Risk Chartis Market Report

IFRS 9
Risk.net October 2016

Lead sponsor Sponsored by

 Contents
Risk Chartis
IFRS 9
Market report
Hassan Machmouchi, Commercial Director
hassan.machmouchi@incisivemedia.com
Hugh Stewart, Research Director, Chartis
hugh@chartis-research.com
Stuart Willes, Commercial Editorial Manager
stuart.willes@incisivemedia.com
James Hundleby, Commercial Subeditor
james.hundleby@incisivemedia.com
Celine Infeld, Managing Director
celine.infeld@incisivemedia.com
Lee Hartt, Group Publishing Director
lee.hartt@incisivemedia.com
Ben Cornish, Senior Production Executive
ben.cornish@incisivemedia.com
Incisive Media (UK)
Haymarket House, 2829 Haymarket
London SW1Y 4RX
Tel: +44 (0)20 7316 9000,
Incisive Media (US)
55 Broad Street, 22nd Floor
New York, NY 10004-2501
Tel:+1 646 736 1888
Incisive Media (Hong Kong)
14th Floor (Unit 1401-3) Devon House
Taikoo Place, 979 Kings Road
Quarry Bay, Hong Kong,
Tel: +852 3411 4900
Twitter @riskdotnet
Facebook facebook.com/riskdotnet
LinkedIn Risk.net
Cover image
nednapa/Shutterstock
The Chartis content published in this report is fully
independent. The content is based on research
conducted by Chartis over the course of nine months,
by a team of senior analysts. All findings in the report
and published herein remain the intellectual property
of Chartis Research. For further materials and
research please visit www.chartis-research.com
Published by Incisive Risk Information Ltd
Incisive Risk Information (IP) Ltd, 2016
All rights reserved. No part of this publication may be
reproduced, stored in or introduced into any retrieval
system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording or
otherwise, without the prior written permission of the
copyright owners. RISK is registered as a trade mark
at the US Patent Office
1 risk.net October 2016
Overview
2 History of IFRS9 and the three-
stage approach
3 Time is tight Implementing the
three-stage approach
Rohit Verma, head of strategy for
risk analytics at Oracle, discusses
the challenges associated with the
implementation of IFRS 9
Classification &
measurement
5 The IFRS9 classification and
measurement model
6 Interview
Co-operation The new imperative
for risk and finance functions
Jean Bernard Caen, subject matter
expert, AxiomSL
8 Feature
A complex nut to crack
Move to expected loss impairment
regime brings major challenges, say
banks and accountants
Impairment
12 Impairment and the three-stage
approach The expected credit loss
model under IFRS 9
13 A strategic approach to IFRS 9
impairment
Burcu Guner, senior director at
Moodys Analytics, discusses the
factors on which success in impairment
calculation depends
Challenges
15 Building on current infrastructure
17 Roundtable
Data challenges in IFRS 9
Four experts in credit loss explore how
IFRS 9 is set to fundamentally change
the way banks do their accounting, and
the data challenges associated with its
implementatione
Model building
22 Model building for IFRS 9
Incorporating the right techniques
23 Whitepaper
Managing earnings volatility and
uncertainty in the supply and
demand for regulatory capital
The impact of IFRS 9
A novel approach to modelling from
Moodys Analytics allows better
management of the interplay of supply
and demand dynamics for regulatory
capital, combining an economic
framework with regulatory capital and
new loss recognition rules
26 Cutting edge
Loan classification under IFRS 9
IFRS 9 requires classifying non-
defaulted loans in two stages
depending on their credit quality
evolution since initial recognition by
the bank. Vivien Brunel proposes
an optimal way to perform this
classification
funda
IT infrastructure
30 Implications for IT systems
31 Whitepaper
Preparing for the IFRS 9 game-
changer How to reconcile the
demands of risk and finance on a
single, flexible platform
How the challenges to IFRS 9 for
financial firms can be overcome by
using a single, integrated platform with
a flexible data model
Profiles
34 Sponsor profiles
35 About Chartis Research Services
 Overview

History of IFRS 9 and the

three-stage approach
International Financial Reporting Standard 9 (IFRS 9) is a high-impact symbolic, operational, IT and organisational transformation
event for finance and risk: an arranged marriage that is turning an uncomfortable courtship and good intentions into a powerful
successful partnership that is greater than the sum of its parts. It is one of a few interlinked, unavoidable initiatives in finance,
regulation, compliance and risk management that are catalysts to invest in sustainable best practice

IFRS 9 has foundations in common with a number of other key regulatory

trends. Therefore, the foundations for an easier implementation of IFRS 9 can be
achieved if an organisation has performed well, for example, in implementing:
Basel Committee on Banking Supervision regulation 239 (BCBS 239) for
data management
Comprehensive Capital Analysis and Review, Dodd-Frank Act Stress Test and
European Bank Authority stress testing
Rigorous enterprise credit and counterparty risk management that is internal
ratings-based (IRB)

Imagentle/Shutterstock
Close working practices, with risk management and finance sharing a common
culture with regard to risk-adjusted performance management

Searching for solutions

Organisational support for implementing and running IFRS 9 will require
change, and through greater involvement of different departments that so
far have not been as directly active in finance activities, particularly risk and
regulatory reporting.
The marketplace, including large Tier 1 financial institutions, is turning to
software vendors for solutions. However, this new marriage of finance and risk
is not reflected in most of the software vendors previous experience. There are
very few one-stop shops that encompass the whole process from transaction
origination to audited profit and loss and balance sheet. Therefore, there are also
many integrated, multi-vendor solutions.
There are few fully complete software packages that reflect the target state
required by 2018, with deliverables still occurring during 2016, which makes
some proofs of concepts reliant on vendor credibility and trust or successes for
the early deliverers.
The complex structures of large financial institutions demand large in-house
development, implementation and operations teams, as well as extra support
from external professional advisers. All other financial institutions can rely on the
packaged software marketplace, but they require close support from the large
audit firms, as well as extra consultancy, development and integration resources.
Throughout 2016 and 2017, there will be a shortfall in suitably qualified and
experienced support services teams in this market sector, which, as mentioned
earlier, faces new methodological and organisational challenges.
Data to support impairment modelling and calculations is a critical success
factor. If not assembled comprehensively, aggregated and normalised rigorously
within a formal data management and well-engineered IT architecture,
companies results will be negatively affected. Many companies, particularly
those that have not been through the IRB experience, will have to upgrade their
IT architecture or rely on a vendors software-as-a-service infrastructure.
History and status of the IFRS 9 standard
During the financial crisis, the Group of 20 tasked global accounting standard-
setters with working towards creating a single set of high-quality global
standards. In response to this request, the International Accounting Standards
Board (IASB) and the Financial Accounting Standards Board (FASB) began
working together on the development of new financial instruments standards.
The IASB decided to accelerate its project to replace International Accounting
Standard 39 (IAS 39), and subdivided it into three main phases:
Classification and measurement
I mpairment
Hedge accounting
At the beginning of the project, the FASB and IASB worked jointly on both the
classification and measurement and the impairment projects. However, due to
lack of support for a three-stage approach for the recognition of impairment
losses in the US, the FASB developed a single measurement model, while the
IASB decided to continue with the three-stage model. In addition, the FASB
decided it would not pursue a classification and measurement model similar to
the IASB. As a consequence, IFRS 9 is not a converged standard, and therefore
not applicable under US Generally Accepted Accounting Principles US financial
firms should refer to the guidelines from the FASB.
On 24 July, 2014 the IASB published the complete version of IFRS 9,
Financial instruments, which replaced most of the guidance in IAS 39 and is
applicable to all jurisdictions operating under IFRS. IFRS 9 is effective for annual
periods beginning on or after 1 January, 2018, subject to endorsement in
certain territories.

risk.net 2
 The three-stage approach
Time is tight
Implementing the three-stage approach
A cultural shift is occurring beneath banks as they will be expected to accurately handle vast amounts of granular data and adopt
a new flexible and forward-thinking outlook. Rohit Verma, head of strategy for risk analytics at Oracle, discusses the challenges
associated with the implementation of IFRS 9
One of the aims of IFRS 9 was to reduce pro-cyclicality in
provisioning for losses on loans and other assets. How far do you
think it has achieved that ambition, or will it once it is implemented?
Rohit Verma: I think the jury is still out on that particular question, but all the
experts agree it is going to make the financial statements much more relevant
and much closer to reality than they were prior to implementation. Whether or
not it reduces pro-cyclicality, whether or not it makes it less volatile, is something
only time will tell once IFRS 9 is implemented and there are numbers available to
analyse and to see whether it has, in fact, reduced cyclicality.
It is a big standard a lot is involved and there is not much time
to implement it. Can you outline the broader standard, which splits
into classification, impairment and hedge accounting?
Rohit Verma: Classification is essentially the phase where banks identify
which assets go for fair valuation and which assets go for amortised cost. This is
important because the whole idea is that there has to be minimal opportunity
for banks to move assets around and gain some kind of an arbitrage benefit.
Phase two is impairment, where the maximum impact has been in coming up
with a completely new model for calculating the provisions. We had the incurred-
loss model that was being followed, and now banks are required to move
to an expected-loss model, which is more forward-looking and incorporates
projections about the economy into the calculation process.
The third phase is hedge accounting, where the effectiveness of a hedge is
essentially made one of the key criteria of deciding whether or not the hedge
gets used in the accounting process.
To focus on impairment thats also split into three stages?
Rohit Verma: Yes, so the first step of the impairment process is to classify
each and every account into one of the three stages: stage one is an account
that is active and is going fine; stage two is an account that is showing signs
of distress; and stage three is an account or asset that has objective evidence
of impairment.
This entire classification process is critical because the stage to which an
account gets mapped will determine the provision calculated for that account.
That is crucial you have to bring qualitative aspects into the assessment and
forward-looking projections into the assessment of a stage, as well as actual
3 risk.net October 2016
information about the counterparty that goes to determine whether an account
is stage one, stage two or stage three.
There are many manual workflow processes that need to be implemented. Its
not just a rules-based process there are people who will have to actually step into
the process and validate that the stage classification has gone as per the model.
With IFRS 9 being implemented at the start of 2018, how well
understood at this time is the three-stage process?
Rohit Verma: I think most banks have understood it pretty well; they know
what needs to be done. The challenge is in implementing the understanding,
because there are so many aspects and 18 months is just not enough time. But
from a purely understanding perspective, I think most organisations know what
they want to do.
How will it be handled if people are not already uniformly at the
starting line? Will there potentially be a kind of phased approach?
Rohit Verma: Absolutely. I think the stage identification process is going
to start off with some basic rules-based approach. The stage identification
process establishes whether there has been deterioration in the quality of the
asset from the date of its inception, except for assets that are purchased or
originated credit impaired.
This rules-based approach to identify a stage is going to evolve over time. Its
going to become more sophisticated; its going to bring more workflows into
the process so that the classification becomes consistent over time and across
portfolios. To begin with, most banks may implement a rather more simple
version of the model. Once they are comfortable they have the systems in place,
theyre going to start tweaking the model and make it more sophisticated.
How would you summarise the changes that banks will need to
make to their data internally, and the way they project to meet
that requirement?
Rohit Verma: The biggest change to data that banks will have to make is to
start handling granular account-level data. Finance organisations are typically
not used to granular data; theyre very comfortable with aggregated data, but
with the stage classification and expected loss calculations of IFRS9 they will
have to learn to handle granular data. That will be a cultural shift.

