III MODELLING WEEK UCM Master in Mathematical
Engineering -
UCM Madrid, June 22-30 , 2009PDa
PDb
VaR
Contents1.
Introduction
2
.
Values
of
constants
3.
VaR
for single units
4
.
Combining
the
units
5.
Macroeconomics
6.
Conclusions
Complubank has two businesses: credit cards and mortgages. Therefore, it has two RU’, each composed of a number of contracts.
IT staff at CompluBank keep in their database each contract of the bank with its LGD, LossGiven Default, and EAD, Exposure at default.
Along the problem LGD and EAD will be considered deterministic variables. That’s to say, along the horizon of risk measurement they will not change.
However, they did not keep track of the PD of each contract. All we know is that, at themoment of measure, all contracts in a given RU have same PD, which corresponds to theprobability of default within 1 year horizon.
This PD should be an unconditional default probability. That means this probabilityshould not be conditionned to any particular state of the economy realization.
Introduction
Moreover, for each RU we have a time series of conditional PDz for a whole economic cycle(from 1990 to 2007). In each of these series, for every three month period, we have the PD of the RU conditionned to the macroeconomic situation in that moment.
We assume that the relation between quarterly known PDz and the unconditionnal PD (unknown) is given by the Vasicek function:
where:
Z is assumed to be a N(0,1), common to all contratcs in a RU. It is called systemicfactor, and represents the state of the economy in given time horizon. It is thefundamental risk driver of the loss.
ρ is an unknown parameter between 0 and 1.
Correlation Ө between the Z factors of risk units can be derived from the correlation betweensystemic factors of each RU.
Introduction
( )⎟⎟⎠
⎞⎜⎜⎝
⎛
−−
=−
ρρ
1
1 zpdNNpd Z
Tha aims of the week have been:
-
Be able
to
stablish
a causal relation
between
the
systemic
factor os each
RU
and some
macroeconomic
factors.
-
Be able
to
compute the
loss
distribution
of
both
A (credit
cards)
and
B (mortgage
loans)
risk units
at CompluBank.
-
Be able
to
compute aggregated
loss
distribution
of
CompluBank, taking
into
account
the
diversification
/ correlation
of
both
risk units.
Once the above is reached:
-
Analyze
the
sensibility
of
percentile
of
loss
distribution
of
both
RUs
A
and
B
and
global, to
changes
in PD, ρ
and Ө.
-
Estimate
losses
in macroeconomic
stress scenarios.
Introduction
To reach our aims, we base our study in Vasicek model, which aim is to measure the loss associated with the risk of an amount of loans, and other definitions.
Assuming EAD and LGD are deterministic, in a given horizon, each RU has a lossdistribution given by:
where:
N is the number of contratcs in the risk unit.
PDiz is the probability of default along the horizon, conditionned to a state of the
economy. Such state of the economy is the risk factor that codifies the evolution ofthe losses.
( )∑=
⋅⋅=N
iii
zi LGDEADPDBerL
1
Introduction
Loss distribution of M risk units of a bank will be given by:
In order to properly measure the risk, it is essential to take into account the correlationbetween contracts within the same risk unit and the correlation between different risk units.
Introduction
MLLL ++= ...1
Contents1.
Introduction
2
.
Values
of
constants
3.
VaR
for single units
4
.
Combining
the
units
5.
Macroeconomics
6.
Conclusions
PD values are the probability of default independentof the economic situation (i.e. unconditionalprobability of default).
