To my mother for her support.

1
Contents
1 Introduction 1
1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1. Credit Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2. Defaultable Securities . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3. Dependent Defaults and Credit Migrations . . . . . . . . . . . 3
1.2. Credit and Types of Credits . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1. Probability of Default and Credit Rating Models . . . . . . . . 5
1.2.2. Credit Rating Agencies . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.3. Risk and Banking Risks . . . . . . . . . . . . . . . . . . . . . . 8
1.2.4. Risk in general . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.5. Banking Risks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.6. Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.7. Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.8. Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.9. Signicance of the study . . . . . . . . . . . . . . . . . . . . . . 11
1.2.10. Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.11. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Literature Review 1
2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.1.1. Credit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.1.2. Credit Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1.3. Credit Risk Management . . . . . . . . . . . . . . . . . . . . . 4
2.1.4. The Structural Form Models . . . . . . . . . . . . . . . . . . . . 5
2.1.5. The Reduced Form Models . . . . . . . . . . . . . . . . . . . . . 7
2.1.6. Factor Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.7. Other Relevant Approaches . . . . . . . . . . . . . . . . . . . . 10
2.1.8. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 Methodology 12
3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.1.1. Mathematical Background . . . . . . . . . . . . . . . . . . . . . 13
3.1.2. Transition Probabilities . . . . . . . . . . . . . . . . . . . . . . 15
3.1.3. Discrete Time Markov Process with an Absorbing State . . . . 15
3.1.4. Problems of markov models . . . . . . . . . . . . . . . . . . . . 15
3.1.5. Assumptions based on the problems of Markov problems above. 16
3.1.6. Non-homogeneous semi-Markov processes (NHSMP) . . . . . 17
2
3.1.7. Non-homogeneous semi-Markov reliability model . . . . . . . 19
3.1.8. The homogeneous semi-Markov reliability credit risk model . 20
3.1.9. The transient analysis: Relevant credit risk indicators . . . . 22
3.1.10. The asymptotic analysis . . . . . . . . . . . . . . . . . . . . . . 23
3.1.11. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Data Analysis and Simulations 27
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.1. POSB Rating Systems . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.2. Normality testing and data transformations. . . . . . . . . . . 29
4.1.3. Calibration of Default Probabilities to Ratings . . . . . . . . . 31
4.1.4. Regression Analysis: Rates versus Q
3;11
. . . . . . . . . . . . . 35
4.2. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5 Conclusions and Recommendations 49
5.0.1. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.0.2. Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.0.3. Area of further study. . . . . . . . . . . . . . . . . . . . . . . . . 56
3
List of Figures
4.1 Probability Plot for January 2010 . . . . . . . . . . . . . . . . . . . . . 29
4.2 Probability Plot for February 2010 . . . . . . . . . . . . . . . . . . . . 30
4.3 Probability Plot for May 2011 . . . . . . . . . . . . . . . . . . . . . . . 31
4.4 Probability of default versus credit ratings . . . . . . . . . . . . . . . 34
4.5 Transition probabilities(t = 2) . . . . . . . . . . . . . . . . . . . . . . . 42
4.6 Transition probabilities(t = 4) . . . . . . . . . . . . . . . . . . . . . . . 44
4.7 Transition probabilities (t = 6) . . . . . . . . . . . . . . . . . . . . . . . 46
4
List of Tables
1.1 S and P rating scale system . . . . . . . . . . . . . . . . . . . . . . . . 8
4.1 Credit Scores for companies as from 2010 to 2011 (Quarterly) . . . . 32
4.2 Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 Calibration of POSB Ratings to Default Probabilities . . . . . . . . . 37
4.4 Correlations of credit ratings under 3 portifolios . . . . . . . . . . . . 39
4.5 Probability of migrating from one state to another within 2 years as
a percent (
ij
(2)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.6 Probability of migrating from one state to another within 2 years as
a percent (
ij
(4)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.7 Probability of migrating from one state to another within 2 years as
a percent (
ij
(6)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.8 Reliability for companies rated AAA, AA, A, BBB, BB, B, CCC and D
at t = 2, t = 4 and t = 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5
Chapter 1
Introduction
1.1. Introduction
Credit risk problem is one of the most important problems that are faced in -
nance. Fundamentally, it consists of computing the default probability of a rm
that is in debt. For credit risk evaluation there are international organizations
such as Fitch, Moodys and Standard and Poors that give different ranks to the
examined rms. Each rm is given a rating that is a vote to the reliability on the
capacity to reimburse the debt. The rating level changes in time and one way to
follow the time evolution of ratings is by means of Markov processes (Jarrow et al,
1997).
In recent times, credit risk analysis has grown to become one of the most important
problems dealt with in mathematical nance. Fundamentally, the problem deals
with estimating the probability that an obligor defaults on their debt in a certain
time. To obtain such a probability, several methods have been developed which are
regulated by the Basel II Accord. This establishes a legal framework for dealing
with credit and market risks, and empowers banks to perform their own method-
Introduction 2
ologies according to their interests under certain criteria. Credit risk analysis is
founded on the rating system, which is an assessment of the capability of an obligor
to make its payments in full and on time, in order to estimate risks and make the
investors decisions easier.
1.1.1. Credit Risk
Credit risk embedded in a nancial transaction is the risk that at least one of the
parties involved in the transaction will suffer a nancial loss due to default or de-
cline in the credit worthiness of the counter-party to the transaction, or perhaps of
some third party. The most extensively studied form of credit risk is the default
risk, that is, the risk that counterparty in a nancial contract will not fulll a con-
tractual commitment to meet her/his obligations stated in the contract. For this
reason, the main tool in the area of credit risk modeling is a judicious specication
of the random time of default. A large part of the present text is devoted to this
issue.
1.1.2. Defaultable Securities
Vulnerable (defaultable) options are options that themselves have default risk. Al-
though option writers are required to deposit initial margins and their credit qual-
ity is checked before they are allowed to write options, they might, in some cases,
not honor their obligations, or only partially honor it. Since the option value at ma-
turity is equal to its payoff, any portion of the payoff not paid by the writer can be
interpreted as a percentage loss in the value of the option at that time; hence the
recovery of market value assumption is directly applicable. Historically, rms that
issue such bonds appear to entail very little default risk yet their credit spreads are
sizable and positive (Amato and Remolona, 2003). A natural explanation is that
these rms are exposed to large sudden and unforeseen movements in the nan-
cial markets. In other words, the spread accounts for exposure to market jump risk.
Introduction 3
1.1.3. Dependent Defaults and Credit Migrations
Financial institutions such as banks normally take on credit risk by granting loans.
They also take on debt of other companies or the debt of governments. By contrast,
commercial companies normally acquire credit risk by delivering products in ex-
change for payments. Credit ratings are useful because they help the process of
issuing and purchasing bonds and other debt issues by acquiring a qualied mea-
sure of the relative credit risk. As a general rule, the higher this measure is, of an
issuer or an issue indifferently, the lower the interest rate the issuer has to pay to
attract investors. Conversely, an issuer with a lower credit risk measure has to pay
a higher interest rate to balance the greater risk assumed by investors. The rating
assessed are monitored and reviewed by the agencies.
In conducting this surveillance, an organisation may consider many factors includ-
ing: identication of the issues either upgraded or downgraded changes in business
climate, changes in credit markets, new technologies that may affect the issuers
earnings, new facts that may inuence the projected revenues, growing or shrink-
ing debt burdens. Specic factors may inuence specic issuers and debt issues:
for example the creditworthiness of a state or municipality may be affected by pop-
ulation shifts or lower incomes of taxpayers.
1.2. Credit and Types of Credits
Credit is a means by which one party, the obligor, obtains a resource, in order to get
an economic benet either in the present or in the future. It is provided by another
party, the creditor, in exchange of a future promise of payment equal to the amount
borrowed plus an additional amount. Such a credit can be a corporate, consumer,
investment, public, real state, etc. One of the most popular credit products are
xed income securities. They are instruments by which a rm or a government ob-
Introduction 4
tains funds in exchange for xed payments, that include regular interest plus the
amount borrowed. If the obligor misses making any such payments the borrower
may force it into bankruptcy, although this can vary depending on the legal frame-
work and the conditions of the security.
A xed income security is made up of the following elements: the indenture, which
is the document that states all the terms of the instrument; the obligor, which
is the entity who issues the instrument and is therefore responsible for making
payments; the creditor, which is the entity who invests in the instrument; the prin-
cipal, which is the amount borrowed; the face value, which is the amount to paid to
the creditor at maturity; the coupon, which is the nominal rate to pay each year in
terms of a percentage of the principal; the maturity, which is the date of the end of
the instrument, in which the issuer must return the face value. Nowadays, there
are a variety of such instruments, but the most used are:
Treasuries: are instruments issued by the U.S. government and represent the
safest investment on the market and, therefore, their coupons are the lowest.
Their maturities can be up to 30 years.
Municipal bonds: are issued by county, state or city governments, and their
goal is funding to develop or improve public infrastructure, but sometimes
they are issued to fund their ongoing operations. They usually pay interest
periodically and at maturity pay a face value.
Corporate bonds: they are issued by public rms to nance their expan-
sion projects or ongoing operations and are traded on public markets. Their
coupons mainly depend on the rating given by an agency, although some ex-
ternal factors can inuence the credit rating. They may pay interest periodi-
cally and pay the face value at maturity or, in the simplest form, a payment
at maturity, this instrument is called a zero coupon bond.
Asset-backed securities: They refer to a bundle of assets that are not possible
to be sold individually, but used to generate regular income for rms. They
Introduction 5
are traded in public markets and examples include home equity loans, auto
loans, credit card receivables, student loans, etc.
Preferred stock: are instruments that pay regular interest quarterly and a
value at maturity but, unlike corporate bonds, in case a rmgoes to bankruptcy
they can return their investment before a regular bondholder.
1.2.1. Probability of Default and Credit Rating Models
Credit rating models include structural form models (SFM), reduced form models
(RFM) and factor models (FM). SFM are based upon the Black and Scholes theory
for option pricing and the Merton model (Bluhm et al, 2002). This, in turn, can
be divided into three generations. The rst generation assumes a risky bond in
order to approach the PD (Probability Density) of a rm, which occurs if its mar-
ket value is lower than its liabilities at the maturity time. The second generation
relaxes the default at maturity time assumption, such that this event can take
place at anytime during the bonds lifetime. The third generation represents the
VaR and assumes that LGD (Loss Given Default) is a random variable related to
PD. RFM mainly focuses on the accuracy of the PD, such that it is more important
than an intuitive economical interpretation. It can be classied as an individual
level reduced form model (ILRFM) and portfolio reduced form model (PRFM). The
former is based on a credit scoring system (two-state or multistate), and the latter
assumes an intensity jump process.
1.2.2. Credit Rating Agencies
A credit rating is an evaluation of the worthiness of an obligor (rm or govern-
ment), that issues a xed income security, to be able to achieve its payments in full
and on time. The rating is assigned by a credit rating agency (CRA) which, through
a standardized process, provides a rating to an obligor according to a dened scale.
This rating is translated into the cost of capital, so that a better rated rm can get
Introduction 6
a lower interest rate. Thus, CRA play a crucial role in the xed income securities
market because it contributes to the decision of investors, by summarizing a lot
of data in a single rating. Therefore, they are forced to have a high accuracy in
ratings assigned and to review them in dened periods of time; they also should
not involve capital markets transactions in order to maintain investor condence.
They are regulated by a technical committee whose main function is to state the
principles of the activities of a CRA. The Technical Committee of the International
Organization of Securities Commissions, 2005 gives the code of conduct fundamen-
tal of a CRA.
There exists a number of CRA all around the world, most of them are focused
on specic industries or markets, the most important are Standard and Poors,
Moodys Investor Service, and Fitch Group. The Securities and Exchange Commis-
sion, in 1975, designated them as the three Nationally Recognized Statistical Rat-
ing Organizations (U.S. Securities and Exchange Commission, 2008) in the USA,
and are considered the big three credit rating agencies. Each agency has its own
methodology in order to express its rating opinion of an instrument issued by a
rm. Such opinion is typically expressed in a scale of letters grades, for example,
from AAA to D where the former represents the best state and the latter the de-
fault state.
In order to assess a rm, a CRA considers current and historical information in
order to identify strengths and weaknesses events that could affect the credit qual-
ity of a debt issuer. Information can be quantitative or qualitative, such as legal
structures, business plans, nancial statements and indicators, forecasting, macro
economic environment, industry trends, position in industry and competitiveness,