Rohit Verma
In that sense, this also provides an opportunity for banks to align risk and
finance, because risk has operated on granular data for years. Its the finance
organisation that now has to adapt, and it provides an opportunity for banks to
bring risk and finance data together into a single environment.
This is obviously going to impact on finance departments because
theyre going to need to have a more granular view of data. What
impact will the standard have on banks other business functions
such as risk, accounting, audit and governance arrangements?
Rohit Verma: Regulations such as BCBS 239 talk about bringing data together.
Not just for the sake of data, but to ensure that there is consistency in the use
of data across the organisation and that results are much more accurate and
reconciled than they were previously.
There are short-term implications and then there is the long-term strategic
impact of such a project. In the short term, what I think were going to see
is the beginning of this convergence of data across different streams risk
and finance. There is going to be more emphasis on the governance aspect
of data.
In the long term we will see the calculations or the IFRS 9 project starting
to have an impact on the business aspect as well. So, product design will
have to incorporate the impact of IFRS 9 and banks will have an opportunity
to fine-tune product pricing because the expected loss numbers are closer to
reality; there is going to be an impact on the capital plans because of the new
numbers that come out of IFRS 9. It is going to have a long-term impact on all
three of these areas.
The three-stage approach
What role will technology play in the transition to IFRS 9, and how
should banks be approaching their vendors?
Rohit Verma: Now that you have to handle millions of records as an institution,
there is no option other than having a really robust technology solution in
place to go live with IFRS 9. In the old world it was fine you were running on
largely aggregated data and you could probably run your entire provisioning
calculations on a spreadsheet. But now, having to perform granular-level
calculations, you need to have a fairly advanced solution in place if you have any
ambition of going live with the 2018 date.
As you implement this solution, a couple of things need to be kept in mind.
First is that IFRS 9 itself is going to evolve over time. There could be changes
brought in by the regulator or by the International Accounting Standards Board;
there could be changes in the model that you, as an organisation, will bring
to the calculation. As you implement the solution, you must ensure you have a
solution in place that can evolve as you, as an organisation, change the model,
or changes are pushed onto you from the external world.
So, flexibility and customisation are going to be key?
Rohit Verma: Yes, the ability to make changes and tweaks to the methodology
as we move along. The third thing of some relevance is the fact that a black box
point solution for IFRS 9 would really be a waste of an opportunity. I recommend
a fairly strategic approach to the whole project so that once you are past the
2018 deadline you can start thinking of leveraging this ecosystem to do things
such as enterprise stress testing, regulatory capital and economic capital, and in
that process comply with regulations like BCBS 239.
The way accounting standards are going to have to change and
with the impact on finance and risk, it will be quite easy to get
bogged down in compliance. From a business perspective, how
should banks be positioning themselves? What are the changes
they might need to make ahead of this standard?
Rohit Verma: I think it will give them an opportunity to make improvements
in their businesses. Take product design and product pricing as an example, in
the old world you were dependent on an incurred-loss model, which was really
backward-looking. Now we are talking about a forward-looking expected-loss
model, which is a lot closer to reality.
That means you can make the pricing a lot finer for your customers than it
was under the old approach, because the number you have for your expected
loss is a lot closer to reality and is also reflective of the kind of economy that you
may expect in the future, rather than backward-looking.
Rohit Verma
Rohit Verma is senior director at Oracle, where he is responsible for risk analytics
strategy. He currently manages strategy and customer relationships for applications
covering liquidity risk, IFRS 9 and FRTB. Previously, he was product manager
for capital adequacy, enterprise stress testing and asset/liability management
applications. In his tenure at Oracle, Rohit has also consulted on various risk and
finance projects spanning complex multi-jurisdictional enterprise programs to tactical
deployments. Prior to Oracle, Rohit held a number of positions in capital markets and
corporate banking organisations.
risk.net 4
 Classification & measurement
The IFRS 9 classification
and measurement model
IFRS 9 provides a new model for the classification and measurement of debt financial assets driven by the business model in which
the assets are held and their cash flow characteristics thus dictating the applicable accounting mechanism
Any equity financial instruments are required to be carried on the balance sheet
at fair value. The rules relating to the classification and measurement of financial
liabilities contained in IAS 39 are carried almost unchanged into IFRS 9, with one
exception relating to the recognition of own credit gain, whereby any fair value
gains and losses on liability measurement are reported.
For the mixed-measurement model, the three main accounting mechanisms are:
Amortised cost
Fair value through profit and loss (P&L)
Fair value through other comprehensive income
Organisational support for implementing and running IFRS 9 will require change,
through greater involvement of different departments that, until now, have not
been as directly active in finance activities. These particularly include risk and
regulatory reporting.
The marketplace, including some large Tier 1 financial institutions, is turning
towards the software vendors for solutions. However, this new marriage between
risk and finance is not reflected in most software vendors previous experience.
There are very few one-stop shops that encompass the whole process from
transaction origination to audited P&L and balance sheet. Therefore, there are
also many integrated, multi-vendor solutions.
5 risk.net October 2016
mighty chiwawa/Shutterstock
There are few fully complete software packages that reflect the target state
required by 2018, with deliverables still occurring during 2016, which make
some proofs of concepts reliant on vendor credibility and trust or successes for
the early deliverers.
Large financial institutions complex structures demand large in-house
development, implementation and operations teams, as well as extra support
from all of their external professional advisers. All other financial institutions
can rely on the packaged software marketplace, but require close support
from the large audit firms as well as extra consultancy, development and
integration resources. Throughout 2016 and 2017, there will be a shortfall
in suitably qualified experienced support services teams in this market
sector, which as mentioned earlier has some new methodological and
organisational challenges.
Data to support impairment modelling and calculations is a critical
success factor. If this data is not assembled comprehensively, aggregated and
normalised rigorously within a formal data management and well-engineered
IT architecture, then firms results will be negatively affected. Many firms
particularly those that have not been through the internal ratings-based
experience will have to upgrade their IT architecture or rely on a vendors
software-as-a-service infrastructure.

Co-operation The new
imperative for risk and
finance functions
In the face of pressures from the incoming IFRS 9 regime, walls between risk and finance divisions must tumble and co-operation
and communication take centre stage, says Jean-Bernard Caen, subject matter expert at AxiomSL, who fleshes out this challenge in a
conversation with Risk.net
IFRS9 has been designed to value assets and
liabilities in a more risk-sensitive manner than under
the prevailing IAS 39 framework. The pressures of
the incoming regime can only be met if the risk and
finance divisions within banks learn to work more
closely together. Presently, risk and accounting data
is aggregated separately and obeys different rules
and norms. Yet inputs from across these divisions
are needed to generate IFRS9 reports.
How would you characterise the challenge of
implementing IFRS 9 for financial institutions?
Jean-Bernard Caen: Banks have assigned a
high priority to their implementation projects
and allocated significant budgets to them.
However, these projects are not proceeding as
efficiently as they should. The major reason is the
difficulty getting risk and finance professionals
to communicate. They have been accustomed to Jean-Bernard Caen
working independently, and they have their own
language, culture and data frameworks that do not
fit easily together. This is a real constraint.
In this market landscape of analysis and regulatory compliance, financial
institutions that are successfully implementing BCBS 239 and data management
within their organisations are better prepared for IFRS 9. AxiomSL enterprise
data management capabilities include the ability to map, aggregate and enrich
source data and models across risk and finance, automate workflow process,
reconcile and validate, as well as delivering visualisation, collaboration and
reporting tools.
How will the relationship between the chief risk officer (CRO) and chief
financial officer (CFO) need to change in an IFRS 9 environment?
Jean-Bernard Caen: They will have to learn to communicate. At present,
they have very clear and separate responsibilities. The CFO signs off on the
accounts and is used to talking in terms of assets and liabilities precise
values. In contrast, the CRO is in charge of ensuring that the firm is compliant
Classification & measurement
with regulatory requirements and he speaks the
language of probabilities and statistics. Up to now,
accounting has been mainly backward-looking;
with the introduction of risk sensitivity, forward-
looking risk items will change the nature of
accounting. Under IFRS 9 they will have to
co-operate. The CRO will be delegated
responsibility to set up the credit risk models to
assess asset impairments, while the CFO will take
charge of mapping those model outputs onto the
published accounts.
At present, accounting is straightforward for an
asset you book the initial value and, under certain
conditions, show amortisation. Under IFRS9,
accounting becomes much more complex. The
value of an asset will be the discount of projected
cashflows. Moreover, these projections, will need to
exhibit a granularity beyond what the risk function
is used to producing under the exposure-at-default
approach. Both functions will have to embrace this
new logic and come to a shared understanding.
What are the challenges involved in producing the correct assessment of
expected credit loss?
Jean-Bernard Caen: With IFRS 9, the International Accounting Standards
Board is demanding that firms take an economic approach to credit risk.
This means calculating the risk of loss on an asset up to maturity with
proper recovery estimates and a forward-looking stance, rather than relying
on simplistic models, such as the regulatory internal ratings-based model.
Essentially, banks must completely review how they assess credit risk an
annual review of credit ratings will no longer be sufficient.
The second challenge will be to assess the probability of default to maturity,
rather than simply to a one-year horizon, which requires access to vast amounts
of data to generate these estimates. The third challenge will be for banks to
include forward-looking scenarios, as under IFRS 9 they will have to be explicit
about how they see the economy evolving.
risk.net 6
 Classification & measurement
What should financial institutions prioritise: modelling expected losses
under IFRS 9 or ensuring data governance is up to scratch?
Jean-Bernard Caen: Data. Without the right data, model outputs will be
flawed. Modelling is more sexy. Banks have plenty of people skilled at
modelling and too few dedicated to data management. The large amounts of
data needed to satisfy IFRS 9 requirements are either not stored in firms existing
databases or, if they are, have not been used for years and may prove unreliable.
A focus on modelling can set banks down the wrong path. For example,
they have to transition from using their one-year through-the-cycle probability-
of-default models to point-in-time models and lifetime models for assessing
impairments. However, what I have learned from industry surveys is that many
banks are considering making this transition by merely installing a multiplier into
their one-year models. That is not the right way to go about this. The right way
is to start with the data analysing the term structure of projected defaults by
product and counterparty type, and sourcing useful economic data to produce
forward-looking scenarios. Some banks are considering asking rating agencies to
provide this data, which is a possible first step.
What is the relationship between BCBS 239 and IFRS 9, and how should
institutions factor this into their IFRS 9 implementation plans?
Jean-Bernard Caen: BCBS 239 sets out rules and principles that banks must
apply to their risk data aggregation processes and reporting. To be compliant,
banks must adopt processes to ensure the quality and reliability of their risk
data, conduct more granular analyses of this data and increase the frequency of
data reviews. In short, firms have to update their IT infrastructures.
However, this revamp does not apply to accounting data. Yet IFRS 9 will
require the use of data that is being constrained by BCBS 239. Once again, this
means there will need to be greater co-operation between risk and accounting
functions. The ideal scenario is one where BCBS 239-constrained data is easy for
the accounting function to access and any questions as to the origin of the data
are readily answerable.
How can risk and accounting data be reconciled to fulfil IFRS 9
requirements?
Jean-Bernard Caen: This data will always be articulated differently. The goal
should be for banks to be able to articulate accounting data with risk data.
This requires a uniform segmentation of assets and liabilities into products and
portfolios one that is coherent across risk and finance.
From a technology perspective, the IFRS 9 requirement to converge risk and
finance environments is now more compelling than ever. Duplication of work
such as parallel processes making the same set of calculations without co-
ordination with other teams should be minimised, and these processes should
be automated to improve the soundness and reliability of the banks results.
AxiomSLs data-driven structure enables banks to adopt a strategic
posture beyond merely complying with minimum reporting and compliance
requirements. AxiomSLs change management platform acts as a catalyst to
move to a converged risk and finance data environment, where close co-
ordination with other data-related initiatives such as Basel III, stress testing,
BCBS 239, Comprehensive Capital Analysis and Review regulation, and the
Fundamental Review of the Trading Book provides senior managers with the
ability to monitor redesigned processes and enforce control frameworks at
business unit and product levels.
7 risk.net October 2016
How important is it for institutions to be able to trace the lineage of data
inputs into their IFRS 9 systems?
Jean-Bernard Caen: The way most banks work today is to originate data in
their production systems and run them through an extract, transform and load
(ETL) process. The data is then fed into an intermediary database from which the
bank produces its regulatory and management reports. The problem with this
approach is that it is difficult to maintain data quality and lineage when data is
transformed by the process. Furthermore, if there is a change at the production
system level, then the ETL processes have to modified accordingly, meaning that
frequent and painful reviews are the norm.
This also means that the data trail can go cold. If a production system
is changed, and data generated under the previous system is warehoused
separately, you lose the ability to data mine and trace that data to its original
source. This limits the flexibility and resilience of the data environment.
AxiomSL delivers a change management approach that demonstrates full
data and process lineage across enterprise core financial, treasury, operations
and risk management functions. Generated reports can be programmed so
a user can trace a data point straight to the initiating production system. Put
simply, you can take any current or historical report and determine where the
data originally comes from, allowing for seamless data mining. You must still
reconcile changes in the production system with the reporting system, but this
automation makes isolating a data points lineage much simpler.
Today, banks face important choices about how to respond strategically to
the convergence of risk and finance data environments. They have to reduce
business-as-usual costs and achieve regulatory and accounting compliance, all
the while enhancing business decision processes; adopting temporary tactical
solutions will impose structural drags on a firms ability to meet its expected
results and manage its future obligations.
About the author
Jean-Bernard Caen is a subject matter expert at AxiomSL specialising in risk and
finance. He is responsible for supporting AxiomSLs client base in analytical and
reporting issues on IFRS 9 and providing direction and insights into the firms
professional services and product development teams.
Prior to joining AxiomSL, Jean-Bernard was head of economic capital and
strategy for Dexia Group for 12 years, where he was in charge of Basel II and Pillar 2
implementation, and risk/finance co-operation.
In 1990, Jean-Bernard founded Finance & Technology Management, a management
consulting firm he ran for 12 years as chief executive officer. As such, he directed
numerous assignments for European financial institutions in the areas of shareholder
value, risk management, capital allocation and asset-liability management.
Jean-Bernard is a member of the Professional Risk Managers International
Association France executive committee, of the Association Francaise de la
Gestion Financire management board and teaches at the French National School
of Economics and Statistics. He is a French civil engineer and graduated from
Massachusetts Institute of Technology.

