JJ Mois Année
SMILOVICE
Jan NeckařDana Chromíková
Scoring Unit (SCOR)2 April 2008
Basel II concept differs two types of loss:
0%
1%
2%
3%
4%
5%
6%
7%
8%
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Years
los
s
Unexpected loss
Expected loss
Frequency
UNEXPECTED LOSS
EXPECTED LOSS
B2 - CONCEPT
LOSSES IN TIME
Scoring Unit (SCOR)3 April 2008
Probability 0,1%
Probability 99,9%
Creation of Provisions
-Covered by
SRC
CAPITAL RESERVESMAINTENANCE
STRESS TESTING
BASEL II CONCEPT – distribution of the losses
VALUE AT RISK
LOSS
FR
EQ
UE
NC
YB2 - CONCEPT
Scoring Unit (SCOR)4 April 2008
B2 - CONCEPT
EXPECTED LOSS
EL = PD * LGD * EAD
PD – probability of default- estimation of probability then client is longer than 90 days delayed with payments, insolvency, …
LGD – loss given default- estimation of the resulting economic loss after the recovery process- conditional estimation in case of client is in default
EAD – exposure at default- conditional estimation of exposures in case of client is in default- average drawing at default is higher than outside default
EL = PD*E(loss|default) + (1-PD)*E(loss|nedefault) = PD * LGD * EAD
Scoring Unit (SCOR)5 April 2008
N(x) – distribution function of normalized normal distribution of random quantityG(x) – inversion function to distribution function of normalized normal distributionScaling factor – according to direction of ČNB is equal to 1,06Maturity (M) – Average maturity of the expected cash-flows (repayments)Factor of maturity
Correlation factor R for retail exposures (excl. Mortgages = 0,15, qualifying revolving = 0,04):
Correlation factor for non-retail exposures:
S – Annual sales for the consolidated group (million EUR)
B2 - CONCEPT
UNEXPECTED LOSS – CAPITAL REQUIREMENT
Factor Scaling12,50PDb1,51
PDb2,5M1PD0,999G
R1
RPDG
R1
1NLGDRW
2ln 0,05478 - 0,11852 b(PD) PD
45
5)5),50,max(min(10,04-
e1
e110,24
e1
e10,12R
50
PD50
50
PD50
S
Tier1 + Tier2 ≥ 8% * Σ RW * EAD Tier1 ≥ Tier2 &
53
PD53
53
PD53
e1
e1116,0
e1
e10,03R
Scoring Unit (SCOR)6 April 2008
B2 - CONCEPT
UNEXPECTED LOSS – CAPITAL REQUIREMENT
Scoring Unit (SCOR)7 April 2008
Behavior under stress is not easy to predict
STRESS-TESTING
Scoring Unit (SCOR)8 April 2008
Frequency0%
2%
4%
6%
8%
10%
12%
14%
16%
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Years
loss
BASELINE
DEPRESSION
LOSSES IN TIME
STRESS-TESTING
Scoring Unit (SCOR)9 April 2008
STRESS-TESTING
ECONOMETRICMODEL
STRESSSCENARIOS
STRESS-TESTINGMODELS
STRESSED CHARACTERISTICS
Scoring Unit (SCOR)10 April 2008
Econometric model predicts the macroeconomic characteristic as:
These models have usually 50 – 100 formulas and above 200 parametersThere are several various of predictions, called as scenarios:
The most probable scenario is selected for development of the model.
STRESS-TESTING
GDP unemployment interest rates inflation / deflation price of oil …
baseline depression deep depression high inflation …
ECONOMETRICMODEL
STRESS-TESTINGSCENARIOS&
Scoring Unit (SCOR)11 April 2008
STRESS-TESTING
STRESS-TESTINGMODELS
Stress testing= a way how to measure risk of
Modeled via scenarios for macroeconomics characteristics
extreme but realistic events
Two type of models:Logistic regression
Factor model based on Merton’s model
We assume that portfolio depends on macroeconomic situation and we need to find relation between stressed variable (PD, LGD, CCF) and macroeconomic characteristics:
PDt = f (Mt1)t ≥ t1, f (Mt1) function of macroeconomic characteristics
Example for stressing PD:
Scoring Unit (SCOR)12 April 2008
STRESS-TESTING
STRESS-TESTINGMODELS
Y is explained variable (indicator of default), EY is probability of default
is vector of explanatory variables (macro-economic indicators).
k
iiiXexp
YE
0
)(1
1
Main advantages of this model:
Basic statistical model used for modelling 0-1 variable with good mathematical properties
),...,( 1 kXXX
Logistic regression
Scoring Unit (SCOR)13 April 2008
STRESS-TESTING
STRESS-TESTINGMODELS
Factor model based on Merton’s model
ittit UFR 1
Where
is logarithmic change of client’s asset
is systematic factor
is specific factor
itR
tF
itU
)1(
log
ti
itit A
AR
)1,0(NFt
)1,0(NU it
),( jtit RRcor),( tit FRcor
Scoring Unit (SCOR)14 April 2008
STRESS-TESTING
STRESS-TESTINGMODELS
Factor model based on Merton’s model
)()1( CRPYP itit
itY
J
jjtjxC
10
1 in case of default
0 in case of non-default
J
jjtj
J
jjtjitt
ititit
xxUFP
CRPYPPD
1100 )()1(
)()1(
Probability of default
Scoring Unit (SCOR)15 April 2008
STRESS-TESTING
STRESS-TESTINGMODELS
1)|1()( 1
0 t
J
jjtj
itttittit
fx
UPfFYPfPD
T
ttt
dnJ
jtjtj
dJ
jtjtj
t
tJ dff
fxfx
d
nl
ttt
1
10
10
0 )(1
11
log),,...,(
Conditional probability of default:
Factor model based on Merton’s model
Likelihood function derivated from binomial distribution of default rate: