Professional Documents
Culture Documents
Rat i ng Model s
and Val i dat i on
.
25
See EIGERMANN, J., Quantitatives Credit-Rating unter Einbeziehung qualitativer Merkmale, p. 102.
26
In the probit model, the function used is j
LLENBECK, M./STEIN, S., Risikofruherkennung im Kreditgeschaft mit kleinen und mittleren Unter-
nehmen, p. 14.
Rating Models and Validation
44 Guidelines on Credit Risk Management
logistic :cg:cssio: |s :uu|c: oi cv:tgcs ovc: \DA li:st, logistic
:cg:cssio: cocs :ot :cqui:c :o:ul cist:i|utio: i: i:ut i:cicto:s I|is llows
logistic :cg:cssio: uoccls to :occss qulittivc c:ccitwo:t|i:css c|:ctc:istics
wit|out :cvious t::sio:utio: cco:c, t|c :csult oi logistic :cg:cssio: c:
|c i:tc::ctcc ci:cctl, s t|c :o||ilit, oi g:ou ucu|c:s|i I|is u|cs it
ossi|lc to ssig: o:c,c: cciult :tc to cc| :csult, io: cxulc |, :cscli:g
t|c vluc
logistic :cg:cssio: uoccls :c oitc: c|:ctc:izcc |, uo:c :o|ust
!
:c
uo:c ccu:tc :csults t|: t|osc gc:c:tcc |, cisc:iui::t :l,sis
l: :ccc:t ,c:s, logistic :cg:cssio: |s scc: uo:c wiccs:cc usc |ot| i:
cccuic :csc:c| :c i: :cticc I|is c: |c tt:i|utcc to t|c lowc: ccu:cs
it u|cs o: ct utc:il s wcll s its uo:c :o|ust :csults cou:cc to cisc:iu
i::t :l,sis
:c cxulc oi t|c :cticl usc oi logistic :cg:cssio: i: |:|s is t|c lVkll
:ti:g uoccl uscc |, t|c lccc:l Associtio: oi Gc:u: ooc:tivc l:|s to
:tc sull :c ucciuusizcc c:tc::iscs
!-
3.2.3 Artificial Neural Networks
Structure of Artificial Neural Networks
A:tiiicil :cu:l :ctwo:|s usc i:io:utio: tcc|:olog, i: : ttcut to siuultc
t|c w, i: w|ic| t|c |uu: |:i: :occsscs i:io:utio: l: siuliiicc tc:us,
t|c |uu: |:i: co:sists oi l:gc :uu|c: oi :c:vc cclls ;:cu:o:s, co::cctcc
to o:c :ot|c: |, :ctwo:| oi s,:scs :cu:o:s :cccivc sig:ls t|:oug| t|csc
s,:scs, :occss t|c i:io:utio:, :c ss :cw sig:ls o: t|:oug| ot|c: :cu
:o:s I|c sig:iiic:cc oi :ticul: iccc oi i:io:utio: is cctc:ui:cc |, t|c
t,c :c st:c:gt| oi t|c li:|s |ctwcc: :cu:o:s l: t|is w,, i:io:utio: c: |c
cist:i|utcc :c :occsscc i: :llcl c:oss t|c c:ti:c :ctwo:| oi :cu:o:s I|c
|uu: |:i: is |lc to lc:: cuc to its ccit, to c|ust t|c wcig|ti:g oi li:|s
|ctwcc: :cu:o:s
A:tiiicil :cu:l :ctwo:|s ttcut to uoccl t|is |iologicl :occss A: :ti
iicil :cu:l :ctwo:| co:sists oi : i:ut l,c:, t|c i::c: l,c:s :c : outut
l,c: ;scc c|:t 1-,
I|c i:ut l,c: sc:vcs t|c u:osc oi t|i:g i: i:io:utio: ;cg scciiic
i:cicto: vlucs, :c ssi:g it o: to t|c cow:st:cu :cu:o:s vi t|c co::cc
tio:s s|ow: i: t|c cig:u |clow I|csc li:|s :c ssig:cc wcig|ts i: : :ti
iicil :cu:l :ctwo:| :c t|us co:t:ol t|c ilow oi i:io:utio:
l: t|c :cu:o:s, ll i:coui:g i:io:utio: i
;
is ii:st li:|cc wit| vluc . I|is
is co:c |, uc:s oi siulc suu iu:ctio: lc| iccc oi i:io:utio: is t|c:
ssig:cc co::cctio: wcig|t n I|c cou:csscc vluc . is t::sio:ucc i:to
vluc o |, :o:li:c: iu:ctio: I|c iu:ctio: uscc io: t|is u:osc ccc:cs
o: t|c scciiic uoccl :c cxulc is t|c iollowi:g logistic iu:ctio:
o
1
1 c
.
.
28
Cf. KALTOFEN, D./MO
LLENBECK, M./STEIN, S., Risikofruherkennung im Kreditgeschaft mit kleinen und mittleren Unter-
nehmen, p. 14.
29
Cf. STUHLINGER, MATTHIAS, Rolle von Ratings in der Firmenkundenbeziehung von Kreditgenossenschaften, p. 72.
Rating Models and Validation
Guidelines on Credit Risk Management 45
|:t 1- A:c|itcctu:c oi : A:tiiicil :cu:l :ctwo:|
|:t !0 low :cu:o:s Wo:|
I|is t::sio:ucc vluc is sscc o: to ll cow:st:cu :cu:o:s, w|ic| i:
tu:: c::, out t|c suc :occcu:c wit| t|c outut icto:s i:ou t|c ust:cu
:cu:o:s |:t !0 givcs sc|cutic ccictio: oi |ow :cu:o: wo:|s l: gc:
c:l, ot|c: :o:li:c: iu:ctio:s c: |c uscc i:stcc oi t|c logistic iu:ctio:
:cc t|c i:io:utio: |s sscc t|:oug| t|c i::c: l,c:s, it is cclivc:cc to
t|c :cu:o: i: t|c outut l,c: I|is i:io:utio: :c:csc:ts t|c :ctwo:|`s out
ut, o: t|c :csult gc:c:tcc |, t|c :tiiicil :cu:l :ctwo:| I|c i::c: l,c:s
:c lso :cic::cc to s |iccc: l,c:s |ccusc t|c sttc oi t|c :cu:o:s i: t|csc
l,c:s is :ot visi|lc i:ou t|c outsicc
Training of Artificial Neural Networks
A: :tiiicil :cu:l :ctwo:| lc::s o: t|c |sis oi t:i:i:g ct scts io: w|ic|
t|c ctul co::cct outut is l:cc, |:ow: l: t|c t:i:i:g :occss, t|c :tiiicil
:cu:l :ctwo:| cou:cs t|c outut gc:c:tcc wit| t|c ctul outut :c
Rating Models and Validation
46 Guidelines on Credit Risk Management
cts t|c :ctwo:| cco:ci:g to :, ccvitio:s it ii:cs l:o||l, t|c uost cou
uo:l, uscc uct|oc oi u|i:g suc| c|:gcs i: :ctwo:|s is t|c c|ustuc:t oi
wcig|ts |ctwcc: :cu:o:s I|csc wcig|ts i:cictc |ow iuo:t:t iccc oi
i:io:utio: is co:sicc:cc to |c io: t|c :ctwo:|`s outut l: cxt:cuc cscs,
t|c li:| |ctwcc: two :cu:o:s will |c cclctcc |, sctti:g t|c co::cso:ci:g
wcig|t to zc:o
A clssic lc::i:g lgo:it|u w|ic| ccii:cs t|c :occcu:c io: c|usti:g
wcig|ts is t|c |c|:ogtio: lgo:it|u I|is tc:u :cic:s to g:cic:t
ccscc:t uct|oc w|ic| clcultcs c|:gcs i: wcig|ts cco:ci:g to t|c c::o:s
ucc |, t|c :cu:l :ctwo:|
!0
l: t|c ii:st stc, outut :csults :c gc:c:tcc io: :uu|c: oi ct :cco:cs
I|c ccvitio: oi t|c clcultcc outut o
d
i:ou t|c ctul outut t
d
is ucs
u:cc usi:g : c::o: iu:ctio: I|c suuoisqu:cs c::o: iu:ctio: is i:cquc:tl,
uscc i: t|is co:tcxt
c
1
2
d
t
d
o
d
2
I|c clcultcc c::o: c: |c |c|:ogtcc :c uscc to c|ust t|c :clcv:t
wcig|ts I|is :occss |cgi:s t t|c outut l,c: :c c:cs t t|c i:ut l,c:
!1
W|c: t:i:i:g : :tiiicil :cu:l :ctwo:|, it is iuo:t:t to voic w|t is
:cic::cc to s ovc:iitti:g vc:iitti:g :cic:s to situtio: i: w|ic| : :tiiicil
:cu:l :ctwo:| :occsscs t|c suc lc::i:g ct :cco:cs gi: :c gi: u:til it
|cgi:s to :ccog:izc :c ucuo:izc scciiic ct st:uctu:cs wit|i: t|c sulc
I|is :csults i: |ig| cisc:iui:to:, owc: i: t|c lc::i:g sulc uscc, |ut low
cisc:iui:to:, owc: i: u:|:ow: sulcs I|c:cio:c, t|c ovc:ll sulc uscc
i: ccvcloi:g suc| :ctwo:|s s|oulc ccii:itcl, |c civiccc i:to lc::i:g, tcsti:g
:c vlictio: sulc i: o:cc: to :cvicw t|c :ctwo:|`s lc::i:g succcss usi:g
u:|:ow: sulcs :c to sto t|c t:i:i:g :occcu:c i: tiuc I|is :ccc to
civicc u t|c sulc lso i:c:cscs t|c qu:tit, oi ct :cqui:cc
Application of Artificial Neural Networks
:cu:l :ctwo:|s :c |lc to :occss |ot| qu:tittivc :c qulittivc ct
ci:cctl,, w|ic| u|cs t|cu csccill, suit|lc io: t|c ccictio: oi coulcx :t
i:g uoccls w|ic| |vc to t|c v:ious i:io:utio: ctcgo:ics i:to ccou:t
Alt|oug| :tiiicil :cu:l :ctwo:|s :cgul:l, ccuo:st:tc |ig| cisc:iui:to:,
owc: :c co :ot i:volvc sccil :cqui:cuc:ts :cg:ci:g i:ut ct, t|csc :t
i:g uoccls :c still :ot vc:, :cvlc:t i: :cticc I|c :cso:s io: t|is lic i: t|c
coulcx :ctwo:| uoccli:g :occcu:cs i:volvcc :c t|c |lc| |ox :tu:c oi
t|csc :ctwo:|s As t|c i::c: wo:|i:gs oi :tiiicil :cu:l :ctwo:|s :c :ot
t::s:c:t to t|c usc:, t|c, :c csccill, susccti|lc to ccct:cc :o|lcus
:c cxulc oi : :tiiicil :cu:l :ctwo:| uscc i: :cticc is t|c llk
;lctgclil:zkti:g
SER, K., Mittelstandsrating mit Hilfe neuronaler Netzwerke, p. 372; cf. HEITMANN, C. , Neuro-Fuzzy, p. 20.
33
Cf. SCHIERENBECK, H., Ertragsorientiertes Bankmanagement Vol. 1.
34
Adapted from GERDSMEIER, S./KROB, B., Bepreisung des Ausfallrisikos mit dem Optionspreismodell, p. 469ff.
35
Cf. KIRMSSE, S., Optionspreistheoretischer Ansatz zur Bepreisung.
Rating Models and Validation
48 Guidelines on Credit Risk Management
I|c :uctc:s :cqui:cc to clcultc t|c otio: :icc ; :is| :cuiuu, :c
t|c cu:tio: oi t|c o|sc:vtio: c:ioc s wcll s t|c iollowi:g
!o
lco:ouic vluc oi t|c cc|t
!
lco:ouic vluc oi t|c cquit,
Voltilit, oi t|c sscts
Duc to t|c :cqui:cc ct i:ut, t|c otio: :ici:g uoccl c::ot cvc: |c
co:sicc:cc io: lictio:s i: :ctil |usi:css
!
lowcvc:, gc:c:ti:g t|c ct
:cqui:cc to usc t|c otio: :ici:g uoccl i: t|c co:o:tc scguc:t is lso
:ot wit|out its :o|lcus, io: cxulc |ccusc t|c cco:ouic vluc oi t|c cou
:, c::ot |c cstiutcc :clisticll, o: t|c |sis oi u|licl, vil|lc i:io:u
tio: lo: t|is u:osc, i:|ousc l::i:g ct :c usull, :cqui:cc i:ou t|c
cou:, itscli I|c cou:,`s vluc c: lso |c clcultcc usi:g t|c ciscou:tcc
cs| ilow uct|oc lo: cxc|:gclistcc cou:ics, voltilit, is i:cquc:tl, csti
utcc o: t|c |sis oi t|c stoc| :icc`s voltilit,, w|ilc :cic:c:cc vlucs scciiic
to t|c i:cust:, o: :cgio: :c uscc i: t|c csc oi u:listcc cou:ics
l: :cticc, t|c otio: :ici:g uoccl |s o:l, |cc: iulcuc:tcc to liu
itcc cxtc:t i: Gc:u:sc|i:g cou:t:ics, ui:l, s : i:st:uuc:t oi c:ccit
sscssuc:t io: cxc|:gclistcc cou:ics
!-
lowcvc:, t|is uoccl is lso |ci:g
uscc io: u:listcc cou:ics s wcll
As c:ccit cciult o: t|c :t oi t|c cou:, is ossi|lc t :, tiuc cu:i:g
t|c o|sc:vtio: c:ioc ;:ot |ust t t|c c:c,, t|c :is| :cuiuu :c cciult :tcs
clcultcc wit| lu:oc:st,lc
+0
otio: :ici:g uoccl :c co:sc:vtivcl,
i:tc::ctcc s t|c lowc: liuits oi t|c :is| :cuiuu :c cciult :tc i: :cticc
Qulittivc cou:, vlutio: c:itc:i :c o:l, i:cluccc i: t|c otio: :ic
i:g uoccl to t|c cxtc:t t|t t|c u:|ct :iccs uscc s|oulc t|c t|is i:io:utio:
;ii vil|lc to t|c u:|ct :tici:ts, i:to ccou:t lc,o:c t|t, t|c otio:
:ici:g uoccl cocs :ot covc: qulittivc c:itc:i lo: t|is :cso:, t|c lic
tio: oi otio: :ici:g uoccls s|oulc |c :cst:ictcc to l:gc: cou:ics io: w|ic|
o:c c: ssuuc t|t t|c u:|ct :icc :cilccts qulittivc icto:s suiiicic:tl, ;ci
scctio: +!!,
3.3.2 Cash Flow (Simulation) Models
s| ilow ;siuultio:, uoccls :c csccill, wcll suitcc to c:ccit sscssuc:t io:
sccilizcc lc:ci:g t::sctio:s, s c:ccitwo:t|i:css i: t|is co:tcxt ccc:cs :i
u:il, o: t|c iutu:c cs| ilows :isi:g i:ou t|c sscts ii::ccc l: t|is csc, t|c
t::sctio: itscli ;:c :ot scciiic |o::owc:, is sscsscc cxlicitl,, :c t|c
:csult is t|c:cio:c :cic::cc to s transaction rating
s| ilow|scc uoccls c: lso |c :csc:tcc s v:itio: o: otio: :ic
i:g uoccls i: w|ic| t|c cco:ouic vluc oi t|c cou:, is clcultcc o: t|c
|sis oi cs| ilow
36
The observation period is generally one year, but longer periods are also possible.
