events, like

selected types of risk, by means of increasingly more powerful mathematical tools –

and of sophisticated enterprise architectures, whose theme was treated in Chapter 4.

13.6 Prerequisites to a study of high frequency events

As explained in the previous section, some risks have high frequency but low impact,

while others have low probabilities but potentially large financial impact. An example

of a high impact operational risk event is Andersen risk,

6

or the risk of deception.

The risk of deception can have one or more origins:



Incompetence,



Repeated errors,



Lack of transparency, and



Outright conflict of interest.

Whether we talk of market risk, credit risk, legal risk, technological risk or other

types, low frequency events require a methodology for a priori experimental analysis