Image: Lightspring/Shutterstock
A complex nut to crack
Move to expected loss impairment regime brings major challenges, say banks and accountants. By Michael Hegarty
Need to know
F rom January 2018, IFRS9 will usher in a
forward-looking expected loss
accounting regime for assets subject to
impairment, suchas loans.
IFRS9 will increase banks loan-loss
provisions, but it is also proving tough to
implement due to the wide-ranging
changes needed and a lack of detail on
how this should be done.
Broadly, banks approaches fit into two
categories: some are choosing to use
complex models involving Monte Carlo
simulations, while others are assigning
weights to future scenarios using expert
judgement.
The whole thing is crystal ball gazing at
theend of the day, complains one
IFRS9 project manager at a large
European bank. The longer you go out,
the more uncertain it becomes, and if
everyone had perfect hindsight wed all
be on a beach drinking tequila.
Some banks and policy experts speculate
that a three-stage system for classifying
assets under IFRS9 could increase
secondarymarket liquidity in loans.
With perfect hindsight, banks might have been
spared the pain of the financial crisis. But in the
absence of a crystal ball, magic mirror or time
machine, such a level of clairvoyance is hard to
achieve. In its absence, banks and regulators have
searched for ways of making sure the possibility
of future downturns is considered for instance,
by using counter-cyclical capital buffers, more
pessimistic modelling assumptions and regular
supervisory stress tests.
Accountants are also doing their bit. From
January 2018, IFRS9 will usher in a forward-looking
expected loss accounting regime for assets subject
to impairment, such as loans. Theidea is to force
banks to consider the impact of potential adverse
scenarios before they occur and ensure adequate
reserves are set aside to cover them. That would be
a contrast to banksresponse to the 2008 financial Its frustrating in some sense that were at this level
crisis, when they were criticised for being too slow
to recognise losses.
Once implemented, IFRS9 is expected to
significantly increase banks loan-loss provisions
(see box: The capital crunch). But even getting
to that point is going to be tough, say banks.
Forone thing, there is confusion over what will
pass muster under the rules, with the largest
global accounting firms said to be offering varying
interpretations. Putting the standard in place will
Classification & measurement
mean revamping accounting systems and the
way banks are organised, and it could also entail
modifying credit risk models, hiring quants and
improving governance.
I dont see much of an industry convergence,
says Wolfgang Reitgruber, head of credit risk
modelling at UniCredit in Austria. I would expect
accountants will see pretty diverse environments
after 2018 so lots of different approaches, which
will make it increasingly difficult to compare one
bank to the other.
Sources close to the International Accounting
Standards Board (IASB) say the principles-based
nature of the standard means it is up to banks,
regulators and auditors to agree on best practice.
But for one IFRS9 project manager at a large
European bank, that approach brings little comfort.
of the project and suddenly these issues are still
dangling like this with 18 months to go-live, he says.
The new expected loss regime is among a
number of changes in IFRS9 that are expected
to have a big impact on banks. The new standard
is the culmination of a decade-long project to
reform accounting for financial instruments, which
covers three areas: classification and measurement;
impairment; and hedge accounting. Impairment is
the last ofthese to beaddressed.
risk.net 8
 Classification & measurement
Originally, IFRS9 was a joint project between
the two major accounting standard-setters the
IASB and the US Financial Accounting Standards
Board (FASB) and part of a wider attempt at
convergence between US and global accounting
rules. But following disagreements between the two
groups, the FASB has since been working on its own
version of the standard that includes its own rules
on impairment, which are expected to be unveiled in
the first half of this year.
Under the IASBs prior standard on accounting
for financial instruments, known as IAS 39, losses
on financial assets subject to impairment are not
recognised until there is evidence that they have
become impaired. IFRS9 represents a radical
departure from this philosophy, forcing banks to
make greater and earlier provisions against losses.
Under IFRS9, banks will have to immediately
set aside 12-month expected credit losses from
the time any unimpaired asset is originated or
purchased. They must then track the assets
changing credit risk at each financial reporting date
using a three-stage process.
For stage one assets, banks would set aside
12-month expected credit losses and calculate
interest revenue based on the gross carrying amount.
If a stage one asset were to undergo a significant
increase in credit risk, it would move to stage two and
the bank would have to begin setting aside expected
losses for the entire lifetime of the asset. If the asset
were to become credit impaired as reflected by a
missed payment or a broken covenant, for example
it would move into stage three. In addition to setting
aside lifetime expected losses, stage three requires
that interest revenue is calculated on the net carrying
amount, meaning expected losses must be taken into
account (see table A).
Once IFRS9 is in place, some banks and policy
experts believe this three-stage approach could
encourage banks to trade their loan portfolios,
because stage two assets may revert back to stage
one by being purchased by a rivalbank, allowing
the buyer to take lower lossprovisions (see box:
IFRS9: good news for loan liquidity?).
Where things get complicated is when working
out expected losses. IFRS9 says banks should
provide an estimate of expected credit losses
to reflect an unbiased and probability-weighted
amount that is determined by evaluating a range
of possible outcomes. Moreover, banks will need
to back their estimates up by considering all
reasonable and supportable information, including
that which is forward-looking.
It is not clear what this means in practice. Some
9 risk.net October 2016
A year and a half before go-live there was a three-inch-thick Basel II bible
that everybody had sitting on their desk, and they could go and check
the specific requirements. IFRS9 is not as far along and the key parts are
written down on two pieces of paper Scott Aguais, credit risk consultant
had hoped a single best estimate of expected losses A year and a half before go-live there was a
might fit the bill, but those expectations were dashed three-inch-thick Basel II bible that everybody had
at a December 2015 meeting of the IFRS Transition sitting on their desk, and they could go and check
Resource Group for Impairment of Financial the specific requirements, he says. IFRS9 is not as
Instruments a discussion forum for banks, auditors far along and the key parts are kind of written down
and the IASB, which seeks to tackle implementation on two pieces of paper.
challenges. The group concluded that banks should It means banks are pursuing different ways of
incorporate multiple scenarios, rather than the most generating unbiased and probability-weighted
probable future one, because loan losses would be estimates of expected losses. Broadly, the
distributed asymmetrically around the most likely approaches fit into two categories: some are
outcome. In other words, a slightly more negative choosing to invest in complex models using Monte-
economic outlook would imply a larger increase in Carlo simulations to estimate losses using forward-
defaults than the decline that would be experienced looking information, while others are assigning
with an equally positive forecast. weights to various future scenarios, using expert
Its going to be unlikely in many cases that judgement and techniques borrowed from scenario
you can just have a single best estimate forecast, analysis and stress-testing.
affirms Chris Spall, London-based global leader Aguais and other credit risk modellers advocate
on IFRS financial instruments at accountancy firm using Monte Carlo simulations as the purists
KPMG and a member of the IFRS group. Youre solution to IFRS9. They argue the techniques ability
going to have to think about what the other to simulate a huge number of potential future
possibilities are and probabilities that attach to outcomes and come up with best estimates of credit
them, and to what extent those different outcomes risk metrics stands a better chance of meeting the
could have an asymmetric impact on your estimate. standards language.
The group reiterated the principles-based Monte Carlo simulations involve random
nature of IFRS9, and said banks should consider sampling being carried out many times over, based
information from different sources and look at on historical data, allowing thousands of potential
several different scenarios. However, it stopped future scenarios to be mapped, with expected losses
short of specifying exactly how many scenarios they calculated for each one. The probability of these
should use, saying instead that arepresentative expected losses would be incorporated and all
sample would be needed. the losses aggregated to give a single, probability-
It is the lack of detail in IFRS9 that is weighted average.
confounding banks, says Scott Aguais, a London- Notwithstanding this, most banks that spoke to
based credit risk consultant who helped to Risk.net had gone down the second, judgement-
build Basel II credit risk models at Barclays and based route. Both types of approach entail
Royal Bank of Scotland. Compared with the challenges. There are lots of ideas and lots of
implementation of Basel II, for example, thelevel things being tried, but I dont think there is any
of guidance being offered to banks on IFRS9 industry consensus yet on the right way to do it,
compares poorly, he says. says Tom Millar, London-based accounting partner
A The three stages of IFRS9
Stage 1 Stage 2 Stage 3
12-month expected losses Lifetime expected losses Lifetime expected losses
Interest calculated on gross Interest calculated on gross Interest calculated on net
carrying amount carrying amount carrying amount
Source: IASB

