Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

ASSET CORRELATION, PORTFOLIO DIVERSIFICATION

AND REGULATORY CAPITAL IN THE BASEL CAPITAL
ACCORD
Sylvia Gottschalk*

Abstract

In this paper, we analyze the properties of the KMV model of credit portfolio loss. This theoretical
model constitutes the cornerstone of Basel II’s Internal Ratings Based(IRB) approach to regulatory
capital. Our results show that this model tends to overestimate the probability of portfolio loss when
the probability of default of a single firm and the firms’ asset correlations are low. On the contrary,
probabilities of portfolio loss are underestimated when the probability of default of a single firm and
asset correlations are high. Moreover, the relationship between asset correlation and probability of
loan portfolio loss is only consistent at very high quantiles of the portfolio loss distribution. These are
precisely those adopted by the Basel II Capital Accord for the calculations of capital adequacy
provisions. So, although the counterintuitive properties of the KMV model do not extend to Basel II,
they do restrict its generality as a model of credit portfolio loss.

Keywords: Assets, Basel Accord, Regulatory Capital

*Middlesex University Business School, Economics and Statistics department, The Burroughs, London NW4 4BT
Email: [email protected]

This work has benefited from discussions with Ricardo Gottschalk. The usual disclaimer applies.

1 Introduction Jones et al. (2002a) and Griffith-Jones et al. (2002b),

for instance, suggest that the overall risk of a
The idea that regulatory capital requirements should geographically diversified portfolio is lower than that
be risk sensitive is at the core of the second Basel of a geographically concentrated portfolio. Garc´ıa
Capital Accord (Basel II), BIS (2005a). Even the (2002), Garc´ıa et al. (2006), show that the credit risk
Basel Committee‘s current proposals for reform of of a portfolio based on a two-factor model is lower
the Capital Accord (Basel III), which were prompted than that of a single-factor portfolio. Tasche (2005)
by the 2008 Credit Crunch, are firmly built on the generalised this result to a multi-factor setting. So
risk-sensitive framework of Basel II (See BIS, far,the theoretical and empirical research has taken
2009a,b). At the conceptual level there clearly is two main directions. The first approach consists of
widespread support for the idea of risk-based capital the empirical estimation of asset correlations and
provisions. However, in order to move this support default correlations (Dietsch and Petey, 2004;
from the conceptual to the practical level, it is D¨ullmann and Scheule, 2003; Erlenmaier and
essential that capital provisions accurately reflect Gerbach, 2001; Servigny and Renault, 2002). Other
credit risk. papers focus on the theoretical result of the main
The Basel II‘s Internal Ratings Based (IRB) credit risk model showing that lower asset
framework of capital adequacy was built on a credit correlations imply lower default correlations (Garc´ıa,
risk model developed by the KMV Corporation, 2002; Garc´ıa et al., 2006; Tasche, 2005, amongst
which was acquired by Moody‘s in 2002. The KMV others).
model is an extension of Merton (1974) to credit risk In this paper, we analyse the properties of the
(Vasicek, 1987), and more importantly, to loan KMV model of credit portfolio loss which constitutes
portfolio risk (Vasicek, 2002). A significant issue in the cornerstone of the IRB approach to regulatory
credit risk analysis is how default and asset capital. We find that the KMV model represents the
correlations are taken into account. It is generally probability of portfolio losses very poorly at low p,
accepted that the overall risk of a portfolio can be and low p. More specifically, it tends to overestimate
reduced by diversifying its assets either sectorally or the probability of portfolio loss when the probability
geographically. There is growing evidence that the of default of a single firm and the firms‘ asset
same principle applies to credit portfolios. Griffith- correlations are low. On the contrary, probabilities of

31
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

portfolio loss are underestimated when the probability debt Ti. Formally, AiTi < Di, where AiTi
of default of a single firm and asset correlations are represents the value of firm i‘s assets at the loan
high. maturity time Ti and Di the value of firms i‘s
The paper is organised as follows. After this liabilities. AiTi is the latent variable of the model.
Introduction, Section 2 the KMV model is presented 2. The value of a firm‘s assets is described by the
and its properties analysed. In Section 3 special stochastic differential equation dAi = Ai(μidt +
emphasis is placed on the relationship between the σidxit) where xit is a standard Brownian Motion.
KMV model and Basel II‘s IRB approach to Moreover, E[dxi]18 = dt, and E[dxi][dxj ] = pdt
regulatory capital provisions. Section 4 concludes. for i ≠ j. μi and σi are constants. They may be
interpreted as the drift and the volatility of the
2 Latent variable models and Basel II’s asset value of firm i, respectively. Finally, a
IRB Approach Brownian motion has a Normal distribution with
mean 0 and variance dt.
The theoretical foundations of the IRB approach of 3. (Portfolio homogeneity) For the sake of
the 2005 Basel Capital Accord was first developed by simplicity, Vasicek (1987, 1991, 2002) assumes
KMV Corporation2 as an extension of Merton that all borrowers are identical. This implies that
(1974)‘s model of corporate debt pricing, and was (i) all loans have the same maturity, Ti = Tj = T,
later published in Vasicek (1987, 1991, 2002). This i, j = 1, . . . ,M, (ii) asset correlations are
model belongs to the class of latent variable models identical, pi=pj=p (iii) debt values Di are
and one of its results is a factor representation of the identical, Di=Dj=D.
determinants of individual default. The main property
of this factor representation is the independence of In the Vasicek setting, the point in time where
individual defaults relative to each other, given the the occurrence of default is considered is the maturity
occurrence of the determinants of individual defaults. of the debt, Ti in Assumption (1). In Assumption (2),
The KMV model is thus considered as an example of p represents the two-by-two correlation of borrowing
conditionally independent credit risk models. firms‘ assets, but not necessarily default correlation.
However, as we will show below, it does not Default correlation and asset correlation can differ
adequately capture dependencies between individual significantly, as shown in Sch¨onbucher (2000), Zhou
default probabilities, and that this failure extends to (2001), and mainly, Frey et al. (2001). Finally, it
Basel II‘s IRB approach. should be emphasized that asset correlation is
exogenous in the KMV model. It is assumed to exist
2.1 KMV model but its value is not obtained from the model. As a
result, it may take any arbitrary value. In a
The main objective of the KMV model is the consultative paper published by the Basel Committee
derivation of the probability distribution function of in 1999, the asset correlation was set to 20%.
the loss of a portfolio of loans. The model first Theoretically, p is the usual correlation formula
derives the probability distribution function of a
single default. Single loans are then aggregated into a
portfolio of loans, and assumptions concerning
default correlations are made at this stage. Finally,
Vasicek (2002) presents Monte Carlo simulations of
the limiting distribution function of portfolio loss
distribution function.
It is worth emphasizing at this stage that we are
modeling corporate rather than retail (consumer
credit) default, and that we take the perspective of the
borrowing firms rather than that of the lender when
using the term ‖assets‖. In the literature on financial
regulation or in the documents published by the Basel
Committee on Banking Supervision, assets usually
refer to loans, which are the firms‘ liabilities.
Throughout this paper, banking loans will always be
referred to as debts or liabilities.
Assumptions
A portfolio consists of M loans of equal amounts, one
loan per firm. For each firm i = 1, ..., M the following
assumptions hold

1. Default occurs when the value of a firm‘s assets

fall below the value of its debt, at maturity of its
18
Kealhofer, McQuown, Vasicek Development, L. P.

32
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

where is the covariance of and at time T and if firm i does not default at time
and the square root of is its t = T. Then, the probability of default of a single firm
standard deviation. The covariance is given by is given by

is obtained by solving the stochastic differential

where is the expectation operator. equation in Assumption (2) above,
Given the assumptions above, let be an
indicator variable such that if firm i defaults

where is the value of the firm‘s assets at time 0, T Using (4), we can write (3) as
is maturity time and Xit is a variable with Normal
(0,T), i.e., with mean 0 and variance T. where pi is the probability of individual default.

The only random element in the left-hand side of this inequality is Xit. So we can re-write this expression as

Since Xit is a Normal distribution with mean 0 and variance 1, the expression above becomes

where and is the cumulative normal distribution function. is the

‖distance-to-default‖ of firm i (Merton, 1974; Vasicek, 1987, 1991, 2002).

Expression (8) clearly shows that the probability 2.2 Probability of loan portfolio loss
of default of a single firm i does not depend in any
way on the probability of default of a single firm j, From result (8), Vasicek (2002) proceeds to derive
since the two-by-two correlation of firms‘ assets, p, the probability of the loss of a loan portfolio. Let Li
does not appear in it. Asset correlations can be denote the gross loss on the i-th loan. The gross loss
endogenized by assuming that the asset value of a represents the loss before recoveries. Li = 1 if the i-th
single firm follows a multidimensional Brownian firm defaults and Li = 0 if the i-th firm does not
motion. In Assumption (2) above the stochastic default. Let L be the portfolio percentage gross loss,
differential equation describing the asset value defined as the weighted sum of each individual
becomes . Recent portfolio percentage gross loss,
papers have pursued this direction, for instance,
Kafetztaki-Boulamatsis and Tasche (2001), and
Nyfeler (2000). Although this approach is
conceptually more adequate for modelling joint where M is the total number of loans in the
defaults or default dependencies, it suffers from a portfolio. We wish to calculate the probability of
major shortcoming. Estimating the asset correlation n defaults out of the M loans
matrix is practically impossible (see Gottschalk,
2011, for details).

33
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

for n = 1,...,M. From Assumption 2, each firm‘s asset

follows a Brownian motion, and the two-by-two Substituting XT in factor representation gives
correlations are identical, i.e., , for any
firm , ,...,M. This implies that the
Brownian motion variables Xit are jointly
equicorrelated standard normal variables. A property Re-arranging the terms in the left-hand side of
of this type of probability distribution is the the inequality, we obtain
representation

where, Ft and Z1t,...,ZMt are mutually

As was seen above, since Zit, i = 1,...,M, is
independent standard normal distributions. (11) is a
Normally distributed
factor representation, the factor being Ft. Vasicek
(1991) explicitly points out that expression (11)
derives necessarily from the assumption of normality
of asset returns, which in turn is a necessary outcome
of the hypothesis that asset returns follow a Brownian
motion. In many subsequent papers in the literature, Note that we have replaced Ft by its value u.
(11) is presented as the starting point of the KMV Substituting (16) in (12) yields the probability of n
credit risk model, with the assumption of normality defaults in the portfolio
replaced by the more convenient hypothesis that the
factor Ft follows a Student‘ t distribution.19 Vasicek
(2002) suggested that Ft can be interpreted as a
common macroeconomic factor affecting the whole
portfolio of loans. Each firm‘s sensitivity to this
A limiting distribution of portfolio loss can be
factor is given by . stands for the obtained by assuming the number of loans in the
firm‘s specific risk. portfolio tends to infinity. If we maintain the
In order to evaluate the probability of n defaults
assumptions of homogeneity, = now
in the portfolio, it is necessary to determine the
number of possible combinations of n individual becomes the fraction of defaulted loans in the
defaults in a portfolio of M loans. Since individual portfolio. By the law of large numbers, the fraction of
defaults are independent given the occurrence of the defaults is (almost surely) equal to the individual
factor Ft,20 the number of possible combinations of default probability, .
defaults is given by the Binomial factor .
Moreover, we now assume for the sake of simplicity,
and following Vasicek (1991, 2002), that the
portfolio is homogeneous. This implies that
individual probabilities of default are identical, as are
the distances-to-default DD and as before the
maturities of the debts.
By the law of iterated expectations, the
probability of having exactly n defaults is the average
of the conditional probabilities of n defaults, averaged
over the possible realizations of Ft and weighted by
the probability density function of Ft evaluated at u,

Once the individual defaults can be assumed to

occur independently, the Vasicek model reduces to a
Binomial model of default.
From (8), p(u) is given by

19
The probability of extreme events is higher in the Student’s
t distribution than in the Normal distri- bution. Distributions
with higher probabilities of extreme events capture more
adequately the empirical distributions of financial variables.
20
See Gottschalk (2011) for proof.