credit history (Standard and Poors, 2011). This evaluation is usually paid for by
the obligor and is made public immediately thereafter, but sometimes it requests a
private rating of an instrument.
Standard and Poors Rating Services is a rating agency with ofces in 23 countries
Introduction 7
and is one of the oldest companies of its kind, being more than 150 years old. It is a
division of McGraw-Hills, and is focused on providing index information, risk anal-
ysis and research in the world markets in order to give condence to investors. To
rate a rm, the agency uses a scale from AAA to D, where the former is best state
and the latter is the default state. Every intermediate state has three positions,
the upgrade (+), downgrade (-), and uncertain.
Moodys is a credit rating agency founded in 1909 and offers credit ratings ser-
vices,development of risk analysis tools and research in order to contribute to the
efcient operation of xed income securities market. Moodys rating system is
based on the LGD in case of default and its aim is to provide a simple mechanism
that can predictthe credit worthiness of a security in order to facilitate decision-
making of investors. The long term xed income securities use a scale of states
from Aaa to C, where the former is the highest and the latter is the lowest. In
turn, intermediate states are added the numeric index 1,2 and 3, where the lowest
number represents the highest rating.
Fitch Group is a rm that provides products and services to the xed income secu-
rities market in order to make more timely and informed business decisions (Fitch
Group, 2012),it was founded in 1913, and is divided into Fitch Rating and Fitch
solution. It is the smallest of the three big rating agencies, and its main function
in the market is that of a tie-breaker if Moodys and S and P ratings differ.
In order to evaluate long term income securities, Fitch uses the S and P rating
scale system (see table below).
Introduction 8
Rating Interpretation Grade
AAA Extremely strong capacity Investment Grades
AA+ Very strong capacity
AA Very strong capacity
AA- Very strong capacity
A+ Susceptible to the adverse effects
A Susceptible to the adverse effects
A- Susceptible to the adverse effects
BBB+ Weakened capacity
BBB Weakened capacity
BBB- Weakened capacity
BB+ Uncertainties and exposure
BB Uncertainties and exposure
BB- Uncertainties and exposure
B+ More vulnerable than the obligors rated BB Speculative Grades.
B More vulnerable than BB
B- More vulnerable than BB
CCC+ Currently vulnerable
CCC Currently vulnerable
CCC- Currently vulnerable
CC+ Currently highly vulnerable
CC Currently highly vulnerable
CC- Currently highly vulnerable
C Highly vulnerable
D defaulted on obligations
Table 1.1: S and P rating scale system
1.2.3. Risk and Banking Risks
This part plays a role as the foundation that supports the main theories of credit
risk management. Knowledge of risk and banking risks in general forms the foun-
dation of understanding credit risk among the banking risks and hence,the signif-
icance of good credit risk management.
Introduction 9
1.2.4. Risk in general
Credit risk is the risk of loss due to a debtors non-payment of a loan or other
line of credit. (Wikipedia.org, as of March 2009). Risks in banking business, risks
in trading activities, risks in our normal life or whatever kind of risks are what
potentially happen sometime in the future and will have unexpected impacts on
risk recipients. Risk actually has been dened in multiple ways but there are two
distinguishable styles of explanation.
1.2.5. Banking Risks
The banking business, compared to other types of business, is substantially ex-
posed to risks, especially in this ever-changing competitive environment. Banks no
longer simply receive deposits and make loans. Instead, they are operating in a
rapidly innovative industry with a lot of prot pressure that urges them to create
more and more value-added services to offer to and better satisfy the customers.
Risks are much more complex now since one single activity can involve several
risks. Risks are inside risks. Risks overlap risks. Risks contain risks.
Scholars and analysts in recent decades have been trying to group banking risks
into categories. The Basel Accords issued by the Basel Committee on Bank Su-
pervision mention three broadest risk types in the rst pillar: credit, market and
operational risks. Then the second pillar deals with all other risks. Anthony et al
(2002) divides risks into six generic kinds: systematic or market risk (interest rate
risk), credit risk, counterparty risk, liquidity risk, operational risk, and legal risks.
This categorization is based on types of services offered by banks.
Introduction 10
1.2.6. Problem Statement
The main event concerning the companies/rms lending money is the event that
the other party defaults on their loan. This can result in severe losses for the
lender. Thus, its probability of occurrence must be estimated, and from this a risk
rate can be obtained. The problem we attempt to tackle is how to optimally min-
imise the default risk for loans given to companies by Post Ofce Savings Bank in
Zimbabwe.
These sudden defaults may occur due to a change in the basic market variables,
and other economic factors. While credit risk management has been addressed to
some extent in the past by some researchers, their approach has been one of quite
simplistic constraints and has been generally silent on the challenges presented by
defaults. Most of the researchers to this date have used a risk-neutral approach
consisting of normal copulas. A semi Markov reliability model will be used in this
study to solve the problem of credit migrations leading to default.
1.2.7. Aim
To develop a model for valuating defaultable securities.
1.2.8. Research Objectives
The research objectives are as follows:
To model dependent defaults on securities between changes in credit ratings
using a reliability technique, discrete time homogenous semi-Markov relia-
bility credit risk model.
To assess changes in credit ratings.
Introduction 11
1.2.9. Signicance of the study
This study will establish a framework of controls/ measures to ensure credit
risk-taking is based on sound credit risk management principles.
The study will enable identication, assessment and measurement of credit
risk clearly and accurately across the companies and within each separate
business, from the level of individual facilities up to the total portfolio.
The control and plan credit risk-taking in line with external stakeholder ex-
pectations and avoiding undesirable concentrations will be enabled.
Monitoring of credit risk and adherence to agreed controls; and ensuring that
risk-reward objectives are met can be done through this study.
Credit exposures typically cut across geographical locations and product lines.
The use of credit risk models offers banks a framework for examining this
risk in a timely manner, centralizing data on global exposures and analyz-
ing marginal and absolute contributions to risk. These properties of models
may contribute to an improvement in any nancial institute overall ability to
identify, measure and manage risk.
Credit risk models may provide estimates of credit risk (such as unexpected
loss) which reect individual portfolio composition; hence, they may provide a
better reection of concentration risk compared to non-portfolio approaches.
By design, models may be both inuenced by, and be responsive to, shifts
in business lines, credit quality, market variables and the economic environ-
ment. Consequently, modelling methodology holds out the possibility of pro-
viding a more responsive and informative tool for risk management.
In addition, models may offer: the incentive to improve systems , a more
informed setting of limits and reserves; more accurate risk- and performance-
based pricing, which may contribute to a more transparent decision-making
process; and a more consistent basis for economic capital allocation.
Introduction 12
1.2.10. Limitations
Credit risk management is a sensitive area. It was costly and a challenge to access
the information that was needed for model calibration although Post Ofce Savings
Bank released the information. The sample which was used was selected by the
company itself therefore it may be biased.
1.2.11. Conclusions
This study is meant to solve the problem of credit risk by the use of Semi Markov
reliability models. The model that will be used is meant to detect the credit migra-
tions of various companies over a period of time.
Chapter 2
Literature Review
graphicx
Introduction
2.1. Introduction
2.1.1. Credit
Credit is dened by the Economist Dictionary of Economics as the use or posses-
sion of goods or services without immediate payment and it enables a producer to
bridge the gap between the production and sale of goods and virtually all exchange
in manufacturing, industry and services is conducted on credit, (Colquitt, 2007).
Credit generates debt that a party owes the other. The former is called a debtor or
borrower. The latter is a creditor or lender. Certainly the debtor will have to pay
an extra amount of money for delaying the payment.
Both debtor and creditor expect a return which is worth their paying more and
waiting, respectively.the debtor pays the creditor extra money earned from re in-
vestments of the credit amount. Creditor gives the debtor time and takes back a
return for supplying the credit. So now it is clear why credit exists and how im-
portant it is to the economy. Firms or individuals that run short of capital need
credit to continue or expand their businesses/investments. The ones that have ex-
Literature Review 2
cess money, on the other hand, never want to keep it in the safes. As a result, all
are growing and making more money.
Demand and supply together exist but do they meet each other? Here nancial
intermediaries who act as the bridge between credit suppliers and clients. Now
in this innovative phase of the global nancial-services industry, numerous types
of nancial institutions have joined the credit supplier group: insurance compa-
nies, mutual funds, investment nance companies, etc (Colquitt, 2007). Neverthe-
less, banks are still the dominant source that both individuals and corporates seek
credit from.
In banking specically, two primary kinds of credit services based on customer cat-
egories are offered: retail credit and wholesale credit. Lending in retail or personal
banking are subject to individuals and may fall under: home mortgages, install-
ment loans (e.g. consumer loans, educational loans, auto loans), credit card revolv-
ing loans, revolving credits (e.g. overdrafts), etc. (Crouhy et al, 2006). Wholesale
lending, on the other hand, involves rms as the borrowers and therefore is of much
higher value, more complicated and poses more threats to the banks.
2.1.2. Credit Risk
Quite often in the previous sections of this paper, credit risk has been mentioned
or even dened. However, it still needs to be repeated from a deeper point of view.
Basically, it is understandable that credit risk occurs when the debtor cannot repay
part or whole of the debt to the creditor as agreed in the mutual contract. More
formally, credit risk arises whenever a lender is exposed to loss from a borrower,
counterparty, or an obligor who fails to honor their debt obligation as they have
agreed or contracted. This loss may derive from deterioration in the counterpartys
credit quality, which consequently leads to a loss to the value of the debt (Colquitt,
2007). Or in the worst case, the borrower defaults when he/she is unwilling or un-
Literature Review 3
able to fulll the obligations (Crouhy et al, 2006).
In banks, credit failures are not rare and they critically affect the banks liquidity,
cash ows and eventually, prot and shareholders dividends. Banks call them bad
debts. Modern banking no longer experiences credit risk solely in its traditional
activity of loan making. In reality, credit risk falls in a broader scope. For instance,
a well-known British banking group sees that the groups credit risk may take the
following forms:
Lending: funds are not repaid
Guarantees or bonds: Funds are not ready upon collection of the liability
Treasury products: payments due from the counterparty under the contract
is not forthcoming or ceases
Trading businesses: settlement will not be effected
Insurance risks reinsured: the re insurance counterparty will be unwilling or
unable to meet its commitments
Cross-border exposure: the availability and free transfer of currency is re-
stricted or ceases
Holdings of assets in form of debt securities: value of these falls e.g. after a
downgrading of credit rating
Lending: funds are not repaid
Guarantees or bonds: Funds are not ready upon collection of the liability
Treasury products: payments due from the counterparty under the contract
is not forthcoming or ceases
Trading businesses: settlement will not be effected
Insurance risks reinsured: the reinsurance counterparty will be unwilling or
unable to meet its commitments
Literature Review 4
Cross-border exposure: the availability and free transfer of currency is re-
stricted or ceases
Holdings of assets in form of debt securities: value of these falls e.g. after a
downgrading of credit rating
Again it is necessary to stress that credit risk has always been the biggest threat
to any banks performance and the principal cause of bank failure (Greuning et
al, 2009). Therefore, a sound credit risk management framework is indispensable
to a healthy and protable banking institution. The following part will give the
audience a clearer view of how credit risk management looks like in a bank.
2.1.3. Credit Risk Management
In banking, credit risk is taken for granted as a fundamental feature of the in-
stitutions. If an organization refuses to acknowledge the inherent risk, it is not
in the lending industry. Wherever risk survives its enemy, risk management, will
also exist and ght against it. Credit risk management is simply the procedures
implemented by organizations with the aim of diminishing or avoiding credit risk.
Credit risk management has been a hot topic of debate as it is one of the fastest
evolving practices, thanks to institutional developments in the credit market, di-
versication of nancial institutions participating in the lending business and mod-
ern technologies (Caouette et al, 2008). As also discussed by the four authors above,
credit risk management lies in an expert analysis system, whose objective is to look
at both the borrower and the lending facility being proposed and to assign a risk
rating(Caouette et al, 2008). Among the evaluated data, nancial ratios are per-
ceived to be very important. The philosophy that Caouette et al (2008) presents is
that credit risk management is a form of engineering in which models and struc-
tures are created that either prevent nancial failure or else provide safeguards
against it. However most of the models are for enterprises in general. For nancial
institutions or banks, perhaps the credit analysis system is of much larger help.
Literature Review 5
Banking credit risk management in the eyes of Crouhy et al (2006) can be divided
into retail and commercial credit risk management. Credit risk management based
upon portfolio management is highlighted for both retail and commercial. This
means that the customers are categorized into different portfolios, each of which
is homogenous in several characteristics. Instead of managing every single client,
the bank will handle them in groups and therefore usually saves time, effort and