and head of the global IFRS banking survey at
Deloitte. Its extremely difficult to practically do
something that is as detailed, granular and data-
based as they might like to do, partly because of
the time, and partly because information and the
methodologies simply dontexist yet.
While both approaches require changes to
accounting systems, firms choosing to use Monte
Carlo simulations must ensure they have the right IT
in place to do the computational heavy lifting. The
use of complex modelling techniques may appear
to be a black box to senior management, says
the IFRS9 programme manager, who warns that a
reliance on historical data could make the models
vulnerable to future market crises.
On the other hand, banks going down the
judgement-based route also have work to do. Vivien
Brunel, head of risk and capital modelling at Societe
Generale in Paris, says there is a debate going on
about how to blend several different scenarios into
the banks models with the required probability
weighting. In theory [the forward-looking
information] should account for all the possible
outcomes in any type of scenario, but its very difficult
to implement like this, because you need to assign
probabilities to the future economic scenarios, which
is difficult in terms of model accuracy, hesays.
CAPITAL IMPACT
The immediate impact of IFRS9 will be to increase the
provisions held by banks against future losses. In a
speech to a London conference in September last year,
Hans Hoogervorst, chairman of the International
Accounting Standards Board, said the increase would
be of the order of around 35%, although this is
expected to vary widely between banks.
Critically, the effect will also depend on how the
regulators respond to the new standard.
Banks using the Basel II internal ratings-based (IRB)
approach to credit risk can obtain limited relief against
loan-loss provisions. Using a mechanism known as the
provision miss, banks must compare their expected
losses over a 12-month period with any accounting
provisions that correspond to those losses. Banks with
provisions larger than their expected losses can recog-
nise the difference in Tier 2 capital for up to 0.6% of
their credit risk-weighted assets, although this number
may be constrained by their national regulator. Con-
versely, if banks 12-month expected losses are larger
than the relevant provisions, they must put aside
additional Tier 1 capital to cover the difference.
Under IFRS 9, banks say the requirement to set
There are huge amounts of new risk and estimation processes and revised
governance procedures that need to be added, because of the judgemental
nature of forward-looking inputs David Schraa, Institute of International Finance
At the moment, banks tend to base their loan-
loss estimates on a single baseline scenario, so
working with multiple visions of the future is a
new frontier for bank economists and one that
makes some uncomfortable. Our economists
are saying they are unable to assign these
probabilities, says Brunel.
Once banks have come up with probabilities
for various scenarios, they must decide how
toincorporate these as expected losses. One
approach being taken is to attach probabilities
toloan-loss metrics, such as probability of default
(PD) and loss-given default (LGD). Another is the use
of probabilities to weight macroeconomic variables.
To obtain unbiased estimates from a panel of
economists, a strict challenge process is needed,
say experts. There are huge amounts of new risk
and estimation processes and revised governance
procedures that need to be added, because of the
judgemental nature of forward-looking inputs,
aside lifetime expected losses for assets that are im-
paired could leave them out of pocket, because only
provisions applying to 12-month expected losses are
considered. Banks using the standardised approach to
credit risk would fare even worse, because there is no
provision-miss calculation in the simpler rules.
What we will have is no recognition of these
additional provisions that have been taken, other than
the potential under the current rules for excess provi-
sions, says Adrian Docherty, head of the financial
institutions advisory team at BNP Paribas in London.
My current understanding is that IFRS9 will deplete
common equity Tier 1 capital dollar for dollar.
Banks say the Basel Committee on Banking Super-
vision should change its credit risk rules to account for
the mismatch, but the committee is not expected to
take any action until after the Financial Accounting
Standards Board (FASB) has released its own impair-
ment rules in the US expected in the first half of
thisyear.
On January 27, the European Banking Authority
(EBA) launched its own impact assessment of IFRS9.
David Grnberger, head of IFRS9 enforcement at the
Classification & measurement
to have unbiased estimates, says David Schraa,
regulatory counsel at the Institute of International
Finance in London. Its a very complex nut.
Potentially, one way to make IFRS9 compliance
easier is to reuse existing credit risk models.
Banks following the internal ratings-based (IRB)
approach to Basel II use several standard credit
risk parameters, such as PD, LGD and exposure-at-
default. The expected credit loss is the mathematical
product of these measures.
One problem is that the numbers used for
regulatory capital may not be entirely suited to
IFRS9. KPMGs Spall admits it is a massive effort to
re-engineer the systems to get banks from existing
Basel II IRB models to what is required by IFRS9.
For instance, the IRB approach requires the
use of through-the-cycle PD, which attempts to
average out the peaks and troughs in default rates
caused by the credit cycle. Under IFRS9, banks
will have to use point-in-time numbers, which can
Austrian regulator, Finanzmarktaufsicht, who is closely
involved with the EBAs work on IFRS9, says the exer-
cise will involve hundreds of Europes largest banks
and will be used by the Basel Committee to help
inform its decision on future capital rules. The first re-
sults are expected in mid-2016, he says, with a second
round of results probably in early 2017.
Grnberger is adamant that any changes to cred-
it risk capital rules must be finished by the time
IFRS9 is implemented, and says the European Union
should beprepared to go it alone if necessary. If
the Basel Committee does not do it in time we will
do it on a European scale, he says. There is pres-
sure on the Basel Committee to solve the issues dur-
ing the next year or so, because IFRS9 is going to be
implemented in less than two years time and we
need a solution.
If the committee fails to act, a possible EU solution
might involve narrow amendments to the Capital
Requirements Regulation or guidance from the EBA
on how regulators should interpret existing rules,
saysGrnberger.
The Basel Committee declined to comment.
risk.net 10
 Classification & measurement

IFRS9: GOOD NEWS FOR LOAN LIQUIDITY?

Some banks and policy experts speculate that the
three-stage system used under IFRS9 could increase
secondary market liquidity in loans.
Under IFRS 9, banks must set aside 12-month
expected credit losses from the moment an unim-
paired asset is originated or purchased. If it undergoes
a significant increase in credit risk, the asset moves
from stage one to stage two, and the bank will have to
begin setting aside expected losses for the entire life-
time of the asset.
If the asset were to become credit impaired as
reflected by a missed payment or a broken covenant,
for example it would move into stage three.
Moving from stage two to stage three has no im-
pact on expected losses, but means interest revenue
must be calculated on the net, rather than gross, carry-
ing amount.
Theoretically, if a bank held a loan in stage two and
it was purchased by another bank, it would be reset to
stage one. This would allow the purchasing bank to
setaside 12-month expected losses and not the full
lifetime provisions that would have been held by the
selling institution.
The innovation here is that if someone steps in
the new owner of the asset potentially has a dramati-
cally different expected loss calculation than the origi-
nator, says a US-based policy expert and former
government official.
Some say the resultant increase in loan liquidity
could be a positive development for European banks,
which have struggled to shed their non-performing
loans (NPLs) in recent years.
If youre looking to incentivise banks to unload
non-performing assets there has to be a mechanism
to ensure the acquisition of NPLs and related provi-
sioning has some benefit. That 12-month reset on
acquisition could be very useful in incentivising asset
acquisition at a discounted rate, adds the
policy expert.
Adrian Docherty, head of the financial institutions
advisory team at BNP Paribas in London, says some
more sophisticated European banks are already
preparing loan portfolios to be traded as a result of
IFRS 9. However, he refuses to be drawn on which
particular firms are involved. He says the rules on life-
time provisions will incentivise banks to hold loans in
the trading book rather than the banking book, and he
expects some trades involving these assets to emerge
as early as next year.
Greater liquidity in NPLs is something that would
be welcomed by European regulators, says David
Grnberger, head of IFRS 9 enforcement at Finanz-
marktaufsicht, the Austrian regulator. But he does not
believe the impact will be as significant as others hope.
High transaction costs are likely to impede liquidity, he
says, while there is also a problem of information
asymmetry between buyers and sellers.
For us, as banking supervisors, this would be a nice
result, because we like higher liquidity on loan mar-
kets, but I am not sure whether this is going to hap-
pen, Grnberger says. It might, in some areas or
niches, produce a market, but I dont expect this to be
a really striking effect of IFRS9.

make a huge difference.
If those commercial and corporate [through-
the-cycle] PD models that everybody is using for
Basel II dont reflect the credit cycle, they arent fit
for purpose for IFRS9, says Aguais. Your credit
losses can vary across the credit cycle by a large
factor of up to 10 times relative to using [through-
the-cycle] models.
Obtaining point-in-time numbers from through-
the-cycle models means building the credit cycle
into them by altering the core econometrics or
overlaying cyclical variations on to the model by
adjusting its outputs. Both will require more work
for banks, but Aguais says adjusting the core
econometrics is a tougher job than adding
an overlay.
In a similar way, some banks taking the
judgement-based route are looking to make
use of work they have already completed for
supervisory stress tests, such as those required
annually by the US Federal Reserve Board and
the European Banking Authority. Such firms
arecentralising their pool of future scenarios so
they can dip into them whenever they need to
whether for stress testing or IFRS9, says Burcu
Guner, a London-based director in the stress-
testing team at Moodys Analytics.
But here, as elsewhere, many of the figures used by
banks are through-the-cycle, as opposed to point-in-
time. I spent most of my career trying to document
what I think is a point-in-time versus a through-the-
cycle probability of default, but not all banks have the
same definition, saysthe IFRS9 project manager.
And, for a lot of banks, the whole concept of point-in-
time risk measures is quite alien tothem.
Whichever path banks choose, it is clear there
is a lot of work to do, and its important to find
the right people to do it. But this is a struggle, say
firms. Most banks are investing tens of millions
of pounds over this four-year effort, and most of
that is going on talent and skills, says the IFRS9
project manager. Resourcing is one of the biggest
challenges at the moment, because everyones
competing at the same time for what is a very niche
pool of quantitative skills.
Once IFRS9 is implemented, will all of the
effort be worthwhile? Banks and industry
observers seem sceptical. Although the aim of the
IFRS9 impairment rules is to encourage faster
recognition of possible future losses, UniCredits
Reitgruber warns not to expect too much from the
resulting predictions.
My personal opinion is that accountants are
overly optimistic with respect to achievable model
quality, he says. They strongly focus on forward-
looking [scenarios], as if anybody would know if the
next year is going to be a crisis or not.
The large European banks IFRS9 project
manager puts it more bluntly: The whole thing
is crystal ball gazing at the end of the day. The
trouble with forecasting is the longer you go out,
the more uncertain it becomes, and if everyone
had perfect hindsight wed all be on a beach
drinking tequila.
Now the IASB has spoken, banks say they
urgently need more clarification from regulators
on which approaches to modelling would be
acceptable. With the January 2018 deadline
looming, firms say they do not want to invest time
and money developing new systems and processes
only to find they dont meet the supervisors
interpretation of IFRS9.
Clearly, the timelines are very short, says
theIFRS9 project manager. Thats the big concern
of banks at the moment: there isnt much room
for manoeuvre if the goalposts were to change
significantly over the course of the next six months.
Deloittes Millar agrees there is more that
regulators and accountants can do. What I hope
we are going to see is more guidance from the
accounting firms, [and] hopefully a slightly more
practical approach from the regulator on what
needs to be done in order to comply.
Previously published on Risk.net