34
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

The cumulative distribution function of the fraction L portfolio loss is thus

The notation is used to emphasized that

this distribution is valid only when the number of
loans in the portfolio becomes infinitely large. The α-
percentile of (19), denoted , is given by inverting
Finally, the derivative of (19) with respect to x
. The α -percentile of the loss distribution,
gives the density distribution function
is thus

where is the exponential function.

Vasicek (2002) and Gordy (2003) show that and 91%. The fixed parameter is the individual
(20) is also valid when the weights of single loans in probability of default p. It is worth remembering at
the portfolio are allowed to differ, i.e., when this stage that (19) hinges on the assumption that all
, where . However, a the firms in the portfolio have the same probability of
necessary and sufficient condition is that no single default.
loan may dominate the portfolio, which implies that Figure 1 clearly shows that for a probability of
default equal to 1% always collapses to p,
. This result is particularly important in when asset correlations are quite low (1% to 41%).
the light of Basel II‘s formulae to calculate regulatory Figure 1 also shows that at low p and low , the
capital. probability of any fraction of the portfolio defaulting
The properties of the cumulative distribution is 100%, irrespective of the level of asset correlation.
function of portfolio loss (19) are summarized in This is quite counterintuitive since one would expect
Vasicek (2002), and a couple illustrative plots are the probability of portfolio loss to be low when the
presented in Sch¨onbucher (2000) and Garc´ıa probability of individual default and the asset
(2002). A more useful reference is Bluhm et al. correlations are low.
(2003), where the properties of (19) are more
thoroughly described. When ,
converges to a one-point distribution concentrated at
L = p. When the distribution flattens and
converges to a zero-one distribution with probabilities
and .

2.3 Is the KMV model adequate for

modeling the probability of portfolio
losses?

In Figures 1 and 2 we simulate (19) to illustrate some

of these properties.21 In all figures, the left-hand
graph shows for asset correlations
between 1% and 41%. The right-hand side graph
plots for asset correlations between 51%
21
Results for other values of and can be found in
Gottschalk (2011)

35
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

Figure 1. Probability of portfolio loss - p=1%

Figure 2. Probability of portfolio loss - p=50%

The KMV model performs a bit better at 3 The KMV model and regulatory capital
higher levels of asset correlation (40% to 91%).
Portfolio default probabilities are more spread out, The IRB approach assumes heterogeneous
and more dependent on the level of asset portfolios of loans. This implies that each borrower
correlation. Nonetheless, it is evident from the two may have a distinct probability of default pi, each
figures that the KMV model represents the loan has a distinct maturity Mi, the weight of each
probability of portfolio losses very poorly at low p, loan in the portfolio is different , and that the
and low . This fact was pointed out by percent loss on each loan can be different, . In the
Schonbucher (2000), and Bluhm et al. (2003), Basel Committee‘s publications, is the referred to
amongst others. as the loss given default, and is equal to 1 minus the
In Figure 2 the probability of portfolio loss recovery rate. For the sake of simplicity, it assumed
when the probability of default of a single firm is in the KMV model that the loss is total, so ,
now p=50%. The left-hand side figure shows a for all loans in the portfolio.
more interesting result. When the asset correlation According to the rules of Basel II, regulatory
is 1%, , up to 40% of the firms in the capital is needed only to cover unexpected losses,
portfolio do not default, even though the individual given that banks are supposed to cover for expected
probabilities of default are quite high. The fraction losses as part of their on-going activities.
of the portfolio that does not default obviously Regulatory capital is thus given by
decreases inversely with asset correlation. When , where
asset correlation is 41%, all the portfolio defaults.

36
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

and is given by expression (20) in the main text, factored by the loss given default .

To take into account the impacts of differing , but not when the components of the
maturities on credit risk, the IRB rules require that the portfolio are perfectly correlated .
following expression be multiplied to the regulatory Moreover, increasing the number of components in
capital formula (22) the portfolio decreases its overall variance, regardless
of the sign of cross-correlations,23 since in a large
portfolio, cross-correlations among assets determine
the portfolio variance. The variance of each asset then
contribute little to portfolio risk (see Ingersoll, 1987).
where Extending this setting to credit portfolios is not
. M is the straightforward. In the case of credit, the concept of
maturity of the loan for which regulatory capital is risk is not solely associated from that of variance.
being calculated, and pi is the borrower‘s probability Risk is an inherent characteristic of a loan, and can be
of default. In the KMV context, is given by (3)22 proxied by the probability of default. In the Vasicek
It is clear from (22) that regulatory capital in the model the probability of default is determined by the
IRB approach is fundamentally the α-percentile of the behaviour of the latent variable, and increasing the
asymptotic cumulative distribution function of the number of loans may not necessarily lead to lower
portfolio loss in the KMV model, (20). In the IRB probability of default.
approach, α is set to 99.99 percent. The inclusion of First, default correlation is positively related to
asset correlation in (22) raises the issue of portfolio asset correlation. Second, asset correlations in the
diversification on the IRB approach. A growing Vasicek model can only assume positive values,
number of papers throw some light on the unlike conventional portfolios. As a result, the
relationship between the KMV model and Basel II‘s opportunities for credit risk diversification are limited
IRB approach. Amongst those, Kjersti (2005), to assets presenting cross- correlations converging to
Hamerle et al. (2003) are the most straightforward zero.
without compromising technicality. Gordy (2003) is Morever, Figures 1 and 2 suggest that the
an important paper showing that Basel‘s capital Vasicek model yields a negative relationship between
adequacy rules can be reconciled with a class of asset correlation and the probability of portfolio loss,
credit risk models which are portfolio-invariant. for certain levels of probability. For instance, for
Gordy (2003) is explicitly mentioned in the Basel , Figure 1 shows that the fraction of the
Committee‘s official documents as one of the portfolio that is lost is lower when (dashed
theoretical cornerstone of their IRB approach, along curve) than when (solid curve). An
with Vasicek (2002) BIS (see 2005a). analogous result can be seen for . The
dashed curve corresponds to , whilst the
3.1 Default correlation and portfolio solid curve is associated to . This
diversification counterintuitive result is much more pronounced for
higher values of the individual probability of default,
Following Markowitz(1952) a portfolio is as Figure 2 illustrates. For the fraction of
efficient if there is no other portfolio with lower risk portfolio lost is lower when (dashed line)
and an at least equal expected return, and no portfolio than when (solid line). This implies that at
with a higher expected return and at most equal risk. these levels of probability, increasing asset
In this context diversification is a means to change correlation actually reduces the overall risk of the
the risk of the portfolio. The portfolio risk is portfolio.
measured as the standard deviation from expected
returns, and, by definition, is the sum of the variances
of each component of the portfolio from the expected
return and the correlations between components
. Risk diversification can be achieved if

22
Note that there is a discrepancy between two Basel
Committee’s publications regarding the Normal distribution
used in expression (22). In BIS (2005a, p.7) Ф(.) is the
Normal distribution function N(.). However, in BIS (2005b,
p.60 footnote 71) Ф(.) is the cumulative Normal distribution.
Since BIS (2005a) is the main document of the Basel Capital
Accord and thus supercedes BIS (2005a), we use its formula
23
here. provided the assets are not perfectly correlated.

