Professional Documents
Culture Documents
9 2>
Reject Inference: DEA: Banks' Operating
A Comparison Revenue on the Banking Products
of Methodologies Production Function and Services Pricing
Felipe Domingues Araujo Eduardo Borges da Silva David F. Hastings
Guilherme Barreto Fernandes Dany Rafael Fonseca Mendes 55
9 772 177 60 311 7 30 Giovanna Valadares Borges
Michel Angelo C. de Oliveira
44
3
07
Impairment Models Based on Expected The new rules set by the IFRS 9 and accepted
Loss and Compliance to IFRS 9 by the BCBS require recognizing losses over the
course of the maturity of exposures. In addition,
the new rules require acknowledging signs of cre-
Wolfgang Reitgruber
dit quality deterioration while estimating those los-
Carlos Antonio Campos Nogueira ses, even if this occurs prior to their recognition as
incurred. This is what has become known as the
Impairment Model. The article outlines a comple-
te methodology for compliance with these requi-
rements, revolving around the concept of Risk Im-
pact.
30
44
DEA: Banks Operating Revenue This article uses DEA (Data Envelopment Analysis)
on the Production Function to evaluate the efficiency of the main financial ins-
titutions active in Brazil since the 2008 internatio-
nal crisis. Taking Operating Return as an output,
Eduardo Borges da Silva
the findings favor domestic (private- and public-
Dany Rafael Fonseca Mendes -sector) banks, which maintained a lead over fo-
Giovanna Valadares Borges reign-owned banks. It is worth noting that domes-
Michel Angelo Constantino de Oliveira tic public-sector banks ranked first in 2009, playing
a leverage role in market financial operations du-
ring a period of international financial crisis.
55
Banking Products and Services Pricing Setting the selling price for the good or service
offered is crucial to the success of any enterprise.
David F. Hastings It is no different with banks. However, when it co-
mes to extending bank loans (and taking bank de-
posits), the task is different and more complex
than for other business activities, even though the
rationale is similar.
7
Wolfgang Reitgruber1
Carlos Antonio Campos Nogueira
1 The presented opinions and methods in this paper are solely the responsibility of the authors and should not be interpreted as
reflecting those of UniCredit Bank Austria AG or of any of the authors former employers.
The concepts and terms Impact of Risk , IoR , iACV , PL Dashboard and NPL Dashboard are copyright by Wolfgang
Reitgruber, all rights reserved. For commercial, profit-oriented or capital-optimizing applications of these concepts (either within
a corporation or for consulting/training purpose), written permission by Wolfgang Reitgruber is required.
8
Abstract
Even though the Basel Accord requires a one-ye-
ar horizon on the risk parameters, and local provisioning prac-
tices and rules take into account already incurred losses, the
new IFRS 9 standards, as released in July 2014, and already ac-
cepted by the Basel Committee, demand the recognition of ex-
pected future losses during the whole exposures maturity (the
forward looking principle). Moreover, and in tune with that
principle, the same standards require that banks figure out any
signs of impaired creditworthiness that might impact the futu-
re behavior of otherwise (currently) performing loans. That is
what has come to be named impairment model. In this arti-
cle we outline a comprehensive methodology to tackle such re-
quirements, centered in the Impact of Risk (IoR) concept.
1. Introduction
After many years of preparation, development of drafts and feedback
sessions, the final standard of IFRS 9 impairment was released in July 2014. The
change from an incurred loss model to an expected loss model for loan loss pro-
visioning was kicked-off. Methodological and stochastic concepts of credit risk,
so far mostly used in internal steering, in regulatory context for IRBA segments
and in modeling economic capital, will enter the area of accounting and provisio-
ning. And local supervisors such as Banco Central do Brasil will have to apply
some important changes to the standard chart of accounts (COSIF) and current
account practices in order to comply to IFRS 9 and, at the same time, avoid or at
least reduce the need of inconvenient (and error prone) double booking.
This article presents methodological arguments and proposes ways how
to approach certain issues in IFRS 9 impairment modelling.
It starts with a detailed discussion of the structural conservatism in the
final standard. The exposure value outlined in the first exposure draft 2009 (ED
2009) will be specified and introduced as iACV (idealized Amortized Cost Va-
lue). It is fair under the economic theory of amortized cost accounting and
consequently provides a valid benchmark for the IFRS 9 net exposure value. The-
9
mentation phase of the final standard. After rollout these key risk indicators will
support the understanding of portfolio dynamics on an ongoing basis when used
in monitoring and reporting. Overall, the use of these measures for credit risk
portfolios before and after default will have a positive contribution to the general
understanding of credit risk processes and their efficiency.
This work is mostly based on the first two references written by Wolfgang Rei-
tgruber (2013, 2015) listed at its end. Carlos Nogueira as co-author integra-
ted content about Brazilian local rules (banco Central and CVM) as well about
practices and current stage of local implementation of the Basel Accord. Be-
sides, when writing these first two references, Wolfgang Reitgruber was dee-
ply inspired from Antifragility by Taleb (2012) and The Known, The Unkno-
wn, The Unknowable in Financial Risk Management by Diebold et al (2010).
Financial institutions will have to develop a better understanding of the sto-
chastic properties of day-to-day credit risk results. The supporting role of ca-
pital as cushion for small and large credit risk shocks is crucial to welcome
these events. They are important input to improve internal processes on ba-
sis of objective quality standards. The concept Impact of Risk and its ele-
ments Performing Loans (PL) Dashboard and non-performing loans (NPL)
Dashboard may provide a methodological approach to move credit risk from
the unknown or even unknowable at least one step closer to the known.
due to its stochastic properties (rare, large failures with complex dependence
structures) makes an empirical analysis even more challenging.
Strategic objective: Management tends to use cost of risk in order to
achieve different strategic goals
smoothing the P&L, displaying financial health (Signaling) or execution of
big cleanup exercises, if budgets or market expectations cannot be achieved
anyway (Big Bath Accounting). An overview of the different managerial stra-
tegies can be found in Gaber (2013). In Domikowsky et al (2014) the situation in
Germany is empirically analyzed, where certain types of impairment are used
for tax optimization. Because of these distortions the empirical analysis based
on cost of risk is challenging.
The methodological link from provisioning to capital requirements under
the IRBA is represented by the shortfall, defined by the difference between the
expected loss and the provision amount. For capital requirements, this leads to
an effective replacement of the provision amount by a parameter-based assess-
ment of the expected credit risk (the expected loss). However, changes in capi-
tal requirements caused by the shortfall receive much lower management atten-
tion than changes to provisions. For P&L, relatively small changes already recei-
ve high attention. For capital however, the absolute magnitude is more relevant.
Significant large changes are mostly driven by external or regulatory factors.
Comparably small monthly changes, triggered by portfolio developments and
shortfall adjustments, tend to receive less attention. The underlying dynamic of
monthly capital requirements caused by expected loss changes is quantified in
the Impact of Risk methodology proposed by Reitgruber (2013). A detailed in-
troduction is provided in Appendix B of the first reference (see bibliographic re-
ferences at the end of this article).
Basically, the increase in provisioning levels, to be realized with the in-
troduction of the final standard caused primarily by the extension of the forecast
period to one year or lifetime, will be significant. These amounts will most like-
ly affect certain methodological decisions in the preparation phase. In the con-
text of the ongoing application however, other effects will prevail: By reducing
flexibility in the provisioning process, the volatility of monthly, quarterly or annu-
al actually reported risk costs will increase. There are already working papers fo-
cusing on increased volatility and timing of loss recognition under various provi-
sioning regimes, eg Grnberger (2012). Although this effect should not come to
a surprise for analysts, it may have unexpected consequences in the context of
disclosure and market perception. Deeper methodological understanding of the
volatility of credit risk will have to be developed in near future to provide reaso-
nable mitigation.
Currently P&L and capital requirements are seen as largely separate dis-
ciplines. The application of the final standard will lead to strong dependencies:
the increase in volatility will lead to a reduction of predictability of credit risk and
less possibilities to micro-control results. On the other hand, fluctuations will
13
rying hazard rates (non-neutral risk profiles) effectively cause distortions of ac-
counting values over time.
Graphic 1: Example with different risk profiles and same risk level
meses. GCA ) neste caso maior no primeiro ano, com valor equivalente perda esperada em 12
in years 2 to0 5 to end up at the same overall risk level. The difference between
meses.
GCAt and iACVt (as % of GCA0) in this case is highest in the first year with a va-
Comlue base em um determinado
equivalent to the perfil de
12-month risco, podemos
expected loss. derivar o teste de calibragem esttica.
Com
O perfil de risco base em
propriamente um determinado
dito perfil de risco, podemos derivar o teste de calibragem esttica.
Based
O perfil on
de arisco
given riskpode
profiser
propriamente
estimado
le dito
the static
pode
por meio
calibration
ser estimado
de test
uma
por isanlise
meio derived.
de
de pool
uma The risk de pool
anlise
esttico (anlise por safra) de uma carteira homognea com caractersticas de risco parecidas.
profile itself
esttico can(anlise
be estimated bydemeans of a static pool analysis (vintage analy-risco parecidas.
Tendncias atuais da carteira por safra)
e outras uma carteira
informaes homognea
prospectivas com caractersticas
(mudana de qualidade de da
sis) efor
carteira aTendncias
homogenous
panorama atuais
econmico) dapodem
carteira
portfolio withe outras
ser similarinformaes prospectivas
risk characteristics.
incorporadas. O conservadorismo (mudana
Current portfo-
(calibragem de de
qualidade da
qualidade) do NCAt no bucket 1 quantificado por meio de comparao direta com iACVt: (calibragem de
lio trendscarteira
and e panorama
other econmico)
forward-looking podem
informationser incorporadas.
(portfolio O conservadorismo
quality changes and
economic qualidade)
outlooks) domay
NCAbe t noincorporated.
bucket 1 quantificado por meio de(calibration
The conservatism comparaoquality) direta com iACVt:
of NCA in bucket 1 is quantified by direct comparison with iACVt:
t EVt0 t NCA0t , onde todos os valores so descontados taxa iED at o momento da
t iACVt0 NCA0t , where onde todos os valores
all values so descontados
are discounted with rate taxa iED ito at o momento da
ED loan
originao (t=0).
origination (t=0). (t=0).
originao
In case of a neutral risk profile with risk level r,
No caso de um perfil de risco neutro com nvel de risco r,
No caso de um perfil de risco neutro com nvel de risco r,
t = iACVt0 NCA 0
= GCA0t 0 NCA
=t iACV 0 NCA
0
t 0=
= GCA
EL0t 0== rELGCA
NCA 0
0
= rt GCA0
t t t t t t t
corresponde corresponde
ao valor anualizado
ao valordaanualizado
perda esperada
da ao esperada
perda longo do prazo
ao da operao.
longo Esse
do conservatism
prazo da operao. Esse
corresponds
conservadorismo to
causado pelathe annualized
estrutura lifetime
da estrutura expected
norma definitiva loss. This
sobdefinitiva
condies decondies
risco estveis
is caused conservadorismo
by the structure causado
of the pela
fiseja
nal ostandard da
under stable risk conditions andde risco
norma sob
e perfil de risco neutro em qualquer que caso.
estveis e perfil de risco neutro em qualquer que seja o caso.
neutral
No caso riskperfil
de um profile indeany case.
No caso geral risco,
de um perfil geral de risco,
t = EVt0 NCA 0
= (EV 0 0
0 t ) + (GCA
0 t GCA
0 0 0
0 t ) = EL0
0 t NCA
t1(r
t 0 0 rT )0GCA0Tt1 0
t = tiACVt NCAt = (iACVt GCAt ) + (GCAt NCA t ) = ELt 0 (r rT ) GCAT (Prova
no Apndice
(Prova no Apndice A de Reitgruber
A de Reitgruber (2015)).
(2015)). ComNCA
Com isso, isso,t se
NCA t se mantm
mantm conservador
conservador desdedesde que
Vt0 NCA0t , onde todos os valores
corresponde so anualizado
ao valor descontados da perda
taxa iED at ao
corresponde
esperada oaomomento
valor da da
longoanualizado
do prazo daoperao.
perda esperada
Esse ao longo do prazo d
conservadorismo causado pela estrutura da 17
norma definitiva sob cond
o (t=0). conservadorismo causado pela estrutura da norma definitiva sob condies de risco
estveis e perfil de risco neutro em qualquere perfilque
de risco
seja oneutro
caso. em qualquer que seja o caso.
Noneutro
o de um perfil de risco caso In
decom
umnvel
caseperfil geral
of ade riscoder,risco,
general No
risk profile,caso de um perfil geral de risco,
GCA
onde
where GCAGCA0T = (1+i)TT.
onde GCA0T = (1+i)TT.