11 risk.net October 2016


Impairment and the
three-stage approach
A summary of the expected credit
loss model under IFRS 9
The revision of hedge accounting rules, the restructuring of credit risk models and a new approach to impairment measurement
market participants must prepare for the transformational effects of IFRS 9
The expected credit loss (ECL) model constitutes
a significant change, which seeks to address the
criticisms of the incurred loss model. Entities will be
required to record impairment almost immediately
equal to the 12-month expected loss after the
initial recognition of financial assets that are not
credit-impaired.
ECL forecasts a probability-weighted estimate
of credit losses. A credit loss is the difference
between the cashflows that are due to an entity
in accordance with the contract and the cashflows
that the entity expects to receive discounted at
the original effective interest rate. Firms should
discount the cashflows that they expect to receive
at the effective interest rate determined at initial
recognition, or an approximate figure, in order to
calculate ECL.
An ECL estimate of loan commitments should be
consistent with expectations of drawdowns on that
loan commitment. Management should consider
the expected portion of the loan commitment
that will be drawn down within 12 months of the
reporting date when estimating 12-month ECL. It
should also consider the expected portion of the
loan commitment that will be drawn down over the
expected life of the loan commitment.
IFRS 9 contains a three-stage approach based on
the change in credit quality of financial assets since
initial recognition. Assets move through the three
stages as credit quality changes, and the stages
dictate how an entity measures impairment losses
and applies the effective interest rate method.
Where there has been a significant increase in
credit risk, impairment is measured using lifetime
ECL rather than 12-month ECL with operational
simplifications for lease and trade receivables
(figure 1).
The standard requires qualitative management
when determining whether the credit risk on a
financial instrument has increased significantly
by considering all reasonable and supportable
Impairment
1 The three-stage approach
Stage 2
Assets where there Assets where there is Assets where there is
is no identified credit deterioration in credit more than significant
deterioration since quality since initial deterioration in credit
initial recognition recognition, but where quality since initial
there may not be an recognition, and there
objective evidence of is objective evidence
Source: Chartis, IFRS 9 Technology Solutions, 2016
impairment of impairment
Interest on gross Interest on gross
carrying amount carrying amount
Stage 3
Stage 1
12-month ECL Lifetime ECL Lifetime ECL
Increase in the probability of default
information available. Financial institutions are Firms can group financial instruments on the
required to gather significant historical data about basis of shared credit risk characteristics such
their credit exposures to enable application of the as instrument type, credit risk ratings, remaining
relative credit quality assessment. term to maturity, industry, and so on. Determining
It is important to note that the credit risk of appropriate segmentation of credit exposures
the instrument needs to be evaluated without based on shared risk characteristics is a very
consideration of collateral. This means that important element of the application of IFRS
financial instruments are not considered to have requirements.
low credit risk simply because that risk is mitigated Extensive disclosures are required to conform to
by collateral. IFRS 9 ECL, including:
The standard requires the use of both forward- Reconciliations from opening to closing amounts
looking and historical information to determine if of the ECL provision, assumptions and inputs.
a significant increase in credit risk has occurred. A reconciliation on transition of the original
Lifetime ECL is expected to be recognised before a classification categories IAS 39 to the new
financial asset becomes delinquent. classification categories in IFRS 9.
The model can be applied granularly at an
individual or portfolio level. However, some Auditors and main boards are emphasising that the
factors or indicators may not be identifiable at required data and systems for all these needs will be
an instrument level. In such cases, the factors or critical to ensure the completeness of IFRS9 project
indicators should be assessed at a portfolio level. planning, implementation and production.
risk.net 12
 Impairment
A strategic approach to
IFRS 9 impairment
Success in impairment calculation depends on flexible modelling techniques, a scenario-driven approach to forecasting and the
transition towards a traceable and controlled technology architecture, writes Burcu Guner, senior director at Moodys Analytics
What is IFRS 9 and why is it important?
IFRS 9, and particularly the new impairment standards, are a
response to the last financial crisis and reflect the intention
of the International Accounting Standards Board (IASB) to
overcome the too little, too late recognition of credit losses
inherent in IAS 39.
IFRS 9 impairment introduces new forward-looking
expected credit loss (ECL) models that will require more
timely recognition of changes in ECLs, and require institutions
to account for them from the point at which a significant
deterioration of the credit quality occurs.
From our experience on IFRS 9 engagements, the
importance of IFRS 9 is twofold:
While provisions are expected to increase significantly, we Burcu Guner
anticipate the impact on earnings level and volatility to
be non-trivial, leading to implications on loan pricing and
availability of credit. Firms will need to think about stakeholder management
ahead of the January 2018 deadline, as well as in an
ongoing capacity.
S ignificant multi-year implementation efforts will require a rethink of and
changes to the data, systems, models and validation as well as to future
monitoring, reporting and business decision-making.
What are the biggest operational challenges in IFRS 9?
Based on our observations, there are two key operational challenges:
The tactical challenges in achieving compliance by the January 2018 deadline.
Banks are addressing this by leveraging and ensuring scalability of existing
internal or external capabilities, such as using existing off-the-shelf modelling,
data and calculation capabilities to meet the tight deadlines.
The strategic challenges to meet the ongoing monitoring, reporting and
validation requirements of the new standard, as well as to manage the
anticipated impact on business and earnings volatility through advanced
analytics and infrastructure.
In an age of increasing regulatory demands and new accounting standards,
coupled with declining margins and low profitability, many banks of all sizes
are seeing the benefits of the Moodys Analytics suite of IFRS 9 solutions as an
opportunity to minimise manual intervention and improve cost savings while
enhancing analytics.
13 risk.net October 2016
What broad approaches are firms taking to
incorporate forward-looking information about
expected losses and, in particular, the required use
of multiple economic scenarios?
The success in IFRS 9 forward-looking impairment calculation
depends on flexible modelling techniques, forward-looking
models and data available for model development and
benchmarking, all of which need to be compliant with the
regulatory and audit requirements. To ensure consistency
across all scenario-based aspects of IFRS 9 and other risk
management concerns, we propose that firms put in place a
flexible modelling approach where models and scenarios can
be leveraged for both IFRS 9, the internal capital adequacy
process (ICAAP), stress testing and other strategic, capital and
business forecasting and planning purposes.
There is increasing scrutiny of the ongoing validation and
independency requirements as to the multiple scenarios used for IFRS 9, and this
is making firms seek external help in the construction of such scenarios as well
as their validation.
To ensure forward-looking estimations, firms should initially start by
estimating point-in-time (PIT) credit measures. Typically, firms tend to have the
through-the-cycle (TTC) measures that they wish to convert to PIT measures by
credit cycle-driven conversion/adjustment factors.
Moodys Analytics has worked on wholesale portfolios where it was able to
convert TTC measures into PIT by using country- and industry-specific credit
cycles. In the case of retail portfolios, vintage elements can be further analysed
to apply conversions. The vintage approach can be used at both granular and
portfolio segment level, depending on data availability.
When firms do not have TTC measures due to data scarcity, we overcome the
issue by using industry standard off-the-shelf models. In most cases, we added
expert-driven elements to the off-the-shelf scorecards, as well as specific factors
affecting the credit quality through the clients internal data, when available.
For forward-looking lifetime measures, we observe banks using a scenario-
based model to convert 12-month probability of default/loss-given default (PD/
LGD) to conditional PD and LGD term structure for a single and/or multiple set
of macroeconomic scenarios. These conditional PD/LGD values can be applied
for IFRS 9 stage allocation and ECL calculations. Most clients capture portfolio-
level default and migration dynamics across different macroeconomic conditions
when projecting the lifetime ECL estimations.

IFRS 9 requires firms to use multiple scenarios to produce probability-
weighted lifetime ECLs. To help firms comply with this requirement, firms should
produce multiple, fully fledged, upside and downside economic scenarios that
align with the scenarios probability distribution and our deep understanding of
the global economy and potential key threats. These scenarios need to extend
through long future horizons to satisfy the IFRS 9 lifetime requirements.
Another approach being taken is the scenario-driven approach to
forecasting this begins with our baseline forecast. We define this as the most
likely outcome based on current conditions and our view of where the economy
is headed. From this, we develop the basic outlines of our alternative scenarios
by running multiple simulations to develop a probability distribution of economic
outcomes. As a result, this allows for the identification of scenarios that are
associated to customer-defined percentiles.
What questions surround the three-stage impairment approach?
One of the most important dimensions of the new accounting standard is the
definition of the transferring criteria or bucketing allocation.
The general principle is to incorporate both quantitative and qualitative
assessments within the determination of significant deterioration of credit
risk and to use lifetime PD as the primary quantitative measure for such
transferring criteria. The questions vary across firms from the weighting between
the quantitative and qualitative parameters to the consistency with the rating
systems, credit policies, monitoring and forbearance processes, and credit
strategies. This, in itself, presents certain governance challenges, especially in a
context of all changes needing to be traceable and justifiable.
Firms are exploring various approaches at a tactical level, clients have
recently considered embedding the criteria within other relevant processes at
the organisation. Moodys Analytics has been involved in formulating a strategic
approach, where clients transition towards a more traceable, controlled and
automated workflow and technology architecture.
How will IFRS 9 affect banks incentives to do certain types
of business?
Stakeholders pay close attention to earnings as they have significant impacts on
stock prices. Typically, they prefer higher earnings with lower earnings volatility.
With IFRS 9, the earnings volatility is expected to increase significantly across
the portfolio and firms will be looking to do one of the following:
Minimise the portfolios earnings volatility given a certain level of expected
earnings equivalent to maximising expected earnings to earnings volatility ratio.
Minimise the loss in portfolio earnings under extreme conditions given a
certain level of expected earnings under normal conditions equivalent to
maximising expected earnings to earnings tail risk.
As a result, firms need to explore ways to enable business in the new era of earnings
of volatility by advancing their analytic capabilities as well as their infrastructure.
Does the three-stage impairment model create the opportunity for
regulatory arbitrage?
Given the principle-based nature of IFRS 9, there is much room for interpretation,
especially in the adoption of staging criteria. This is expected to narrow down
as supervisors and auditors review firms IFRS 9 practices, including the
determination of staging criteria.
According to the Basel Committee on Banking Supervisions guidance on
accounting for expected credit losses (GAECL), when assessing capital adequacy,
supervisors consider how a banks accounting and credit risk assessment policies
and practices affect the quality of the banks reported earnings and, therefore, its
capital position.
Impairment
When credit risk assessments and parameterisations are widely varied or
deficient and not remedied on a timely basis, the supervisors can consider
whether such deficiencies and variations should be reflected in supervisory
ratings or through a higher capital requirement under ICAAP Pillar 2 of the Basel
capital framework.
This in itself encourages the firms to go through ongoing independent
validation process of the IFRS 9 models and frameworks and ensures the staging
criteria is appropriate, considering the existing supervisory guidelines, accounting
firm interpretations and evolving industry practices.
Can Basel II credit risk models be redeployed for use with IFRS 9?
According to the Basel Committees GAECL, firms should use information
consistently across the bank, and common models and data should be used
for both capital and provisioning purposes. This will ensure consistency in the
interpretation and application of the new IFRS 9 standard and reduce the
extent of bias. Therefore, banks typically take Basel credit risk models and make
adjustments to these PIT calibrations, scaling up to lifetime measures that
take into account multiple scenarios, and so on. This approach is compliant and
consistent with that applied across the industry and is in line with guidelines
from the supervisory authorities.
Beyond credit risk models, the guidelines encourage firms to ensure
consistency in macroeconomic forecasts used across the organisation,
from ICAAP scenario analysis, stress testing business forecasting, strategic
planning to IFRS 9 models and leveraging existing governance processes in
place for macroeconomic scenarios in ICAAP. This will ensure that economic
forecasts are disclosed transparently and consistently among ICAAP reports
and IFRS 9 disclosures.
If the firms choose not to extend existing credit risk models, systems and
information for the purposes of IFRS 9 ECL calculations, then the underlying
rationale for differences should be documented and approved by senior
management. In our opinion, this will create additional unnecessary burden for
clients and should be avoided.
Can IFRS 9 be viewed as an opportunity to revamp your modelling,
model governance and data?
IFRS 9 can be viewed as a catalyst to optimise and seek synergies across the
firms modelling landscape and data frameworks.
Leveraging the multi-year IFRS 9 programmes, many firms have started to
review and extend their modelling capabilities, revise their governance structure
and enhance data and technical infrastructure capabilities in this regard.
For example, in developing multiple scenario-based credit loss estimates
for financial reporting, many of our clients are starting to initially consider the
experience and lessons learnt from similar exercises they have conducted for
regulatory purposes and then extend the forward-looking information and
related credit risk factors for IFRS 9 ECL purposes.
We anticipate further advancements in the analytics space as firms move
away from the tactical mind-set towards a strategic one and begin to focus on
the management of earnings volatility and capital implications.
Contact
Burcu Guner Senior Director
T +44 (0)20 7772 1344
E burcu.guner@moodys.com
www.moodysanalytics.com
risk.net 14
 Challenges

Challenges in IFRS 9
Building on current infrastructure
Upgrading to the latest IFRS 9 is a significant transformational event for all financial institutions, regardless of their size and complexity
Miss Kanithar Aiumla-Or/Shutterstock

15 risk.net October 2016


Market commentators have compared the changes required as being similar
in scale to the re-engineering required for Basel, but in practice financial
institutions will be affected unevenly. For example, a bank that has an advanced
BCBS239 project is likely to have less trouble mining for historic data and
linking it back to the general ledger than a bank that has invested less in its data
infrastructure and governance processes.
Similarly, financial institutions with a mature portfolio of internal ratings-
based (IRB) credit risk models will be better able to evaluate IFRS9 exposures
and re-engineer these models than financial institutions following the Basel
standardised approach with its simpler rules and governance.
Due to the significance of the changes expected, the International Accounting
Standards Board (IASB) provided a multi-year period to facilitate the necessary
changes and parallel test the new provisions and monitoring systems ahead of
formal adoption in January 2018.
Point-in-time probability of default
All financial institutions need to repurpose or build their main credit models to
incorporate probability of default (PD), loss-given default (LGD) and exposure
at default (EAD). Currently, credit loss provisions are posted on an incurred loss
basis. Now models will need to predict credit exposure at point-in-time (PIT)
rather than through-the-cycle (TTC), which is the basis for Basel IRB.
Twelve-month expected credit losses used for regulatory purposes are
normally based on TTC probabilities of a default in cycle-neutral economic
conditions. PD used for IFRS 9 should be PIT PD in current economic conditions,
and will therefore change as an entity moves through the economic cycle.
Historic data will be required especially origination data to build
12-month and lifetime estimates of PD, LGD and EAD.
IFRS 9 model validation
Model validation will follow many existing IRB processes, but will diverge from
them in these key areas:
There is likely to be more diversity in the models that require testing for
example, the complexities in validating low default portfolios and expert
judgement models, especially LGD calculations.
IRB models are tested as TTC and IFRS 9 models are PIT, so validation for
IFRS9 will be a parallel and separate process to IRB.
IRB does not require full coverage of the balance sheet; however, IFRS9
coverage is much higher, so more models will require validation.
Model validation will be required at a minimum to cover the following:
Review of model documentation methodology, delivery of models
and testing.
Governance process status, compliance and appropriateness.
Methodology review challenges to the techniques used and focus on
weaknesses and limitations. IRB models often cater for weaknesses by being
conservative; however, IFRS 9 models are not meant to be conservative, but
best estimates.
Review of model performance through backtesting, historic model testing
for each period under review, reperforming models, comparison of model
performance using other models, etc.
IRB calibration tests will need to be enhanced for IFRS 9 calibration, including:
C  alibration for a maximum of 90 days past due (DPD) for IFRS 9 (with some
exceptions) versus possible 180 DPD for IRB.
Conservatism in IRB models to adjust for model error or uncertainty versus
IFRS 9 measures that are meant to be the current best estimates.
Challenges
E xpected life can be greater than contractual life for example, revolving
credit facilities which can affect the quantum in the EAD calculation.
Overview of the hedge accounting rules under IFRS 9
IFRS 9 has revised the existing rules relating to hedge accounting contained in
IAS 39, viewed by some as disconnected from the practice of risk management.
The rules on hedge accounting in IAS 39 have frustrated many, as the
requirements have often not been linked to common risk management practices,
and have made the process impossible or very costly.
IFRS 9 improves this by better aligning hedge accounting with the risk
management activities of an entity. IFRS 9 addresses many of the issues
in IAS39 that have frustrated treasury and asset-liability management
departments. In doing so, it makes fundamental changes to the current
requirements by removing or amending some of the key prohibitions and rules
under IAS 39. The main changes in IFRS 9 in relation to hedge accounting are
presented in figure 1.
It is important to note that the IFRS 9 hedge accounting rules do not
apply to fair-value hedges of the interest rate exposure of a portfolio of
financial assets or financial liabilities that is, fair-value macro hedges. This
is because the IASB carved out the macro hedge accounting part of the
overall hedge accounting project, which will be issued separately outside
of IFRS 9. At the moment there is no clarity over when the rules relating to
macro hedge accounting will be finalised. In the meantime, until the macro
hedge accounting rules are finalised, companies applying the IFRS 9 hedge
accounting framework can continue to apply IAS 39 requirements for fair-
value macro hedges. This is an important issue for the banking sector as banks
generally take a portfolio view of interest rate risk that is, when hedging the
interest rate risk on mortgages.
1 Hedge accounting changes
Non-derivatives can be hedging instruments when
Hedging hedging fair-value risk
instruments Simplifications in hedge accounting, using options and
forwards as hedging instruments
More components of risks can be hedged
Hedge items Group/net and aggregate exposures can also be hedged
Hedge No bright lines when assessing hedge effectiveness
effectiveness Testing more qualitative then quantitative in nature
Only performed prospectively
testing
Voluntary De-designation of hedging relationships is only allowed
de-designation
Source: Chartis, IFRS 9 Technology Solutions, 2016
in the case of a change in risk management objectives
not allowed
Link with risk Direct link established between the practice of risk
management management and hedge accounting thereof
Enhanced disclosures to provide more meaningful
Disclosures information about the hedging strategies applied and
their financial impact
risk.net 16
 Challenges
Data challenges in IFRS9
Amid the unrelenting focus on regulation in recent years, it is sometimes easy to forget that new accounting standards may have
an equal, if not greater, impact on the way in which banks operate. In a roundtable discussion convened by Risk.net and sponsored
by Oracle, four experts in credit loss explore how the introduction of IFRS9 is set to fundamentally change the way banks do their
accounting, and specifically the data challenges associated with its implementation
The implementation of IFRS9 early in 2018 will be split into three principal
buckets classification and measurement, impairment, and hedge accounting
and moves to a forward-looking approach to calculating expected credit losses.
A research paper published by Chartis Research in March 2016 states:
IFRS9 is a high-impact, symbolic, operational IT and organisational
transformational event for risk and finance, an arranged marriage that is turning
an uncomfortable courtship and good intentions into a powerful, successful
partnership that is greater than the sum of its parts. Our panel aims to
determine just how well this arranged marriage of finance and risk is working
in the face of IFRS9.
With little more than a year until IFRS9 is implemented, how well
is the project going and what are the challenges that lie ahead?
David Grnberger, Financial Market Authority Austria: The project is
well ahead and we, from the supervisors perspective, have started to align and
enter into an intensive dialogue with the banks, both from a national level and
from the EU level. With the European Banking Authority (EBA) and the European
Central Bank, we have all set up our national and European projects. We are
looking into the banks, querying their implementation and asking for numbers,
impacts and challenges.
Our activities have given us a pretty good picture of where we stand and
where the banks stand. The projects are advancing well; we are seeing some
impacts, usually not large but still considerable, and we are seeing the challenges
ahead quite clearly, which relate mostly to model performance and the reaction
of supervisors with regard to banks IFRS9 implementation.
As supervisors, we are currently designing and drafting strategies about how
to react and how to approach banks during the next couple of months.
Wolfgang Reitgruber, UniCredit: Regulatory internal ratings-based (IRB)
frameworks are now expanding nicely into IFRS9 impairment concepts on the
one hand, to enforce backtesting expected loss measurements and, on the other,
to link the new upcoming provision quantities and steering concepts with an
improved understanding of credit risk measures.
17 risk.net October 2016
THE PANEL
David Grnberger, Head of Accounting Enforcement,
Financial Market Authority Austria
Wolfgang Reitgruber, FVP, Group Credit Risk Modelling, UniCredit
Hugh Stewart, Research Director, Chartis Research
Rohit Verma, Head of Strategy for Risk Analytics, Oracle
Rohit Verma, Oracle: In terms of the projects, there have been many
developments in implementing IFRS9 at different banks. All the banks are
paying very close attention to these projects, simply because they realise that this
is a game-changer it fundamentally alters the earnings and the balance sheet
that banks report.
There are challenges that are being addressed today in terms of methodology,
data availability, data quality and governance. These challenges will continue
to be faced; it is not a one-time effort that will no longer be required after
implementation in 2018. These challenges will continue, but most banks realise
if they implement this project correctly it provides a huge opportunity for them to
improve the transparency of their financial statements which, in turn, is good
as most of them report their numbers to the Street and will also improve the
capital planning processes within their banks in the long term.
Hugh Stewart, Chartis Research: Reality has set in, and there is that sense
when you go walking and you think youre going to hit the summit. Then you
realise that many of your targets are false summits. New impact analysis is
identifying there is more to do with regard to modelling, finding missing data,
and so on.
Therefore, there is talk of a little more forgiveness in 2018 when IFRS9
becomes live, but there is also a sense of excitement with regard to how
economic capital, regulatory capital and the loan loss provisions can be worked
together and optimised a sense of threat, but also a sense of opportunity.

The concept of moving to forward-looking provisioning for
expected losses was a major change to accounting standards in
the wake of the financial crisis. Will it succeed in dampening the
pro-cyclicality of accounting standards, just as weve seen similar
efforts in regulatory capital? How confident are you that the new
standards are going to achieve that?
David Grnberger: This is a common misunderstanding I find when talking to
practitioners. The aim of the IFRS9 project is not really to dampen pro-cyclicality.
In fact, our simulations show that it will increase the volatility and probably also
the pro-cyclicality of accounting provisions.
The main aim and this should be the merit of IFRS9 is to increase the
relevance of accounting provisions to transfer information from the bank to
the investors and to the regulators about those events that can be reliably
forecast, reducing the noise from arbitrary effects from errors and really making
accounting provisions relevant, as far as possible, to investors.
Relevance means statistical relevance and model performance, but it also
means practical relevance and judgement. All of a banks efforts should be
steered towards making numbers and forecasts as reliable as possible, reducing
all arbitrary effects from personal estimates and human judgement, and
improving provisioning to the maximum relevant extent.
Investors and banking supervisors will hold banks responsible for preparing
those forecasts based on data and experience, having a high standard of models,
improving the data and using high-quality data for their estimates because,
whenever a bank deviates in practice from its forecasts, banking supervisors
and analysts will dig into the issue and ask why. It would present a pretty bad
scenario if the bank cannot explain the deviations, and bad model quality and
bad data quality are revealed.
What role are regulators playing now in helping banks to prepare
for the standard?
David Grnberger: Right now the regulators role is more or less an advisory one.
We are trying to pool the data we receive from the banks from the initial trial rounds
of IFRS9. We are feeding what has come out of the analysis back to the banks, at
least on a qualitative level, from the EBA impact studies. We are talking with the
banks and showing them where they deviate from European sample averages.
Challenges
We discuss the policy choices they have made and provide initial feedback on
whether or not, from a supervisory perspective, they would be in line with policy
choices and whether nor not we would accept simplifications.
The most important contribution we will be making in the coming months is
how we expect them to validate and backtest IFRS9 results in the future because,
as supervisors, we at least have some experience in validation and backtesting
from our IRB-model inspections. We want to extend this experience and the
measures we are familiar with from the IRB approach to IFRS9 implementation.
Can you give us a general sense of the role of data in IFRS9
obviously its a big issue, but what is the overall picture here?
Rohit Verma: Data is going to play a crucial role in the success or failure of
any IFRS project. There are three aspects of data that Id like to discuss, the first
of which is model calibration. These new-age, forward-looking models are only
as good as the data used to calibrate them. If you dont have the right quality
or the right amount of data, the calibration of these models is not going to be
relevant and will therefore affect things like backtesting, validation, and so on.
The second aspect is from a processing perspective. IFRS9 requires processing
to happen at an instrument level, so millions of records will be processed on a
periodic basis. You are not only obliged to process this under baseline conditions,
but also under different scenario conditions, which creates a huge demand in
terms of data processing. Running millions of records through multiple scenarios
on a monthly or quarterly basis is a huge task for any banking organisation.
The third aspect is convergence between risk and finance, and how
organisations can leverage it for addressing not just the immediate IFRS9
requirements, but also some of the other requirements such as regulatory
capital, economic capital and loss forecasting. Once you have this account level
of granular data available and reconciled to both risk and finance organisations,
you can do a lot more to it than just IFRS9.
What is your perspective on the additional granularity of data
that banks will be required to collect to fulfil the impairment
requirements in this standard?
Hugh Stewart: Its very interesting, not just on the financial engineering and
data management side, but also on the cultural side. It has been alluded to
risk.net 18
 Challenges
Rohit Verma
already, but I think the accountancy and finance world is less forgiving than the
risk management reporting world, and that a sense of detail, provability and
verification, absence of error, and the locating of missing data are much more
important within IFRS9. There needs to be a lot more historical data provided
than there is currently.
Some of the less sophisticated banks will either be required to mine their
data more thoroughly probably manual data or very unstructured data or
will need to make use of various types of proxy data.
A different set of bucketing is needed for IFRS9 compared with some of
the regulatory and economic capital reporting areas. The data pack should be
reshuffled, but should still run in harmony with other operational and regulatory
activities. This vast increase has performance issues and data management issues,
but I think the biggest issue is a change in culture with regard to accuracy.
Looking into this over recent months, how would you say banks are
dealing with that? How far advanced are they, and do they have
the resources they need?
Hugh Stewart: There are elements of denial and hope within some
organisations. I think the big Tier 1 banks understand the principles but have the
challenge of pulling their organisation together; they have the problems of size,
complexity and scale, but are addressing them well.
The large Tier 2 banks are well on target. They copycat the sophistication
of Tier 1s, but do not have the same levels of size and complexity. Some of
the smaller organisations are way ahead of target or are in denial and hope,
relying on either existing or third-party data which are not proving as granular
and complete as they hoped and on core banking systems that are less risk
sophisticated. I think there is an element of danger here. These are the people
that are negotiating now for forgiveness throughout 2018.
Can you give some internal perspective on the challenges of
IFRS 9 impairment?
Wolfgang Reitgruber: I would say, as a main advanced IRB bank in the
market, like many of the large companies, we are well on the way. We have been
working for over 10 years in Basel IRB environments with different regulators,
recently contributing actively to regulatory benchmarking exercises and various
quantitative impact studies.
19 risk.net October 2016
On the micro-data level, there was a lot happening in the past, and IFRS 9 is
therefore not something where we are starting from zero. We have already been
building up the infrastructure, which is continuously being improved and enhanced.
Im more concerned about managing the macro data, the forecasting element
of IFRS 9. We will enter an environment where forecasts may drive provisions,
and proper governance and data quality are required to ensure the ongoing
soundness of provisions.
We all know that economic forecasts tend not to be very reliable, but you need
the framework and the governance in place to justify that whether one or five
years later at a certain point in time, we had reason to believe in this forecast.
The world is changing, but there are real data challenges on the macro side.
Obviously, theres no point generating the data for this standard if
its not trusted or robust data. How can banks have confidence in
the data theyre generating for IFRS9?
Rohit Verma: Banks need to establish a couple of things so that, over a period
of time, they become more confident about the data that has been used, not just
for IFRS9 but for any regulatory or reporting purposes.
First, banks can establish a single repository of raw data that has the capability
to ingest data from multiple sources and at multiple levels of granularity. For
IFRS9, you need account-level data, which is literally millions of records for large
banks. You also need economic forecasts, which could be a few lines or a few
hundred records, each of them a different forecast period and forecast horizon.
So, you need to set up a repository that can input data from different sources,
keeping in mind these business requirements. Then, this repository becomes the
source of data for all of the regulatory requirements and reporting.
Second, establish strong data governance processes. This includes a data
quality process that identifies trends in data quality from different source
systems or a reconciliation process to ensure your granular data is reconciled
with your aggregated general ledger information. You could also establish a data
repository [of definitions], so that everyone in the organisation refers to the same
data in a uniform manner.
These processes will help improve the quality of data over a period of time
and ensure that, going forward, many data quality issues are addressed at
source, not at an intermediate level.
So, you should ensure the data is of good quality to start with,
rather than having to go back for it.
Rohit Verma: Exactly if you know there is a portfolio in which there is a
secular trend of data problems, you could go back to the source and fix it there,
rather than having to fix it when the data lands in the repository, which is a lot
more expensive.
What role are validation and backtesting going to play here? How
should the quality of data be checked and validated?
David Grnberger: Validation, or designing models that are accessible to
validation, is crucial to the entire project. You should start from the top down,
and decide in the first step, when you have a good set of data: what kind
of data, which factors and which parameters performed well in forecasting
expected losses, especially over a 12-month horizon.
If you select those data sets variables that have a good performance there
will also be a need to differentiate between time horizons. Which data sets and
which indicators have a good model performance or explanatory value for six
months, 18 months or 48 months in the future? Then you can design your model
in that way.