37
For more information on the fundamental circular logic of the option pricing model due to the mutual dependence of the market
value of debt and the risk premium as well as the resolution of this problem using an iterative method, see JANSEN, S, Ertrags-
und volatilitatsgestutzte Kreditwurdigkeitsprufung, p. 75 (footnote) and VARNHOLT B., Modernes Kreditrisikomanagement, p.
107 ff. as well as the literature cited there.
38
Cf. SCHIERENBECK, H., Ertragsorientiertes Bankmanagement Vol. 1.
39
Cf. SCHIERENBECK, H., Ertragsorientiertes Bankmanagement Vol. 1.
40
In contrast to an American option, which can be exercised at any time during the option period.
Rating Models and Validation
Guidelines on Credit Risk Management 49
l: :i:cil, it is ossi|lc to ccii:c cs| ilow i:ou v:ious c:scctivcs
w|ic| :c co:sicc:cc cqull, suit|lc Acco:ci:gl,, suit|lc vlutio: uoccl
c: |c sclcctcc cco:ci:g to t|c i:civicul :cqui:cuc:ts oi t|c o:g:iztio:
c:io:ui:g t|c vlutio: :c i: li:c wit| t|c u:osc oi t|c vlutio: l: t|is
co:tcxt, t|c totl i:cc cs| ilow vil|lc is :ticul:l, :clcv:t to vlutio:
+1
lo: t|c u:osc oi cou:, vlutio: o: citl u:|cts, i:cc cs| ilow is
clcultcc s lllIDA
+!
ui:us i:vcstuc:ts
+!
I|c vc:gc i:cc cs| ilowovc: t|c
lst iivc ,c:s gc:c:ll, sc:vcs s t|c oi:t oi cc:tu:c io: clculti:g cou
:,`s vluc Divici:g t|c vc:gc i:cc cs| ilow |, t|c wcig|tcc citl costs
io: cquit, :c cc|t ii::ci:g
++
,iclcs cou:, vluc w|ic| c: |c uscc s :
i:ut :uctc: i: t|c otio: :ici:g uoccl I|c voltilit, oi t|is vluc c: |c
clcultcc i: : :logous w, |scc o: t|c tiuc sc:ics uscc to cctc:ui:c t|c
vc:gc i:cc cs| ilow
Iwo t,cs oi uct|ocs c: |c uscc to cxt:oltc iutu:c cs| ilow o: t|c
|sis oi st cs| ilow ct
Analytical methods :c |scc o: tiuc sc:ics :l,sis uct|ocs, w|ic| couc i:
two io:us
Regression models c:ctc iu:ctio:l uoccl oi t|c tiuc sc:ics :c otiuizc
t|c uoccl :uctc:s |, ui:iuizi:g t|c ccvitio:s o|sc:vcc
Stochastic time series models ccict t|c tiuc sc:ics s t|c :cliztio: oi sto
c|stic :occss :c clcultc otiuuu cstiutcs io: t|c :occss :uctc:s
Simulation methods gc:c:tc :c wcig|t ossi|lc iutu:c :cliztio:s oi cs|
ilow o: t|c |sis oi |isto:icl ct o: i: t|c :oc| uo:c couuo:l, uscc
i: :cticc |, ccvcloi:g uc:occo:ouic uoccls w|ic| ccict t|c i:ut vl
ucs io: cs| ilow i: :cltio: to cc:ti: scc::ios ;cg t|c ovc:ll ccvclouc:t oi
t|c cco:ou,,
3.4 Hybrid Forms
l: :cticc, t|c uoccls ccsc:i|cc i: t|c :cvious scctio:s :c o:l, ::cl, uscc i:
t|ci: u:c io:us kt|c:, |cu:istic uoccls :c gc:c:ll, cou|i:cc wit| o:c oi
t|c two ot|c: uoccl t,cs ;sttisticl uoccls o: cusl uoccls, I|is :oc|
c: gc:c:ll, |c scc: s ivo:|lc, s t|c v:ious :oc|cs coulcuc:t cc|
ot|c: wcll lo: cxulc, t|c cv:tgcs oi sttisticl :c cusl uoccls lic i:
t|ci: o||cctivit, :c gc:c:ll, |ig|c: clssiiictio: c:io:u:cc i: cou:iso:
to |cu:istic uoccls lowcvc:, sttisticl :c cusl uoccls c: o:l, :occss
liuitcc :uu|c: oi c:ccitwo:t|i:css icto:s Wit|out t|c i:clusio: oi c:ccit
cxc:ts` |:owlccgc i: t|c io:u oi |cu:istic uoculcs, iuo:t:t i:io:utio:
o: t|c |o::owc:`s c:ccitwo:t|i:css woulc |c lost i: i:civicul cscs l: cci
tio:, :ot ll sttisticl uoccls :c c|lc oi :occssi:g qulittivc i:io:utio:
ci:cctl, ;s is t|c csc wit| cisc:iui::t :l,sis, io: cxulc,, o: t|c, :cqui:c
l:gc uou:t oi ct i: o:cc: to iu:ctio: :oc:l, ;cg logistic :cg:cssio:,,
t|csc ct :c i:cquc:tl, u:vil|lc i: |:|s l: o:cc: to o|ti: coulctc ic
41
Cf. JANSEN, S., Ertrags- und volatilitatsgestutzte Kreditwurdigkeitsprufung.
42
EBITDA: earnings before interest, tax, depreciation and amortization.
43
Cf. KEMPF, M., Dem wahren Aktienkurs auf der Spur. More detailed explanations on the general conditions of the cash flow
method can be found in JANSEN, S., Ertrags- und volatilitatsgestutzte Kreditwurdigkeitsprufung.
44
This interest rate is required in order to discount future earnings to their present value.
Rating Models and Validation
50 Guidelines on Credit Risk Management
tu:c oi t|c |o::owc:`s c:ccitwo:t|i:css i: suc| cscs, it t|us u|cs sc:sc to
sscss qulittivc ct usi:g sulcuc:t:, |cu:istic uoccl
I|is |cu:istic couo:c:t lso i:volvcs c:ccit cxc:ts uo:c |cvil, i: t|c
:ti:g :occss t|: i: t|c csc oi utoutcc c:ccit sscssuc:t usi:g sttisticl
o: cusl uoccl, uc:i:g t|t cou|i:i:g uoccls will lso sc:vc to i:c:csc usc:
ccct:cc
l: t|c scctio:s |clow, t|:cc ciiic:c:t :c|itcctu:cs io: t|c cou|i:tio: oi
t|csc uoccl t,cs :c :csc:tcc
3.4.1 Horizontal Linking of Model Types
As sttisticl :c cusl uoccls ccuo:st:tc :ticul: st:c:gt| i: t|c sscss
uc:t oi qu:tittivc ct, :c t t|c suc tiuc uost oi t|csc uoccls c::ot
:occss qulittivc ct wit|out sig:iiic:t ccitio:l ciio:t, t|is cou|i:tio:
oi uoccl t,cs c: |c c:cou:tc:cc i:cquc:tl, i: :cticc A sttisticl uoccl
o: cusl uoccl is uscc to :l,zc ::ul ii::cil sttcuc:ts o: ;i: |:occ:
tc:us, to cvlutc |o::owc:`s ii::cil situtio: Qulittivc ct ;cg u:
gcuc:t qulit,, is cvlutcc usi:g |cu:istic uoculc i:cluccc i: t|c uoccl
lI is t|c: ossi|lc to uc:gc t|c outut :ocuccc |, t|csc two uoculcs to gc:
c:tc : ovc:ll c:ccit sscssuc:t
|:t !! lo:izo:tl li:|i:g oi kti:g \occls
Applied example:
:c :cticl cxulc oi t|is cou|i:tio: oi :ti:g uoccls c: |c iou:c i: t|c
Dcutsc|c lu:ccs|:|`s c:ccit sscssuc:t :occcu:c ccsc:i|cc i: scctio: !1+
A::ul ii::cil sttcuc:ts :c :l,zcc |, uc:s oi sttisticl cisc:iui::t
:l,sis I|is qu:tittivc c:ccitwo:t|i:css :l,sis is sulcuc:tcc wit| cci
tio:l qulittivc c:itc:i w|ic| :c sscsscc usi:g iuzz, logic s,stcu I|c
ovc:ll s,stcu is s|ow: i: c|:t !!
Rating Models and Validation
Guidelines on Credit Risk Management 51
|:t !! A:c|itcctu:c oi Dcutsc|c lu:ccs|:|`s :ccit Asscssuc:t l:occcu:c
+.
3.4.2 Vertical Linking of Model Types Using Overrides
I|is :oc| li:|s :til :ti:gs to gc:c:tc :ooscc clssiiictio:, w|ic|
c: t|c: |c uociiicc |, c:ccit cxc:t to ,iclc t|c ii:l c:ccit :ti:g I|is
cow:st:cu uociiictio: couo:c:t |scc o: cxc:t |:owlccgc co:stitutcs
sc:tc t,c oi |,|:ic uoccl I|is cou|i:tio: ii:st sscsscs qu:tittivc
s wcll s qulittivc c:ccitwo:t|i:css c|:ctc:istics usi:g sttisticl o: cusl
uoccl I|c :csult oi t|is sscssuc:t is :ooscc clssiiictio: w|ic| c: t|c:
|c uociiicc ;wit|i: cc:ti: liuits, |, c:ccit :l,sts o: t|c |sis oi t|ci: cxc:t
|:owlccgc
l: t|csc cou|i:cc uoccls, it is iuo:t:t to ccii:c :cciscl, t|c cscs :c
t|c ::gc i: w|ic| ovc::iccs c: |c uscc
l: :ticul:, icts w|ic| |vc l:cc, |cc: uscc i: sttisticl o: cusl :l
,sis uoculc s|oulc :ot sc:vc s t|c |sis io: ltc: uociiictio:s |, c:ccit :
l,sts l:stcc, t|c |cu:istic couo:c:t is iuo:t:t i: o:cc: to i:clucc icto:s
:clcv:t to c:ccitwo:t|i:css w|ic| :c o:l, |:ow: to t|c c:ccit :l,st :c
w|ic| coulc :ot |c covc:cc |, t|c ust:cu uoculc
li t|c ust:cu uoculc is uocclcc :oc:l,, |owcvc:, ovc::iccs s|oulc o:l,
|c :cccss:, i: souc cscs I|c cxccssivc usc oi ovc::iccs u, i:cictc lc| oi
usc: ccct:cc o: lc| oi u:cc:st:ci:g oi t|c :ti:g uoccl :c s|oulc t|c:c
io:c |c :cvicwcc c:ciull, i: t|c cou:sc oi vlictio:
45
Adapted from. BLOCHWITZ, STEFAN/EIGERMANN, JUDITH, Bonitatsbeurteilungsverfahren der Deutschen Bundesbank.
Rating Models and Validation
52 Guidelines on Credit Risk Management
|:t !+ Vc:ticl li:|i:g oi kti:g \occls Dsi:g vc::iccs
3.4.3 Upstream Inclusion of Heuristic Knock-Out Criteria
I|c co:c clcuc:t oi t|is cou|i:tio: oi uoccl t,cs is t|c sttisticl uoculc
lowcvc:, t|is uoculc is :cccccc |, |:oc|out c:itc:i ccii:cc o: t|c |sis oi
t|c :cticl cxc:ic:cc oi c:ccit cxc:ts :c t|c |:|`s i:civicul st:tcg, li
otc:til |o::owc: iuliills |:oc|out c:itc:io:, t|c c:ccit sscssuc:t :occss
cocs :ot co:ti:uc cow:st:cu to t|c sttisticl uoculc
|:t !. l,|:ic lo:us Dst:cu l:clusio: oi lcu:istic l:oc|ut :itc:i
l:oc|out c:itc:i io:u : i:tcg:l :t oi c:ccit :is| st:tcgics :c :o
vl :cticcs i: c:ccit i:stitutio:s :c cxulc oi |:oc|out c:itc:io: uscc i:
:cticc is :cgtivc :co:t i: t|c co:suuc: lo:s :cgistc: li suc| :co:t is
iou:c, t|c |:| will :c|cct t|c c:ccit lictio: cvc: |cio:c cctc:ui:i:g cii
ic:c:titcc c:ccit :ti:g usi:g its i:|ousc :occcu:cs
Rating Models and Validation
Guidelines on Credit Risk Management 53
4 Assessing the Models Suitability for
Various Rating Segments
l: gc:c:l, c:ccit sscssuc:t :occcu:cs |vc to iuliill :uu|c: oi :cqui:c
uc:ts :cg:clcss oi t|c :ti:g scguc:ts i: w|ic| t|c, :c uscc I|csc :cqui:c
uc:ts :c t|c :csult oi |usi:css co:sicc:tio:s licc to c:ccit sscssuc:t s
wcll s cocuuc:ts u|lis|cc o: t|c lkl :oc|cs u:cc: lscl ll I|c iu:c
uc:tl :cqui:cuc:ts :c listcc i: c|:t !o :c cxli:cc i: cctil iu:t|c: |clow
|:t !o lu:cuc:tl kcqui:cuc:ts oi kti:g \occls
4.1 Fulfillment of Essential Requirements
4.1.1 PD as Target Value
I|c :o||ilit, oi cciult :cilcctcc i: t|c :ti:g io:us t|c |sis io: :is| u:gc
uc:t lictio:s suc| s :is||scc lo: :ici:g lculti:g lD s t|c t:gct
vluc is t|c:cio:c |sic :c:cquisitc io: :ti:g uoccl to u|c sc:sc i: t|c
|usi:css co:tcxt
I|c ct sct uscc to clcultc lD is oitc: uissi:g i: heuristic models :c
uig|t |vc to |c ccuuultcc |, usi:g t|c :ti:g uoccl i: :cticc :cc t|is
:cqui:cuc:t is iuliillcc, it is ossi|lc to cli|:tc :csults to cciult :o||ilitics
cvc: i: t|c csc oi |cu:istic uoccls ;scc scctio: .!,
Statistical models :c ccvclocc o: t|c |sis oi : cui:icl ct sct, w|ic|
u|cs it ossi|lc to cctc:ui:c t|c t:gct vluc lD io: i:civicul :ti:g clsscs
|, cli|:ti:g :csults wit| t|c cui:icl ccvclouc:t ct li|cwisc, it is os
si|lc to cli|:tc t|c :ti:g uoccl ;cx ost, i: t|c cou:sc oi vlictio: usi:g t|c
ct gi:cc i:ou :cticl cclo,uc:t
:c cssc:til |c:ciit oi logistic regression is t|c ict t|t it c:|lcs t|c ci:cct
clcultio: oi cciult :o||ilitics lowcvc:, cli|:tio: o: :cscli:g u, lso
Rating Models and Validation
54 Guidelines on Credit Risk Management
|c :cccss:, i: t|is csc ii t|c cciult :tc i: t|c sulc ccvitcs i:ou t|c vc:
gc cciult :tc oi t|c :ti:g scguc:t ccictcc
l: t|c csc oi causal models, t|c t:gct vluc lD c: |c clcultcc io: i:ci
vicul :ti:g clsscs usi:g ct gi:cc i:ou :cticl cclo,uc:t l: t|is csc,
t|c uoccl ci:cctl, oututs t|c cciult :uctc: to |c vlictcc
4.1.2 Completeness
l: o:cc: to c:su:c t|c coulctc:css oi c:ccit :ti:g :occcu:cs, lscl ll
:cqui:cs |:|s to t|c ll vil|lc i:io:utio: i:to ccou:t w|c: ssig:i:g :t
i:gs to |o::owc:s o: t::sctio:s
+o
I|is s|oulc lso |c uscc s guiccli:c io:
|cst :cticcs
As :ulc, classic rating questionnaires o:l, usc sull :uu|c: oi c|:ctc:
istics :clcv:t to c:ccitwo:t|i:css lo: t|is :cso:, it is iuo:t:t to :cvicw t|c
coulctc:css oi icto:s :clcv:t to c:ccitwo:t|i:css i: c:iticl lig|t li t|c
uoccl |s :occsscc suiiicic:t qu:tit, oi cxc:t |:owlccgc, it c: covc:
ll i:io:utio: ctcgo:ics :clcv:t to c:ccit sscssuc:t
li|cwisc, qualitative systems oitc: usc o:l, sull :uu|c: oi c:ccitwo:t|i
:css c|:ctc:istics As cxc:t |:owlccgc is lso :occsscc ci:cctl, w|c: t|c
uoccl is licc, |owcvc:, t|csc s,stcus c: covc: uost i:io:utio: ctcgo:ics
:clcv:t to c:ccitwo:t|i:css
I|c couutc:|scc :occssi:g oi i:io:utio: c:|lcs expert systems :c
fuzzy logic systems to t|c l:gc :uu|c: oi c:ccitwo:t|i:css c|:ctc:istics i:to
co:sicc:tio:, uc:i:g t|t suc| s,stcu c: covc: ll oi t|osc c|:ctc:istics ii
it is uocclcc :oc:l,
As l:gc :uu|c: oi c:ccitwo:t|i:css c|:ctc:istics c: |c tcstcc i: t|c
ccvclouc:t oi statistical models, it is ossi|lc to c:su:c t|c coulctc:css oi
t|c :clcv:t :is| icto:s ii t|c uoccl is ccsig:cc :oc:l, lowcvc:, i: u:,
cscs t|csc uoccls o:l, :occss icw c|:ctc:istics oi |ig| cisc:iui:to:,
owc: w|c: t|csc uoccls :c licc, :c t|c:cio:c it is :cccss:, to :cvicw
coulctc:css c:iticll, i: t|c cou:sc oi vlictio:
Causal models cc:ivc c:ccit :ti:gs usi:g t|co:cticl |usi:css|scc uoccl
:c usc o:l, icw cxclusivcl, qu:tittivc i:ut :uctc:s wit|out cxlic
itl, t|i:g qulittivc ct i:to ccou:t I|c i:io:utio: :clcv:t to c:ccitwo:
t|i:css is t|c:cio:c o:l, coulctc i: cc:ti: scguc:ts ;cg sccilizcc lc:ci:g,
l:gc co:o:tc custouc:s,
4.1.3 Objectivity
l: o:cc: to c:su:c t|c o||cctivit, oi t|c uoccl, classic rating questionnaire
s|oulc co:ti: qucstio:s o: c:ccitwo:t|i:css icto:s w|ic| c: |c :swc:cc
clc:l, :c wit|out :oou io: i:tc::cttio:
Ac|icvi:g |ig| cisc:iui:to:, owc: i: qualitative systems :cqui:cs t|t t|c
:ti:g g:ccs gc:c:tcc |, qulittivc sscssuc:ts usi:g :cccii:cc sclc :c s
o||cctivc s ossi|lc I|is c: o:l, |c c:su:cc |, :ccisc, u:cc:st:c|lc :c
lusi|lc usc:`s u:ul :c t|c :o:itc t:i:i:g ucsu:cs
As expert systems :c fuzzy logic systems cctc:ui:c t|c c:ccitwo:t|i:css
:csult usi:g ccii:cc lgo:it|us :c :ulcs, ciiic:c:t c:ccit :l,sts usi:g t|c
46
Cf. EUROPEAN COMMISSION, draft directive on regulatory capital requirements, Annex D-5, No. 8.
Rating Models and Validation
Guidelines on Credit Risk Management 55
suc i:io:utio: :c t|c co::cct uoccl i:uts will :cccivc t|c suc :ti:g
:csults l: t|is :cscct, t|c uoccl c: |c co:sicc:cc o||cctivc
l: statistical models, c:ccitwo:t|i:css c|:ctc:istics :c sclcctcc :c
wcig|tcc usi:g : cui:icl ct sct :c o||cctivc uct|ocs I|c:cio:c, wc
c: :cg:c t|csc uoccls s o||cctivc :ti:g :occcu:cs W|c: t|c uoccl is su
licc :oc:l, wit| t|c suc i:io:utio:, ciiic:c:t c:ccit :l,sts will |c |lc
to gc:c:tc t|c suc :csults
li causal models :c sulicc wit| t|c co::cct i:ut :uctc:s, t|csc uoc
cls c: lso |c :cg:ccc s o||cctivc
4.1.4 Acceptance
As heuristic models :c ccsig:cc o: t|c |sis oi cxc:t oi:io:s :c t|c cxc
:ic:cc oi :ctitio:c:s i: t|c lc:ci:g |usi:css, wc c: ssuuc t|t t|csc uoccls
will ucct wit| |ig| ccct:cc
I|c cxl:to:, couo:c:t i: : expert system u|cs t|c clcultio: oi t|c
c:ccit sscssuc:t :csult t::s:c:t to t|c usc:, t|us c:|:ci:g t|c ccct:cc
oi suc| uoccls
l: fuzzy logic systems, ccct:cc u, |c lowc: t|: i: t|c csc oi cxc:t
s,stcus, s t|c io:uc: :cqui:c g:ctc: ccg:cc oi cxc:t |:owlccgc cuc to
t|ci: uoccli:g oi iuzzi:css wit| li:guistic v:i|lcs lowcvc:, t|c co:c oi
iuzz, logic s,stcus uoccls t|c cxc:ic:cc oi c:ccit cxc:ts, w|ic| uc:s it
is ossi|lc io: t|is uoccl t,c to tti: t|c :cccss:, ccct:cc ccsitc its
i:c:cscc coulcxit,
Statistical rating models gc:c:ll, ccuo:st:tc |ig|c: cisc:iui:to:, owc:
t|: |cu:istic uoccls lowcvc:, it c: |c uo:c ciiiicult to gi: uct|ocicl
ccct:cc io: sttisticl uoccls t|: io: |cu:istic uoccls :c cssc:til :cso:
io: t|is is t|c l:gc uou:t oi cxc:t |:owlccgc :cqui:cc io: sttisticl uoccls
l: o:cc: to i:c:csc ccct:cc :c to c:su:c t|t t|c uoccl is licc i: :
o||cctivcl, :oc: u::c:, usc: t:i:i:g scui::s :c i:cisc:s|lc
:c scvc:c ciscv:tgc io: t|c ccct:cc oi artificial neural networks is
t|ci: |lc| |ox :tu:c
+
I|c i:c:csc i: cisc:iui:to:, owc: c|icvcc |,
suc| uct|ocs c: ui:l, |c tt:i|utcc to t|c coulcx :ctwo:|`s |ilit, to lc::
:c t|c :llcl :occssi:g oi i:io:utio: wit|i: t|c :ctwo:| lowcvc:, it is
:cciscl, t|is coulcxit, oi :ctwo:| :c|itcctu:c :c t|c cist:i|utio: oi i:io:
utio: c:oss t|c :ctwo:|s w|ic| u|c it ciiiicult to cou:c|c:c t|c :ti:g
:csults I|is :o|lcu c: o:l, |c cou:tc:cc |, t|c :o:itc t:i:i:g ucs
u:cs ;cg sc:sitivit, :l,scs c: u|c t|c :occssi:g oi i:io:utio: i: :tiiicil
:cu:l :ctwo:| c: uo:c lusi|lc :c cou:c|c:si|lc to t|c usc:,
Causal models gc:c:ll, ucct wit| ccct:cc w|c: usc:s u:cc:st:c t|c
iu:cuc:tls oi t|c u:cc:l,i:g t|co:, :c w|c: t|c i:ut :uctc:s :c
ccii:cc i: : u:cc:st:c|lc w, w|ic| is lso :o:itc to t|c :ti:g scg
uc:t
Accct:cc i: i:civicul cscs will ccc:c o: t|c ccou:,i:g ucsu:cs
t|c: i: t|c cou:sc oi i:t:ocuci:g t|c :ti:g uoccl, :c i: :ticul: o: t|c
t::s:c:c, oi t|c ccvclouc:t :occss :c cttio:s s wcll s t|c qulit,
oi t:i:i:g scui::s
47
Cf. FU
2
p
in an ideal procedure.
83
The relevant literature contains various definitions of the Pietra Index based on different standardizations; the form used here
with the value range [0, 1] is the one defined in LEE: Global Performances of Diagnostic Tests.
Rating Models and Validation
106 Guidelines on Credit Risk Management
Interpretation of the Pietra Index using Probability Theory
l: t|c cxl:tio: |clow, 41 :c:csc:ts t|c vc:gc |solutc ciiic:c:cc i:
cciult :o||ilitics i: t|c sulc wit| :c wit|out |:owlccgc oi t|c clssiii
ctio: oi t|c cscs i: t|c c :ti:g clsscs
41
c
j
c
j11jc 11j.
Wit| suuutio: ovc: ll :ti:g clsscs c, t|c v:i|lc j
c
:cic:s to t|c :cltivc
i:cquc:c, oi t|c ssig:uc:t oi cscs to i:civicul :ti:g clsscs, 11jc cc:otcs
t|c cciult :tc i: clss c, :c 11 st:cs io: t|c cciult :tc i: t|c sulc
cxui:cc I|us t|c iollowi:g :cltio: is t:uc
+
41 2 11 1 11 Pietra Index.
l: : iccl :occcu:c w|ic| lccs ll t:ul, |c cscs ;:c o:l, t|osc cscs,
i: t|c wo:st :ti:g clss, t|c iollowi:g lics
41
ior
2 11 1 11.
I|c:cio:c, wc c: siulii, t|c :cltio: |ovc s iollows
Pietra Index
41
41
ior
.