No
Nocaso
casodo No caso
doiACV
iACV dorespectiva
eeda
da iACV e da
respectiva respectiva
taxa
taxa de taxaide
dedesconto
desconto iED, , desconto iED,
No caso do iACV e da respectiva taxa de desconto iED,
ED
No caso In case
do iACV of
e da iACV and
respectiva the
taxa respective
dededesconto discount rate iED,
No caso do iACV e da respectiva taxa iED, ,
descontoiED 10
iACV
iACV ++iACV 00
1000+
iACV
iACV ++
0 iACV 0 00
++iACV + +++
iACV iACV00 +
1 TT + iACV T+ + iACV 0 +
DDiACV
iACV
mod = = DiACV
iACV + = 0011+ iACV
iACV +
iACV
iACV 0 0
0 1 T
mod
iACV
mod
iACV
0 0 + iACV
(1
(1+1 iD
1+ +
i )
mod) +=iACV
(1
iACV
0
T+
+00iED
iACV T )+iACV
Dmod
Dmod==
iACV EDED (1 + iED ) iACV0
(1 iED))iACV
(1++iED iACV00
AAduration A durationpode
durationmodificada
modificada modificada
pode ser pode ser como
serinterpretada
interpretada interpretada
comosemielasticidade como semielasticidade
semielasticidade (ex.:
(ex.:Schwaiger,
Schwaiger, (ex.: Schwaiger, 2014),
2014),
2014),
relacionando A duration
a variao modificada
relativa do valor pode
presenteser interpretada
GCA ou como
iACVabsoluta semielasticidade
absoluta(ex.: Schwaige
A relacionando
relacionando
A duration
duration The
modificadamodifi
aavariao
variao
modificada ed
pode duration
relativa
relativa
pode ser ser do can
dovalor
valor
interpretada
interpretada be interpreted
presente
presente
como
como GCA
GCA as
00ou semi-elasticity
ou iACV
semielasticidade
semielasticidade iACV variao
00(ex.:0variao
(ex.: (e.g.
Schwaiger,
Schwaiger, 0 2014),
variao
Schwai-
absoluta 2014),da
dataxa
taxa da taxa
relacionando a variao relativa do valor presente GCA0 ou iACV0 variao abso
de desconto
deger,
desconto
relacionando dea desconto
2014),aaplicada
relacionando aplicada
relating
variao the iaplicada
i irelativa
variao ou . .Um
ourelativa dodo
Um
irelative ivalor
oude
nvel
nvel ide
change
valor .risco
Um in nvel
EDpresente
risco
presente GCAde00ou
crescente
crescente
present
GCA risco
ou (com
value crescente
iACV
(com variao or(com
variao
00variao
GCA variao
absoluta
absoluta
absoluta
iACV tor), absoluta
que
dada
the
r), taxa
que
ab- temr), que tem
tem
iED.iACV variao absoluta taxa
EDED
de desconto o mesmo
aplicada iimpacto
ou i .
de
Um
desconto
sobre
nvel o valor
de
aplicada
risco presente
crescente
i ouque
(com
Um
um nvel
aumento
variao
0
de risco crescente
0
correspondente
absoluta r), que
(com
temda variao
taxa de absoluta
deoodesconto
mesmo impacto
mesmochange
solute impacto
aplicadain sobre
theioEDo
sobre
i ou valor
. valor
applied presente
presente
Umo nvel
ED
mesmo
discount que
que
de impacto
risco um
umaumento
rate i aumento
crescente
sobre o
. An
or iED(com correspondente
valor
correspondente
increasing
variao
presente que
da
risk
absoluta dataxa
um
taxade
level
r), de
(with
que
aumento tem
correspondente da tax
o mesmo
o desconto,
desconto,
mesmo impacto
leva
leva
impacto
absolute aauma
change umasobre
desconto,
sobre r),oleva
reduo o
reduo
valorvalor
which a uma presente
relativa
presente
has reduo
relativa de
de
thequeque
Dsame
DiACVum
relativa
um
iACVmod aumento
de DiACV
sobre
sobre
aumento
impact
mod
correspondente
iACV
iACV
on 00no
the
mod sobre
no
correspondente valor
valor
present iACV
dededa
r
0r
da taxa
no
valueD valor
DiACV
taxa de
iACV
asde
modade
mod r DiACVmod. Essa
. .cor-
Essa
Essa
desconto, leva a uma reduo desconto,
relativa de leva
D iACVamod
umasobrereduo
iACV relativa
no valordo de
de D
r
iACV
DDaomod sobre . Essa iACV no valor
ao0 longo da de r DiACV
duration aduration
modificada podemodificada
ofrepresentar pode umrepresentar
papel um
crucial papel
no crucial
contexto no contexto
EAD longodo EAD
da vida vida
iACV
duration
desconto,
responding modificada
leva uma
increasepode
reduo representar
relativa
the discountde um
D iACVpapel
rate, crucial
sobre
modleads no
iACV
to a D0contexto
no
0 iACV
valor do
de EAD
r
(relative) ao
iACV longo
mod
mod. Essa
decrease daofvida
duration modificada pode durationum
representar modificada
papel pode
crucial norepresentar
contexto do umEADpapelao longo crucial no contexto do EAD ao lon
da vida
da operao,
daiACV
operao,
duration modificada da
temaoperao,
tema que
que
pode est
est tema
alm
alm
representar que
do est
escopo
doedescopo
um alm
papel do
deste
deste escopo
crucialartigo.
artigo. deste mod artigo.
by
da operao,
0
r D iACV
tema mod . This
que est alm modifi
da operao, duration
do escopo tema deste quemay
artigo.est alm do escopo deste artigo.vida
no
playcontexto
a crucial do EAD
role ao
in longo
the da
context
da operao,
of lifetimetema EAD, quewhich
est alm is out doofescopo
scopedeste of thisartigo.
work.