Using
the
data we
have
obtained
PD’s values
as the
average of
the
PD at each
year
for each
risk
unit. 6,67%
Year Conditional PD A Conditional PD B1990 1.40% 7.84%1991 2.16% 5.44%1992 3.08% 6.38%1993 4.56% 9.57%1994 5.03% 5.56%1995 1.88% 4.86%1996 2.41% 3.21%1997 1.88% 1.80%1998 1.74% 0.99%1999 1.52% 0.97%2000 1.63% 0.87%2001 1.38% 0.54%2002 1.18% 0.44%2003 1.04% 0.40%2004 0.63% 0.47%2005 0.52% 0.43%2006 0.44% 0.44%2007 0.89% 0.62%
Average (PD) 1,85% 2,82%
Values of constants
6,67%
Values of constants
( 1 ) ( )P z c N cρ ρ ε+ − < =
( 1) ( )P Default P X c= = <
1( )c N PD=
RU A RU B
C -2.5850 -1.97020
Values of constants
As we now the unconditional probability of default, PD, we can calculate ρ:
( )
( ) ( )( )ρ
ρρρ
ρρρ
ρ
−=−+
−=
=−−
=⎟⎟⎠
⎞⎜⎜⎝
⎛
−−
=
−
−−
−
1)(
11
)()1(
11
)())((
1
12
11
zVarianzapdNVarianza
zpdNVarianzazpdN
VarianzapdNVarianza z
( )
( )))((1))((
)))((1())((1))((
1
1
11
1
z
z
zz
z
pdNVarianzapdNVarianza
pdNVarianzapdNVarianzapdNVarianza
−
−
−−
−
+=
+⋅=
=−⋅
ρ
ρ
ρρ
ρ values are a measure of the independence of the debtors. ρ = 1 implies alldebtors are 100% correlated (if one defaults, everybody will) but if ρ = 0 all debtorsdefault independently.
We
use the
following
relation
to
obtain
ρ
values
for both
RU
giving
( )1
1
( ( ))1 ( ( ))
z
z
Varianza N pdVarianza N pd
ρ−
−=
+6,67%
RU A RU B
ρ 6,67% 19,38%
6,67%6,67%6,67%
Values of constants
Values of constants
One of the last variables we have to know is Ө, that we estimate:
( ) ( )( ) ( )
( ) ( )( )ZB
ZA
BBAA
BABA
pdNpdNnCorrelatio
zpdNzpdNnCorrelatio
ZZnCorrelatioZZianceCo
11
11
,
1,
1
,,var
−−
−−
=
=⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛
−−
⎟⎟⎠
⎞⎜⎜⎝
⎛
−−
=
===Θ
ρρ
ρρ
Θ value
We
have
estimated
θ, that
in our
model
represents
correlation
between
the
probabilities
of
default
of
the
two
risk
units, through
the
following
formula
where and represents
the
contitioned
probability, obtaining
the
value
6,67%( ) ( )( )1 1,Z ZA BCorrelation N pd N pd− −
ZApd Z
Bpd
Θ=79,7%
Values of constants
Contents1.
Introduction
2
.
Values
of
constants
3.
VaR
for single units
4
.
Combining
the
units
5.
Macroeconomics
6.
Conclusions
VaR
for single units
Procedure
Generate a random (N(0,1)) variable z to describe a macroeconomic situation
Get the PD for this z
Generate a Bernoulli distributed random variable epsilon (N(0,1)) based on PD for each client like a individual feature of the client
•
If PD is higher than epsilon then the default event occurs, not
in the other case
1 1
( 1/ ) ( 1 )
( ) ( )( ) ( )1 1
t t
zt t
P D z z P z c
N PD z N PD zP PD N
ρ ρ ε
ρ ρε
ρ ρ
− −
= = = + − <
− −< = =
− −
Continue Procedure
The loss generated for this client is
The total loss of the portfolio is the sum of the loss of each client
where
n denotes the
number
of
clients
for
each
units
( )zClientLoss Ber PD EAD LGD= ⋅ ⋅
( )1
nzi i i
iL Ber PD EAD LGD
=
= ⋅ ⋅∑
VaR
for single units
VaR
for single units
We simulate the process 10000 times and get the distribution show below
VaR
for single units
PORTFOLIO A PORTFOLIO B
Results
for
100.000 Simulations.
99.90
99.97
799.21891.1975
VaRVaR
=
=
99.90
99.97
1684.51890.2
VaRVaR
=
=
VaR
for single units
Varying the rho we get how VaR varies in the graph below.
99.9%€8990AVaR
AA ρρ∂
=∂
99.97%€9810AVaR
AA ρρ∂
=∂
This also gives us the numerical derivative with respect to var around rhoA = 0.067
VaR
for single units
99.9%€9140BVaR
BB ρρ∂
=∂
99.97%€11040BVaR
BB ρρ∂
=∂
This also gives us the numerical derivative with respect to var around rhoA = 0.067
Contents1.
Introduction
2
.
Values
of
constants
3.
VaR
for single units
4
.
Combining
the
units
5.
Macroeconimics
6.
Conclusions
Combining the units
What if portfolios are correlated ?
If so we need two normal distributions with a given correlation
Where Θ is the correlation between portfolios
2
( (0,1))
1 ( (0,1))
PortfolioA
PortfolioB PortfolioA
Z random N
Z Z random N
=
= Θ + −Θ
Combining the units
Like we did for a single units we can do now. And we get the Loss distribusion
Var99.9 Var99.97RU A
810.5599 1050.1
RU B 1902.7 2297.5
RU AB 2532.9 2982.9
Combinig
the units
Rho a
Rho b
VaR
99.9
Rho a
Rho b
VaR
99.97
Variation of rho respect the whole portfolio
Combinig
the units
We now estimate the first derivate of VaR (99.97% and 99.9%) with respect of Θ.The plotobtained is shown below:
This also gives us the numerical derivative with respect to
99.97%€698A BVaR +∂
= Θ∂Θ
99.9%€501A BVaR +∂
= Θ∂Θ
0.8Θ ≈
Combining the units
We see the VAR obtained for the different PD
PDa
PDb
VaR
99.9
PDa
PDb
VaR
99.97
Contents1.
Introduction
2
.
Values
of
constants
3.
VaR
for single units
4
.
Combining
the
units
5.
Macroeconomics
6.
Conclusions
Macroeconomics
The aim is to establish a relationship between the macro economic factors and the systemicsfactors for each risk unit.
For this purpose we decided to use a linear regression model after a long study of theproblem.
Firstly, we calculated the systemic factors and then normalized both the systemics and themacro economic factors.
Macroeconomics
Using Excel, we calculate the regression coefficients and the errors that are involved since weneed them in our further work.
Risk
Unit A Risk
Unit Bb1 0,266197489 -0,077627772
b2 -0,79495506 -0,621544539
b3 0,08264054 -0,12181275
b4 0,048293754 -0,564335208
b5 0,193449818 -0,039757321
b6 -0,021242591 -0,073168947
b_epsilon 0,362743864 0,15134132
Macroeconomics
So we conclude that the systemic factors depend of the values of all our macro ecomonic factors.
where,
has a Normal(0,1) as distribution function.
If we fix one of the factors, we have:
but
the
macroeconomic
factors
don’t follow
a N(0,1) distribution
anymore
}{6
1, ,j j j j
i ii
z F j A Bεβ β ε=
= + ∈∑
}{6
1 12
, ,j j j j ji i
iz F F j A Bεβ β β ε
=
= + + ∈∑
}{, 1, 2, 3, 4, 5, 6 ,jiF i j A Bε ∀ = ∈
Macroeconomics
For simulating the systemic factor, we apply the Cholesky factorization using the correlationmatrix between the macro economic factors.
GDP Unemployme
nt
House_Price E3M IBEX Inflation
GDP 1 -0,37297 -0,30282 -0,40533 -0,11382 -0,02879
Unemployment -0,37297 1 -0,31707 0,57914 -0,06146 -0,39557
House_Price -0,30282 -0,31707 1 0,26396 -0,11747 0,58063
E3M -0,40533 0,57914 0,26396 1 -0,38748 -0,11351
IBEX -0,11382 -0,06146 -0,11747 -0,38748 1 -0,19195
Inflation -0,02879 -0,39557 0,58063 -0,11351 -0,19195 1
We now fix the values of the change of GDP and do a simulation to obtain a plot of VaR as a function of GDP. The plot obtained is shown below:
Macroeconomics
99.9% 99.97%€ €-16500 -18600A B A BVaR VaR
GDP GDPGDP GDP+ +∂ ∂
= =∂ ∂
This also gives us the numerical derivative around GDP growth around 2.5%:
99.9% 99.97%€ €89 1053 33 3
A B A BVaR VaRE M E ME M E M
+ +∂ ∂= =
∂ ∂
Macroeconomics
Contents1.
Introduction
2
.
Values
of
constants
3.
VaR
for single units
4
.
Combining
the
units
5.
Macroeconimics
6.
Conclusions
The variation of VAR is influenced by the combination of many factors.
To define the state of the economy the highest ifluence is given by the unemployment.
The most importan factor is the correlation of each portfolio to the economy, represented byRho.
This shows that diversification reduces risk
Conclusions