37
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

Figure 3. Regulatory Capital

However, for and , IRB‘s assumption of a negative relationship between

respectively, the more intuitive relationship between asset/factor correlation and default probability stems
asset correlation and risk holds. The higher asset from Lopez (2004). However, several subsequent
correlation, the higher default risk. In the context of studies have produced results that contradict this
Basel II‘s IRB Approach, these findings have hardly assumption. Amongst other, Dietsch and Petey
any implications. First, the Basel Committee decided (2004), D¨ullmann and Scheule (2003), and Hamerle
that banks should make capital provisions for losses et al. (2003).
occurring with a probability of less than 0.01%. Figure 3 plots the regulatory capital given by
Second, the formulae adopted by the Basel (22), for various levels of individual probability of
Committee for the asset correlation restrict its value default , a loan maturity of 2.5 years
to the interval [0.12;0.24], (see BIS, 2005a). (standard value in BIS (2005b)), and a
In the IRB formulae, the correlation between loss given default of 10%. Clearly, the higher the
individual assets and the macroeconomic factor is probability of default, the higher the necessary
given by the expression (24) for bank and sovereign amount regulatory capital, for asset correlations lower
borrowers, than 0.65.

4 Conclusions

In this paper, we present the theoretical model that

The correlation between individual asset and the constitutes the cornerstone of Basel II‘s Internal
macroeconomic factor for corporate borrowers are Ratings Based (IRB) approach to regulatory capital.
derived from (24). for . This model was developed by KMV Corporation, and
for , later published in Vasicek (1987, 1991, 2002). We
then analyse the properties of the KMV model of
and for . is the correlation
credit portfolio loss, for distinct value of single firm
coefficient for corporate borrowers and is given by
default probability and asset correlations. Our results
(24). Si are annual sales of firms i. Clearly, is
show that this model tends to overestimate the
reduced for small firms. (24) assumes a downward
probability of portfolio loss when the probability of
relationship between the asset/factor correlation and
default of a single firm and the firms‘ asset
default probability. The higher the individual
correlations are low. On the contrary, probabilities of
probability of default, the lower the asset/factor
portfolio loss are underestimated when the probability
correlation, since in this case, the idiosyncratic
of default of a single firm and asset correlations are
factors are assumed to dominate the high. Moreover, the relationship between asset
macroeconomic factor. In other words, the higher the correlation and probability of loan portfolio loss is
probability of default the higher the likelihood that only consistent at very high quantiles of the portfolio
default will be determined by factors specific to the loss distribution. These are precisely those adopted by
borrower rather than macroeconomic conditions. The