Kolmogorov-Smirnov Test for the Pietra Index
As c: |c :ovc:, it is lso ossi|lc to i:tc::ct t|c lict: l:ccx s t|c uxi
uuu ciiic:c:cc |ctwcc: t|c cuuultivc i:cquc:c, cist:i|utio:s oi gooc :c
|c cscs
Pietra Index max 1
qood
cni
1
/od
cni
_
.
l:tc::cti:g t|c lict: l:ccx s t|c uxiuuu ciiic:c:cc |ctwcc: t|c cuuu
ltivc i:cquc:c, cist:i|utio:s io: t|c sco:c vlucs oi gooc :c |c cscs u|cs it
ossi|lc to c:io:u sttisticl tcst io: t|c ciiic:c:ccs |ctwcc: t|csc cist:i|u
tio:s I|is is t|c loluogo:ovui::ov Icst ;l Icst, io: two i:ccc:cc:t su
lcs lowcvc:, t|is tcst o:l, ,iclcs vc:, :oug| :csults, s ll |i:cs oi ciiic:c:ccs
|ctwcc: t|c cist:i|utio: s|cs :c ctu:cc :c cvlutcc, :ot o:l, t|c ciiic:
c:ccs i: vc:gc :ti:g sco:cs io: gooc :c |c csc ;w|ic| :c csccill, :cl
cv:t to cisc:iui:to:, owc:, |ut lso ciiic:c:ccs i: v:i:cc :c t|c |ig|c:
uouc:ts oi t|c cist:i|utio:, t|t is, t|c s|c oi t|c cist:i|utio:s i: gc:c:l
I|c :ull |,ot|csis tcstcc is I|c sco:c cist:i|utio:s oi gooc :c |c cscs
:c icc:ticl I|is |,ot|csis is :c|cctcc t lcvcl ii t|c lict: l:ccx cquls o:
cxccccs t|c iollowi:g vluc
1
1
` j1 j
_
` cc:otcs t|c :uu|c: oi cscs i: t|c sulc cxui:cc :c j :cic:s to t|c
o|sc:vcc cciult :tc li t|c lict: l:ccx 1, t|c:cio:c, sig:iiic:t ciiic:c:
ccs cxist |ctwcc: t|c sco:c vlucs oi gooc :c |c cscs
I|c vlucs 1
|
j
c
j
|
j1jc 11j|j.
w|c:c io: suuutio: ovc: ll :ti:g clsscs c :c |, t|c v:i|lcs j
c
:c j
|
:cic:
to t|c :cltivc i:cquc:c, oi t|c ssig:uc:t oi cscs to i:civicul :ti:g clsscs,
11jc :c 11j| cc:otc t|c cciult :tcs i: clsscs c :c |, t|c iollowi:g :cl
tio: is t:uc
41
2 11 1 11 Gini coeffizient.
As wc c: :c:ocucc i: scvc:l stcs, t|c iollowi:g lics to : iccl :o
cccu:c w|ic| clssiiics ll |c cscs i: t|c wo:st :ti:g clss
41
ior
2 11 1 11.
I|is uc:s t|t wc c: siulii, t|c :cltio: |ovc s iollows
Gini coeffizient
41
41
ior
.
Confidence Levels for the Gini Coefficient and AUC
As o:cciuc:sio:l ucsu:cs oi cisc:iui:to:, owc:, t|c Gi:i ociiicic:t :c
AD :c sttisticl vlucs su||cct to ::cou iluctutio:s l: gc:c:l, two :o
cccu:cs c: |c uscc to clcultc co:iicc:cc lcvcls io: t|csc vlucs
A:l,ticl cstiutio: oi co:iicc:cc lcvcls |, co:st:ucti:g co:iicc:cc |:cs
:ou:c t|c Al o: k cu:vc
-
lcu:istic cstiutio: oi co:iicc:cc lcvcls |, uc:s oi :csuli:g
-0
88
Cf. LEE, Global Performances of Diagnostic Tests.
89
Cf. FAHRMEIR/HENKING/HU
1
`
p
to t|c x vlucs :c t|c vluc
1
`
p
to t|c , vlucs oi t|c oi:ts o: t|c k cu:vc `
:c `
c: |c iou:c i:
t|c |:ow: loluogo:ov cist:i|utio: t|lc
|:t o1 loluogo:ov Dist:i|utio: I|lc io: clcctcc o:iicc:cc lcvcls
|:t o! s|ows siuult:cous co:iicc:cc |:cs :ou:c t|c k cu:vc t
t|c co:iicc:cc lcvcl ;1q,
!
-0 io: t|c cxulc uscc |c:c I|c co:iicc:cc
i:tc:vl io: t|c vluc AD ! ;clcultcc usi:g t|c sulc, is |ctwcc:
o-- :c -1.
I|c t|lc |clow ;c|:t o!, s|ows t|c co:iicc:cc i:tc:vls oi t|c :uctc:
AD clcultcc :l,ticll, usi:g siuult:cous co:iicc:cc |:cs io: v:ious
co:iicc:cc lcvcls
lcu:istic cstiutio: oi co:iicc:cc i:tc:vls io: t|c :uctc: AD is |scc
o: :csuli:g uct|ocs l: t|is :occss, l:gc :uu|c: oi su|sulcs :c c:w:
i:ou t|c cxisti:g sulc I|csc su|sulcs c: |c c:w: wit|out :clccuc:t
;cc| csc i: t|c sulc occu:s cxctl, o:cc o: :ot t ll i: su|sulc, o: wit|
:clccuc:t ;cc| csc i:ou t|c sulc c: occu: uultilc tiucs i: su|
sulc, I|c :csulti:g su|sulcs s|oulc cc| co:ti: t|c suc :uu|c: oi
cscs i: o:cc: to c:su:c sou:c cou:|ilit, I|c k cu:vc is c:w: :c
t|c :uctc: AD is clcultcc io: cc| oi t|c su|sulcs
Rating Models and Validation
110 Guidelines on Credit Risk Management
|:t o! k u:vc wit| iuult:cous o:iicc:cc l:cs t t|c -0 lcvcl io: t|c Dt lxulc
|:t o! I|lc oi Dc: :c lowc: AD liuits io: clcctcc o:iicc:cc lcvcls i: t|c Dt lxulc
Givc: suiiicic:t :uu|c: oi su|sulcs, t|is :occcu:c ,iclcs cstiutcs oi
t|c vc:gc :c v:i:cc oi t|c AD vluc lowcvc:, t|is :occcu:c oitc: o:l,
s|ows : :c:tl, cxct AD :uctc: clcultcc i:ou t|c suc sulc, s
csccill, i: |ouogc:ous sulcs t|c AD iluctutio:s i: su|sulcs :c
:t|c: low lowcvc:, t|c :occcu:c is usciul io: sull sulcs w|c:c t|c co:
iicc:cc |:cs :c vc:, wicc cuc to t|c sull :uu|c: oi cscs
Bayesian Error Rate
As ucsu:c oi cisc:iui:to:, owc:, t|c l,csi: c::o: :tc is ccii:cc s t|c
ui:iuuu c::o: :tc occu::i:g i: t|c sulc cxui:cc ;a c::o: lus b c::o:,
l: t|is tcc|:iquc, t|c ui:iuuu is sc:c| io: uo:g ll cutoii vlucs ;t|c sco:c
vluc |c,o:c w|ic| t|c cc|to: is clssiiicc s |c,
11 min
C
p cC 1 p uC.
I|c a :c b c::o:s io: cc| cutoii vluc :c wcig|tcc wit| t|c sulc`s
cciult :tc o: its coulcuc:t ;1, lo: t|c cxulc wit| 10 :ti:g clsscs
uscc |c:c, t|c t|lc |clow ;c|:t o+, s|ows t|c a :c b c::o:s io: ll cutoii
vlucs s wcll s t|c co::cso:ci:g l,csi: c::o: :tcs io: v:ious vlucs oi
I|c l,csi: c::o: :tc uc:s t|c iollowi:g l: t|c otiuuu usc oi t|c :t
i:g uoccl, :oo:tio: 11 oi ll cscs will still |c uisclssiiicc
Rating Models and Validation
Guidelines on Credit Risk Management 111
|:t o+ I|lc oi l,csi: l::o: ktcs io: clcctcc ulc Dciult ktcs i: t|c Dt lxulc
As t|c t|lc s|ows, o:c scvc:c c:w|c| oi usi:g t|c l,csi: c::o: :tc to
clcultc :ti:g uoccl`s cisc:iui:to:, owc: is its |cv, ccc:cc:cc o: t|c
cciult :tc i: t|c sulc cxui:cc I|c:cio:c, ci:cct cou:iso:s oi l,csi:
c::o: :tc vlucs cc:ivcc i:ou ciiic:c:t sulcs :c :ot ossi|lc l: co:t:st,
t|c ucsu:cs oi cisc:iui:to:, owc: uc:tio:cc t|us i: ;AD, t|c Gi:i oci
iicic:t :c t|c lict: l:ccx, :c i:ccc:cc:t oi t|c cciult :o||ilit, i: t|c
sulc cxui:cc
I|c l,csi: c::o: :tc is li:|cc to : otiuuu cutoii vluc w|ic| ui:i
uizcs t|c totl :uu|c: oi uisclssiiictio:s ;a c::o: lus b c::o:, i: t|c :ti:g
s,stcu lowcvc:, s t|c otiuuu lw,s occu:s w|c: :o cscs :c :c|cctcc
;a 100, b 0,, it |ccoucs clc: t|t t|c l,csi: c::o: :tc |:cl, llows
t|c ciiic:c:titcc sclcctio: oi : otiuuu cutoii vluc io: t|c low cciult :tcs
occu::i:g i: t|c vlictio: oi :ti:g uoccls
Duc to t|c v:ious costs oi a :c b c::o:s, t|c cutoii vluc cctc:ui:cc |,
t|c l,csi: c::o: :tc is :ot otiul i: |usi:css tc:us, t|t is, it cocs :ot ui:
iuizc t|c ovc:ll costs :isi:g i:ou uisclssiiictio:, w|ic| :c usull, su|st:
till, |ig|c: io: a c::o:s t|: io: b c::o:s
lo: t|c cciult :tc .0, t|c l,csi: c::o: :tc cquls cxctl, |li t|c
suu oi a :c b io: t|c oi:t o: t|c ab c::o: cu:vc w|ic| is closcst to oi:t
;0, 0, wit| :cg:c to t|c totl a :c b c::o:s ;ci c|:t o., l: t|is csc, t|c
iollowi:g is lso t:uc ;a :c b c::o:s :c cc:otcc s a :c b,
-1
11 'minc u 'max1 c u 1 'maxF
bad
cum
F
good
cum
'
'1 Pietra Index
lowcvc:, t|is cquivlc:cc oi t|c l,csi: c::o: :tc to t|c lict: l:ccx o:l,
lics w|c:c .0
-!
91
The Bayesian error rate 11 is defined at the beginning of the equation for the case where p 50%. The error values a and b
are related to the frequency distributions of good and bad cases, which are applied in the second to last expression. The last
expression follows from the representation of the Pietra Index as the maximum difference between these frequency distributions.
92
Cf. LEE, Global Performances of Diagnostic Tests.
Rating Models and Validation
112 Guidelines on Credit Risk Management
|:t o. l:tc::cttio: oi t|c l,csi: l::o: ktc s t|c lowcst vc:ll l::o: w|c:c .0
Entropy-Based Measures of Discriminatory Power
I|csc ucsu:cs oi cisc:iui:to:, owc: sscss t|c i:io:utio: gi:cc |, usi:g
t|c :ti:g uoccl l: t|is co:tcxt, i:io:utio: is ccii:cc s vluc w|ic| is
ucsu:|lc i: |solutc tc:us :c w|ic| cquls t|c lcvcl oi |:owlccgc |out
iutu:c cvc:t
lct us ii:st ssuuc t|t t|c vc:gc cciult :o||ilit, oi ll cscs i: t|c scg
uc:t i: qucstio: is u:|:ow: li wc loo| t : i:civicul csc i: t|is scc::io
wit|out |ci:g |lc to cstiutc its c:ccit qulit, usi:g :ti:g uoccls o: ot|c:
ssuutio:s, wc co :ot osscss :, i:io:utio: |out t|c ;gooc o: |c, iutu:c
cciult sttus oi t|c csc I|c uxiuuu i: t|is scc::io is t|c i:io:utio: :
o|sc:vc: gi:s |, witi:g io: t|c iutu:c sttus
li, |owcvc:, t|c vc:gc :o||ilit, oi c:ccit cciult is |:ow:, t|c i:io:
utio: gi:cc |, ctull, o|sc:vi:g t|c iutu:c sttus oi t|c csc is lowc: i: t|is
scc::io cuc to t|c :cviousl, vil|lc i:io:utio:
I|csc co:sicc:tio:s lcc to t|c ccii:itio: oi t|c i:io:utio: c:t:o,
vluc, w|ic| is :c:csc:tcc s iollows io: cic|otouous cvc:ts wit| :o||ilit,
oi occu::c:cc io: t|c 1 cvc:t ;i: t|is csc t|c c:ccit cciult,
H
0
fj log
2
j 1 j log
2
1 jg
H
0
:cic:s to t|c |solutc i:io:utio: vluc w|ic| is :cqui:cc i: o:cc: to cctc:
ui:c t|c iutu:c cciult sttus, o: co:vc:scl, t|c i:io:utio: vluc w|ic| is
gi:cc |, o|sc:vi:g t|c c:ccit cciult:o c:ccit cciult cvc:t I|us c:t:o,
c: lso |c i:tc::ctcc s ucsu:c oi u:cc:ti:t, s to t|c outcouc oi :
cvc:t
H
0
:cc|cs its uxiuuu vluc oi 1 w|c: .0, t|t is, w|c: cciult :c
:o:cciult :c cqull, :o||lc H
0
cquls zc:o w|c: j t|cs t|c vluc 0 o: 1,
t|t is, t|c iutu:c cciult sttus is l:cc, |:ow: wit| cc:ti:t, i: cv:cc
o:citio:l c:t:o, is ccii:cc wit| co:citio:l :o||ilitics jjc i:stcc oi
|solutc :o||ilitics j, t|c co:citio:l :o||ilitics :c |scc o: co:citio: c
Rating Models and Validation
Guidelines on Credit Risk Management 113
lo: t|c u:osc oi vlicti:g :ti:g uoccls, t|c co:citio: i: t|c ccii:itio: oi
co:citio:l c:t:o, is t|c clssiiictio: i: :ti:g clss c :c cciult cvc:t to |c
ccictcc 1 lo: cc| :ti:g clss c, t|c co:citio:l c:t:o, is /
c
/
c
fj1jc log
2
j1jc 1 j1jc log
2
1 j1jcg.
I|c co:citio:l c:t:o, /
c
oi :ti:g clss t|us co::cso:cs to t|c u:cc:
ti:t, :cui:i:g wit| :cg:c to t|c iutu:c cciult sttus itc: csc is ssig:cc
to :ti:g clss Ac:oss ll :ti:g clsscs i: uoccl, t|c co:citio:l c:t:o, H
1
;vc:gcc usi:g t|c o|sc:vcc i:cquc:cics oi t|c i:civicul :ti:g clsscs` j
c
, is
ccii:cc s
H
1
c
j
c
/
c
.