Com
Combase
basenessaCom
nessa base
of nessa
relao,
relao, relao,
ooteste
teste de o teste
decalibragemthede
calibragem calibragem
esttica
estticada esttica
daseo
seo da from
anterior
anteriorseo
para
para anterior
nveis
nveis para nveis variveis
variveis
variveis
Com baseOn basis
nessa this
relao, relationship,
Com
o teste de base nessa
calibragem static
relao,
esttica calibration
ode
da teste
seo de test
calibragem
anterior last section
esttica
para nveis da seo
variveis anterior para nve
de
de
Comtorisco,
risco,
base pode
pode
nessa de risco,
ser
ser pode
generalizadoser
generalizado
relao, o teste generalizado
para
depara perfis
perfis
calibragemde para
derisco perfis
risco
esttica risco
no-neutros
no-neutros
da seoparaeno-neutros
e variaes
variaes doe
do variaes
nvel
nvelde
de do nvel
risco,
risco, de risco,
de variable
risco, poderisk
serlevels can de
generalizado be generalized
para perfis
risco, podedeser for
risco non-neutral
no-neutros
generalizado eanterior
risk
variaes para
profi
perfis de nveis
doles and
nvel
risco devariveis
chan-
risco,
no-neutros e variaes do nve
dena
naforma como nasegue:
forma como segue:perfis de risco no-neutros e variaes do nvel de risco,
naforma
risco,
ging pode
forma
riskcomo
ser
como
levelsegue:
generalizado
as follows: para
segue: na forma como segue:
na forma como segue:
0
tt== iACVt0t0(nvel
iACV
t = iACV
t = iACV
0(nvel de t (nvel
derisco,
risco,
t (nvel de risco,
de de
perfil
perfil
perfil risco,
de perfil
risco)
derisco)
risco) NCA
0 (nvel de
NCA
NCA 00t=
0
risco)
t== NCA0t =
t = iACV t de risco,
t perfil de risco) NCA0t =
t = iACVt000(nvel de risco,
0 perfil de risco) NCA 0
= 0
t (nvel t (nvel
= (iACV de de
risco, perfil de t000(nvel
risco) original,
iACV original,
tt(nvel 0 perfil de risco)) +
===(iACV 0(nvel de (nvel
(iACV
(iACV t (nvel dederisco,
risco, perfil
perfil
perfil derisco)
de risco)
risco) iACV
0 (nvel iACV
iACV t(nvel
t original,
original, perfil
perfil
perfil de
derisco))
deiACV + ++
risco))
risco))
= (iACVt de risco, perfil de risco) t (nvel original, perfil de
t
0 (nvel 0 (nvel
= (iACVt 000 (iACV de risco, 0 perfil de risco) iACV0t00 original,
0 perfil de risco)) +
+++ (iACV
(iACV
(iACV + original,
tt(nvel
t(nvel
(nvel t (nvel
original,
original, original,
perfil
perfil
perfil de
derisco)
de perfil
risco)
risco)
0 (nvel de
NCA
NCA
NCArisco)
t t)t))=== NCAt ) =
+ (iACV t original, perfil de risco) NCA0t ) =
+ (iACVt0 (nvelt1t1
t1 original, t1 perfil de risco) NCA 0)
t =
t1
=== ELEL0
EL0
tt
0t1
t(r(r
(r = EL
0
rtrTr)))GCA
TT (r
GCA 00
0
GCATTT r
rr )0D
Tr GCA
D iACV
iACV
iACV
Dmod,t
mod,t
0
T rtttDiACV
GCA
GCA
GCA
000
mod,t GCA
0
t
= EL t mod,t
(r rT ) GCA 0
T r DiACV
mod,t GCAt
0
= EL0t (r 000 rT ) GCA 0 0T r DiACV mod,t GCA 0
t
0
0
iACV
Onde
Onde D
DDiACV
iACV
aaaduration
Onde duration
DiACV modificada
athe
durationmedida no
notempo
modificada t.t.t. no tempo
medida t.
Onde Where
mod,t mod,t modificada
duration modificada
is medida
medida
modifi no tempo
tempo
ed duration measured at time t.
mod,t
mod,t
iACV Onde DiACV
mod,t a duration modificada medida no tempo t.
Onde Dmod,t
a duration modificada medida no tempo t.
Consequentemente, permanececonservador
Consequentemente, permanece ( 0) se, e somente
conservador 0)se,
( only
se,
Consequentemente,
Consequentemente, t ttpermanece
Consequently, permanece conservador
tconservador
t remains (
(0)
conservative se,
se,eifesomente
0)(0) somente
and ife somente se,
se,
se,
Consequentemente, t permanece conservador ( 0) se, e somente se,
Consequentemente,
0 t1 t permanece 0 conservador ( 0) se, e somente se,
EL (rr ) GCA
r EL 0 TT.(rrT ) GCA
t t10 T
0 t1
00T 0
EL0t0t
0t1 EL
(rrTTt))
(rr GCA
0GCA
r
r r
Dmod,t0iACV
0 .r
GCA00t iACV . GCA ELt . t1 0 T 0
0 (rrT ) GCA T
EL0t Dt1
DiACV
iACV
(rr )
GCA
GCAGCAD 0
.
r mod,t T tt mod,t
0mod,t
iACV 0
T
. t
DiACV
mod,t GCAt
0
Dmod,tpouco
Se permanecer GCAt conservadorismo aps o realizar o ajuste para a distribuio temporal
Se permanecer
Sedopermanecer Sepouco
risco (pequenospermanecer
pouco
If little pouco
conservadorismo
conservadorismo
conservatism
valores conservadorismo
0 remains aps
t1(raps ro
after 0 aps
orealizar
realizar)),ooou
adjusting oem
realizar
ajuste
ajuste
for para
para
the
caso deo ajuste
aauma para
distribuio
distribuio
temporal atemporal
distribuio temporal
temporal
distribution
(EL t
Se permanecer
0 T ) GCATconservadorismo
0 pouco t10(r 0 apsgrande duration
o realizar o ajuste para a distribuio
Sedo
do risco
permanecer
of
modificada do
risk(pequenos
risco (pequenos risco
Dpouco
(small (pequenos
valores
valores
amounts
iACV (EL 00 valores
conservadorismo t1
t1 (EL
(r
(ELt t 00(ex.:
(r
aps )
)GCA
otrrrealizar
0 aTo
GCA 0
)), r
), ou
ou
ajuste
)), or )
Tem
emGCA
0para
in caso
caso
caseTa de
)), ou em
deauma
uma
distribuio
of large caso
grande
grandede umaed grande duration
0 duration
duration
temporal
modifi
mod,t do ativo financeiro
do risco logo
(pequenos TTaps
valoresoriginao,
T
(EL no
caso
t1 (r de
contratos
r ) GCA de
)), ou em caso de uma grand
iACV t 0 T T
domodificada
modificada
risco
longo(pequenos
duration modificada
iACV
iACV
DDmod,t
prazo), do
do
valores
o NCA
mod,t ofpodeD(EL
ativo
ativo
the fi t do
0 0ativo
financeiro
nancial
tornar-se
mod,t
t1
financeiro (r financeiro
(ex.:
(ex.:
asset
no logo
logo aps
) GCA
T(eg
riACV
conservador. 0
aps(ex.:
right
T )), logo
aaafter aps
originao,
originao,
ou emloan ano
no originao,
caso
umade
caso
casoorigination
de de
grande nocase
caso
contratos
contratos
in durationde
decontratos de
de
of
modificada Dmod,t do ativo financeiro (ex.: logo aps a originao, no caso de cont
longo
longprazo),
longo
modificada term longo
doprazo),
ooNCA
DiACV
prazo), NCA pode
contracts),podeNCA
ativo o NCA
tornar-se
tornar-se
financeiro
maypode
no
notornar-se
conservador.
(ex.: no
conservador.
logo aps
become conservador.
a originao,
non-conservative. no caso de contratos de
No bucket
mod,t
2, aps a deteco longo
de um prazo),
aumento o significativo
NCA pode tornar-se
do risco, ano conservador.
quantificao do
longo prazo), o NCA pode tornar-se no conservador.
