Transcript MODERN RISK MANAGEMENT

INTRODUCTION TO MODERN
RISK MANAGEMENT
A brief survey
by
Juhani Raatikainen
University of Jyväskylä
Introduction
• Often risk management is defined as measures
to hedge against such large unexpected losses,
that might threaten existence of a bank or a
non-financial company. However, we should
probably define risk management as a part of
usual business decision making, including, of
course, hedging against the risks, but also
hedging against all other unexpected losses.
The target of risk management is to contribute
Juhani Raatikainen 5.6.2002
2
directly to profitability of a bank or a
corporation.
• We can loosely define modern risk management as any quantitative approach applying
probabilistic measurement and forecasting
methods in decision making with regard of
risks. However, when accepting this definition, we have to bear in mind, that the area
of risk management is much wider than just
the technical methodology, including as its
core decision making at the board level.
Juhani Raatikainen 5.6.2002
3
• There are several types of risks which any
bank or corporation has to face. Today and
tomorrow we are focusing especially on two of
them, namely market risk (risk caused by
changes in market prices) and credit risk (risk
caused by changes in creditworthiness of our
debtors or counterparties).
• Origins of the modern quantitative risk
management, especially the Value-at-Risk
analysis, is in the late 1980’s and late 1990’s,
when the first applications were born.
Juhani Raatikainen 5.6.2002
4
• During the 90’s Value-at-Risk (VaR) became
the standard tool used by (almost without
exceptions) all large or medium sized banks.
• Also since the mid 90’ VaR has won popu-larity
among large international (non-financial)
companies.
• Today, credit risk analysis, especially credit risk
at portfolio level, is the hot topic. The first
general (commercial) solutions or approaches
have been on the market about five years,
however most of the banks are just now
building their credit risk analysis systems.
Juhani Raatikainen 5.6.2002
5
• The Basle Accord gives banks (under certain
conditions) an opportunity to use their own inhouse models to calculate minimum capital
requirement instead of a calculation rule given
by the Basle Group. This has contributed
much to the fast speed with which VaR models have gain popularity inside the
banking community. With the New Basle
Accord now changes, the same will happen
also with credit risk models.
Juhani Raatikainen 5.6.2002
6
What the “New” Approach Offers
• It forecasts probability of outcomes (instead
of ad hoc scenario calculations)
• Measures portfolio effect (gains offered by
diversification)
• Makes all different types of risks comparable (“integrated risk management”)
• Statistically testable models
Juhani Raatikainen 5.6.2002
7
Risk Management System
• Target setting and planning: selection of
efficient target return and risk combination
(Figure 1)
• “Risk Policy” (usually defined by the board)
• Risk limit systems
• Monitoring and planning
• Risk reduction by hedging
• Risk-Adjusted-Return-Measurement (RAPM)
Juhani Raatikainen 5.6.2002
8
Figure 1. The Efficient Set of
Investment Opportunities
Expected
return
Expected risk
Juhani Raatikainen 5.6.2002
9
Value-at-Risk
• In Value-at-Risk analysis we are forecasting
probability distribution of returns at a given
horizon, often with special emphasis on size
of the possible losses with given probability.
• Typical type of answer to be sought is what
is the maximum one week loss at 95 %
confidence level, or what is the minimum
loss with 5 % probability.
Juhani Raatikainen 5.6.2002
10
Figure 2. Forecasted Return Distribution
Todennäköisyys
Probability
Profit/Loss Millions of EUR
Juhani Raatikainen 5.6.2002
SAS RiskDimensions
11
• In the figure 2, there is 95 % probability to
have return higher than -2 million eur, and 5
% probability to have a loss bigger than or
equal to 2 .
Juhani Raatikainen 5.6.2002
12
Figure 3. Structure of a Value-atRisk -model
Structure of
Market Prices
Portfolio
Statistical
Simulation
Models
Engine
Describing
Instrument
Behaviour of
Valuation
the Risk
Factors
VaR -Forecast
Juhani Raatikainen 5.6.2002
13
Traditional VaR Techniques
• Analytical Approach (“Delta-Normal
Approach, RiskMetrics Approach)
– usually base on the multi-Normal probability
distribution and approximation of pricing
functions of the financial instruments
• Historical Simulation
– Assumes, that returns are generated by a
unknown probability distribution, which does
not change in time
Juhani Raatikainen 5.6.2002
14
• Monte Carlo Simulation
– very flexible, can be built on a huge variety of
different types of statistical models
– traditionally built on EWMA covariance
estimate or simplest forms of GARCH models,
and assumption of multi-normally distributed
returns
Juhani Raatikainen 5.6.2002
15
Modern VaR Techniques
• The above traditional VaR techniques do not
seem to have good forecasting ability. One of
the problems is, that return distributions have
fatter tails than the Normal distribution. Much
statistical modelling is done to overcome this
problem.One of the popular approaches is to
use Extreme Value Theory (EVT) to estimate
the fat tails (Smith (1990) offers a short
introduction).
Juhani Raatikainen 5.6.2002
16
• In standard approaches dependence is
measured by a covariance matrix. However,
more accurate and flexible approach is offered
by Copula functions (copula is a function
connecting two or more marginal probability
distribution functions to a multidimensional
probability distribution function).
• It is interesting to note, that use of EVT and
Copula functions is very close to models
applied in medicine, engineering, physics, and
geology.
Juhani Raatikainen 5.6.2002
17
• A good introduction to Copulas is offered by
Nelsen (1998), Joe (1997) discusses
multivariate copulas including also copulas for
extreme value distributions.
• Also other types of model, as co-integration
and/or models allowing regime shifts (for
example Switching Markov GARCH models,
or STAR-GARCH models) look promising (for
a survey see Pagan (1996), and van Dijk,
Teräsvirta, and Franses (2002))
Juhani Raatikainen 5.6.2002
18
• Later today Mr. Esa Vilhonen (OKO Bank)
will present us an example of a very good
banking solution (both VaR and credit risk).
One interesting point to notice is that the
OKO Bank is one of the first in the world to
use Copula functions in a business solution.
Juhani Raatikainen 5.6.2002
19
Corporate Risk Management
• Value-at-Risk approach applied to
corporations differ from the above “banking
models”, because
– forecasting horizon is longer
– nature of the exposure items differ: the exposure
of the included business items have to be forecasted (for example sales volume)
Juhani Raatikainen 5.6.2002
20
– The target exposure is different
• Cash-Flow-at-Risk (CFaR) (value of
business)
• Profit-at-Risk (profitability)
• Earning-at-Risk (EaR), (earning of a
company)
• Industries using Value-at-Risk
– Pulp and Paper production
– Energy companies, and especially companies
on the electricity market
Juhani Raatikainen 5.6.2002
21
– Aviation
– Metal industry
– Large international high-tech companies
• After a couple of minutes Mr. Jean-Marc
Servat (Nokia Corporation) will discuss more
thoroughly Corporate Risk Management while
presenting the Nokia Risk Management
solution.
Juhani Raatikainen 5.6.2002
22
Credit Risk Management
• Focus of credit (or counterparty) risk management is portfolio level risk. A bank can not
run a profitable business without having also
clients, who will default. But, if a bank is able
to forecast total amount of the losses (portfolio
level risk) and price that risk in an appropriate
way, it will have a flourishing business.
Juhani Raatikainen 5.6.2002
23
• Market risk is very often incorporated into
credit risk: the question to be asked is what
is probability of a default, and in that case
what is market value of the instrument
• Credit risk analysis consists of two parts
– “micro analysis” (for example
“creditworthiness of a counterparty)
– portfolio level analysis
Juhani Raatikainen 5.6.2002
24
“Micro Analysis”
• Use of statistical techniques or neural
networks to estimate
– credit quality of a counterparty
– recovery rate
– conditional or unconditional credit migration
probabilities
• Key concept is the Loss Given Default: the
amount of loss when value of the collateral is
subtracted and recovery is taken into account
Juhani Raatikainen 5.6.2002
25
Portfolio Level Analysis
• May be based on
– unconditional credit migration probabilities and
correlations (CreditMetrics)
– conditional credit migration probabilities and
correlations, key dependece is between
portfolio credit risk and macroeconomic
fluctuations
– both the above “micro analysis” and portfoliolevel analysis may be based on option theory
(this is the traditional academic approach)
Juhani Raatikainen 5.6.2002
26
Credit Risk Forecast
Todennäköisyys
Probability
Loss Millions of EUR
Juhani Raatikainen 5.6.2002
SAS RiskDimensions
27
• The presentation by Ph.D. Esa Jokivuolle
(Bank of Finland) tomorrow at 10.00 o’clock
will discuss credit risk measurement especially
with regard of the new Basel Accord.
Juhani Raatikainen 5.6.2002
28
Future Trends
• The greatest change during the 90’ has been
in business philosophy: financial community has adopted systematic “technical”
(portfolio) approach to business and risk
management (applied also in pricing of risk)
• Use of modern risk management concepts
in business decision making continue to
increase
Juhani Raatikainen 5.6.2002
29
• One of the great challenges is to develop more
accurate forecasting models both for market
risk and credit risk.
• From technical point of view, use of nonGaussian statistical distributions and “new”
dependence measures (Copulas) will be the
direction of future research work.
• The presentation by Ph.D. Esa Mangeloja
University of Jyväskylä) tomorrow at 12.30
o’clock will discuss dependence between the
Nordic Stock exchanges.
Juhani Raatikainen 5.6.2002
30
• My own presentation tomorrow will discuss
how to test forecasting accuracy of VaR
models using as an example test results of
some popular models.
Juhani Raatikainen 5.6.2002
31
References
Joe, H. (1997) Multivariate Models and
Dependence Concepts, Chapman & Hall,
London.
Nelsen, R. (1998) An Introduction to
Copulas, Lecture Notes in Statistics,
Springer, New York.
Pagan, A. (1996) The Econometrics of
Financial Markets, Journal of Empirical
Finance, 3: 15 - 102.
Juhani Raatikainen 5.6.2002
32
Smith, R. (1990) Extreme Value Theory, in
Handbook of Applicable Mathematics, (ed.
W. Lederman), vol 7, Wiley, Chichester, 437 472.
Van Dijk, D., Teräsvirta, T. and Franses, H.
(2002) Smooth Transition Autoregressive
Models - A Survey of Recent Developments,
Econometric Reviews, vol. 21, 1 - 47.
Juhani Raatikainen 5.6.2002
33