I|c vc:gc co:citio:l c:t:o, l1 co::cso:cs to t|c u:cc:ti:t, :cui:
i:g wit| :cg:c to t|c iutu:c cciult sttus itc: lictio: oi t|c :ti:g uoccl
Dsi:g t|c c:t:o, H
0
, w|ic| is vil|lc wit|out l,i:g t|c :ti:g uoccl ii
t|c vc:gc cciult :o||ilit, oi t|c sulc is |:ow:, it is ossi|lc to ccii:c
:cltivc ucsu:c oi t|c i:io:utio: gi:cc cuc to t|c :ti:g uoccl I|c co:
citio:l i:io:utio: c:t:o, :tio ;llk, is ccii:cc s
-!
C111
H
0
H
1
H
0
1
H
1
H
0
.
I|c vluc llk c: |c i:tc::ctcc s iollows
li :o ccitio:l i:io:utio: is gi:cc |, l,i:g t|c :ti:g uoccl, H
1
H
0
:c C111 0
li t|c :ti:g uoccl is iccl :c :o u:cc:ti:t, :cui:s :cg:ci:g t|c cciult
sttus itc: t|c uoccl is licc, H
1
0 :c C111 1
I|c |ig|c: t|c llk vluc is, t|c uo:c i:io:utio: :cg:ci:g t|c iutu:c
cciult sttus is gi:cc i:ou t|c :ti:g s,stcu
lowcvc:, it s|oulc |c :otcc t|t i:io:utio: o: t|c :oc:tics oi t|c :ti:g
uoccl is lost i: t|c clcultio: oi llk, s is t|c csc wit| AD :c t|c ot|c:
o:cciuc:sio:l ucsu:cs oi cisc:iui:to:, owc: As : i:civicul i:cicto:,
t|c:cio:c, llk |s o:l, liuitcc uc:i:g i: t|c sscssuc:t oi :ti:g uoccl
|:t oo l:t:o,lscc \csu:cs oi Disc:iui:to:, lowc: io: t|c Dt lxulc
93
The difference H
0
H
1
is also referred to as the Kullback-Leibler distance. Therefore, CIER is a standardized Kullback-
Leibler distance.
Rating Models and Validation
114 Guidelines on Credit Risk Management
I|c t|lc |clow ;c|:t o, s|ows cou:iso: oi Gi:i ociiicic:t vlucs
:c llk cisc:iui:to:, owc: i:cicto:s i:ou stuc, oi :ti:g uoccls io:
Auc:ic: co:o:tcs
|:t o Gi:i ociiicic:t :c llk Vlucs i:ou tuc, oi Auc:ic: o:o:tcs
-+
6.2.2 Back-Testing the Calibration
I|c ssig:uc:t oi cciult :o||ilitics to :ti:g uoccl`s outut is :cic::cc to
s cli|:tio: I|c qulit, oi cli|:tio: ccc:cs o: t|c ccg:cc to w|ic| t|c
cciult :o||ilitics :ccictcc |, t|c :ti:g uoccl utc| t|c cciult :tcs
ctull, :clizcc I|c:cio:c, :cvicwi:g t|c cli|:tio: oi :ti:g uoccl is i:c
quc:tl, :cic::cc to s |c|tcsti:g
I|c |sic ct uscc io: |c|tcsti:g :c t|c cciult :o||ilitics io:ccst
ovc: :ti:g clss io: scciiic tiuc |o:izo: ;usull, 1! uo:t|s,, t|c :uu|c:
oi cscs ssig:cc to t|c :cscctivc :ti:g clss |, t|c uoccl, :c t|c cciult st
tus oi t|osc cscs o:cc t|c io:ccsti:g c:ioc |s clscc, st:ti:g i:ou t|c tiuc
oi :ti:g ;ic usull, 1! uo:t|s itc: t|c :ti:g ws ssig:cc, li|:tio:
i:volvcs ssig:i:g io:ccst cciult :o||ilitics to t|c i:civicul :ti:g clsscs
;ci scctio: .!, l: t|is :occss, it is lso ossi|lc to usc lo:gc: io:ccsti:g |o:i
zo:s t|: t|c 1!uo:t| |o:izo: :cqui:cc oi lkl |:|s, t|csc ot|c: tiuc |o:i
zo:s lso |vc to u:cc:go |c|tcsti:g
I|c :csults oi v:ious scguc:tscciiic :ti:g :occcu:cs :c i:cquc:tl,
ccictcc o: u:iio:u ustc: sclc oi cciult :o||ilitics
l: t|c cou:sc oi qu:tittivc vlictio:, sig:iiic:t ciiic:c:ccs u, |c icc:
tiiicc |ctwcc: t|c cciult :tcs o: t|c ustc: sclc :c t|c cciult :tcs ctull,
:clizcc io: i:civicul :ti:g clsscs i: scguc:tscciiic :ti:g :occcu:c l:
o:cc: to co::cct t|csc ccvitio:s, two ciiic:c:t :oc|cs :c ossi|lc
l: iixcc ustc: sclc, t|c :cccii:cc cciult :o||ilitics :c :ot c|:gcc,
i:stcc, o:l, t|c ssig:uc:t oi :csults i:ou t|c :ti:g :occcu:c u:cc:
:cvicw to :ti:g clsscs o: t|c ustc: sclc is c|ustcc
l: v:i|lc ustc: sclc, t|c :cccii:cc cciult :o||ilitics :c c|:gcc,
|ut t|c ssig:uc:t oi :ti:g :csults i:ou t|c :ti:g :occcu:c u:cc: :cvicw
to :ti:g clsscs o: t|c ustc: sclc is :ot c|ustcc
As :, c|:gcs to t|c ustc: sclc will iicct ll oi t|c :ti:g :occcu:cs
uscc i: |:| i:cluci:g t|osc io: w|ic| :o ;o: o:l, ui:o:, c::o:s i: cli|:
tio: |vc |cc: icc:tiiicc iixcc ustc: sclcs :c gc:c:ll, :cic:|lc I|is is
csccill, t:uc i: cscs w|c:c t|c cciult :o||ilit, oi :ti:g clsscs sc:vcs s
t|c |sis io: :is||scc :ici:g, w|ic| woulc |c su||cct to i:cquc:t c|:gcs ii
v:i|lc ustc: sclc wc:c uscc
94
Cf. KEENAN/SOBEHART, Performance Measures.
Rating Models and Validation
Guidelines on Credit Risk Management 115
lowcvc:, c|:gcs to t|c ustc: sclc :c still c:uissi|lc uc| c|:gcs
uig|t |c :cccss:, i: cscs w|c:c :cw :ti:g :occcu:cs :c to |c i:tcg:tcc
o: ii:c::oug|c: clssiiictio:s :c :cqui:cc io: cciult :o||ilitics i: cc:ti:
vluc ::gcs lowcvc:, it is lso :cccss:, to c:su:c t|t ct |isto:, :cco:cs
lw,s i:clucc :ti:g clss`s cciult :o||ilit,, t|c :ti:g :csult itscli, :c
t|c ccou:,i:g cciult :o||ilit, i: ccitio: to t|c :ti:g clss
|:t o s|ows t|c io:ccst :c :clizcc cciult :tcs i: ou: cxulc wit|
tc: :ti:g clsscs I|c vc:gc :clizcc cciult :tc io: t|c ovc:ll sulc is
1!, w|c:cs t|c io:ccst |scc o: t|c i:cquc:c, wit| w|ic| t|c i:civicul
:ti:g clsscs wc:c ssig:cc s wcll s t|ci: :cscctivc cciult :o||ilitics ws
10 I|c:cio:c, t|is :ti:g uoccl u:cc:cstiutcc t|c ovc:ll cciult :is|
|:t o ou:iso: oi lo:ccst :c kclizcc Dciult ktcs i: t|c Dt lxulc
Brier Score
I|c vc:gc quc:tic ccvitio: oi t|c cciult :tc io:ccst io: cc| csc oi t|c
sulc cxui:cc i:ou t|c :tc :clizcc i: t|t csc ;1 io: cciult, 0 io: :o
cciult, is |:ow: s t|c l:ic: co:c
-.
1o
1
`
`
i1
j
)oicco:t
i
y
i
2
where y
i
1 for default in n
0 for no default in n
_
l: t|c csc cxui:cc |c:c, w|ic| is civiccc i:to :ti:g clsscs, t|c l:ic:
co:c c: lso |c :c:csc:tcc s t|c totl io: ll :ti:g clsscs c
1o
1
`
1
c1
`
/
j
o/:ci.cd
c
1 j
)oicco:t
c
2
1 j
o/:ci.cd
c
j
)oicco:t
c
2
_
l: t|c cqutio: |ovc, `
c
cc:otcs t|c :uu|c: oi cscs :tcc i: :ti:g clss c,
w|ilc j
o/:ci.cd
:c j
)oicco:t
:cic: to t|c :clizcc cciult :tc :c t|c io:ccst
cciult :tc ;|ot| io: :ti:g clss c, I|c ii:st tc:u i: t|c suu :cilccts t|c
cciults i: clss c, :c t|c scco:c tc:u s|ows t|c :o:cciultcc cscs I|is
l:ic: co:c cqutio: c: |c :cw:ittc: s iollows
1o
1
`
1
c1
`
c
j
o/:ci.cd
c
1 j
o/:ci.cd
c
j
)oicco:t
c
j
o/:ci.cd
c
2
_
95
Cf. BRIER, G. W., Brier-Score.
Rating Models and Validation
116 Guidelines on Credit Risk Management
I|c lowc: t|c l:ic: co:c is, t|c |cttc: t|c cli|:tio: oi t|c :ti:g uoccl
is lowcvc:, it is lso ossi|lc to civicc t|c l:ic: co:c i:to t|:cc couo:c:ts,
o:l, o:c oi w|ic| is ci:cctl, li:|cc to t|c ccvitio:s oi :cl cciult :tcs i:ou
t|c co::cso:ci:g io:ccsts
-o
I|c cv:tgc oi t|is civisio: is t|t t|c cssc:til
:oc:tics oi t|c l:ic: co:c c: |c sc:tcc
1o j
o/:ci.cd
1 j
o/:ci.cd
..
Uncertainty,variation
1
`
1
c1
`
c
j
)oicco:t
c
j
o/:ci.cd
c
2
_
..
Calibration,reliability
1
`
1
c1
`
c
j
o/:ci.cd
j
o/:ci.cd
c
_ _
2
_ _
..
Resolution
I|c ii:st tc:u 1o
ic)
j
o/:ci.cd
1 j
o/:ci.cd
ccsc:i|cs t|c v:i:cc oi t|c
cciult :tc o|sc:vcc ovc: t|c c:ti:c sulc j
o/:ci.cd
I|is vluc is i:ccc:c
c:t oi t|c :ti:g :occcu:c`s cli|:tio: :c ccc:cs o:l, o: t|c o|sc:vcc su
lc itscli lt :c:csc:ts t|c ui:iuuu l:ic: co:c tti:|lc io: t|is sulc wit|
c:icctl, cli|:tcc |ut lso trivial rating model w|ic| io:ccsts t|c o|sc:vcc
cciult :tc :cciscl, io: cc| csc |ut o:l, cou:iscs o:c :ti:g clss low
cvc:, t|c t:ivil :ti:g uoccl cocs :ot ciiic:c:titc |ctwcc: uo:c o: lcss gooc
cscs :c is t|c:cio:c u:suit|lc s :ti:g uoccl u:cc: t|c :cqui:cuc:ts oi t|c
lkl :oc|
I|c scco:c tc:u
1
`
1
c1
`
c
j
)oicco:t
c
j
o/:ci.cd
c
_ _
2
_ _
:c:csc:ts t|c vc:gc quc:tic ccvitio: oi io:ccst :c :clizcc cciult :tcs
i: t|c :ti:g clsscs Awcllcli|:tcc :ti:g uoccl will s|ow lowc: vlucs io:
t|is tc:u t|: oo:l, cli|:tcc :ti:g uoccl I|c vluc itscli is t|us lso
:cic::cc to s t|c cli|:tio:
I|c t|i:c tc:u
1
`
1
c/1
`
c
j
)oicco:t
j
o/:ci.cd
c
_ _
2
_ _
ccsc:i|cs t|c vc:gc quc:tic ccvitio: oi o|sc:vcc cciult :tcs i: i:civicul
:ti:g clsscs i:ou t|c cciult :tc o|sc:vcc i: t|c ovc:ll sulc I|is vluc is
:cic::cc to s :csolutio: W|ilc t|c :csolutio: oi t|c t:ivil :ti:g uoccl is
zc:o, it is :ot cqul to zc:o i: cisc:iui:ti:g :ti:g s,stcus l: gc:c:l, t|c :cs
olutio: oi :ti:g uoccl :iscs s :ti:g clsscs wit| clc:l, ciiic:c:titcc
o|sc:vcc cciult :o||ilitics :c cccc kcsolutio: is t|us li:|cc to t|c cis
c:iui:to:, owc: oi :ti:g uoccl
I|c ciiic:c:t sig:s :cccci:g t|c cli|:tio: :c :csolutio: tc:us u|c it
uo:c ciiiicult to i:tc::ct t|c l:ic: co:c s : i:civicul vluc io: t|c u:osc
oi sscssi:g t|c clssiiictio: ccu:c, oi :ti:g uoccl`s cli|:tio: l: cci
tio:, t|c :uuc:icl vlucs oi t|c cli|:tio: :c :csolutio: tc:us :c gc:c:ll,
i: lowc: t|: t|c v:i:cc I|c t|lc |clow ;c|:t o-, s|ows t|c vlucs oi t|c
l:ic: co:c :c its couo:c:ts io: t|c cxulc uscc |c:c
96
Cf. MURPHY, A. H., Journal of Applied Meteorology.