cuts cost. For retail banking, the book introduces the credit scoring model that is
customized for personal banking.
Commercial lending, on the other hand, will utilize the helpfulness of internal risk
rating system established, based on the rating system of professional credit rating
agencies such as Moodys, Fitch Ratings or Standard and Poors. Credit scoring
and internal rating are based upon nancial and non-nancial assessment. Sev-
eral credit models and credit derivatives are also presented as new approaches
to, respectively, measure and to mitigate credit risk management (Crouhy et al,
2006). But the risks seem to be insufcient and some overlapping can be found.
Counterparty risk and credit risk are quite alike or the list lacks country risks, for
example.
2.1.4. The Structural Form Models
The Merton Model (Merton, 1974) represented the rst attempt at a credit risk
model and is based on the option pricing theory (Black, 1973). Merton derived a
formula to estimate the probability of default (PD) by assuming rm issues a zero-
coupon bond that represents its entire debt. In this model the probability of default
is equivalent to the probability that the value of the rms assets are less than the
face value of a given bond, or similarly, the probability that market value of a rm
is lower than its liabilities. Derived from the Merton model, other models were
developed subsequently by modifying the original one by relaxing one or more of
the assumptions.
Literature Review 6
So, Black et al (1976) provide a more complete model by assuming rms have
complex structures of capital and debt. Whilst, Geske (1977) extended the model
to a bond in which interest is paid periodically and the default events can happen
either at maturity time or every time the coupon is paid, whilst Vasicek (1984)
makes a distinction between short and long term debts. These models are known
as the rst generation of structural form models. Jones, et al (1984) found that
these approaches are very useful in qualitative aspects of credit risk, but that in
practical applications they are not realistic because of the following reasons:
The default event can occur just at maturity of the bond.
The default risk of a rm should be estimated according to its composition in
nancial statements rather than to assuming the issue of a zero coupon bond.
Meanwhile, Julian et al (1994) found, with empirical evidences, that the absolute
priority rule is often violated and the lognormal distribution assumption tends to
overstate the loss given default (LGD) in the event of default. In response, a second
generation of SFM was developed by Kim et al (1993). These approaches assume
that default may occur at anytime during the bonds lifetime, so the event of default
happens when the rms assets values are less than a reference indicator. Eom et
al (2001) found that despite these improvements, these approaches have relatively
poor empirical performances that are caused by:
Asset value of a rm is a non-observable variable and we are required to
estimate it.
A xed income security uses to have downgrades when it is closed to the
default event and by SFM it is not possible to model it.
Most SFM assume that the market value of a certain rm can be seen at any time.
As a result, the event of default can be identied just before it occurs. Darrell et
al (2000) concluded that it was better to use an approach based on a multistate
system that estimates the default event before it occurs.
Literature Review 7
The third generations of the SFM are the credit value at-risk models, whose de-
velopment has been encouraged by the greater signicance of credit risk analysis.
For example, banks and consultants have developed and patented their own model.
These include J.P Morgans Credit Metric (Morgan, 1997), Credit Suisse Financial
Products Credit Risk+, Mckinseys Credit Portfolio View (Wilson, 1998) and Ka-
mukura Risk Manager (Kamukura Corporation, 2008).
2.1.5. The Reduced Form Models
Individual Level Reduced Form Models
Individual level reduced form models (ILRFM) were proposed by Altman (1968),
who used linear or binomial models (such as logit or probit) for differentiating de-
fault from non default rms. The approach focuses on estimating coefcients and
assigning a Z-score to an obligor. After its proposal, many models based on credit
scoring system have been widely used in credit risk analysis.
Altman et al (1997) analyzed various predictor variables in order to identify the
most signicant ones on the event of default and found that indicators related to
protability, liquidity and leverage were the most representative. Jacobson et al
(2003) proposed a method to estimate portfolio credit without bias, and Lin (2009)
developed a new approach by three kinds of two stage hybrid models of logistic
regression articial neural network. Finally, Luppi et al (2007) , showed that tra-
ditional accounting based credit scoring had less explanatory power in non-prot
rms than in prot-making ones.
Akalu et al (2008) used data from the UK in order to compare the RFM and SMF,
and they found that there were few differences between SFM and scoring models
in terms of predictive power with Das et al (2009) conrming the results. The main
weakness of these approaches is that they are founded in nancial statements and
indicators as forecasting variables, such that they do not usually adjust in a fast
Literature Review 8
way to the changing conditions of markets Cheng et al (2009). So, tendencies are
focused on dynamic portfolio reduced models more than a purely static and indi-
vidual level credit scoring model.
Portfolio Reduced Form Models
Portfolio reduced form models (PRFM) are an attempt to overcome the weaknesses
of SFM and, in empirical studies, are reported to perform better in capturing the
properties of credit risks (Cheng et al, 2009). These approaches were originally
introduced by Unal (1992), in which they decomposed the credit spread into two
components: PD and LGD. So this approach is based on the distributions of such
variables. The simplest model was proposed by Jarrow (1995) who modelled the
default event as a Poisson process with a instantaneous transition rate and de-
fault time t exponentially distributed. However, it assumed that was constant
over the time and across the states, which in reality is not easy to accept.
In order to relax this assumption, Madan (1998) states the assuption of the ex-
cess return on the issuers equity, and Carling (2007) worked on duration mod-
els. Similarly, Lando (1998) modelled the intensity as a Cox process. Jarrow et al
(1997) modelled the PD by the DTMP (Discrete Time Markov Processes) approach,
where the default event was considered as an absorbing state, such that from a his-
torical database was obtained a transition matrix in order to estimate migration
probabilities at any time. This work served as the basis for other studies, includ-
ing Kalotychou et al (2008), Lucas et al (2001) applied both discrete time Markov
process (DTMP) and continuous time Markov process (CTMP) models in empirical
studies. Meanwhile, Lu (2009) studied differences between DTMP and CTMP for
estimating migration probabilities. The conclusions taken were:
DTMP contained the embedding problems whereas the CTMP did not.
DTMP tends to underestimate migration probabilities since it is not consid-
Literature Review 9
ering changes occurred within a year, which can be often in non-investment
states.
Meanwhile, some works about the suitability of Markov processes in the credit risk
environment have been done. They mainly focus on the following:
the waiting time in a state: the time that a rm will remain in a certain state
is related with the time that it has been in such a state (Duffe et al 2003).
The rating assignment and time,that is, assigning the rating to a rm is re-
lated to the time it is made and in particular with the business cycle (Nickell
et al 2000).
Dependence of the current state: the current state may depend of various
previous states assigned to a rm, not only in the previous one (Carty et al,
1994).
Damico et al (2005) and Monteiro (2006) dealt with the rst problem by means
of semi-Markov process, that relaxes the exponential waiting time distribution
assumption. The second problem has been solved with the addition of a non-
homogeneous environment by Damico (2009). The third problem was dealt with by
the backward recurrence time (Damico et al, 2009) and by a hidden semi-Markov
process (Banachewicz, 2007). Other PRFM were also performed for the mortality
analysis of survival time of a loan, and was introduced by Altman (2000), who ap-
plied actuarial methods to study mortality rates of obligations. Subsequently, the
analysis was based on the recovery rates and LGD values when the event of default
occurs, and the correlation between these variables and the frequencies of such an
event. Nickell (2000) studied the probability of survival through time, based in a
representative set of banks, and Dermine (2006) applied a mortality analysis to
LGD.
Literature Review 10
2.1.6. Factor Models
The simplest version of a factor model (FM) is considered to be the one proposed
by Perraudin (2003), who assumed a systematic risk factor and an individual error
term that followed the standard normal distribution. Alternatively, other single
FM were performed. These include Dietsch et al (2009), who developed it from the
internal rated base approach and used FM to calibrate risk weights. Kupiec (2007)
considered two vectors of explanatory variables for latent variables where a set of
macro-economical variables was considered, such as GDP; and again a set of rm-
specic variables which account for individual risk, usually for the return rates.
Pederzoli (2005) considered multifactor models, and Ebnother (2007) extended the
standard single FM to multi-period framework from a set of return rates in differ-
ent time periods. On the other hand, Gordy (2002) showed a factor model based on
a gamma distribution with mean 1 and
2
.
2.1.7. Other Relevant Approaches
Recently, new approaches have been developed mainly focusing on the correlation
between the PD and the LGD, these include (Ebnother, et al 2001). Ebnother
(2001) took an approach to make more exible the correlation between the PD and
LGD, empirical results found evidence to support a negative correlation between
these variables. Previously, Frye et al (2000) also found an inverse relationship in
PD and LGD. Its model is founded on the model developed by Ebnother (2007) and
Gordy et al (2002), and the approach assumes that certain economical indicators
can determine the event of default. Meanwhile, Pederzoli et al (2005) created a
model that combined features of SFM and RFM by using the option pricing frame-
work and assuming that the borrowing rms total assets determines the PD, and
also found an inverse correlation in PD and LGD. Pederzoli (2005), based on some
historical databases, analysed correlation patterns between PD and LGD.
LGD measures and PD are found and correlation is close to zero. However they
were more in line with Fryes results Pederzoli (2005) when the sample was lim-
Literature Review 11
ited to the period 1988-1998. Moreover,Pederzoli (2005) studied correlations be-
tween LGD and PD based on Moodys historical database, by using the probability
techniques of extreme value analysis and non parametric statistics. Similarly, they
studied the patterns of correlation on VaR measures.
2.1.8. Conclusion
Economic cycles and other factors can impact rating performance over time. The
challenge is understanding how to adjust transition ratings and probability of de-
fault rate forecasts to reect potential future economic conditions. Semi Markov
Reliability models allow investment companies and banks to easily predict how
obligors are likely to be in terms of credit rating from time to time (forecasts).
The model is based on history of ratings information , which allows to come up
with transition patterns and accurate forecasts. Unlike other methodologies Semi
Markov Reliability models enable banks and investment companies to track or
monitor credit migrations from time to time.
Chapter 3
Methodology
3.1. Introduction
In this chapter we will present an overview of Markov and semi-Markov processes
in order to apply them to credit risk models. The discrete time Markov process
(DTMP) and its transition probabilities are introduced. The discrete time semi-
Markov process (DTSMP) is dened. Fundamentally it consists of computing the
default probability of a company going into debt. The problem can be studied by
means of Markov transition models. The generalization of the transition models
by means of homogeneous semi-Markov models is presented in this study. This
chapter provides a short description of HSMP. The reliability semi- Markov model
is then explained.
Putting the hypothesis that the next transition depends only on the last one (the
future depends only on the present) the problem can be faced by means of Markov
processes. In discrete time Markov chain environment the time transition is given,
but in the reality the transition between two states in a mechanical system usually
happens after a random duration. This is the reason why the semi-Markov envi-
Methodology 13
ronment ts better than the Markov one in reliability problems. Another relevant
phenomenon in the time evolution of a system can be the system age. The intro-
duction of non-homogeneity (due to changing credit ratings) gives the possibility
to take into account this problem. All the highlighted aspects can be faced using
homogeneous semi-Markov models.
The data comes from the internal rating system of Post Ofce Saving bank. This
rating system is composed of assessment of creditworthiness of obligors, security/
collateral, average prot generated by the company, credibility, integrity, premises,
authenticity of information, account history / performance, gearing / level of in-
debtedness and purpose of loan. The bank provided with the rating history of some
of its borrowers in its loan portfolio as at December 2009.In addition, rating in-
formation for all borrowers who defaulted and did not default within the period
2010-2011 of paying back the loan was obtained. Besides rating and default data
we collected some additional information on creditworthiness of obligors, security/
collateral, average prot generated by the company, credibility, integrity, premises,
authenticity of information, account history / performance, gearing / level of in-
debtedness and purpose of loan. Minitab 16 and Matlab 6.1 will be used to analyse
the data collected from Post Ofce Savings Bank in Zimbabwe.
3.1.1. Mathematical Background
Discrete Time Markov Process
A system S with m possible states is introduced with space state I = [1,2... m]. The
system S performs randomly in the discrete time set T = [t
1
, t
2
.....]. X
t
n
is dened
as the state of S at time t
n
T. Assuming that in small intervals of time there are
no changes, we say that the random sequence (X
t
n
, t
n
) T is a Markov process if
and only if for all (X
t
0
, X
t
1
, ...., X
t
n
) T.
P[X
t
n
= x
t
n
| X
t
0
= x
t
0
|, ...., | X
t
n1
= x
t
n1
] = P[X
t
n
= x
t
n
| X
t
n1
= x
t
n1
] (3.1)
Methodology 14
Similarly, the random sequence (X
t
n
, t
n
) T is homogeneous if and only if:
P[X
t
n
= x
t
n
| X
t
n1
= x
t
n1
] = P[X
t
1
= x
t
1
| X
t
0
= x
t
0
] (3.2)
That is, jumps between states are independent of the time; otherwise it is non-
homogeneous. Now, we introduce the migration probability in one step:
P[X
t
n
= j | X
t
n1
= i] =
ij
i, j I, t
n
T. (3.3)
Consider the transition matrix P dened as
P = [
ij
]i, j I (3.4)