38
 Risk governance & control: financial markets & institutions / Volume 1, Issue 3, 2011

the Basel II Capital Accord for the calculations of 14. Griffith-Jones, S., S. Spratt, and M. Segoviano
capital adequacy provisions. So, although the (2002a) ‗Basel II and developing countries:
counterintuitive properties of the KMV model do not Diversification and portfolio effects.‘ Discussion
extend to Basel II, they do restrict its generality as a Paper 437, LSE Financial Market Group.
15. ___(2002b) ‗The onward march of Basel II: Can the
model of credit portfolio loss. interests of developing countries be protected?‘
Institute of Development Studies mimeo.
References http://www.ids.ac.uk/ids/global/Finance/ifpubs.htlm.
16. Hamerle, A., T. Leibig, and D. R¨osch (2003) ‗Credit
1. BIS (2005a) ‗An explanatory note on Basel II risk risk factor modelling and the Basel II approach.‘
weight functions.‘ Basel Committee on Banking Discussion Paper.Series 2: Banking and Finance
Supervision Supervision 02/2003, Deutsche Bundesbank.
2. ___(2005b) ‗International convergence of capital 17. Ingersoll, J. E. (1987) Theory of financial decision
measurements and capital standards: Arevised making (Rowman & Littlefield).
framework.‘ Basel Committee on Banking 18. Kafetztaki-Boulamatsis, M., and D. Tasche (2001)
Supervision. ‗Combined market and credit risk stress testing based
3. ____(2009a) ‗Comprehensive response to the global on the merton model.‘ RiskLab.
financial crisis.‘ Basel Committee on Banking http://www.risklab.ch/ftp/papers/MarketCreditRiskStr
Supervision. essTest.pdf.
4. ___(2009b) ‗Consultative proposals to strengthen the 19. Kjersti, A. (2005) ‗The Basel II IRB approach for
resilience of the banking sector.‘ Basel Committee on credit portfolios: A survey.‘ Note SAMBA/33/05,
Banking Supervision. Norsk Regnesentral.
5. Bluhm, C., L.Overbeck, and C. Wagner (2003) An 20. Lopez, A. (2004) ‗The empirical relationship between
introduction to credit risk modelling (Chapman & average asset correlation, firm probability of default
Hall/CRC). and asset size.‘ Journal of Financial Intermediation
6. Dietsch, M., and J. Petey (2004) ‗Should SMEs 13(3), 265–283.
exposures be treated as retail or corporate exposures? 21. Merton, R.C. (1974) ‗On the pricing of corporate
a comparative analysis of default probabilities and debt: The risk structure of corporate debt.‘ Journal of
default correlations in French and German SMEs.‘ Finance 29(2), 449–470.
Journal of Banking and Finance 28, 773–778. 22. Nyfeler, M.A. (2000) ‗Modelling dependencies in
7. Dullmann, K., and H. Scheule (2003) ‗Determinants credit risk
of asset correlations of german corporations and management.‘http://www.risklab.ch/ftp/papers/Diplo
implications for regulatory capital.‘ University of maThesisNyfeler.pdf.
Regensburg Working Paper. http://www.cofar.uni- 23. Schonbucher, P.J. (2000) ‗Factor models for portfolio
mainz.de/dgf2003/paper/paper53.pdf. credit risk.‘
8. Erlenmaier, U., and H. Gerbach (2001) ‗Default http://www.schonbucher.de/papers/portfolio fo.pdf.
probabilities and default correlations.‘Research Notes 24. Servigny, A, and O. Renault (2002) ‗Default
in Economics and Statistics December, Deutsche correlation: Empirical evidence.‘ Standard and Poor‘s
Bank. Working Paper.
9. Frey, R., A. McNeil, and M. A. Nyfeler (2001) http://www3.imperial.ac.uk/pls/portallive/docs/1/568
‗Modelling dependent defaults: Assets correlations 015.PDF.
are not enough.‘ http://www.defaultrisk.com/pp corr 25. Tasche, D. (2005) ‗Risk contributions in an
17.htm. asymptotic multi-factor framework.‘ mimeo.
10. Garc´ıa, J.C. (2002) ‗Credit risk modelling and Basel Deutsche Bundesbank.
II.‘ Algo Research Quarterly 5(1), 57–66. 26. Vasicek, O.A. (1987) ‗Probability of loss on loan
11. Garc´ıa, J.C, J.A. de Juan, A. Kreinin, and D. Rosen portfolio.‘ KMV Corporation.
(2006) ‗A simple multifactor‖factor adjustment‖ for http://defaultrisk.com/pp model 60.htm.
the treatment of credit capital diversification.‘ mimeo. 27. ___(1991) ‗Limiting loan loss probability
http://www.fields.utoronto.ca/ drosen/ Papers/Multi- distribution.‘ KMV Corporation.
factor. http://www.moodyskmv.com/research/whitepaper/Li
12. Gordy, M.B. (2003) ‗A risk-factor model foundation miting Loan Loss Probability Distribution.pdf.
for ratings-based bank capital rules.‘Journal of 28. ___(2002) ‗Probability of loss on loan portfolio.‘
Financial Intermediation 12, 199–232. KMV Corporation.
13. Gottschalk, S. (2011) ‗Asset correlation, portfolio http://www.moodyskmv.com/conf04/pdf/papers/dist
diversification and regulatory capital in the Basel loan port val.pdf.
Capital Accord.‘ Middlesex University Working 29. Zhou, C. (2001) ‗An analysis of default correlations
Paper 144, Middlesex University. and multiple defaults.‘ Review of Financial Studies
14(2), 555–576.

39