Rating Models and Validation
Guidelines on Credit Risk Management 117
|:t o- lcultio: oi l:ic: co:c io: t|c Dt lxulc
l: :cticc, st:c:cizcc ucsu:c |:ow: s t|c l:ic: |ill co:c ;l, is
oitc: uscc i:stcc oi t|c l:ic: co:c I|is ucsu:c sclcs t|c l:ic: co:c to t|c
v:i:cc tc:u
1oo 1
1o
j
o/:ci.cd
1 j
o/:ci.cd
.
l: t|c t:ivil, iccl uoccl, t|c l:ic: |ill co:c t|cs t|c vluc zc:o lo: t|c
cxulc |ovc, t|c :csulti:g vluc is l ~ +0+
Reliability Diagrams
Accitio:l i:io:utio: o: t|c qulit, oi cli|:tio: io: :ti:g uoccls c: |c
cc:ivcc i:ou t|c :cli|ilit, cig:u, i: w|ic| t|c o|sc:vcc cciult :tcs :c
lottcc gi:st t|c io:ccst :tc i: cc| :ti:g clss I|c :csulti:g cu:vc is oitc:
:cic::cc to s t|c cli|:tio: cu:vc
|:t 0 kcli|ilit, Dig:u io: t|c Dt lxulc
Rating Models and Validation
118 Guidelines on Credit Risk Management
|:t 0 s|ows t|c :cli|ilit, cig:u io: t|c cxulc uscc |c:c l: t|c
illust:tio:, cou|lc log:it|uic :c:csc:ttio: ws sclcctcc |ccusc t|c
cciult :o||ilitics :c vc:, closc togct|c: i: t|c gooc :ti:g clsscs i: :tic
ul: :otc t|t t|c oi:t io: :ti:g clss 1 is uissi:g i: t|is g:| I|is is
|ccusc :o cciults wc:c o|sc:vcc i: t|t clss
I|c oi:ts oi wcllcli|:tcc s,stcu will ill closc to t|c cigo:l i: t|c
:cli|ilit, cig:u l: : iccl s,stcu, ll oi t|c oi:ts woulc lic ci:cctl, o: t|c
cigo:l I|c cli|:tio: tc:u oi t|c l:ic: co:c :c:csc:ts t|c vc:gc
;wcig|tcc wit| t|c :uu|c:s oi cscs i: cc| :ti:g clss, squ:cc ccvitio:
oi oi:ts o: t|c cli|:tio: cu:vc i:ou t|c cigo:l I|is vluc s|oulc |c s
low s ossi|lc
I|c :csolutio: oi :ti:g uoccl is i:cictcc |, t|c vc:gc ;wcig|tcc wit|
t|c :uu|c:s oi cscs i: t|c i:civicul :ti:g clsscs, squ:cc ccvitio: oi oi:ts
i: t|c :cli|ilit, cig:u i:ou t|c |:o|c: li:c, w|ic| :c:csc:ts t|c cciult :tc
o|sc:vcc i: t|c sulc I|is vluc s|oulc |c s |ig| s ossi|lc, w|ic| uc:s
t|t t|c cli|:tio: cu:vc s|oulc |c s stcc s ossi|lc lowcvc:, t|c stcc:css
oi t|c cli|:tio: cu:vc is :iu:il, cctc:ui:cc |, t|c :ti:g uoccl`s cisc:iu
i:to:, owc: :c is i:ccc:cc:t oi t|c ccu:c, oi cciult :tc cstiutcs
A: iccl t:ivil :ti:g s,stcu wit| o:l, o:c :ti:g clss woulc |c :c:c
sc:tcc i: t|c :cli|ilit, cig:u s : isoltcc oi:t loctcc t t|c i:tc:scctio:
oi t|c cigo:l :c t|c cciult :o||ilit, oi t|c sulc
li|c cisc:iui:to:, owc: ucsu:cs, o:cciuc:sio:l i:cicto:s io: cli
|:tio: :c :csolutio: c: lso |c ccii:cc s st:c:cizcc ucsu:cs oi t|c :c
|ctwcc: t|c cli|:tio: cu:vc :c t|c cigo:l o: t|c sulc cciult :tc
-
Checking the Significance of Deviations in the Default Rate
l: lig|t oi t|c ict t|t :clizcc cciult :tcs :c su||cct to sttisticl iluctutio:s,
it is :cccss:, to ccvclo i:cicto:s to s|ow |ow wcll t|c :ti:g uoccl csti
utcs t|c :uctc: lD l: gc:c:l, two :oc|cs c: |c t|c:
Assuutio: oi u:co::cltcc cciult cvc:ts
o:sicc:tio: oi cciult co::cltio:
lui:icl stucics s|ow t|t cciult cvc:ts :c gc:c:ll, :ot u:co::cltcc
I,icl vlucs oi cciult co::cltio:s ::gc |ctwcc: 0. :c ! Dciult co:
:cltio:s w|ic| :c :ot cqul to zc:o |vc t|c ciicct oi st:c:gt|c:i:g iluctu
tio:s i: cciult :o||ilitics I|c tolc::cc ::gcs io: t|c ccvitio: oi :clizcc
cciult :tcs i:ou cstiutcc vlucs u, t|c:cio:c |c su|st:till, l:gc: w|c:
cciult co::cltio:s :c t|c: i:to ccou:t l: o:cc: to c:su:c co:sc:vtivc csti
utcs, t|c:cio:c, it is :cccss:, to :cvicw t|c cli|:tio: u:cc: t|c i:itil
ssuutio: oi u:co::cltcc cciult cvc:ts
I|c sttisticl tcst uscc |c:c c|cc|s t|c :ull |,ot|csis I|c io:ccst cciult
:o||ilit, i: :ti:g clss is co::cct gi:st t|c ltc::tivc |,ot|csis I|c
io:ccst cciult :o||ilit, is i:co::cct usi:g t|c ct vil|lc io: |c|tcsti:g
I|is tcst c: |c o:csiccc ;c|cc|i:g o:l, io: sig:iiic:t ovc::u:s oi t|c io:ccst
cciult :tc, o: twosiccc ;c|cc|i:g io: sig:iiic:t ovc::u:s :c u:cc::u:s oi
t|c io:ccst cciult :o||ilit,, l:ou u:gcuc:t st:coi:t, |ot| sig:iii
c:t u:cc:cstiutcs :c ovc:cstiutcs oi :is| :c :clcv:t A o:csiccc tcst c:
97
See also HASTIE/TIBSHIRANI/FRIEDMAN, Elements of statistical learning.
Rating Models and Validation
Guidelines on Credit Risk Management 119
lso |c uscc to c|cc| io: :is| u:cc:cstiutcs :csiccc :c twosiccc tcsts
c: lso |c co:vc:tcc i:to o:c :ot|c:
-
Calibration Test using Standard Normal Distribution
:c siulc tcst io: t|c cli|:tio: oi cciult :tcs u:cc: t|c ssuutio: oi
u:co::cltcc cciult cvc:ts uscs st:c:c :o:ul cist:i|utio:
--
l: t|c io:uuls
|clow,
1
cc:otcs t|c i:vc:sc cuuultivc cist:i|utio: iu:ctio: io: t|c st:c:c
:o:ul cist:i|utio:, `
c
st:cs io: t|c :uu|c: oi cscs i: :ti:g clss c, :c j
c
:cic:s to t|c cciult :tc
;o:csiccc tcst, li
_
j
o/:ci.cd
c
j
)oicco:t
c
_
1
j
)oicco:t
c
1 j
)oicco:t
c
`
c
.
t|c cciult :tc i: clss c is sig:iiic:tl, u:cc:cstiutcc t t|c co:iicc:cc
lcvcl
;o:csiccc tcst, li
_
j
jioqio:c
c
j
o/:ci.cd
c
_
1
j
)oicco:t
c
1 j
)oicco:t
c
`
c
.
t|c cciult :tc i: clss c is sig:iiic:tl, ovc:cstiutcc t t|c co:iicc:cc lcvcl
;twosiccc tcst, li
j
o/:ci.cd
c
j
)oicco:t
c
1
1
2
_ _
j
)oicco:t
c
1 j
)oicco:t
c
`
c
.
t|c cciult :tc i: clss c is sig:iiic:tl, uiscstiutcc t co:iicc:cc lcvcl
I|c t|lc |clow ;c|:t 1, s|ows t|c :csults oi t|c o:csiccc tcst io: :is|
u:cc:cstiutcs ig:iiic:t ovc::u:s oi t|c io:ccst cciult :tcs :c s|ccc i:
g:, io: i:civicul sig:iiic:cc lcvcls I|c cciult :tcs i: clsscs . to wc:c
sig:iiic:tl, u:cc:cstiutcc t lcvcl oi -., w|ilc t|c cciult :tcs i: clsscs
- :c 10 wc:c o:l, u:cc:cstiutcc t t|c -0 lcvcl I|c vc:gc cciult :o|
|ilit, oi t|c ovc:ll sulc ws u:cc:cstiutcc cvc: t t|c lcvcl oi ---
100
I|is s|ows t|t |ig|l, sig:iiic:t uiscstiutc oi t|c vc:gc cciult :tc
io: t|c c:ti:c sulc c: :isc cvc: ii t|c cciult :tcs i: t|c i:civicul clsscs
:cui: wit|i: t|ci: tolc::cc ::gcs I|is is cuc to t|c i:clusio: oi t|c :uu|c:
oi cscs ` i: t|c cc:oui:to: oi t|c tcst sttistic, t|is :uu|c: :ccuccs t|c tcst
sttistic w|c: ll :ti:g clsscs :c cou|i:cc, t|us u|i:g t|c tcst uo:c sc:si
tivc
98
The limit of a two-sided test at level q is equal to the limit of a one-sided test at the level ! 1. If overruns/underruns are
also indicated in the two-sided test by means of the plus/minus signs for differences in default rates, the significance levels will
be identical.
99
Cf. CANTOR, R./FALKENSTEIN, E., Testing for rating consistencies in annual default rates.
100
The higher the significance level q is, the more statistically certain the statement is; in this case, the statement is the rejection of
the null hypothesis asserting that PD is estimated correctly. The value (1-q) indicates the probability that this rejection of the
null hypothesis is incorrect, that is, the probability that an underestimate of the default probability is identified incorrectly.
Rating Models and Validation
120 Guidelines on Credit Risk Management
|:t 1 lcc:tiiictio: oi ig:iiic:t Dcvitio:s i: li|:tio: i: t|c Dt lxulc
;Icst wit| t:c:c :o:ul Dist:i|utio:,
lo: t|c u:osc oi i:tc::cti:g co:iicc:cc lcvcls, t:iiic lig|ts :oc|
|s |cc: :ooscc io: :cticc i: Gc:u:,
101
l: t|is :oc|, ccvitio:s oi
:clizcc :c io:ccst cciult :tcs |clow co:iicc:cc lcvcl oi -. s|oulc
:ot |c :cg:ccc s sig:iiic:t ;g:cc: ::gc, Dcvitio:s t co:iicc:cc lcvcl
oi t lcst --- :c t|c: co:sicc:cc sig:iiic:t :c s|oulc ccii:itcl, |c co:
:cctcc ;:cc ::gc, Dcvitio:s w|ic| :c sig:iiic:t t co:iicc:cc lcvcls
|ctwcc: -. :c --- u, :ccc to |c co::cctcc ;,cllow ::gc,
lo: t|c cxulc |ovc, t|is uc:s t|t t|c ovc:ll cciult :tc w|ic| ws
u:cc:cstiutcc |, t|c :ti:g uoccl s|oulc |c co::cctcc uw:c, :cic:|l,
|, u|i:g t|c :o:itc c|ustuc:ts i: :ti:g clsscs . to
Binomial Calibration Test
I|c tcst usi:g :o:ul cist:i|utio: ;ccsc:i|cc |ovc, is gc:c:liztio: oi t|c
|i:ouil tcst io: i:cquc:cics oi u:co::cltcc |i::, cvc:ts I|c |i:ouil tcst
is ccsc:i|cc i: cctil |clow
lo: low cciult :o||ilitics :c low :uu|c:s oi cscs i: t|c i:civicul
:ti:g clsscs, t|c :c:cquisitcs io: usi:g :o:ul cist:i|utio: :c :ot lw,s
uct I|c t|lc |clow ;c|:t !, lists t|c ui:iuuu :uu|c: oi cscs :cqui:cc
io: sou:c :oxiutio: oi tcst vlucs wit| st:c:c :o:ul cist:i|utio:
w|c: tcsti:g v:ious cciult :o||ilitics
10!
li i:civicul clsscs co:ti: icwc: cscs t|: t|c ui:iuuu :uu|c: i:ci
ctcc, t|c |i:ouil tcst s|oulc |c c::icc out l: t|c io:uul |clow, `
c
cc:otcs t|c :uu|c: oi cciults o|sc:vcc i: clss c, :c `
c
:cic:s to t|c :uu|c:
oi cscs i: clss c uuutio: is c:io:ucc io: ll cciults i: clss c
;o:csiccc tcst, li
c
i0
`
c
i
_ _
j
)oicco:t
c
i
1 j
)oicco:t
`
c
i
.
t|c cciult :tc i: clss c is sig:iiic:tl, u:cc:cstiutcc t co:iicc:cc lcvcl
10!
101
Cf. TASCHE, D., A traffic lights approach to PD validation.
102
The strict condition for the application of standard normal distribution as an approximation of binomial distribution is
`
j
1j 9, where N is the number of cases in the rating class examined and j is the forecast default probability (cf.
SACHS, L., Angewandte Statistik, p. 283).
103
This is equivalent to the statement that the probability of occurrence 1i < `
/
jj
jioqio:c
/
; `
/
(assumed to be binomially
distributed) of a maximum of `
c
defaults among the `
c
cases in class c must not be greater than q.