ij
0i, j I (3.5)

ij
= 1i, j I (3.6)
The vector p = [p
1
, p
2
.....p
m
] as the initial condition probabilities, with
p
i
= P[X
0
= i], i I. This probability should satisfy the following conditions:
p
ij
0 ; i, j I (3.7)

pij = 1 ; i, j I (3.8)
Thus, a homogeneous DTMP is characterised by the couple (p; P). If X
0
= i, i.e.if
the probability that S starts from state i is equal to 1, then the vector p will be:
p
j
=
ij
, j I, where (3.9)

ij
=

1 i j
0 Otherwise
(3.10)
Methodology 15
3.1.2. Transition Probabilities
The transition probabilities are dened as p
ij
= P[X
r+n
= j | X
r
= i] r; r + n
T. Considering the homogeneity assumption and n = 2 gives
ij
(2) =

KI

ik
and
kj
i, j I. These probabilities are contained in the two-step matrix P
2
=
[
ij
(2)]
mm
i, j I. It can be shown that P
(2)
= P
2
i, j I. It implies that
P
(n)
= [
ij
(n)]
mm
and
ij
(n) = P[X
r+n
= j | X
r
= i] is the ij
th
element of P
n
.
3.1.3. Discrete Time Markov Process with an Absorbing State

ij
(n) = 0; i = j, n T (3.11)

ij
(n) = 1; n T (3.12)
The system cannot leave when it is in an absorbing state. A DTMP is called ab-
sorbent if it has at least one absorbing state and the system can migrate from any
non absorbing state to an absorbing state. When we have a process with at least
one absorbing state, the main information that we can get is:
The expected time before the process goes to the absorbing state.
lim
t

ij
(t), i, j I (3.13)
3.1.4. Problems of markov models
1. The duration inside a state. Actually the probability of changing rating de-
pends on the time that a rm remains in the same rating. Under the Markov
assumption, this probability depends only on the rank own at previous tran-
sition (Carty and Fons, 1999).
2. The dependence of the rating evaluation from the epoch of the assessment.
This means that, in general, the rating evaluation depends on when it is
Methodology 16
done and, in particular, on the business cycle Nickell et al (2000). The rating
evaluation done at time t generally is different from the one done at time s, if
s = t.
3. The dependence of the new rating from all history of the rms rank evolution,
not only from the last evaluation.
Actually the effect exists only in the downward cases but not in the case of up-
ward ratings in the sense that if a rm gets a lower rating (for almost all rating
classes) then there is a higher probability that the next rating will be lower than
the preceding one (Carty and Fons, 1994, Nickell et al,2000).
3.1.5. Assumptions based on the problems of Markov prob-
lems above.
The rst problem above will be solved or avoided by means of semi-Markov
processes (SMP). In fact in SMP the transition probabilities are function of
the waiting time spent in a state of the system.
The second problem can be faced in a general approach by means of a non-
homogeneous environment.
The third problem is solved increasing the number of states to differentiate
the case in which the system arrives in a state from a lower or a higher rating
evaluation.
The next part will present a short description of NHSMP. After this the reliabil-
ity non homogeneous semi-Markov model will be shown. In the successive para-
graph the relation between the reliability model and the credit risk problem will
be described. The model enlarges the number of states in this way the downward
problem can be solved.
Methodology 17
3.1.6. Non-homogeneous semi-Markov processes (NHSMP)
In this part the NHSMP will be described; the notation given in Janssen and
Manca (2005)is followed. First the stochastic process is dened. In SMP envi-
ronment two random variables (r.v.) run together. J
n
where n N with state space
{I = 1, 2, ....m}represents the state at the nth transition. T
n
where n N with state
space equal to R
+
represents the time of the nth transition,
J
n
: T
n
: R
+
We suppose that the process (J
n
, T
n
) is a non-homogeneous
markovian renewal process.
The kernel Q = [Q
ij
(s, t)] associated to the process is dened in the following way:
Q
ij
(s, t) = P[J
n+1
= j; T
n+1
t
n
= i; T
n
= s], (3.14)
p
ij
(s) = lim[Q
ij
(s, t)], (3.15)
and it results: where P(s) = [p
ij
(s)] is the transition matrix of the embedded non-
homogeneous Markov chain in the process. Furthermore it is necessary to intro-
duce the probability that process will leave the state i from the time s up to the
time t:
H
i
(s; t) = P[T
n+1
t | J
n
= j; J
n+1
= j; T
n
= s]. (3.16)
Obviously it results that:
H
i
(s; t) =
m

j=1
{Q
ij
s; t}. (3.17)
Now it is possible to dene the distribution function of the waiting time in each
state i, given that the state successively occupied is known:
G
ij
(s; t) = P[T
n+1
t | J
n
= j; J
n+1
= j; T
n
= s]. (3.18)
Obviously the related probabilities can be obtained by means of the following for-
Methodology 18
mula:
G
ij
(s; t) =

G
ij
(s,t)
P
ij
(s)
P
ij
(s)=0
0 otherwise
(3.19)
The main difference between a continuous time non-homogeneous Markov process
and a NHSMP is in the increasing distribution functions G
ij
(s; t). In Markov en-
vironment this function has to be a negative exponential function. Instead in the
semi-Markov case the distribution functions G
ij
(s; t) can be of any type. If we apply
the semi-Markov model in the credit risk environment we can take into account,
by means of the G
ij
(s; t) the problem given by the duration of the rating inside the
states. Now the NHSMP Z = (Z
t

+
)can be dened. It represents, for each wait-
ing time, the state occupied by the process. The transition probabilities are dened
in the following way:

ij
(s, t) =
ij
(1 H
i
(s; t)) +
n

=1

t
s
Q
i
(s; )
;j
(; t). (3.20)
represents the Kronecker symbol. The rst part of relation:

ij
(1 H
i
(s; t)) (3.21)
The probability that the system doesnt have transitions up to the time t given that
it was in the state i at time s. The formula in rating migration case represents
the probability that the rating organization doesnt give any new rating evaluation
from the time s up to the time t. This part has sense if and only if i j. In the
second part:
n

=1

t
s
Q
i
(s; )
;j
(; t). (3.22)
Q
i
(s; ) is the derivative at time of Q
i
and represents the probability intensity
that the system was at time s in the state i and remained in this state up to the
Methodology 19
time and that it went to the state just at time . After the transition the system
will go to the state j following one of the possible trajectories that go from the state
at the time to the state j within the time t. In the credit risk environment
it means that from the time s up the time the rating company doesnt give any
other evaluation of the rm; at time the rating company gave the new rating
for the evaluating rm. After this the rating will arrive to the state j within the
time t following one of the possible rating trajectories.
3.1.7. Non-homogeneous semi-Markov reliability model
Let us consider a reliability system S that can be at every time t in one of the states
of I = (1, ..., m). The stochastic process of the successive states of S is Z = (Z(t), t
0). The state set is partitioned into sets U and D, so that:
I = U D; = U D; = U; U = I (3.23)
The subset U contains all good states in which the system is working and subset
D all bad states in which the system is not working well or is failed. The classical
indicators used in reliability theory are the following ones: The non-homogeneous
reliability function R giving the probability that the system was always working
from time s to time t:
R(s, t) = P[Z(u) U : u (s, t]]. (3.24)
The point wise non-homogeneous availability function A giving the probability that
the system is working on time t whatever happens on (s, t]:
A(s, t) = P[Z(t) U]. (3.25)
Methodology 20
The non-homogeneous maintainability function M giving the probability that the
system will leave the set D within the time t being in D at time s:
M(s, t) = 1 P[Z(u) D : u(s, t]]. (3.26)
It is shown in Blasi et al (2003) that these three probabilities can be computed
in the following way, if the process is a non-homogeneous semi-Markov process of
kernel Q. The point wise availability function A
i
given that Z(0) = i is:
A(s, t) =

ij
(s, t). (3.27)
To compute these probabilities all the states of the subset D are changed in absorb-
ing states. R
i
(s,t) is given by solving the evolution equation of HSMP but now with
the embedded Markov chain having:
p
ij
=
ij
, i D; (3.28)
the related formula is: R
i
(s, t) =

r
ij
(t) where
r
ij
(t) is the solution of main Equa-
tion with all the states in D that are absorbing. The maintainability function M
i
given that Z(0) = i; in this case, all the states of the subset U are changed into
absorbing states. Then M
i
(t) is given by solving the evolution equation of HSMP
with the embedded Markov chain with p
ij
=
ij
, i U; the related formula is:
M
i
(s, t) =

m
ij
(t). (3.29)
3.1.8. The homogeneous semi-Markov reliability credit risk
model
The credit risk problem can be situated in the reliability environment. The rating
process, done by the rating agency, gives a reliability degree of a rm bond. In the
Standard and Poors case there are the 8 different classes of ratings that means to
Methodology 21
have the following set of states:
I = AAA, AA, A, BBB, BB, B, CCC, D (3.30)
To take into account the downward problem we introduce other 6 states. The set of
the states become the following:
I = AAA, AA, AA, A, A, BBB, BBB, BB, BB, B, B, CCC, CCC, D (3.31)
For example the state BBB is divided into BBB and BBB-. The system will be in
the state BBB if it arrived from a lower rating, instead it will be in the state BBB-
if it arrived in the state from a better rating (a downward transition). It is also
possible to suppose that if there is a virtual transition then if the system is in the
BBB- state it will go to the BBB state. The rst 13 states are working states (good
states) and the last one is the only bad state. The two subsets are the following:
U = AAA, AA, AA, A, A, BBB, BBB, BB, BB, B, B, CCC, CCC (3.32)
D = D. (3.33)
In this case, the maintainability function M does not make sense because the de-
fault state D is absorbing and once the system has gone into this state it is no
longer possible to leave it. Furthermore, the fact that the only bad state is an
absorbing state implies that the availability function A corresponds to the relia-
bility function R.Another important result that can be obtained by means of the
semi-Markov approach is the distribution function of the default conditioned to the
rating state at time 0.
Methodology 22
3.1.9. The transient analysis: Relevant credit risk indicators
Among all possible indicators that can be useful to investors and that can be de-
rived from the reliability model.In the credit risk environment the reliability model
is substantially simplied.In fact, to obtain all the results that are relevant in the
credit risk case it is enough to solve the system numerically only once.
On solving this system the following are obtained:

ij
(t) which represents the probabilities of being in state j after a time t starting in
state i at time 0; these results take into account the different transition probabili-
ties during the permanence of the system in the same state (ageing problem);
R
i
(s, t) = A
i
(s, t) =

ij
(t). (3.34)
The reliability function represents the probability that a rm will never go into the
default state in a time t. This function gives very important nancial information
in itself and, moreover, permits to compute the total interest rate overdue by the
rm because of credit risk.
A(s, t) =

ij
(s, t). (3.35)
The availability function gives the probability that the rm has an up rating at
time t. Notice that in the case above the availability coincides with the reliability
function.
M(s, t) = 1 P[Z(u) D : u(s, t)]. (3.36)
The maintainability function gives the probability that a rm will be re-organized
exiting from the down set within the time t given that at the starting time it was in
the default class. Notice that in the case above the maintainability does not make
sense because the state D is absorbing.
1Hi(t) which represents the probability that in a time interval t there was no new
Methodology 23
rating evaluation for the rm.Before giving another result that can be obtained in
an SMP environment,the concept of the rst transition after the time t must be
introduced. More precisely, it is assumed that the system at time 0 was in state i
and that it is known with probability 1 Hi(t) that the system does not move from
state i. Under these hypotheses it is possible to determine the probability of the
next transition being to state j. This probability is denoted by
ij
(t) and is dened
as
ij
(t) = P[X
n+1
= i | X
n
= j; T
n+1
T
n
> t]. This probability can be obtained as:

ij
(t) =
p
ij
Q
ij
1 Hi(t)
. (3.37)
SMP allows us to obtain the following:
which represents the probability of receiving the rank j at the next rating if the
previous state was i and no rating evaluation was done up to the time t; in this
way, for example, if the transition to the default state is possible and if the system
does not move for a time t from the state i,the probability that in the next transition
the system will go to the default state is known.
3.1.10. The asymptotic analysis
In order to study the asymptotic behavior of a Markov or a semi-Markov process
it is important to dene an equivalence relation on the set of the states of the pro-
cess. It is said that the two states i and j are in the same equivalence class if it is
possible to go from state i to state j and from state j to state i. A class of states is
transient if the system can get out of the class and absorbing if, once the system
goes into the class, it cannot get out of it. A process is irreducible if there is only one
equivalence class, unireducible if there is only one absorbing class and reducible if
there is more than one absorbing class. If the process is irreducible or unireducible
there is a limit vector. In the irreducible case:

i
=

i

k
where = {
1
,
2
, .....,
m
} is the limit vector of the embedded Markov
chain and
k
=

p
ik

kj
where
ij
is the mean of the distribution function G
kj
re-
lated to the duration of passage from the state j to the state k. In the unireducible
Methodology 24
case the set of the states can be divided in two parts I = U D, where U contains
the transient states and D the absorbing ones. In this case:

i
=

0 i U

k
i U
(3.38)
where the
i
0 constitute the limit vector of the sub-Markov chain formed by the
absorbing states.In the case of the credit risk model, the semi-Markov process is
unireducible,the set D contains only one state and relation above becomes

i
=

0 i U
1 i U
(3.39)
Furthermore
lim
t

ij
(t) =
i
, (3.40)
and then
lim
t

iD
(t) = 1 i U, (3.41)
It is known that

iD
(t) = P[Z(s) = D | Z(0) = i], (3.42)
moreover, D is the unique absorbing state so that

iD
(s)
iD
(t)i UIfs t. (3.43)
A Markov chain X(t) denoting the rating class occupied at time t (t 0). Fur-
thermore,we assume that the chain has a nite number of states, 1, 2, ..., l
D
. Here
1, ..., l
D
1 are dened by decreasing levels of credit quality and D is the default
state, which is absorbing. Given that class l is entered at calendar time t and is still
Methodology 25
occupied at t +s, the transition out of l is determined by the set of l
D
1 transition
intensities.
q

ll
(t, s) = q

ll
, (3.44)
q
ll
=

ll
, (3.45)
,where l = l
1
In probability theory the attribute expected always refers to an expectation or
mean value, and this is also the case in risk management. The basic idea is as
follows: A default probability (DP) was assigned to each obligor, a loss fraction
called the loss given default (LGD), describing the fraction of the loans exposure
expected to be lost in case of default, and the exposure at default (EAD) subject to
be lost in the considered time period. The loss of any obligor is then dened by a
loss variable.
The current rating (reects the probability of default, PD)
The terms of the loan (reect the loss given default, LGD)
The loan size (reects the exposure at default, EAD)
The closeness of the bank-borrower relationship (reects the moral hazard
component of credit risk).
Thus the structure of the monitoring times will vary for each borrower, depending
on his level of expected loss. For obtaining representation of the EL, we need some
additional assumption that EAD and LGD are constant values.This is not neces-
sarily the case under all circumstances. There are various situations in which, for
example, the EAD has to be modeled as a random variable due to uncertainties in
amortization, usage, and other drivers of EAD up to the chosen planning horizon.
EL = EAD LGD P(D) (3.46)
Let Q denote the l
D
l
D
intensity matrix or generator.
Methodology 26
3.1.11. Conclusions
By means of Semi Markov Reliability models approach a generalisation of the tran-
sition probabilities of an Homogenous Semi Markov Processes is given and we show
how it is possible to dertemime the reliability probability over a period of time.
Semi Markov Reliability models are very useful as they can be used to track or
monitor credit migrations from time to time.
Chapter 4
Data Analysis and Simulations
4.1. Introduction
The objective of this chapter is to provide estimators for reliability indicators of a
system used by Post Ofce Savings Bank . In order to achieve this purpose, we
estimate the basic quantities of a discrete-time semi-Markov system by the use of
Minitab 16 and Matalab 6.1. Data is also analysed in this chapter.
4.1.1. POSB Rating Systems
The Internal rating system assigns both an obligor rating to each borrower (or
group of borrowers), and a facility rating to each available facility. A risk rating
(RR) is designed to depict the risk of loss in a credit facility. A robust risk rat-
ing system should offer a carefully designed, structured, and documented series of
steps for the assessment of each rating. The goal is to generate accurate and con-
sistent risk rating, yet also to allow professional judgment to signicantly inuence
a rating whenever appropriate. A typical risk rating methodology initially assigns
an obligor rating that identies the expected PD by that borrower (or group) in
repaying its obligations in the normal course of business.
Data Analysis and Simulations 28
Risk ratings quantify the quality of individual facilities, credits, and portfolios.
If RR is accurately and consistently applied, then they provide a common under-
standing of risk. Levels of credit management allow for active portfolio manage-
ment. An IRRS also provides the initial basis for capital charges used in the vari-
ous pricing models. It can also assist in establishing loan reserves. The IRRS ()can
be used to rate credit risks in most of the major corporate and commercial sectors,
but it is unlikely to cover all business sectors.
From the data collected, all transitions made by a single borrower over 2 years
observation period and the corresponding exact transition times are known. In or-
der to rate the companies whose data was used for this study, it was ensured that
their data is exact ,that is, exact transition times are available.
Grade 1 corresponds to the lowest and grade 8 to the highest degree of credit risk.
1 (AAA), 2(AA), 3(A), 4(BBB), 5(BB), 6(B), 7(CCC) and 8(D).Below is a scale that is
used by POSB to come up with credit ratings.
Credit score Rate Credit Rate
89%- 100% 1 AAA
76%- 88% 2 AA
64%- 75% 3 A
51%- 63% 4 BBB
39%- 50% 5 BB
26%- 38% 6 B
14%- 25% 7 CCC
00%- 13% 8 D
Data Analysis and Simulations 29
4.1.2. Normality testing and data transformations.
The data collected from POSB was tested for normality. Initially the data was not
normally distributed for January 2010 to December 2012. The data was then trans-
formed using Johnson transformation method and it became normally distributed.
Figure 4.1 below shows January 2010 statistics before and after Johnson transfor-
mation. It can be seen from gue 4.1 that the data is now normally distributed.
Tests and data transformations were done for all months although only 3 months
were used below to show how data was transformed using Johnsons transforma-
tion method.
Figure 4.1: Probability Plot for January 2010
Figure 4.2 below shows the distribution of February 2010 statistics before and af-
Data Analysis and Simulations 30
ter Johnsons transformation.Figure 4.2 below shows that February 2010 data is
normally distributed after Johnsons transformation.
Figure 4.2: Probability Plot for February 2010
Figure 4.3 below shows the distribution of May 2011 statistics before and after
Johnsons transformation.Figure 4.2 below shows that February data is normally
distributed after Johnsons transformation.
Data Analysis and Simulations 31
Figure 4.3: Probability Plot for May 2011
4.1.3. Calibration of Default Probabilities to Ratings
The banks systemhas eight main grades,to construct an eight grade scale,the main
grades are considered. Rating grade 1 corresponds to the lowest and grade 8 to the
highest degree of credit risk.
Data Analysis and Simulations 32
Industry Company 2010:Q1 Q2 Q3 Q4 2011:Q1 Q2 Q3 Q4
Agriculture A 1.00 0.93 0.86 0.92 0.84 0.93 0.93 0.86
B 0.98 0.93 0.93 0.93 0.93 0.77 0.93 0.76
C 0.98 0.93 0.83 0.88 0.82 0.93 0.91 0.88
D 0.98 0.88 0.99 1.00 1.00 1.00 1.00 1.00
E 0.98 0.93 0.93 0.93 0.93 0.93 0.76 0.71
F 0.94 0.76 0.93 0.93 1.00 1.00 0.88 0.93
G 0.93 0.76 0.76 0.85 0.93 0.66 0.59 0.62
Mining H 1.00 0.98 0.94 0.85 0.68 0.75 0.5 0.53
I 0.94 0.88 0.85 0.82 0.73 0.78 0.85 0.93
J 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
K 0.86 0.80 0.73 0.56 0.50 0.67 0.74 0.69
L 1.00 0.79 1.00 1.00 1.00 1.00 1.00 1.00
M 0.67 1.00 1.00 0.83 1.00 1.00 1.00 1.00
N 1.00 0.79 0.67 0.67 1.00 1.00 1.00 1.00
O 0.94 0.95 0.95 0.99 0.84 1.00 1.00 1.00
Production P 0.95 0.95 0.95 0.95 0.95 0.95 0.95 0.95
Q 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90
R 0.94 0.90 0.94 0.94 0.94 0.9 0.95 0.95
S 0.75 0.89 0.9 0.95 1.00 1.00 1.00 1.00
T 0.78 0.57 0.73 0.79 0.95 0.95 0.95 0.95
Table 4.1: Credit Scores for companies as from 2010 to 2011 (Quarterly)
The table above shows historic scores or ratings for the years 2010 up to 2011 in
Quarters.The ratings were derived by POSB by a means of an internal tool that as-
sesses creditworthiness of obligors, Security/ Collateral, average prot generated
by the company, credibility, integrity, premises, Authenticity of Information, Ac-
count History / performance, gearing / level of indebtedness and purpose of loan.
Now, an important observation is that for best ratings few or no defaults at all
should occur.
The credit scored for companies A to T are not constant, that is, there are uc-
tuating due to the factors stated earlier. The review of creditworthiness of obligors,
Security/ Collateral, average prot generated by the company, credibility, integrity,
premises, Authenticity of Information, Account History / performance,Gearing /
level of indebtedness and purpose of loan dertemines the changes in credit scores
Data Analysis and Simulations 33
assigned to companies over a given period of time. Company L under the mining
sector is an exception though as it was given a credit score of 100% throughout the
period of assessment by Post Ofce Savings Bank.
Positive default probabilities are assigned to the ratings on the table above.Calibration
is done in three steps:
1. We denote H
i
(R) =the historic default frequency of rating class R for Quarter
i and let i range from 1 up to 160, that is each entry of a score from 2010 to
2011. For example H
1
(R) = 100% ,H
8
(R) = 86% for company A and H
9
(R) =
98% and H
16
(R) = 76% for company B. Then the mean value is computed and
the standard deviation of these frequencies over the years, where the rating
is xed, namely:
m(R) =
1
160
160