No conservadorismo
Nobucket No
bucket2,2,aps implcito
bucket 2,
apsaadeteco
detecodedireta:
aps a
deum as perdas
deteco
umaumento de originalmente
um aumento
aumentosignificativo esperadas
significativodo (com
significativo
dorisco, do base no
risco,
risco,aaquantificao
quantificao aperfil
do de
quantificao
do do
risco neutro) esto No
implicitamente
conservadorismo bucket 2,
refletidas
implcito aps
no
direta: a
GCAdeteco
da norma
as perdas de um aumento
definitiva.
originalmente O significativo
requisito
esperadas de do risco, a quantificao
Noconservadorismo
conservadorismo
bucket 2, aps aimplcito
implcitode
deteco direta:
direta:
um as
asperdas
perdas
aumento originalmente
originalmente
significativo do esperadas
esperadas
risco, a (com
(com base
basedo
quantificao no(com
no perfilbase
perfil de no perfil de
de
provises que cubram conservadorismo
as perdas esperadas implcito
(atualizadas, no
direta:aoas perdas originalmente esperadas de (com base
risco
risco neutro)
neutro)
conservadorismo risco
esto
estoneutro) esto
direta:implicitamente
implicitamente
implicitamente
implcito refletidas
asrefletidas no
noGCAGCAda
perdas originalmente daefetivas)
refletidas
norma
norma GCA longo
da norma
definitiva. do prazo
(comOObase
definitiva.
esperadas
da de
definitiva.
requisito
requisito O requisito
de
no perfil de
19
6. Consideration of Forecasts in
Expected Loss Estimation
One of the major differences between regulatory and accounting stan-
dards is the objective of the loss estimates. Regulators generally demand loss
estimates to match long-term default rates and expect certain conservatism.
Specifically we find minimum time series requirements in the current regula-
tion and reference to a through-the-cycle approach, which is methodologically
close to an unconditional loss expectation. Conservatism is included by defini-
tion of minimum thresholds, eligibility criteria or it is explicitly requested throu-
gh add-ons in case of unreliable or unrepresentative data. This understanding
was most recently re-enforced by the consultation paper EBA (2014b), article 49.
The focus of standard setters is the economically unbiased estimate and
a fair and objective exposure value on basis of (reasonably) available information
at the time (forward looking,point-in- time conditional loss estimates).
In this sense, regulators try to achieve stability of the minimum capital
requirements while requesting conservatism in case of uncertainty. The gene-
ral tendency of rating systems to show cyclical behavior is considered with cau-
21
tion. In accounting standards, however, the current best prediction (with reaso-
nable effort) has to be taken into account. As immediate consequence, cyclical
market sentiments shall directly impact the balance sheet. This effect is stron-
gest in fair value accounting, but also to be expected when measuring at amor-
tized cost through forward looking loss estimates. Although the final standard
does not specify in detail how to execute this requirement, it is clear that inter-
nal and external factors have to be considered when deriving these estimates.
As internal factors we understand changes to risk-related processes,
which may have a sustainable and significant effect on the overall risk level or
the temporal distribution of losses. These impacts can be considered in form of
an adjustment to the loss estimates. Relevant process measures must be defi-
ned and monitored to reconfirm these adjustments on an ongoing basis. There
may be good arguments for risk improvements (such as improved credit rating
systems, higher collateral requirements, strengthened underwriting to reduce
high-risk customer segments, etc.), but also potential risk increases after long
periods of low losses have to be considered. As an example, significant above-
-average growth in an otherwise stable segment might indicate potential yet hi-
dden risk increase.
External factors are found predominantly in the area of macro economic
developments and tend to be cyclical. Regulatory standards in IRBA segments
clearly focus on long term averages to provide a stable baseline (e.g. as stressed
by EBA 2014b). Accounting standards diverge from the regulatory view: cycli-
cal developments must not be averaged but have to be considered when deri-
ving loss estimates. The challenge will be how to separate overly optimistic or
pessimistic market sentiment from the real underlying significance of predic-
tions. Additionally, accounting standards require probability weighted estima-
tes. Overly cyclical point-in-time estimates seem inconsistent with this approa-
ch, whereas Bayesian type modelling may better fit the picture. The (small but
reasonable) likelihood of a crisis situation needs to be appropriately included in
the estimates even in case of great economic outlook .
With respect to credit risk we are primarily concerned regarding timing
(forecast horizon) and depth (quantity of losses) of these forecasts. An interes-
ting analysis on basis of the 2008/2009 crisis can be found in Ivantsov (2013). On
basis of this crisis he demonstrated that even shortly before the default of Leh-
man Brothers the depth of the subsequent economic impact was seriously unde-
restimated. Although an economic downturn was already expected, its systemic
impact, however, was ultimately not recognized. As a matter of fact, economic
forecasts seem to be only reliable with respect to relatively short-term trend in-
dications (a few quarters), but fail seriously with respect to forecasting the dep-
th of upcoming extreme events.
This uncertainty has to be considered appropriately when implemen-
ting IFRS 9 impairment standards. An uncritical adoption of long-term forecasts
would almost certainly cause excessive cyclicality and thus may well increase
22
the likelihood of credit bubbles and subsequent losses in case the economy fi-
nally turns. Thus, the volatility caused by the predictive aspect in risk estimates
needs to be carefully controlled.
For easier understanding of this effect let us take the following nave
example: In a typical portfolio with annual risk cost of 100 million euros (long-
-term average, regulatory point of view), only 50 million euro were realized annu-
ally in the past two years due to positive economic developments. This may ea-
sily be in line with a general economic capital model, where mode and median
(to be considered typical annual results) are significantly below the mean of the
loss distribution (corresponding to the expected loss). The general economic fo-
recasts continue to expect positive development. However, we know on basis of
not too long history that realized risk costs amounted to around 500 million euro
in recent crisis situations. The loss estimate (which according to IFRS 9 has to
be based on probability-weighted outcomes) must take into account that such
an event will realize with a probability of 5-20%. Thus the fair estimate, even after
taking into account the actual positive economic outlook, is closer to 95 million
euros (= 50 million * 90% + 500 million * 10%) and may be virtually identical to the
long-term regulatory estimate.
Statistical verification of these cyclical factors and economic predictions
in credit risk through backtesting poses the next analytical challenge: risk costs
generally have to serve a variety of managerial objectives, as already mentioned
in Section 2.
To objectively test the validity of economic forecasts, undistorted and
frequently (quarterly or monthly) available estimates of realized credit risk are
necessary. One measurement to satisfy these requirements was derived in Reit-
gruber (2013), through Impact of Risk (IoR). A short introduction can be found
in appendix B of the first reference in the bibliographic reference list. The appli-
cation of this framework allows for a high frequent measurement of credit risk,
which is largely uninfluenced by manual processes. In contrast to the classical
credit risk measurement via direct write-offs or impairment changes, the timing
of risk realization is largely established by the regulatory definition of default and
the LGD parameter. It was proven that for a complete portfolio cycle, from gene-
ration to liquidation, the aggregated Impact of Risk measure is identical (except
for possible time value effects) to traditional cost of risk.
To verify the effect of external (or internal) factors on credit risk, the PL
Dashboard from the Impact of Risk framework will be primarily used. In this
measure, realized credit risk is identified through the regulatory definition of de-
fault and quantified by EAD times LGD. The regulatory definition of default is
largely standardized by definition of specific risk signals and the 90-day (or 180-
day in case) past due threshold. Consequently, we can assume a fairly homo-
genous application across different portfolios and even across financial institu-
tions. Another recent consultation paper by EBA (2014a) is attempting to further
align and strengthen the default definition although this is going to cause in-
23
consistencies in the default time series right after implementation, it will ultima-
tely support an even more objective application of Impact of Risk.
Remark 6.1: For the application of the PL Dashboard the accuracy of the
EAD and LGD estimates is less critical than the default definition: the product
EAD x LGD is practically applied only as a weighting function in a classical ba-
cktest of default rates. The validation of the LGD model itself is performed throu-
gh a separate NPL Dashboard and the EAD model is validated BOP
through the an-
EOP
PL Dashboard = EL
nual version of the PL Dashboard + wo
as outlined below EL
with PL . ou
EAD
In case of an annual observation period the PL Dashboard can be des-
EL EOP
cribed as: = ELBOP
PL wo + PL Dashboard.
BOP
PL Dashboard = ELEOP + wo ELPL ou
EOP BOP
PL Dashboard
Para separar os impactos = de PD,ELEOPEAD a+
e LGD,
wo
verso dividida ELBOP
deriva-se
ou
assim:
PL
EOP PL DashboardBOP = EL + wo EL PL ou
EL = ELPL wo + PL Dashboard.
EOP BOP
ELEOP
= ELBOP PL wo + PL Dashboard.
EL = ELPL wo +BOPPL Dashboard.
PL Dashboard = (EAD BOP
Para separar os impactos de PD, EAD e LGD, a verso dividida deriva-se assim:
LGD PL ) ELBOPPL +
Para separar os impactos de PD, EAD e LGD, a verso dividida deriva-se assim:
+Toseparar
Para separate impactsde
os impactos
(EAD
from PD,
PD,LGD
DEF eEAD
EADDEFLGD,and LGD
a verso
PL EAD
BOPthe split
LGDBOP
dividida versionassim:
deriva-se
PL ) +
is derived
as follows:
PL Dashboard = (EADBOP LGDBOP PL ) ELPL +
BOP
EOP EOP BOP DEFBOP DEF BOP
(EAD
PL+Dashboard = LGD (EAD
NPL BOP
EAD LGDBOPLGD EL
PL ) PL
) BOP
PL +
PL Dashboard = (EAD LGD PL ) EL PL +
+ (EADDEF LGDDEF PL EAD
BOP
LGDBOPPL ) +
DEF DEF BOP BOP
+ o
BOP e EOP representam comeo (EAD e oDEFfim doLGD PL
perodo
DEF de EAD BOP LGD
observao e DEF ) +o momento
PLindica
BOP
do default. +
Emprstimos (EAD
inadimplentes
EOP LGD
EOPso diferenciados
PLDEF EAD
com base
DEF LGD
em PL ) +
estarem ou no
+ (EAD LGDNPL EAD LGDPL )
inadimplentes no comeo do perodo de EOPobservao
EOP(antigos NPL
DEF ou novos
DEF NPL).
+ (EADEOP LGDEOP
NPL EADDEF LGDDEF PL )
+ (EAD LGDNPL EAD LGDPL )
BOP e EOP representam
O primeiro termo o comeo e o fim do perodo de observao e DEF indica o momento
do default. Emprstimos inadimplentes
BOP e EOP representam so diferenciados
o comeo e o fim do com basedeem
perodo estarem ou
observao no indica o momento
e DEF
inadimplentesBOP
dono e EOP
comeo
default.
BOP representam
do perodo
Emprstimos
and EOP o comeo e
de observao
inadimplentes
represent o fim do
so
the beginning perodo
(antigos NPL de
diferenciados
and end observao
oucom
novos
of the e DEF
NPL).
base indica
em estarem
observation o momento
ou
pe- no
do default. Emprstimos
inadimplentes no comeo inadimplentes
do perodo de so diferenciados
observao com NPL
(antigos base em
ou estaremNPL).
novos ou no
riod and DEF denotes the time of default. Non-performing loans are differentia-
inadimplentes no comeo do perodo de observao (antigos NPL ou novos NPL).
O primeiro
ted termo
whether they were non-performing at the beginning of the observation pe-
BOP BOP BOP
O primeiro
riod (old
NPL)termo
PDnot (new NPL).
or (EAD LGD PL ) EL PL
O primeiro termo novos defaults
The first term
descreve um retroteste de EAD com ponderao LGDBOPPL corrigida por retomadas de garantias
EAD
reais alteraes
e outras da LGD (EAD DEF
LGDDEF
imediatamente antesdo
PL EAD BOP
default. ALGD
BOP
LGDDEF )
PLPL precisa ser
haja umahaja uma diferena
diferena metodolgica
metodolgica em ao
em relao relao ao da
clculo clculo da LGD
LGD para para emprstimos
emprstimos
inadimplentes.
inadimplentes. 24
haja uma diferena metodolgica em relao ao clculo da LGD para emprstimos
O termoO termo final
final
inadimplentes.
describes an EAD backtest with weights LGDBOPPL corrected for collate-
ral repossessions and other LGD changes immediately before default. LGDDEFPL
Ohas
termo final
to be in line with the LGD method for performing loans, in case there is a me-
thodological difference to the LGD method for non-performing loans.
EOP EOP DEF DEF DEF
last
The LGD
term (EADEOP(EADLGDEOP LGD EAD EAD
NPL LGDDEF LGD
) PL )
LGD NPL PL
novos defaults
novos defaults EOP
LGD EOP
(EAD LGD
DEF NPL
EADDEF LGDDEF
PL )EOP
comparacompara o LGD imediatamente
o LGD imediatamente
novos defaults antes (LGD antes
DEF (LGD PL) e imediatamente aps
PL) e imediatamente aps (LGD
EOP (LGD NPL) o default e
NPL) o default e
deve, em geral, ser um pequeno termo de correo ou erro.
deve, em geral, ser um pequeno termo de correo ou erro. Um grande desvio pode indicar Um grande desvio podeaindicar a
descontinuidade
descontinuidade do LGD
mtodo do mtodo
deantes de
LGD (LGD LGD
logobefore
aps logo aps a ocorrncia
a imediatamente
ocorrncia do default, ou pode at destacar
compara ocompares
LGD imediatamente immediately DEF
PL) e (LGD DEF
) do
and default,
aps ou
after(LGD
(LGDpode
EOPEOP at destacar
NPL) o )default
de- e
uma em uma estimativa
estimativa incorreta incorreta
da taxa da cura
taxa de cura precoce PLda LGD. NPL
deve,
fault andgeral, ser
should um pequeno
generally adesmall
betermo precoce
de correo
correction da
ouLGD.
erro.
or Umterm.
error grande A desvio pode indicar a
large deviation
descontinuidade do mtodo deofLGD
may indicate discontinuity thelogo
LGD aps a ocorrncia
method do default,
right around andou pode
after at destacar
default, or
uma estimativa
As incorreta
medidas de da taxa
risco de de curaem
crdito, precoce
sua da LGD.se baseiam em taxas anuais de default. No
maioria,
As medidas
may de risco de
even highlight ancrdito,
incorrectem sua maioria,
estimate of se
thebaseiam
early cureem taxas anuais
rate in LGD.de default. No
caso de caso
uma de uma carteira
carteira
Measurements
suficientemente estvel,
suficientemente
of credit riskestvel,
are mostlycontudo, contudo,
based PDson PDs mensais
mensais
annualoudefault
ou trimestrais
trimestrais
rates. podem
In ser
podem ser
derivadas derivadas proporcionalmente (pro-rata) aos valores anuais. Adotada essa premissa, o PL
As medidas
case a proporcionalmente
de risco
of Dashboard
suffi de crdito,
ciently
pode stable
ser
(pro-rata)
em
portfolio
descrito como
aos
sua maioria, valores
however,
segue:
anuais.em
se baseiam
monthly Adotada
ortaxas essa premissa,
anuais
quarterly canobe
de default.