Rating Models and Validation
Guidelines on Credit Risk Management 121
|:t ! I|co:cticl \i:iuuu :uu|c: oi scs io: t|c :o:ul Icst |scc o: V:ious Dciult ktcs
I|c t|lc |clow ;c|:t !, s|ows t|c :csults oi t|c |i:ouil tcst io: sig:ii
ic:t u:cc:cstiutcs oi t|c cciult :tc lt is i:cccc co:sicuous t|t t|c :csults
l:gcl, utc| t|osc oi t|c tcst usi:g :o:ul cist:i|utio: ccsitc t|c ict t|t
scvc:l clsscs cic :ot co:ti: t|c :cqui:cc ui:iuuu :uu|c: oi cscs
|:t ! lcc:tiiictio: oi ig:iiic:t Dcvitio:s i: li|:tio: io: t|c Dt lxulc ;li:ouil Icst,
:c i:tc:csti:g ccvitio: c:s i: :ti:g clss 1, w|c:c t|c |i:ouil tcst
,iclcs sig:iiic:t u:cc:cstiutc t t|c -. lcvcl io: : cstiutcc cciult :o|
|ilit, oi 00. :c :o o|sc:vcc cciults l: t|is csc, t|c |i:ouil tcst cocs
:ot ,iclc :cli|lc :csults, s t|c iollowi:g l:cc, lics w|c: :o cciults :c
o|sc:vcc
1 i 0jj
jioqio:c
1
; `
1
_
1 j
jioqio:c
1
`
1
90%.
l: gc:c:l, t|c tcst usi:g :o:ul cist:i|utio: is istc: :c csic: to c:io:u,
:c it ,iclcs usciul :csults cvc: io: sull sulcs :c low cciult :tcs I|is tcst
is t|us :cic:|lc to t|c |i:ouil tcst cvc: ii t|c ut|cuticl :c:cquisitcs io:
its lictio: :c :ot lw,s uct lowcvc:, it is iuo:t:t to |c: i: ui:c t|t
t|c |i:ouil tcst :c its gc:c:liztio: usi:g :o:ul cist:i|utio: :c |scc o:
t|c ssuutio: oi u:co::cltcc cciults A tcst :occcu:c w|ic| t|cs cciult
co::cltio:s i:to ccou:t is :csc:tcc |clow
Calibration Test Procedure Based on Default Correlation
I|c ssuutio: oi u:co::cltcc cciults gc:c:ll, ,iclcs : ovc:cstiutc oi t|c
sig:iiic:cc oi ccvitio:s i: t|c :clizcc cciult :tc i:ou t|c io:ccst :tc I|is
is csccill, t:uc oi :is| u:cc:cstiutcs, t|t is, cscs i: w|ic| t|c :clizcc
cciult :tc is |ig|c: t|: t|c io:ccst :tc l:ou co:sc:vtivc :is| sscssuc:t
st:coi:t, ovc:cstiuti:g sig:iiic:cc is :ot c:iticl i: t|c csc oi :is| u:cc:
Rating Models and Validation
122 Guidelines on Credit Risk Management
cstiutcs, w|ic| uc:s t|t it is c:ti:cl, ossi|lc to oc:tc u:cc: t|c ssuu
tio: oi u:co::cltcc cciults l: :, csc, |owcvc:, c:sistc:t ovc:cstiutcs oi
sig:iiic:cc will lcc to uo:c i:cquc:t :ccli|:tio: oi t|c :ti:g uoccl, w|ic|
c: |vc :cgtivc ciiccts o: t|c uoccl`s st|ilit, ovc: tiuc lt is t|c:cio:c :cc
css:, to cctc:ui:c t lcst t|c :oxiutc cxtc:t to w|ic| cciult co::cl
tio:s i:iluc:cc lD cstiutcs
Dciult co::cltio:s c: |c uocclcc o: t|c |sis oi t|c ccc:cc:cc oi
cciult cvc:ts o: couuo: :c i:civicul ::cou icto:s
10+
lo: co::cltcc
cciults, t|is uoccl lso u|cs it ossi|lc to cc:ivc liuits io: sscssi:g ccvi
tio:s i: t|c :clizcc cciult :tc i:ou its io:ccst s sig:iiic:t t cc:ti: co:
iicc:cc lcvcls
l: t|c :oxiutio: io:uul |clow, cc:otcs t|c co:iicc:cc lcvcl ;cg
-. o: ---,, `
c
t|c :uu|c: oi cscs o|sc:vcc c: :ti:g clss, j
/
t|c cciult
:tc c: :ti:g clss, t|c cuuultivc st:c:c :o:ul cist:i|utio:, c t|c :o|
|ilit, cc:sit, iu:ctio: oi t|c st:c:c :o:ul cist:i|utio:, :c , t|c cciult
co::cltio:
;o:csiccc tcst, li
j
o/:ci.cd
c
Q
1
2`
c
2Q 1
Q1Q
c
,
1
1t
1,
p
_
_ _
12,
1
1t
,
p
,1,
p
_ _
_
_
_
_
_
_
_
_
t|c cciult :tc i: clss c is sig:iiic:tl, u:cc:cstiutcc t t|c co:iicc:cc lcvcl
l: t|is co:tcxt, t
1
j
)oicco:t
c
; Q
1
_
,
p
1
t
1,
p
_
I|c tcst |ovc c: |c uscc to c|cc| io: ovc:cstiutcs oi t|c cciult :tc i:
i:civicul :ti:g clsscs s wcll s t|c ovc:ll sulc
10.
I|c t|lc |clow ;c|:t +, cou:cs t|c :csults oi t|c cli|:tio: tcst u:cc:
t|c ssuutio: oi u:co::cltcc cciults wit| t|c :csults |scc o: cciult co:
:cltio: oi 001 A ccvitio: i: t|c :clizcc cciult :tc i:ou t|c io:ccst :tc is
co:sicc:cc sig:iiic:t w|c:cvc: t|c :clizcc cciult :tc cxccccs t|c uc:
liuit i:cictcc io: t|c :cscctivc co:iicc:cc lcvcl
|:t + lcc:tiiictio: oi ig:iiic:t Dcvitio:s i: li|:tio: i: t|c Dt lxulc
;ou:iso: oi Icsts io: D:co::cltcc scs :c kcsults io: Dciult o::cltio: oi 001,
104
Vasicek One-Factor Model, cf. e.g. TASCHE, D, A traffic lights approach to PD validation.
105
In this context, it is necessary to note that crossing the boundary to uncorrelated defaults , ! 0 is not possible in the approx-
imation shown. For this reason, the procedure will yield excessively high values in the upper default probability limit for very
low default correlations (< 0.005).
Rating Models and Validation
Guidelines on Credit Risk Management 123
I|c ssuutio: oi wc| cciult co::cltio: is l:cc, suiiicic:t i: t|is cxu
lc to :ciutc t|c ssuutio: oi uiscli|:tio: io: t|c uoccl t |ot| t|c -. s
wcll s t|c --- lcvcl D:cc: t|c ssuutio: oi u:co::cltcc cciults, o: t|c
ot|c: |:c, ccvitio:s i: t|c ,cllow ::gc c: i: :ti:g clsscs . to , :c
ccvitio: i: t|c :cc ::gc ;s ccii:cc i: t|c io:cuc:tio:cc t:iiic lig|ts
:oc|, is s|ow: io: t|c ovc:ll sulc
I|c t|lc |clow ;c|:t ., i:cictcs t|c uc: liuits i: t|c o:csiccc tcst t
-. sig:iiic:cc lcvcl io: v:ious cciult co::cltio:s l: t|is co:tcxt, it is
iuo:t:t to :otc t|t wit| |ig|c: cciult co::cltio:s t|c liuits ,iclccc |, t|is
tcst |ccouc lowc: io: :ti:g clss 1 cuc to t|c vc:, low cciult :tc :c t|c
sull :uu|c: oi cscs i: t|t clss
A :uu|c: oi ccitio:l tcst :occcu:cs cxist io: t|c vlictio: oi :ti:g
uoccl cli|:tio:s I|c :oc| :csc:tcc |c:c is cisti:guis|cc |, its :cltivc
siulicit, lcsiccs :l,ticl uoccls,
10o
siuultio: uoccls c: lso |c iulc
uc:tcc, |ut io: cciult co::cltio:s |ctwcc: 001 :c 00. t|c, s|oulc ,iclc
siuil: :csults to t|osc :csc:tcc |c:c I|c |solutc uou:t oi t|c cciult
co::cltio:s t|cusclvcs is lso su||cct to cstiutio:, |owcvc:, t|is will :ot
|c ciscusscc i: iu:t|c: cctil t t|is oi:t
10
lowcvc:, t|c ssuucc co::cl
tio:`s sig:iiic:t i:iluc:cc o: t|c vlictio: :csult s|oulc sc:vc to illust:tc t|t
it is :cccss:, to cctc:ui:c t|is :uctc: s ccu:tcl, s ossi|lc i: o:cc: to
c:|lc uc:i:giul sttcuc:ts :cg:ci:g t|c qulit, oi cli|:tio:
|:t . Dc: Dciult ktc liuits t t|c -. o:iicc:cc lcvcl io: V:ious Dciult o::cltio:s i: t|c lxulc
6.2.3 Back-Testing Transition Matrices
l: gc:c:l, t|c uct|ocs licc io: cciult :o||ilitics c: lso |c uscc to
|c|tcst t|c t::sitio: ut:ix lowcvc:, t|c:c :c two cssc:til ciiic:c:ccs
i: t|is :occcu:c
lc|tcsti:g t::sitio: ut:iccs i:volvcs siuult:cous tcsti:g oi i: l:gc:
:uu|c: oi :o||ilitics I|is iuoscs su|st:till, |ig|c: ct :cqui:cuc:ts
o: t|c sulc uscc io: |c|tcsti:g
I::sitio: :o||ilit, c|:gcs w|ic| |ccouc :cccss:, cu:i:g |c|tcsti:g
uust :ot |:i:g |out i:co:sistc:cics i: t|c t::sitio: ut:ix
I|c :uu|c: oi ct iiclcs i: t|c t::sitio: ut:ix i:c:cscs s|:l, s t|c
:uu|c: oi :ti:g clsscs i:c:cscs ;ci scctio: .+1, lo: cc| coluu: oi t|c
106
Cf. TASCHE, D., A traffic lights approach to PD validation.
107
Cf. DUFFIE, D./SINGLETON, K. J., Simulating correlated defaults and ZHOU, C., Default correlation: an analytical result,
as well as DUFFIE, D./SINGLETON, K. J., Credit Risk, Ch. 10.
Rating Models and Validation
124 Guidelines on Credit Risk Management
t::sitio: ut:ix, t|c ct :cqui:cuc:ts io: |c|tcsti:g t|c ut:ix :c cquiv
lc:t to t|c :cqui:cuc:ts io: |c|tcsti:g cciult :o||ilitics io: ll :ti:g
clsscs
cciiicll,, |c|tcsti:g t|c cciult coluu: i: t|c t::sitio: ut:ix is
cquivlc:t to |c|tcsti:g t|c cciult :o||ilitics ovc: t|c tiuc |o:izo: w|ic|
t|c t::sitio: ut:ix ccsc:i|cs I|c c:t:ics i: t|c cciult coluu: oi t|c t::
sitio: ut:ix s|oulc t|c:cio:c :o lo:gc: |c c|:gcc o:cc t|c :ti:g uoccl |s
|cc: cli|:tcc o: :ccli|:tcc I|is lics to o:c,c: t::sitio: ut:iccs s
wcll s uulti,c: t::sitio: ut:iccs ii t|c uoccl is cli|:tcc io: lo:gc: tiuc
c:iocs
W|c: |c|tcsti:g i:civicul t::sitio: :o||ilitics, it is ossi|lc to t|c
siuliiicc :oc| i: w|ic| t|c t::sitio: ut:ix is civiccc i:to i:civicul clc
uc:t:, cvc:ts l: t|is :oc|, t|c cic|otouous csc i: w|ic| I|c t::sitio:
oi :ti:g clss x lccs to :ti:g clss , ;lvc:t A, o: I|c t::sitio: oi :ti:g
clss x cocs not lcc to :ti:g clss , ;lvc:t l, is tcstcc io: ll i:s oi :ti:g
clsscs I|c io:ccst :o||ilit, oi t::sitio: i:ou x to , j
)oicco:t
t|c: |s to |c
c|ustcc to lvc:t A`s :clizcc i:cquc:c, oi occu::c:cc
I|c :occcu:c ccsc:i|cc is |ig|l, siulistic, s t|c :o||ilitics oi occu:
:c:cc oi ll ossi|lc t::sitio: cvc:ts :c co::cltcc wit| o:c :ot|c: :c s|oulc
t|c:cio:c :ot |c co:sicc:cc sc:tcl, lowcvc:, lc| oi |:owlccgc |out t|c
sizc oi t|c co::cltio: :uctc:s s wcll s t|c :l,ticl coulcxit, oi t|c
:csulti:g cqutio:s :csc:t o|stclcs to t|c :cliztio: oi ut|cuticll, co::cct
solutio:s As i: t|c ssuutio: oi u:co::cltcc cciult cvc:ts, t|c i:civicul
t::sitio: :o||ilitics c: |c su||cctcc to isoltcc |c|tcsti:g io: t|c u:osc
oi i:itil :oxiutio:
lo: cc| i:civicul t::sitio: cvc:t, it is :cccss:, to c|cc| w|ct|c: t|c
t::sitio: ut:ix |s sig:iiic:tl, u:cc:cstiutcc o: ovc:cstiutcc its :o||il
it, oi occu::c:cc I|is c: |c co:c usi:g t|c |i:ouil tcst l: t|c io:uuls
|clow, `
i0
`
1
1
i
_ _
j
)oicco:t
i
1 j
)oicco:t
`
1
i
.
t|c t::sitio: :o||ilit, ws sig:iiic:tl, u:cc:cstiutcc t co:iicc:cc
lcvcl
10
;o:csiccc tcst, li
i0
`
1
1
i
_ _
j
)oicco:t
i
1 j
)oicco:t
`
1
i
< 1 .
t|c t::sitio: :o||ilit, ws sig:iiic:tl, ovc:cstiutcc t co:iicc:cc lcvcl
I|c |i:ouil tcst is csccill, suit|lc io: lictio: cvc: w|c: i:civicul
ut:ix :ows co:ti: o:l, sull :uu|c:s oi cscs li l:gc :uu|c:s oi cscs :c
108
This is equivalent to the statement that the (binomially distributed) probability 1i < `
jj
)oicco:t
; `
`
1
of occur-
rence of a maximum of `
defaults for `
`
1
cases in the original rating class must not be greater than q.