i=1
H
i
(R) (4.1)
s(R) =
1
159
160

i=1
(H
i
(R) m(R))
2
(4.2)
The value m(R) for rating R is our rst guess of the potential default proba-
bility assigned to rating R. The standard deviation s(R) gives us some insight
about the volatility and therefore about the error we eventually make when
believing that m(R) is a good estimate of the default probability of R-rated
obligors.
2. Next, we plot the mean values m(R) into a coordinate system,where the x-axis
refers to the rating classes (here numbered from 1 to 8. 1 (AAA), 2 (AA), 3 (A),
4 (BBB), 5 (BB), 6 (B), 7 (CCC) and 8 (D). One can see in the chart in Figure
4.2 below that on a logarithmic scale the mean default frequencies m(R) can
be tted by a regression line. Here we should add a comment that there is
Data Analysis and Simulations 34
strong evidence from various empirical default studies that default frequen-
cies grow exponentially with decreasing creditworthiness. For this reason we
have chosen an exponential t (linear on logarithmic scale). Using standard
regression theory,by MINITAB software providing basic statistical functions,
the following exponential function tting data was used:
DP(x) = 3 10
5
e
0.5075x
, x = [1, 8] (4.3)
Figure 4.4: Probability of default versus credit ratings
Data Analysis and Simulations 35
3. As a last step, we use our regression equation for the estimation of default
probabilities DP(x) assigned to rating classes x ranging from 1 to 8. Table
4.3 shows our results, which we now call a calibration of default probabilities
to POSB ratings. Note that based on our regression even the best rating Aaa
has a small but positive default probability. Moreover, our hope is that our
regression analysis has smoothed out sampling errors from the historically
observed data.
Let Q
ij
be quarter i in year j, where i = [1, 8] and j = [2010, 2011].Credit ratings (1 to
8) were regressed against scores per quarter for each obligor. Q
4;11
, Q
3;10
, Q
1;11
, Q
4;10
,
Q
2;10
, Q
1;10
, Q
2;11
were eliminated from the regression following the order there are
stated using backward elimination of regression. Only Q
3;11
and a constant were
signicant since the P Values are very small for the stated variable and a constant.
4.1.4. Regression Analysis: Rates versus Q
3;11
We wish to test or come out with variables which are more important or signicant
in dertiming the credit ratings in this study. There are a quite a number of factors
leading to the uctuation of credit scores per quarter but we want to come up with
the quarters that were are more signicant in a regression model.
Hypothesis
H
0
: At least one credit score done per quarter dertemines a credit rating.
H
1
: No credit score done per quarter dertemines a credit rating.
The regression equation is CreditRatings = 11.5 9.46Q
3;11
Predictor Coeff SE Coeff T P
Constant 11.522 1.327 8.68 0.000
Q
3;11
-9.460 1.473 -6.42 0.000
Table 4.2: Regression Analysis
Data Analysis and Simulations 36
From table 4.1 |T| > 2 ,that is, 8.68 for the constant and 6.42 for the variable Q
3;11
and this implies that Q
3;11
and the constant are signicant in the regression model.
The signicancy of the conastant and the variable Q
3;11
is also shown by P values
which are very small,that is,P = 0 for both Q
3;11
and the constant . The credit rat-
ings derived from the credit scores can be predicted by the use Q
3;11
data, meaning
that Q
3;11
is very signicant as it explains the behaviour of credit ratings. It is very
important to Post Ofce Savings Bank to monitor credit ratings as default depends
on credit ratings of a companies (Obligors).
Source DF SS MS F P
Regress 1 34.676 34.66 41.27 0.000
Res Error 18 15.124 0.840
Total 19 49.800
Durbin-Watson statistic = 1.9770
S = 0.916648, R Sq = 69.6% , R Sq(adj) = 67.9%,and R Sq(pred) = 65.54%
The independent variable Q
3;11
and a constant of the model explains 69.6% of the
credit ratiings. In other words variability or uctuations in credit ratings are ex-
plained by 69.6% of the constant and the variable Q
3;11
.
Data Analysis and Simulations 37
Obligor Credit rating Mean Standard-deviation Default Probability
A A 0.04 0.034 0.04%
B BBB 0.04 0.034 0.02%
C A 0.04 0.029 0.01%
D AA 0.05 0.042 0.08%
E BBB 0.04 0.034 0.02%
F A 0.04 0.036 0.01%
G B 0.04 0.027 0.06%
H CCC 0.04 0.027 0.01%
I BB 0.04 0.028 0.03%
J AAA 0.05 0.045 0.04%
K BBB 0.04 0.026 0.02%
L AA 0.05 0.044 0.08%
M A 0.05 0.042 0.01%
N AA 0.04 0.034 0.08%
O AA 0.05 0.039 0.08%
P AAA 0.05 0.037 0.04%
Q AAA 0.04 0.029 0.04%
R A 0.04 0.034 0.01%
S AA 0.04 0.035 0.08%
T BBB 0.04 0.028 0.02%
Table 4.3: Calibration of POSB Ratings to Default Probabilities
The range of default for companies A to T is [0.01%;0.08%]. The listed companies
in table 4.3 above were assessed before there were given loans hence they are con-
ning to the standards set by Post Ofce Savings Bank, that is, the probability
of defaulting is very low. Some of the companies failed to pay required monthly
repayment but they paid Post Ofce Savings Bank a percentage or fraction of the
premium due hence the probability of default is very low.
Investing in different companies falling under different sectors generally reduces
the overall portfolio risk, because usually it is very unlikely to see a large number
of loans defaulting all at once under different industries. POSB invested in Agricul-
tural, Mining and Production companies thereby reducing the credit portifolio risk.
The less the loans in the portfolio have in common, the higher the chance that
Data Analysis and Simulations 38
default of one obligor does not mean a lot to the economic future of other loans in
the portfolio and this case minimizes the overall portfolio risk.In this case if POSB
was investing in 7 agricultural companies only for example, then that means that
the 7 companies A - F are interrelated and default can be very high in such a sce-
nario, as companies A-F will be targeting the same markets and these companies
are all under the agricultural sector and are likely to be affected by the same fac-
tors and conditions. The companies within the same sector increase the likelihood
of default and credit risk.
Data Analysis and Simulations 39
Industry AAA AA A BBB BB B CCC D
Agriculture AAA 1
AA 0.632 1
0.128
A -0.232 0.544 1
0.617 0.207
BBB -0.179 0.648 0.940 1
0.702 0.115 0.002
BB -0.167 -0.258 0.241 -0.105 1
0.721 0.576 0.602 0.823
B -0.248 0.466 0.720 0.825 -0.248 1
0.592 0.291 0.068 0.022 0.592
CCC -0.280 0.230 0.442 0.551 -0.280 0.366 1
0.543 0.620 0.321 0.200 0.543 0.419
D -0.476 -0.176 0.181 0.232 -0.132 0.443 0.714 1
0.280 0.706 0.698 0.617 0.777 0.320 0.072
Mining AAA 1
AA -0.197 1
0.672
A 0.219 0.763 1
0.637 0.046
BBB 0.732 0.339 0.768 1
0.061 0.456 0.044
BB 0.393 -0.109 0.471 0.520 1
0.383 0.816 0.286 0.232
B 0.089 -0.322 0.148 0.234 0.825 1
0.850 0.481 0.751 0.614 0.022
CCC -0.010 -0.329 0.035 0.032 0.753 0.937 1
0.982 0.471 0.941 0.945 0.051 0.002
D 0.143 -0.230 0.199 0.168 0.874 0.877 0.952 1
0.759 0.620 0.669 0.719 0.010 0.009 0.001
Production AAA *
AA * 1
*
A * * 1
* *
BBB * * * 1
* * *
BB * * * * 1
* * * *
B * * * * * 1
* * * * *
CCC * * * * * * 1
* * * * * *
D * * * * * * * 1
* * * * * * *
Table 4.4: Correlations of credit ratings under 3 portifolios
Data Analysis and Simulations 40
The correlation between credit rate AA and AAA is 0.632 under the Agriculture in-
dustry and under Mining credit rate AA and AAA are negatively related (-0.197).
This means that even if obligors running mining companies are to default due to
poor business or markets it does not mean a lot to the economic future loans as all
industries can not collapse at the same time.
The data is aggregated to the portfolio level taking into account diversication
effects. Across different bank customers,signicant correlations among default
events/rating migrations. The nancial condition of rms in the same industry or
within the same country reect similar factors, and so may improve or deteriorate
in a correlated fashion.Similarly, for rms within the same industry,exposures due
to drawdowns of credit lines, increase (decrease) relative to their long-run averages
in periods when the average condition of rms in that sector is deteriorating (im-
proving). While POSB tend to be well aware of these potential relationships, their
ability to model such correlations is often limited in practices.
Production companies are reliable as there are no correlations of credit ratings
displayed in the correlation matrix above for production companies as the ratings
are 100% throughout as from the time POSB started assessing the companies. It
is therefore easy to predict that companies under the production sector are doing
very well and chances are very high that in future if given loans there are likely
to pay all monthly repayments without defaulting. Companies under the Agricul-
tural sector save for a purpose of curbing credit risk as interests charged are likely
to cover for defaults or losses incurred under Production and Mining.
It is a good thing for Post Ofce Savings Bank to accomodate more than one in-
dustry or sector, this is a benet in the sense that if a lot of companies in one sector
for example mining default then the other sector can cover up. This as a result
eliminates or reduces direct loss of earnings and erosion of capital or may result
in imposition of constraints on a banks ability to meet its business objectives. The
Data Analysis and Simulations 41
disadvantage of having obligors from one sector is that it could hinder a Post Of-
ce Savings Bank (POSB) capability to conduct its business or to take advantage of
opportunities that would enhance its business. As such, managers of banking insti-
tutions are expected to ensure that the risks an institution is taking are warranted
as seen with companies falling under the production sector that have a constant
reliability of 100%.
Companies are likely to remain in the same state if they default as banks like
POSB carry out assessments before granting loans thereby making it impossible
for bankrupt companies to access loans, i.e the correlation between D and D is 1
(absorbing state). It is a challenge for companies to migrate in terms of credit risk
from speculative grades like BB, B, CCC and D to investment grades like BBB, A,
AA and AAA and this is shown by low and negative correlations between Invest-
ment grades (higher grades) and speculative grades (lower grades).
The transition matrix was constructed starting from t = 2
AAA AA A BBB BB B CCC D
AAA 82.33 10.93 0.55 0.63 0.32 0.19 0.05 0.02
AA 2.84 86.23 5.47 4.16 0.11 1.13 0.055 0.02
A 0.27 1.59 89.05 7.40 1.48 0.13 0.06 0.03
BBB 1.84 1.89 5.00 84.21 6.51 0.32 0.16 0.07
BB 0.08 2.91 3.29 5.53 78.68 4.05 4.14 1.32
B 0.21 0.36 9.25 8.29 2.31 63.89 10.13 5.58
CCC 1.06 0.25 0.85 2.06 14.34 22.86 38.97 19.60
D 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00
Table 4.5: Probability of migrating from one state to another within 2 years as a
percent (
ij
(2))
Data Analysis and Simulations 42
Figure 4.5: Transition probabilities(t = 2)
The transition matrices were constructed starting from the 2 year transition ma-
trix, that is at t = 2. The matrix shown in table 2.5 shows
ij
(2)s. Element AAA
(82.33%) shows that the probability of companies (obligors) remaining in the same
credit rate, that is, AAA is very high irregardless of the sectors or industries. At
time 0 companies rated D are absorbed in that state as they can not be given any
loans. It might take time for companies rated AA to migrate to rate AAA within a
period of 2 years and this is shown by the probability of 2.84%.
AA rated companies in the bank portifolio at time 0 are likely to remain in the
same state for 2 years and there are likely to migrate to credit rating A (5.47%)and
Data Analysis and Simulations 43
BBB (4.16%). AA is an investment grade so the probability of migrating from AA
to speculative grades is very low. Its very rare for a bank to use a credit rating sys-
tem that may bring in clients with high credit rates only to default 2 years down
the line. As seen in the transition probability matrix in table 2.5 the probability
of migrating from AA to BB is 1.11% (very small) and the probability of companies
migrating from credit rating AA to B, CCC and D approaches zero because its rare
for companies rated AA to default.
AAA AA A BBB BB B CCC D
AAA 90.13 6.04 0.50 0.15 0.16 0.00 0.00 0.00
AA 0.46 94.42 4.33 0.52 0.10 0.17 0.00 0.00
A 0.05 1.51 94.4 29.50 0.70 0.33 0.00 0.06
BBB 0.03 0.30 3.70 4.11 90.38 0.98 0.11 0.40
BB 0.02 0.15 0.57 4.73 5.89 85.62 0.91 2.11
B 0.10 0.00 0.20 0.35 3.38 2.40 89.00 4.58
CCC 0.00 0.00 0.47 0.54 3.46 1.28 85.29 8.96
D 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
Table 4.6: Probability of migrating from one state to another within 2 years as a
percent (
ij
(4))
Data Analysis and Simulations 44
Figure 4.6: Transition probabilities(t = 4)
From time t = 2 to time t = 4 if POSB is to mantain the same number of compa-
nies under each category,that is,Agriculture, Mining and Production and with the
same or similar credit ratings at time t = 0 then the transition probabilities that
are likely to be followed are shown in table 4.6. The probability of remaining in
the same state for AAA rated companies is 90.13% meaning that companies falling
under investment grades mantantain credit worthiness, security/collateral, prot
generated, premises and aunthenticity of information and these are components
used to assess credit rates.
Data Analysis and Simulations 45
The probability of migrating from AAA to B, CCC and D is 0 at time t = 4. Not only
investment companies benet from granting loans to companies but even compa-
nies tend to enjoy the benets of loans if utilised accordingly. Companies which
have investment rates therefore tend to grow with time.The probability of a credit
migration from A to AA is 0.05%, from A to AAA is 1.51% but the probability of re-
maining in the samme rate, that is, A for obligors is very high ,that is, 94.4%. Most
of the companies (obligors)tend to mantain the credit rate assigned at the initial
time or migrate backward because of poor management and the economic situation
in Zimbabwe.
Companies rated BBB are likely to migrate backward in the scale of credit rat-
ing this is shown by the probability of 90.38% of migrating from BBB to BB. The
probability of a company migrating from credit rate BBB to AAA in a period of 4
years is unlikely and this is shown by the probability of 0.03%. For companies with
speculative grades at initial stage migration backwards is likely and this is shown
by the probability of migrating from B to BBB for example which is 85.62%. There
is 89% probability that companies rated B will migrate to credit rating CCC. CCC
rated companies are likely to remain in the same state for a long time or default
and this can be seen in table 4.6 that the probability of remaining in the same state
is 85.29% and 8.96% for migrating from CCC to D.
Data Analysis and Simulations 46
AAA AA A BBB BB B CCC D
AAA 92.01 7.04 0.40 0.14 0.41 0.00 0.00 0.00
AA 1.46 90.42 5.33 0.02 0.15 0.00 0.00 2.62
A 81.33 11.93 0.55 0.64 0.34 0.19 5.00 0.02
BBB 0.84 88.23 7.47 2.16 1.11 0.12 0.07 0.00
BB 0.27 1.59 89.05 7.40 1.48 0.15 0.04 0.02
B 1.84 1.89 5.00 84.21 6.51 0.32 0.16 0.07
CCC 0.08 3.91 2.29 5.53 14.68 8.05 10.14 55.32
D 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
Table 4.7: Probability of migrating from one state to another within 2 years as a
percent (
ij
(6))
Figure 4.7: Transition probabilities (t = 6)
Data Analysis and Simulations 47
Table 4.7 shows
ij
(t), the probability of companies rated AAA remaining at the
same state is 92.01% after a duration of 6 years. The probability of defaulting for
such companies is 0.00% meaning that such companies are established to the ex-
tent that they remain in the same credit grade.Companies rated CCC are likely to
default this is shown by 55.32% probability. Such companies possess speculative
grades and are likely to collapse.Companies with credit rating BBB are likely to
attain a AA credit rate after a certain period of time.
R
i
(s, t) = A
i
(s, t) =