PDs PL
No
Dashboard
caso de uma
derived by pode sersuficientemente
carteira descritoannual
down-scaling como segue:estvel,Under
values. contudo, thisPDs mensais ou the
assumption, trimestrais podem ser
PL Dashbo-
derivadas proporcionalmente (pro-rata) aos valores anuais. Adotada essa premissa, o PL
ard can be described as follows:
Dashboard pode ser descrito como segue:
PL Dashboard
PL Dashboard = ELEOP
M EOP + wo ELBOP ELBOP
PLou/12 ou
M = EL +
wo
PL /12
EOP BOP
PL
ELDashboard
EOP ELEOP
= M
EL=BOP
ELEL
= /12
BOP
PL /12
wo
+ wo wo+
PL
+ EL /12 .ou M.
PLPLDashboard
Dashboard
PL M
BOP
ELEOP
AMain
= EL
principal PL /12
vantagem
advantage
umwo
of de perodo
a shorter de + PL Dashboard
observao
observation mais
period is breve M .cant
uma
a signifi reduo
reduc-significativa do
A principal vantagem de um perodo de observao mais breve uma reduo significativa do
tion
desvio desvio
of causado
the causado
deviation
por EAD por
caused EADbyem
ou LGD ouEAD LGD em
EADor e
LGDEADine LGD
LGD. Assim,
. and
Assim, as. verses
LGD Therefore,mensais ou trimestrais
the mon- do PL do PL
EADas verses mensais ou trimestrais
Dashboard
Athly
principalDashboard
vantagem
or quarterly podem
podemversions
ser ser
desuficientes
um perodosuficientes
of thepara para
de Dashboard
PL observao uma
uma anlisemais anlise
mais mais
detida
maybreve
well be detida
dos
uma dos
efeitos
suffi efeitos
cientcclicos,
reduo cclicos,
em do
significativa
for further em
contraste contraste
com com
ocyclical o
, mais , mais
preciso. preciso.
Caso Caso essa aproximao revele-se inadequada do PL a carteira
para
desvio causado thepor PD
EAD oueffects,
LGD EAD eessa
eminstead ofaproximao
.the
Assim, revele-se
as exact
verses inadequada
mensais ou this para
trimestrais a carteira
PD
analysis of LGD more PD , Should appro-
Dashboard empodem
em anlise, anlise,serser
ser necessrio necessrio
aplicar
suficientes paraaaplicar a representao
representao
umathe anlise em detida
mais emdos
componentes
componentes separados,
efeitos separados,
cclicos, descrita
em descrita
ximation prove to be inadequate for analyzed portfolio, the more detailed
anteriormente.
anteriormente.
contraste
split com o PD, mais
representation as preciso.
described Caso essa aproximao
above has to be applied. revele-se inadequada para a carteira
em anlise, ser necessrio aplicar a representao em componentes
in carteira
a stableestvel
portfolio separados, descrita
Additionally, the inflBOP uence of PLELem uma secomparada
in comparison
BOP
Alm
Alm disso,
anteriormente. a disso, a influncia
influncia de EL de ELBOP
PL em uma carteira estvel se comparada ao componente em
PL ao componente em
to the generally
geral high stochastic
altamente estocstico component
das medidas in risco
de actual de credit
crdito risk measures
efetivo costuma isserPara
baixa. Para
geral altamente estocstico das medidas de risco de crdito efetivo costuma ser baixa.
low. fins
an de
Foranlise
fins de anlise economtrica,
econometric
economtrica, a srie
analysis, temporal
the temporal
monthly mensal
or quarterly outime
trimestral
deseries deofna
perdas
formana forma
losses
Alm disso, a influncia de ELBOPaPLsrie
em uma carteira mensal ou
estvel trimestral
se comparada perdas
ao componente em
geral altamente estocstico das medidas de risco de crdito efetivo costuma ser baixa. Para
Perda
fins Perda
EL
de anlise ELNPL
nova
EOP
economtrica,
EOP
+awo
nova srie
NPL +temporal
nova
wo
NPL nova mensal
NPL ou trimestral de perdas na forma
Perda pode
may
pode ser ser
ELbe suficiente.
nova NPL + wonova NPL
EOPsuffi
suficiente. cient.
On basis of Loss, a test on the loss distribution, and in particular the sig-
nifi
Comcance
pode Com
base
ser ofbase
na Perda,
suficiente. napodemos
Perda,effects,
correlation podemos desenvolver
can
desenvolver testeum
beumdeveloped.datesteTheda distribuio
null das
distribuio dase,
hypothesis
perdas perdas
of
emthe e, em especial,
especial,
nave da significncia
da significncia
portfolio dos
model dos
efeitos
with efeitos
de de correlao.
correlao.
independent A hipteseA hiptese
binomially nula do nula do defaults
modelo
distributed modelo
simplistasimplista de carteira,
de carteira,
provides
com defaults distribudos independente e binomialmente, permite verificar a plausibilidade da
acom
Com defaults
base
check nathe
of distribudos
Perda, podemos
plausibility independente
desenvolver
of eum
binomialmente,
the volatility teste
of permite
da distribuio
actual credit verificar
risk.das
Thisperdas a plausibilidade
e, em especial,
hypothesis re- da
volatilidade
volatilidade do
de risco de efetivo.
crdito efetivo. Essa hiptese representa
pisoum piso para os efeitos
da
presents a do
significncia risco
dos
lower limitcrdito
efeitos correlao.Essa
ofdestochastic hiptese
A hiptese
effects representa
nula
in credit dorisk umsimplista
modelo
and helps para
to deoscarteira,
efeitos
separate
estocsticos
estocsticos em em risco
riscorandom de
de independente
crdito, crdito,
e ajuda e ajuda a
a distinguirdistinguir tendncias
tendncias de
deverificar resultados
resultados mensais mensais
com defaults
trends from distribudos
purely e binomialmente,
statistically signifi cant permite
month results. a plausibilidade
This property da
estatisticamente
estatisticamente significativos,
significativos, mas mas puramente
puramente aleatrios.
aleatrios. Essa Essa propriedade
propriedade a torna a torna altamente
altamente
volatilidade do risco
makes it highly de crdito
relevant for efetivo.
budgetingEssaprocesses
hiptese representa
and actualumvs piso
planpara os efeitos
performan-
estocsticos em risco de crdito, e ajuda a distinguir tendncias de resultados mensais
ce comparisons.
estatisticamente significativos, mas puramente aleatrios. Essa propriedade a torna altamente
17 17
25
plicit conservatism, compared to the total loss expectation for already defaulted
transactions, this structural conservatism in the final standard is relatively small.