Rating Models and Validation
Guidelines on Credit Risk Management 125
vil|lc, t|c |i:ouil tcst c: |c :clccc wit| tcst usi:g st:c:c :o:ul
cist:i|utio:
10-
|:t o |clow s|ows : cxulc oi |ow t|c |i:ouil tcst woulc |c c:
io:ucc io: o:c :ow i: t::sitio: ut:ix lo: cc| ut:ix iiclc, wc ii:st cl
cultc t|c tcst vluc o: t|c lcit sicc oi t|c i:cqulit, Wc t|c: cou:c t|is tcst
vluc to t|c :ig|t sicc oi t|c i:cqulit, io: t|c -0 :c -. sig:iiic:cc lcvcls
ig:iiic:t ccvitio:s |ctwcc: t|c io:ccst :c :clizcc t::sitio: :tcs :c icc:
tiiicc t t|c -0 lcvcl io: t::sitio:s to clsscs 1|, !, !|, :c !c At t|c -.
sig:iiic:cc lcvcl, o:l, t|c u:cc::u:s oi io:ccst :tcs oi t::sitio: to clsscs 1|
:c !| :c sig:iiic:t
l: t|is co:tcxt, clss 1 is sccil csc |ccusc t|c io:ccst t::sitio: :tc
cquls zc:o i: t|is clss l: t|is csc, t|c tcst vluc uscc i: t|c |i:ouil tcst
lw,s cquls 1, cvc: ii :o t::sitio:s :c o|sc:vcc lowcvc:, t|is is :ot co:
sicc:cc uiscstiutc oi t|c t::sitio: :tc li t::sitio:s :c o|sc:vcc, t|c t::
sitio: :tc is o|viousl, :ot cqul to zc:o :c woulc |vc to |c c|ustcc cco:c
i:gl,
|:t o Dt lxulc io: lc|tcsti:g I::sitio: \t:ix wit| t|c li:ouil Icst
li sig:iiic:t ccvitio:s :c icc:tiiicc |ctwcc: t|c io:ccst :c :clizcc t::
sitio: :tcs, it is :cccss:, to c|ust t|c t::sitio: ut:ix l: t|c siulcst oi
cscs, t|c i:ccu:tc vlucs i: t|c ut:ix :c :clccc wit| :cw vlucs w|ic|
:c cctc:ui:cc cui:icll, lowcvc:, t|is c: lso |:i:g |out i:co:sistc:cics
i: t|c t::sitio: ut:ix w|ic| t|c: u|c it :cccss:, to suoot| t|c ut:ix
I|c:c is :o siulc lgo:it|uic uct|oc w|ic| c:|lcs t|c :llcl c|ust
uc:t oi ll t::sitio: i:cquc:cics to t|c :cqui:cc sig:iiic:cc lcvcls wit| cuc
ttc:tio: to t|c co:sistc:c, :ulcs I|c:cio:c, :ctitio:c:s oitc: usc :gutic
solutio:s i: w|ic| :ts oi i:civicul t::sitio: :o||ilitics :c s|iitcc to
:cig||o:i:g clsscs out oi |cu:istic co:sicc:tio:s i: o:cc: to c|c:c to t|c co:
sistc:c, :ulcs
110
li uulti,c: t::sitio: ut:iccs c: |c clcultcc ci:cctl, i:ou t|c ct sct,
it is :ot |solutcl, :cccss:, to c|c:c to t|c co:sistc:c, :ulcs lowcvc:, w|t
is cssc:til to t|c vlicit, :c :cticl vi|ilit, oi :ti:g uoccl is t|c co:sis
109
Cf. the comments in section 6.2.2 on reviewing the significance of default rates.
110
In mathematical terms, it is not even possible to ensure the existence of a solution in all cases.
Rating Models and Validation
126 Guidelines on Credit Risk Management
tc:c, oi t|c cuuultivc :c co:citio:l cciult :o||ilitics clcultcc i:ou t|c
o:c,c: :c uulti,c: t::sitio: ut:iccs ;ci scctio: .+!,
:cc t::sitio: ut:ix |s |cc: c|ustcc cuc to sig:iiic:t ccvitio:s, it is
:cccss:, to c:su:c t|t t|c ut:ix co:ti:ucs to sc:vc s t|c |sis io: c:ccit :is|
u:gcuc:t lo: cxulc, t|c :cw t::sitio: ut:ix s|oulc lso |c uscc io:
:is||scc :ici:g w|c:cvc: lic|lc
6.2.4 Stability
W|c: :cvicwi:g t|c st|ilit, oi :ti:g uoccls, it is :cccss:, to cxui:c two
sccts sc:tcl,
|:gcs i: t|c cisc:iui:to:, owc: oi :ti:g uoccl givc: io:ccsti:g
|o:izo:s oi v:,i:g lc:gt| :c c|:gcs i: cisc:iui:to:, owc: s lo:s
|ccouc olcc:
|:gcs i: t|c gc:c:l co:citio:s u:cc:l,i:g t|c usc oi t|c uoccl :c t|ci:
ciiccts o: i:civicul uoccl :uctc:s :c o: t|c :csults t|c uoccl gc:c:
tcs
l: gc:c:l, :ti:g uoccls s|oulc |c :o|ust gi:st t|c gi:g oi t|c lo:s
:tcc :c gi:st c|:gcs i: gc:c:l co:citio:s l: ccitio:, :ot|c: cssc:til
c|:ctc:istic oi t|csc uoccls is suiiicic:tl, |ig| lcvcl oi cisc:iui:to:, owc:
io: c:iocs lo:gc: t|: 1! uo:t|s
Changes in Discriminatory Power over Various Forecast Horizons
l: scctio: o!1, wc ccsc:i|cc t|c :occss oi tcsti:g t|c cisc:iui:to:, owc:
oi :ti:g uoccls ovc: tiuc |o:izo: oi 1! uo:t|s lowcvc:, it is lso ossi|lc
to ucsu:c cisc:iui:to:, owc: ovc: lo:gc: c:iocs oi tiuc ii suiiicic:t ct
sct is vil|lc l: t|is co:tcxt, :, ucsu:c oi cisc:iui:to:, owc:, suc| s t|c
Gi:i ociiicic:t ;lowc:stt,, c: |c uscc
W|c: :ti:g uoccls :c otiuizcc io: c:ioc oi 1! uo:t|s, t|ci: cisc:iu
i:to:, owc: ccc:cscs io: lo:gc: tiuc |o:izo:s lc:c it is :cccss:, to c:su:c
t|t t|c cisc:iui:to:, owc: oi :ti:g uoccl o:l, cctc:io:tcs stccil,, t|t
is, wit|out c:oi:g |:utl, to cxccssivcl, low vlucs ou:c :ti:g uoccls
s|oulc lso ccuo:st:tc suiiicic:t cisc:iui:to:, owc: ovc: io:ccsti:g |o:i
zo:s oi t|:cc o: uo:c ,c:s
A:ot|c: scct oi t|c tiuc st|ilit, oi :ti:g uoccls is t|c ccc:csc i: cis
c:iui:to:, owc: s lo:s |ccouc olcc: I|is is csccill, :clcv:t i: t|c csc
oi lictio: sco:cs w|c:c t|c cisc:iui:to:, owc: io: : o|sc:vcc qu:tit,
oi :cw |usi:css cscs ccc:cscs :oticc|l, ovc: c:ioc oi o to !o uo:t|s
itc: : lictio: is su|uittcc I|is is cuc to t|c ict t|t t|c ct uscc i:
lictio: sco:i:g |ccouc lcss sig:iiic:t ovc: tiuc I|c:cio:c, :ctitio:c:s
i:cquc:tl, coulcuc:t lictio: sco:i:g uoccls wit| |c|vio: sco:i:g uoc
cls I|c lttc: uoccls cvlutc uo:c :ccc:t i:io:utio: i:ou t|c ccvclouc:t
oi t|c c:ccit t::sctio: :c t|c:cio:c :ovicc |cttc: i:cicto: oi c:ccitwo:
t|i:css t|: lictio: sco:i:g uoccls lo:c lowcvc:, |c|vio: sco:i:g is :ot
ossi|lc u:til c:ccit icilit, |s :cc|cc cc:ti: lcvcl oi utu:it,, t|t is, o:cc
|c|vio::cltcc ct :c ctull, vil|lc
Rating Models and Validation
Guidelines on Credit Risk Management 127
Changes in the General Conditions for Model Use
I|c sscssuc:t oi c|:gcs i: t|c gc:c:l co:citio:s u:cc: w|ic| uoccl is
uscc |s st:o:g qulittivc clcuc:ts : t|c o:c |:c, it is :cccss:, to :cvicw
w|ct|c: ccvclouc:ts i: t|c cco:ouic, oliticl, o: lcgl c:vi:o:uc:t will
|vc : i:iluc:cc o: t|c :ti:g uoccl o: i:civicul uoccl :uctc:s :c c:i
tc:i : t|c ot|c: |:c, i:tc::l icto:s t t|c |:| suc| s c|:gcs i: |usi:css
st:tcgics, t|c cx:sio: oi ctivitics i: cc:ti: u:|ct scguc:ts, o: c|:gcs i:
o:g:iztio:l st:uctu:cs u, lso iicct t|c c:io:u:cc oi :ti:g uoccl su|
st:till,
|:gcs i: t|c cco:ouic c:vi:o:uc:t i:clucc t|c |usi:css c,clc i: :tic
ul:, w|ic| c: cusc u|o: iluctutio:s i: t|c :uctc: lD cu:i:g c:iocs oi
:ccovc:, :c cccli:c lowcvc:, icto:s suc| s tcc|:icl :og:css o: oliticl
:c lcgl ccvclouc:ts c: lso i:iluc:cc t|c ciicctivc:css oi :ti:g uoccls
l: :ticul:, cou:t:, :c :cgio:l :ti:gs ccc:c |cvil, o: c|:gcs i: t|c
oliticl c:vi:o:uc:t lowcvc:, suc| scc::ios s|oulc l:cc, |c i:tcg:tcc
i:to :ti:g uoccl, ot|c:wisc o:c woulc |vc to qucstio: t|c suit|ilit, oi
t|c uoccl itscli
lxulcs i: t|c lcgl c:vi:o:uc:t i:clucc c|:gcs i: couuc:cil lw o:
ccou:ti:g st:c:cs w|ic| u, |vc ositivc o: :cgtivc i:iluc:cc o: t|c
ciicctivc:css :c sig:iiic:cc oi cc:ti: ii::cil i:cicto:s |:gcs i: lcgisl
tio: w|ic| :c :clcv:t i: t|is co:tcxt lso i:clucc :cvisio:s oi t|c ui:iuuu
su|sistc:cc lcvcl ;ic t|c sl:, uou:t w|ic| c::ot |c ttc|cc, o: c|:gcs
i: |:|:utc, :occcu:cs l: :ticul:, t|csc c|:gcs c: cusc s|iits i: t|c
:is| :uctc: lGD
Qu:tii,i:g t|c ciiccts oi c|:gcs i: gc:c:l co:citio:s o: t|c iu:ctio:lit,
oi :ti:g uoccls :cqui:cs : i:cct| :l,sis oi t|c uoccl :uctc:s :c
s|oulc t|c:cio:c ccou:, t|c o:goi:g ccvclouc:t oi t|c uoccl kti:g
uoccls |vc to u:cc:go iu:t|c: ccvclouc:t w|c:cvc: t|ci: c:io:u:cc
ccc:cscs cuc to c|:gcs i: gc:c:l co:citio:s
: t|c ot|c: |:c, t|c |:| u, lso cccicc to ccvclo :cw :ti:g uoccl
ii cxc:ts |clicvc t|t otc:til o: l::cc c|:gc i: gc:c:l co:citio:s woulc
lcc to su|st:til loss i: t|c c:io:u:cc oi t|c cu::c:t uoccl
6.3 Benchmarking
l: t|c qu:tittivc vlictio: oi :ti:g uoccls, it is :cccss:, to cisti:guis|
|ctwcc: |c|tcsti:g :c |c:c|u:|i:g
111
lc|tcsti:g :cic:s to vlictio:
o: t|c |sis oi |:|`s i:|ousc ct l: :ticul:, t|is tc:u ccsc:i|cs t|c cou
:iso: oi io:ccst :c :clizcc cciult :tcs i: t|c |:|`s c:ccit o:tiolio
l: co:t:st, |c:c|u:|i:g :cic:s to t|c lictio: oi :ti:g uoccl to
:cic:c:cc ct sct ;|c:c|u:| ct sct, lc:c|u:|i:g scciiicll, llows qu:
tittivc sttisticl :ti:g uoccls to |c cou:cc usi:g u:iio:u ct sct lt c:
|c :ovc: t|t t|c i:cicto:s uscc i: qu:tittivc vlictio: i: :ticul: cis
c:iui:to:, owc: ucsu:cs, |ut lso cli|:tio: ucsu:cs ccc:c t lcst i:
:t o: t|c sulc cxui:cc
11!
I|c:cio:c, |c:c|u:|i:g :csults :c :cic:|lc
111
DEUTSCHE BUNDESBANK, Monthly Report for September 2003, Approaches to the validation of internal rating systems.
112
HAMERLE/RAUHMEIER/RO
sterreichisches Bank-Archiv
2003, No. 10, 747752
Van de Castle, Karen/Keisman, David, Recovering your Money: Insights into Losses from Defaults, in:
Standard & Poors CreditWeek 1999, 2834
Rating Models and Validation
Guidelines on Credit Risk Management 171
Van de Castle, Karen/Keisman, David, Suddenly Structure Mattered: Insights into Recoveries of
Defaulted Loans, in: Standard & Poors Corporate Ratings 2000
Verband Deutscher Hypothekenbanken, Professionelles Immobilien-Banking: Fakten und Daten,
Berlin 2002
Wehrspohn, Uwe, CreditSmartRiskTM Methodenbeschreibung, CSC Ploenzke AG, 2001
Rating Models and Validation
172 Guidelines on Credit Risk Management