ij
(t)
R
i
(s, t) =

ij
(2) R
i
(s, t) =

ij
(4) R
i
(s, t) =

ij
(6)
AAA 0.80 0.90 0.95
AA 0.88 0.89 0.85
A 0.81 0.72 0.67
BBB 0.90 0.88 0.79
BB 0.60 0.60 0.80
B 0.50 0.40 0.45
CCC 0.20 0.13 0.11
D 0.00 0.00 0.00
Table 4.8: Reliability for companies rated AAA, AA, A, BBB, BB, B, CCC and D at
t = 2, t = 4 and t = 6
The reliability for companies rated AAA are 80% at time 2, 90% at time 4 and 95%
at time 6. The reliability for companies rated AAA is increasing with time. AAA
rated companies expand in terms of prot generation and growth. Investment com-
panies and banks benet most from AAA rated companies since the probability of
such companies defaulting is nil.
For AA rated companies the reliability is not stable it is uctuating, that is, 88% at
t = 2, 89% at t = 4 and 85% at t = 6. Companies falling under this credit rating are
reliable although they lack consistency. Same applies to credt rating BBB whose
reliability is decreasing from 90% to 88% and nally to 70%. From the table above
it can be seen that Credit rating CCCs reliability is decreasing from 20% to 13%
Data Analysis and Simulations 48
and nally to 11%.
Credit rating D is not reliable at all. Post Ofce Savings Bank can not or is not
advised to invest in such companies as it may lead to credit risk. Most companies
rated CCC are bankrupt they can not pay back investing companies. The reliabilty
for t = 2 to t = 6 is 0%
4.2. Conclusions
In POSB there are 8 different classes of rating expressing the creditworthiness
of the rated rm.The creditworthiness is highest for the rating AAA, assigned to
rm extremely reliable with regard to nancial obligations and decrease towards
the rating D which expresses the occurrence of payment default on some nancial
obligation.Figure 4.3 to 4.4,shows the reliability function for rating class i . As it is
possible to see, the reliabilities have a different behaviour depending on the values
of t. This is due to the non homogeneity of the rating process.
The reliability exhibits also variability as a function of the backward process, then
as a result, our model assigns different survival probabilities to rms having the
same rating class but with different age in this state. From the probabilities pre-
sented in g 4.3 to g 4.4 we conclude that companies with investment grades
are more reliable unlike those with speculative grades which are likely to default.
Companies with investment grades are likely to mantain their credit rates/ status
for a long period of time, for speculative grades the opposite is the case as such
companies tend to migrate towards the default state.
Chapter 5
Conclusions and
Recommendations
5.0.1. Conclusions
Rating migration matrices, which measure the expected changes in credit quality
of borrowers,are cardinal inputs to many applications, including credit portfolio
monitoring and management, capital allocation and pricing of credit derivatives.
Bank management and regulators are keenly interested in accurately estimated
transition probabilities associated with internal rating grades for as to curb credit
risk. The ndings of this paper have some important implications for both, risk
management and banking supervision. We fnd that internal ratings are more point
in time ratings, which implies that up and downgrade behaviour of credit ratings
can be done if internal rating is done in the system.
Risk-taking is an inherent element of banking and, indeed, prots are in part the
reward for successful risk taking in business. On the other hand, excessive and
poorly managed risk can lead to losses and thus endanger the safety of a banks
depositors. Accordingly, the Post Ofce Savings Bank should continue to place a
Conclusions and Recommendations 50
signicant management system of risk. Risk in POSB refers to the possibility that
the outcome of an action or event could bring adverse impacts on the institutions
capital, earnings or its viability. For the data gathered POSB loaned out USD
3,987,000.00 to companies under different industries so as to minimise loss due
to defaulting. The amount loaned to companies (obligors) was allocated as follows
USD 1,765,000.00 which makes 44.3% of the total ammount to the agicultural sec-
tor, USD 1,312,000.00 (32.9%) to the mining industry and USD 910,000.00 (22.8%)
to the production sector.
Credit scores assigned after a thourogh assessment of credit worthiness, security/
collateral, prot generated, credibility, the value of premises, aunthenticity of in-
formation, account history/ perfomance, level of indebtedness and purpose of loan
were used to further allocate the stated ammounts to companies under each sector.
Risks are considered warranted when they are understandable, measurable, con-
trollable and within POSB capacity to readily withstand adverse results. Sound
risk management systems like the credit score card system used enable managers
of POSB to take risks knowingly, reduce risks where appropriate and strive to pre-
pare for a future, which by its nature cannot be predicted with absolute certainty.
To further curb or manage risk nancial models need to be used to scrutinise the
data collected by POSB internal tools like the credit score card. POSB attaches
considerable importance to improve the ability to identify, measure, monitor and
control the overall risks assumed by assessing the value of assets that each com-
pany (obligor) possess.
Because of the vast diversity in risk and change of the economic situation POSB
should not just stick to the system in use. Post Ofce Saving Bank should tailor its
risk management program to its needs and circumstances so as to minimise risk.
In order to properly manage risks, POSB must recognize and understand risks
that may arise from both existing and new business initiatives; for example, risks
inherent in lending activity include credit, liquidity, interest rate and operational
risks. Since POSB mainly deals with 3 sectors that is Agricultural, Mining and
Conclusions and Recommendations 51
Production it should be able to made predictions on the perfomance of each sector
in terms of generating prots and establishment. Risk identication should be a
continuing process, and should be understood at both the transaction and portfolio
levels.
Credit scores shown in table 4.1 shows that companies under mining are under-
perfoming hence once risks have been identied, they should be measured in order
to determine their impact on the banking institutions protability and capital. A
model should be deviced to invest the least ammount to such a sector with com-
panies with low score rates. As can be seen from table 4.1 the least credit score
in all 8 quarters is 50% for company K at the rst quarter of 2011. This can be
done using various techniques ranging from simple to sophisticated models. Ac-
curate and timely measurement of risk is essential to effective risk management
systems. An institution that does not have a risk measurement system has lim-
ited ability to control or monitor risk levels. POSB should periodically test their
risk measurement tools to make sure they are accurate for example after maturity
time management can analyse if the tool used to assess companies minimised risk
or not.Good risk measurement systems assess the risks of both individual transac-
tions and portfolios (sectors in this case).
5.0.2. Recommendations
Management information system (MIS) to monitor risk levels and facilitate timely
review of risk positions and exceptions can be of use and they can be used to de-
tect risk earlier than manual forms like the credit score card used by POSB. For
example company A under the Agricultural sector may apply for a loan thereby
providing all the nessecary information that is fed to the information system .
Such system should be able to automatically calculate a credit of 100% for the
rst quarter in 2010.Monitoring reports should be frequent, timely, accurate, and
informative and should be distributed to appropriate individuals to ensure action,
Conclusions and Recommendations 52
when needed.
After measuring risk,POSB should establish and communicate risk limits through
policies, standards, and procedures that dene responsibility and authority. These
limits should serve as a means to control exposure to various risks associated with
the banking institutions activities. Institutions may also apply various mitigating
tools in minimizing exposure to various risks like computirised credit score tools
that automatically pick all the information about the obligor from the system and
provide a score without physical calculation. POSB should have a process to au-
thorize and document exceptions or changes to risk limits when warranted. Some
companies (obligors) may default and as a result POSB should not invest in such
companies even if they are to apply for loans after a long period after they have
defaulted.
Boards of directors have ultimate responsibility for the level of risk taken by their
institutions. Accordingly, they should approve the overall business strategies in-
cluding risk management and enforcing signicant policies of their institution, in-
cluding those related to managing and taking risks, and should also ensure that
senior management is fully capable of managing the activities that their institu-
tions undertake. While all boards of directors are responsible for understanding
the nature of the risks signicant to POSB and for ensuring that management is
taking the steps necessary to identify, measure, monitor, and control these risks,
the level of technical knowledge required of directors may vary depending on the
particular circumstances at the bank.
Directors should have a clear understanding of the types of risks to which their
institutions are exposed and should receive reports that identify the size and sig-
nicance of the risks in terms that are meaningful to them. In fullling this re-
sponsibility, directors should take steps to develop an appropriate understanding
of the risks their institutions face, possibly through briengs from auditors and ex-
perts external to the institution. Using this knowledge and information, directors
Conclusions and Recommendations 53
should provide clear guidance regarding the level of exposures acceptable to their
institutions and have the responsibility to ensure that senior management imple-
ments the procedures and controls necessary to comply with adopted policies.
POSB senior management is responsible for implementing strategies in a manner
that limits risks associated with each strategy and that ensures compliance with
laws and regulations on both a long-term and day-to-day basis. Accordingly, man-
agement should be fully involved in the activities of their institutions and possess
sufcient knowledge of all major business lines to ensure that appropriate poli-
cies, controls, and risk monitoring systems are in place and that accountability and
lines of authority are clearly delineated. Senior management is also responsible for
establishing and communicating a strong awareness of and need for effective in-
