SCENARIO GENERATION FOR OPERATIONAL RISK

• Date: 01 July 2013
• Published date: 01 July 2013
• Author: Sovan Mitra
• DOI: http://doi.org/10.1002/isaf.1341

SCENARIO GENERATION FOR OPERATIONAL RISK

SOVAN MITRA*

Glasgow Caledonian University, Glasgow, UK

SUMMARY

Operational risk is an increasingly important area of risk management. Scenarios are an important modelling tool in

operational risk management as alternative viable methods may not exist. This can be due to challenging

modelling, data and implementation issues, and other methods fail to take into account expert information. The

use of scenarios has been recommended by regulators; however, scenarios can be unreliable, unrealistic and fail

to take into account quantitative data. These problems have also been identified by regulators such as Basel, and

presently little literature exists on addressing the problem of generating scenarios for operational risk.

In this paper we propose a method for generating operational risk scenarios. We employ the method of cluster

analysis to generate scenarios that enable one to combine expert opinion scenarios with quantitative operational

risk data. We show that this scenario generation method leads to significantly improved scenarios and significant

advantages for operational risk applications. In particular for operational risk modelling, our method leads to

resolving the key problem of combining two sources of information without eliminating the information content

gained from expert opinions, tractable computational implementation for operational risk modelling, improved

stress testing, what-if analyses and the ability to apply our method to a wide range of quantitative operational risk

data (including multivariate distributions). We conduct numerical experiments on our method to demonstrate and

validate its performance and compare it against scenarios generated from statistical property matching for

comparison. Copyright © 2013 John Wiley & Sons, Ltd.

Keywords: operational risk; decision analysis; scenarios; advanced measurement approach; cluster analysis;

method of moments; property matching

1. INTRODUCTION AND OUTLINE TO PAPER

Operational risk has become increasingly important, with companies and regulators recognizing that

adequate risk management must incorporate it. Additionally, some financial commentators have argued

that the global credit crunch and other significant industry losses have been partly attributed to

operational risk; for example, long-term capital management (Chaudhury, 2010). Despite the importance

of operational risk, the research literature is scarce, and this has also significantly disadvantaged

practitioner s from effectively man aging operati onal risk.

Scenarios are a recognized (Loader, 2002), well-established approach to operational risk manage-

ment and are categorized as an advanced measurement approach (Loader, 2002). Scenarios are

particularly important in operational risk because alternative methods would not be suitable, in

particular due to modelling, implementation and data issues. Hence, operational risk scenarios are

formed from expert opinion (Peters et al., 2009) but also contain information that would not be

obtained from other sources of data; for example, quantitative data. For instance, expert opinion data

contain forward-looking information that is not available in quantitative data.

* Correspondence to: Sovan Mitra, Glasgow Caledonian University, Glasgow, UK. E-mail: sovan.mitra@gcu.ac.uk

Copyright © 2013 John Wiley & Sons, Ltd.

INTELLIGENT SYSTEMS IN ACCOUNTING, FINANCE AND MANAGEMENT

Intell. Sys. Acc. Fin. Mgmt. 20, 163–187 (2013)

Published online 28 July 2013 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/isaf.1341

A key problem with expert opinion operational risk scenarios is that they are fundamentally based on

opinion (Peters et al., 2009) and do not take into account quantitative data. Consequently, scenarios can

become unrealistic, unreliable and fail to take into account important information. This has also been

observed by regulators such as Basel (Chaudhury, 2010). Moreover, generating scenarios purely by

objective methods (e.g. analytical or computational methods) is generally not feasible since operational

risk scenarios are typically utilized when objective methods are not feasible. Additionally, scenarios

contain expert information that cannot be easily incorporated into other methods.

The generation of operational risk scenarios that takes into account expert opinion information,

quantitative operational risk data and does not impose significant modelling, implementation and data

requirements is therefore a key issue of interest (Peters et al., 2009). However, such problems are

nontrivial (to be discussed in more detail later). In particular, there are challenges in merging data from

operational risk probability distributions with expert opinion scenarios and combining data without loss

of expert opinion information.

Presently, there is little literature on methods for generating operational risk scenarios using both

expert opinion scenarios and quantitative operational risk data. The closest method available is

statistical property matching, but this is a method from stochastic programming and is not typically

used in operational risk scenario generation. Additionally, property matching has significant disadvan-

tages (to be discussed in more detail later).

In this paper we propose a new method of generating operational risk scenarios by applying the

method of clustering. Our scenario generation method modifies scenarios produced from expert

scenarios to take into account quantitative operational risk data and without ‘losing’the information

from expert scenarios, thereby solving the key problems of combining scenario and quantitative data.

Our scenario generation method also provides other significant advantages, by providing what-if and

stress-testing analyses, which are beneficial properties in decision analysis. Additionally, we provide

a computationally tractable method that can be applied to multivariate data, both of which are important

to operational risk applications and not necessarily applicable to statistical property matching. Finally,

we show that the impact of quantitative operational risk data has a significant impact on scenario values

and should be taken into account for operational risk scenario generation in general.

The plan of the paper is as follows: in Section 2 we providea literature review introducing operational

risk, models, scenarios and scenario generation. In Section 3 we explain our scenarios generation method

using clustering analysis. We then elaborate on the advantages and benefits of our scenario generation

method. In Section 4 we conduct numerical experiments comparing our method against property matching

to provide a performance comparison and then end with the conclusions in Section 5.

2. INTRODUCTION AND LITERATURE REVIEW

In this section we introduce and review the literature on operational risk, its models, their scenarios and

scenario generation.

2.1. Introduction to Operational Risk and Models

The essence of operational risk definitions is that it is the risk arising from the operational activities in

conducting business, rather than the business's ‘financial’risk. Examples of operational risk therefore

include (Chorafas, 2004) IT failure (physical or software), damage to physical assets (e.g. through

natural disasters), administration errors (e.g. incorrect data entry), fraud and other operational activities.

164 S. MITRA

Copyright © 2013 John Wiley & Sons, Ltd. Intell. Sys. Acc. Fin. Mgmt. 20, 163–187 (2013)

DOI: 10.1002/isaf