The final accounting standard summarizes the following arguments for
the choice of the discount rate in its basis for conclusions: (a) the effective inte-
rest rate is the conceptually correct rate and is consistent with amortized cost measu-
rement; (b) it limits the range of rates that an entity can use when discounting cash
shortfalls, thereby limiting the potential for manipulation; (c) it enhances comparabili-
ty between entities; and (d) it avoids the adjustment that arises when financial assets
become credit-impaired (interest revenue is required to be calculated on the carrying
amount net of expected credit losses) if a rate other than the effective interest rate has
been used up to that point. The method suggested for calibration of the LGD me-
thod can be found in section 7 of the first reference, at the end of this article.
Nevertheless it should be emphasized that all Brazilian FIs are, so far,
still using standardized approach in Credit Risk, so some changes must be ap-
plied to the methodology, in a way that only EL would be backtested for both NPL
and PL.
Moreover, local rules, set by Banco Central, concerning the booking of
provisions (Resoluo 2682/99) must be updated in order to comply with IFRS 9
requirements. As is, it demands provisions for incurred losses only, without con-
sidering future impact of impaired loans. One simple approach would be to cre-
ate off-balance provisions, but it would undermine the very purpose of IFRS 9,
which is to avoid excessive profit distribution, among shareholders, of amounts
that should be posted to reserves instead. The use of such off-balance provi-
sions could be used to cap the amount of bottom-line profit that could be dis-
tributed, imposing the posting of in-balance reserves, but then other local laws
and rules (including CVM ones) come in the way. Therefore, a full locally com-
pliant methodological approach could be developed only after such updates, in
laws and regulations, effectively take place.
Meanwhile those FIs that eventually apply to IRBA approach (so are alre-
ady calculating LGD for their portfolios) could benefit from the aforementioned
methodology for risk management purposes.
Wolfgang Reitgruber
Holds a degree in Technical Mathematics (Dipl.-Ing. Dr.) from University of Technology in Vienna.
After starting in the research area of econometrics and time series analysis, he accumulated al-
most 20 years experience in analytics in the banking industry. His responsibilities covered mostly
credit risk management (up to successful A-IRB implementation in UniCredit Bank Austria AG),
but extend into finance, process and 6-sigma-quality management as well. Currently he is part-ti-
me research- and consulting oriented, covering EL Backtesting and methodological topics of IFRS
9 impairment, while keeping a senior position in credit risk modelling in the bank.
E-mail: wolfgang.reitgruber@outlook.com
References REITGRUBER, Wolfgang (2015): Methodological thoughts on expected loss estimates for IFRS
9 Impairment: hidden reserves, cyclical loss predictions and LGD backtesting. http://ssrn.com/
abstract=2641688, V2, Jan 4, 2015
REITGRUBER, Wolfgang (2013): Expected loss and Impact of Risk: backtesting parameter-based
expected loss in a Basel II framework. Journal of Risk Model Validation 7(3), 59-84
BENSTON G. J. e WALL L. D. (2005): How Should Banks Account for Loan Losses. Economic
Review Fourth Quarter 2005, 19-38
BERD, Arthur M. (2013): Lessons from the Financial Crisis 2nd Impression, Insights from the
defining economic event of our lifetime. Incisive Media, Risk Books
BIS (2013a): Directive 2013/36/EU of the European parliament and of the council of 26 June 2013
BIS (2013b): Regulation (EU) No 575/2013 of the European parliament and of the council of 26 June 2013
BCKER, Klaus (2010): Rethinking Risk Measurement and Reporting Vol II. Incisive Media, Risk Books
DIEBOLD, Francis X.; DOHERTY, Neil A.; HERRING, Richard J. (2010): The Known, The Unknown,
The Unknowable in Financial Risk Management, Princeton
EBA (2013): Third interim report on the consistency of risk-weighted assets: SME and residential
mortgages, Dec 17, 2013
EBA (2014a): Consultation paper, Draft Regulatory Technical Standards on materiality threshold of
credit obligation past due, Oct 31, 2014
EBA (2014b): Consultation paper, Draft Regulatory Technical Standards on the specification of the
assessment methodology for competent authorities regarding compliance of an institution with
the requirements to use the IRB Approach, Nov 12, 2014
GABER, Thomas (2013): Die Qualitt der Finanzberichterstattung bei Banken: Empirische
Untersuchungen zu Ergebnisqualitt und Wertrelevanz, Dissertation, Graz, August 2013
GRNBERGER, David (2012): Expected Loan Loss Provisions, Business- and Credit Cycles
(December 10, 2012). Available at SSRN: http://ssrn.com/abstract=2187515 or http://dx.doi.
org/10.2139/ssrn.2187515.
29
References HACKL, Rainer (2014): Analysis of ELNPL based on the LGD parameter: A theoretical study and
a practical back-testing approach. Master Thesis, Vienna University of Economics and Business
IASB (2009): Exposure Draft Financial Instruments: Amortized Cost and Impairment (ED 2009)
IVANTSOV, Evgueni (2013): Heads or Tails: Financial Disaster, Risk Management and Survival
Strategy in the World of Extreme Risk, Gower Publishing Ltd
MACLACHLAN, Iain (2005): Choosing the Discount Factor for Estimating Economic LGD. In:
Recovery Risk, The Next Challenge in Credit Risk Management, Edited by Altman E., Resti A.
and A. Sironi, Incisive Media, Risk Books, pp. 285-30
MIU, P., e OZDEMIR, B. (2006). Basel requirements of downturn loss given defaults: modeling and
Estimating probability of default and loss given default correlations. The Journal of Credit Risk 2(2),
43-68
MIU, P., e OZDEMIR, B. (2005). Practical and theoretical challenges in validating Basel
parameters: key learnings from the experience of a Canadian bank. The Journal of Credit Risk 1(4),
89-136
PECDC (2013a). Variation in RWA seen from the hypothetical portfolio study, Analytics Meeting
June 2013 Copenhagen, http://www.globalcreditdata.org/. Ongoing research, available to
member banks only
PECDC (2013c). PECDC Downturn LGD Study, December 17, 2013, http://www.globalcreditdata.org/
TALEB, Nassim N. (2012): Antifragile - Things That Gain from Disorder, Random House