ternal controls and high ethical standards. Meeting these responsibilities requires
senior managers of POSB to have a thorough understanding of banking and -
nancial market activities and detailed knowledge of the activities their institution
conducts, including the nature of internal controls necessary to limit the related
risks.
POSB directors and senior management should tailor their risk management poli-
cies and procedures to the types of risks that arise from the activities the institu-
tion conducts. Once the risks are properly identied, policies and its more fully
articulated procedures should provide detailed guidance for the day-to-day imple-
mentation of broad business strategies, and generally include limits designed to
shield POSB from excessive and imprudent risks. While all institutions should
have policies and procedures that address their signicant activities and risks.Policies
are written statements of the institutions commitment to pursue certain objectives
and results.
For effective risk monitoring POSB requires to identify and measure all material
risk exposures. Consequently, risk monitoring activities must be supported by in-
formation systems that provide senior managers and directors with timely reports
on the nancial condition, operating performance, and risk exposure of the insti-
Conclusions and Recommendations 54
tution, as well as with regular and sufciently detailed reports for line managers
engaged in the day to day management of the bank. POSB should have risk mon-
itoring and management information systems in place that provide directors and
senior management with a clear understanding of the banking positions and risk
exposures.
Besides an ineffective credit risk system at banks there are many other factors that
contribute to the collapse of banks and these include difcult macro economic en-
vironment, inadequate risk management systems, poor corporate governance, non
performing insider loans, chronic liquidity challenges, diversion from core banking
to speculative activities, foreign exchange shortages, rapid expansion drives, un-
sustainable earnings, creative accounting and insufcient regulatory framework.
POSB in the survey measure credit loss over a two-year time horizon. The reasons
put forward for this choice generally favour computational convenience rather than
model optimisation; furthermore, banks do not appear to test the sensitivity of
their model output to the chosen horizon. The reasonableness of this decision rests
on whether one year is indeed a period over which either fresh capital can be raised
to fully offset portfolio credit losses beyond that horizon, or risk mitigating actions,
such as loan sales or the purchase of credit protection, can be taken to eliminate
the possibility of further credit losses. In assessing the capacity of models to meet
various risk management and capital allocation needs, the choice of horizon would
appear to be an important decision variable.
The use of technology related products, activities, processes and delivery channels
exposes a bank to strategic, operational, and reputational risks and the possibil-
ity of material nancial loss. Consequently, a bank should have an integrated
approach to identifying, measuring, monitoring and managing technology risks.
Sound technology risk management uses the same precepts as operational risk
management and includes:
Conclusions and Recommendations 55
governance and oversight controls that ensure technology, including outsourc-
ing arrangements, is aligned with and supportive of the banks business ob-
jectives;
policies and procedures that facilitate identication and assessment of risk;
establishment of a risk appetite and tolerance statement as well as perfor-
mance expectations to assist in controlling and managing risk;
implementation of an effective control environment and the use of risk trans-
fer strategies that mitigate risk; and
monitoring processes that test for compliance with policy thresholds or limits.
Management should ensure the bank has a sound technology infrastructure that
meets current and long-term business requirements by providing sufcient capac-
ity for normal activity levels as well as peaks during periods of market stress;
ensuring data and system integrity, security, and availability, and supporting inte-
grated and comprehensive risk management. Mergers and acquisitions resulting
in fragmented and disconnected infrastructure, cost cutting measures or inade-
quate investment can undermine a banks ability to aggregate and analyse infor-
mation across risk dimensions or the consolidated enterprise, manage and report
risk on a business line or legal entity basis, or oversee and manage risk in periods
of high growth. Management should make appropriate capital investment or oth-
erwise provide for a robust infrastructure at all times, particularly before mergers
are consummated, high growth strategies are initiated, or new products are intro-
duced.
Outsourcing is the use of a third party either an afliate within a corporate group
or an unafliated external entity to perform activities on behalf of the bank. Out-
sourcing can involve transaction processing or business processes. While outsourc-
ing can help manage costs, provide expertise, expand product offerings, and im-
prove services, it also introduces risks that management should address. The
Conclusions and Recommendations 56
board and senior management are responsible for understanding the operational
risks associated with outsourcing arrangements and ensuring that effective risk
management policies and practices are in place to manage the risk in outsourcing
activities. Outsourcing policies and risk management activities should encompass:
1. procedures for determining whether and how activities can be outsourced;
2. processes for conducting due diligence in the selection of potential service
providers;
3. sound structuring of the outsourcing arrangement, including ownership and
condentiality of data, as well as termination rights;
4. programmes for managing and monitoring the risks associated with the out-
sourcing arrangement, including the nancial condition of the service provider;
5. establishment of an effective control environment at the bank and the service
provider;
6. development of viable contingency plans; and
7. execution of comprehensive contracts and/or service level agreements with a
clear allocation of responsibilities between the outsourcing provider and the
bank.
5.0.3. Area of further study.
Further research on credit risk management in these services will be extremely
valuable to the banks. Another interesting aspect is the adoption of credit deriva-
tives in managing credit risk. This is a new concept in some areas of the world. It
would be great to see the analysis and applications of successful credit derivative
cases, or the reasons behind the failures of other cases.
This study may serve the purpose of inviting stochastic modelling engineers
into a new eld.Future work can be in the continuous time version of this model
Conclusions and Recommendations 57
and the related algorithm and computer program. Future work can also be done
for the non homogenous model and the related algorithm and computer program.
In future an attempt to deal with dowward problem while increasing the number
of states can be done.
Bibliography
Sharpe W. , A Simplied Model for Portfolio Analysis Management Science,1963,
489-516.
Reserve Bank of Zimbabwe, Risk Management-Guideline No. 1-2006/BSD, 2006.
Basle Committee, Basle Core Principles, October 2006
Vasicek, O.: Credit Valuation, KMV Corporation, March 1984.
Wilson, T. C.: Portfolio Credit Risk, Federal Reserve Board of New York,Economic
Policy Review, October,1998,71-82.
Witzany, J.: Loss, Default, and Loss Given Default Modeling, Charles University
Prague, Faculty of Social Sciences, Institute of Economic Studies, Working paper
2009.
Nikolaos Limnios and Gheorghe Oprisan, Semi-Markov Processes and Reliabil-
ity,Published by Springer, 2001 .
Janssen and Manca,Janssen, J., Manca, R.. Applied semi-Markov processes for
Finance, Insurance and Reliability. 2005
P.J. Schonbucher. The pricing of credit risk and credit risk derivatives. Preprint,
1997
J Laurent, A Cousin, and J Fermanian. Hedging default risks of cdos in marko-
vian contagion models. Quantitative Finance, 17731791, 2011
Kusuoka, S. : A Remark on Default RiskModels, Advances in Mathematical Eco-
nomics,1999, 6982.
H. Leland. Corporate debt value, bond covenants, and optimal capital structure.
Journal of Finance,1994, 49.
T. Bielecki, M. Rutkowski Dependent defaults and credit migrations. To appear in
Applicationes Mathematicae,2001
F. Black, J.C. Cox, Valuing corporate securities: Some eects of bond indenture
provi- sions. Journal of Finance 1976, 31, 351,367.
A.J. McNeil, R. Frey, and P. Embrechts. Quantitative Risk Management: Con-
cepts, Techniques, and Tools. Princeton University Press, Princeton, 2005
BIBLIOGRAPHY 59
D. Wong, A unifying credit model. Working paper, Research Advisory Services,
Capital Markes Group, Scotia Capital Markets, 1998.
latis, A., Limnios, N. and Le Du, M. Hitting times in a nite non-homogeneous
Markov chain with applications. Appl. Stoch. Models Data Anal,1998, 14, 241253.
latis, A., Limnios, N. and Le Du, M. Hitting times in a nite non-homogeneous
Markov chain with applications. Appl. Stoch. Models Data Anal,1998, 14, 241253.
D. Tasche. Calculating value-at-risk contributions in
CreditRisk+.http://arxiv.org/abs/cond-mat/0112045, 2000.
J. R. Sobehart and S. C. Keenan. An introduction to marketbased credit analysis.
Moodys Risk Management Services, November 1999.
Standard and Poors. Global CBO/CLO Criteria, 1998.
Standard and Poors. Global Synthetic Securities Criteria, 1998.
Standard and Poors. Standard and Poors Corporate Ratings Criteria,1998.
J. Janssen and R. Manca, Semi-Markov Risk Models for Finance, Insurance and
Reliability, Springer, New York, NY, USA, 2007.
N. Limnios and G. Oprisan, Semi-Markov Processes and Reliability Modelling,
Birkhauser Boston Inc., Boston, Mass, USA, 2001.
G. DAmico, J. Janssen, and R. Manca, Duration dependent semi-Markov models,
Applied Mathematical Sciences, vol. 5, no. 4144 2011 pp. 20972108
D. Lando, Credit Risk Modeling, Princeton University Press, Princeton, NJ, USA,
2004
Janssen, J., De Dominicis R., Finite non-homogeneous semi-Markov processes.
Insurance:Mathematics and Economics,1984, 157-165.
Janssen J., Manca R.,A realistic non-homogeneous stochastic pension funds model
on scenario basis. Scandinavian Actuarial Journal 1997, 113-137.
Janssen J., Manca R, Applied semi-Markov processes. Springer, 2006
Janssen J., Manca R.,Semi-Markov Risk Models for Finance, Insurance and Reli-
ability. Springer, 2007.
Limnios, N., Oprian, G,Semi-Markov Processes and Reliability. Birkhauser,
Boston, 2001
Wolthuis, H, Life Insurance Mathematics (The Markovian Model), second edition,
Instituut voor Actuariaat en Econometrie, Amsterdam. 2003