Hugh Stewart
This means only selecting those kinds of data and data sources that have a
good track record in explaining future losses. Deselect those sets of data and
variables that create noise or have been shown to have high autocorrelation,
and try to establish a simple set of variables not too many that have good
performance. Understanding model performance is the first step to creating a
model that can be validated later.
For example, if your macro data or macro variables do not have a good track
record, include them only if they have a very low weight or with regard to short-
term time horizons. Do not include macro projections for a time horizon of, for
example, four years in the future if they are not relevant for that time horizon,
but include them for predicting expected losses and defaults for, for example, a
six-month time horizon.
This should also solve the problem of macro data seeming to have little
relevance, which is always determined based on time horizons. The time horizon
and the historical performance should be the driving factors when you design a
model and validate them later on.
Wolfgang Reitgruber: I would focus much more on backtesting. The purpose
of IFRS 9 impairment is to get a good view of the value of the balance sheet.
What part of this value do we control? The provisioning part, so in the end its
about backtesting the provisions.
That is now the main challenge in IFRS9: there will be a lot of profit-and-loss
volatility driven by risk results, and a lot of point-in-time/forward-looking types
of assumption in our estimates. But we definitely have to ensure that they are
neither overly conservative nor overly aggressive.
In the end, it should be a fair picture of the value and that should basically
drive the decisions of what type of data and what type of data quality are
needed, and how to end up at a fair representation, probably after a couple of
years of running IFRS9.
From a data perspective, how are you planning to incorporate
forward-looking projections into expected credit loss?
Wolfgang Reitgruber: One of the main topics, which is not unique to our
group, is where to take these predictions and these forward-looking opinions
from. And one of the main points was that it must be consistent, at least
across the legal entity of the group. So, if you have an economics department
Challenges
that is making predictions, you really want to avoid using different
predictions for IFRS9. There are stress-testing environments, there are certain
assumptions and there are also asset-liability management frameworks.
All of these are already using and building up certain types of predictions;
therefore, consistency really is the most important aspect of where to get the
data from.
There are two philosophies of how to include data in the models themselves
One philosophy is to use a probability of default (PD) model with factors that
reflect the economic status and the economic cycle. The second is more of an
IRB-type process through the cycle models, and adjusting them for point-in-time/
forward-looking-type features.
I prefer the second option as it allows much easier reconciliation
between Basel frameworks and IFRS9 frameworks. It also allows for much
easier correction. If youre assuming that the economic cycle information
is changing relatively quickly, you probably have a less reliable forecast
than you are typically used to in a PD framework, for example. The relative
risk ranking can be described pretty nicely, but the absolute risk ranking
is probably tricky and every year you have to make, probably significant,
adjustments.
Personally, Id go for this second approach, but I know a couple of banks that
are already including economic information directly into their systems.
What is your view on the need for consistency with regulatory
capital data standards?
Rohit Verma: Its a given that you need consistency in data; looking at a
regulatory capital number versus the provision, they need to be generated from
the same source, otherwise the two numbers are not really in sync and present
an inaccurate scenario to stakeholders to the bank.
Most banks are going with the second approach, taking the regulatory
capital PD and adjusting it for forward-looking measures. Maybe in the
long run the trend would be towards incorporating the forward-looking
projections into the PD model, into the transition matrix that drives the
transition of credit quality for each and every account. This is a kind of
evolution of the model that we might see in the future when incorporating
forward-looking projections.
We have also seen the inclusion of a lot of forward-looking projections in the
identification process stage, as that will alter the provision generated for each
and every account. The forward-looking projection of the economy will have a
big impact on the stage to which each account is classified.
What impact will all this have on legacy business, for which banks
may not have relevant data available?
Hugh Stewart: This is really important. Banks that havent been through the
BCBS 239 or Comprehensive Capital Analysis and Review dont necessarily have
the same discipline as the larger banks. This is a matter of data mining and its a
difficult job. So maybe the models will have to be adjusted or replaced according
to the valid data available.
We havent yet mentioned the word disclosures. This is going to be very
important, especially as this first step transfers from IAS 39 to IFRS9. There will
be a huge body of disclosures to explain opening and closing from one state to
another and provisioning. That is just a vast body of work that will be necessary
in the first year.
IFRS9 Paragraph 35H, a hot topic among auditors and accountants, deals
with the need for very heavy, accurate and forensically validated disclosures,
which are all a part of the data quality initiatives.
risk.net 20
 Challenges
What kinds of technology
might banks need to consider
to overcome some of these
challenges?
Rohit Verma: Banks will definitely
need a technology solution that can
address the data challenges the
www.charakter.photos/Philipp Monihart
ability to store and process large
volumes of data. A technology partner
that is looking not just at the traditional
ways of processing, but also some of
the newer big data technologies for
managing the whole processing aspect,
is definitely required.
Another aspect is that a technology
solution should be ready to meet the
requirements as they stand today, as
Wolfgang Reitgruber
well as be flexible enough to evolve to
meet the specific requirements each
bank has around IFRS9. IFRS9 is not really prescriptive in nature, and every
situation will have its own aspects unique to IFRS 9. You need a technology
solution that can handle these nuances in each bank in terms of methodology,
organisational structure, and so on.
Finally, you need a technology solution that is not just focused on IFRS9,
but able to address IFRS9 projects other areas, such as enterprise stress
testing, liquidity risk, regulatory capital, and so on. These are some of the
ideas around technology that banks should be looking at when selecting
a partner.
This is a huge lift for banks; some have seen it as even bigger and
more impactful than Basel III, for example. To what extent is this an
opportunity to improve modelling, model governance and data as
well as being a compliance burden?
David Grnberger: Whether its an opportunity or a compliance burden
depends mostly on how banks decide to implement IFRS9. The supervisor
expectation and were looking for this in the future is that banks
improve the models not only for the accounting purpose, but also for the risk
management function, for capital purposes and to integrate as far as possible
the IFRS9 improvements into the existing risk management framework a bank
already uses.
A vast majority of the banks told us they expect IFRS9 to be used in product
design and in decisions to grant or extend loans, and that they are actually
trying to integrate the IFRS9 improvements into their risk management and
governance structure. If this is the case, it will be an opportunity to improve
modelling overall, and this is what I expect from most banks.
How should banks be positioning their businesses, as well as their
accounting standards, ahead of IFRS9?
Wolfgang Reitgruber: I agree that a couple of banks seem to use IFRS9
for granting processes. When comparing accounting standards now with the
Basel environment, Basel is actually more focused on the stock and the on-
book portfolio. So there was always a little more pressure towards behaviour-
scoring systems, but not so much on the application scores, if youre just
taking this comparison.
IFRS 9 focuses more on the PD at inception, which is triggering a lot of
21 risk.net October 2016
awareness to this part of the credit
process. We achieve greater focus on
the inception point and need lifetime
loss estimates, which are not required
within the Basel environment. This is
probably laying the groundwork for
better pricing decisions and better
strategy in underwriting, which
will have a positive impact on the
business side.
There is much to be done and it
is a huge project. There are a lot of
new methodological dimensions that
have to be covered, but I do not see
the type of bureaucracy part of the
regulatory framework to approve
these models for IFRS 9. We will need
David Grnberger
strong internal governance to keep
everything under control, but we have
yet to see how the internal procedures are going to look and how the regulators
will review processes and models. It might be easier from that perspective than
the IRB framework.
How is the culture within organisations changing, and what is the
impact of this standard on the business decisions banks are making?
Hugh Stewart: It is definitely having a major effect, and theres a big learning
experience going on just with risk and finance working together, as well as
different IT teams and financial engineering, pricing and risk analytics groups
working together. It is raising the bar holistically across a number of different
disciplines within the bank.
I agree that the focus on lifetime losses is changing things. This is great in
terms of being in parallel to Basel with a similar focus on model governance,
but its definitely a burden. Although its leading to a promised land and were
all going to be better people as a result of it, it is tough medicine. There is going
to be a real sense of resource pressure on all financial services organisations to
reach these summits, or false summits, before planting the flag at the top.
Rohit Verma: David mentioned earlier that IFRS9 makes the financial
statements more relevant. That is a huge benefit that banks will gain if they
implement the project correctly.
Its not like an on/off switch. Its not that, in 2018, your financial statements
will suddenly become more relevant, but rather it is the continuous improvement
in the relevance of the financial statements and the numbers that is going to
have an impact on product design and pricing. Its going to have an impact on
the capital planning process, and also on the credibility of the numbers and how
they are perceived by the markets, which in turn will have an effect on the cost
of capital for banks.
Overall, if handled correctly and there are definitely challenges the key
thing, if you implement it correctly from a data management perspective and a
modelling perspective, is that there are benefits that can be derived in the long
run because the numbers are now much more relevant.
The commentary and responses to this forum are personal and do not necessarily reflect the views
and opinions of the panellists respective organisations.
 Model building

shooarts/Shutterstock

Model building for IFRS 9

Incorporating the right techniques
An illustration of the process for building an IFRS 9 model, outlining approaches for both wholesale/corporate and retail

Wholesale/corporate IFRS 9 model build
The main activities are typically:
Determine the client segmentation.
Define the targeted model structure: group/local/multiple models.
D efine the input data and identify candidate variables for the analysis between
quantitative factors for example, return on assets, return on equity, debt
equity ratio and qualitative factors for example, quality of the management
board, market share, market structure (monopoly versus competitive), hurdles
to entry.
Definition of default: 30, 60 or 90 days past due (DPD).
Perform data quality verification.
Define the historic data population for development and validation.
Development of subsequent modules: univariate or multivariate analyses.
Development of the joint model from separately developed modules.
Analyses of the prepared models and final model selection.
Pre-implementation tests.
Retail IFRS 9 model build
From a risk perspective, retail models are driven more by scorecards and
homogenous groups. Such segmentation has a significant effect on the analytics;
therefore, vital areas for consideration are statistical methods, governance and
model validation/calibration. A summary of variables is offered below.
Credit application models
S ocio-demo variables: income, profession, region, age, marital status,
education, and so on.
R elations with the bank: length of the co-operation, average amount of loans,
product mix, and so on.
Interactions between complex variables: age and income, region and income,
and so on.
Variables based on credit bureau data, describing repayment history: debt level
and number of DPD.
History of loan origination: frequency, level of debt, loan types.
History of queries to credit bureau.
Behavioural models:
Behaviour on current accounts.
History of delinquency.
L evel of exposure, exposure amount divided by exposure as at
origination data.
Frequency of loan origination how frequently the client takes new loans.
Delinquency value to exposure value.
Usage of the available off-balance limit in case of revolving products.
N  umber, value, frequency or share of cash transactions and cashless
transactions.
Repayment patterns for subsequent instalments.
History of co-operation with the bank.
Best practice for choosing a model for retail probability of default, loss-given
default and exposure at default is to use several techniques or a combination of
these. Final selection of the model is then based on its statistics and not on the
assumptions of modelling techniques.

risk.net 22
 Model building: Whitepaper
Managing earnings volatility and
uncertainty in the supply and
demand for regulatory capital
The impact of IFRS 9
This novel approach to modelling from Moodys Analytics allows better management of the interplay of supply and demand dynamics
for regulatory capital, combining an economic framework with regulatory capital and new loss recognition rules
Introduction
The framework is particularly relevant in
understanding the extent to which IFRS 9 can lead
to more aggressive provisioning, which feeds into
earnings volatility. The approach provides guidance
on how organisations can better manage their
capital buffer, considering investment concentration,
its impact on earnings volatility and relationship
to regulatory capital requirements. Imperative to
portfolio management, the framework recognises
the likelihood of a capital shortfall being significantly
impacted by portfolio name, industry, geography and
asset class concentration, as extreme fluctuations
in capital supply and demand occur more often for
institutions holding more concentrated portfolios.
Finally, we discuss integrated investment and
strategic decision-making measures that account for
the full spectrum of economic risks and interactions
with regulatory and accounting rules, as well as the
instruments contribution to earnings volatility and
capital surplus dynamics.
With stringent regulatory and accounting
requirements, risk managers can struggle with
incorporating regulatory capital requirements and
loss accounting rules into investment decisions.
An economically appealing approach considers
stakeholders who maximise return while recognising
the risks and who face regulatory capital constraints;
investment decision rules recognise both regulatory
capital and economic risks1. Intuitively, an investment
with less concentration risk, all else being equal, is
better diversifying and more appealing. Similarly,
an investment that attracts less regulatory capital,
all else being equal, is less constraining and
23 risk.net October 2016
more appealing. Accordingly, institutions can use
integrated metrics that account for both regulatory
capital and economic risks such as regulatory-
adjusted return on risk-adjusted capital (Rorac),
concentration-adjusted return on regulatory capital,
or composite capital measures that reflect regulatory
capital constraints, as well as economic risks.
In addition to considering regulatory requirements
and economic risks related to concentration
effects, the question of being able to fulfil future
regulatory requirements is material. In reality,
credit deterioration flows into earnings along
with increases in regulatory capital, resulting in
a potential capital breach. The likelihood of such
a breach depends very much on a portfolios
composition, the degree to which it is diversified
and the supply of equity. Ideally, investment decision
rules should account for the likelihood and cost of
breaching future regulatory capital requirements.
The key to managing the dynamics in regulatory
capital requirements is to quantify the likelihood
that the supply of capital is sufficient to address
future regulatory capital requirements. This
probability is determined by the dynamics of
regulatory capital supply and demand, with the
supply affected by earnings and loss recognition
rules. Compared to its predecessor, IAS39, the new
IFRS 9 accounting standard for financial instruments
requires institutions to set aside provisions at
origination. In addition, the staging rule for IFRS
9 requires institutions to update loss allowance to
reflect changes in credit quality at each reporting
date, which can increase earnings volatility that
flows into the supply of capital.
This article examines how IFRS 9 affects
regulatory capital supply and demand. We provide
an overview of how an institution can utilise
integrated measures that account for economic
risks and regulatory capital for better capital
management. Our approach leverages an economic
capital framework similar to the one proposed by
Levy, Kaplin, Meng and Zhang (LKMZ) in 20121,
in which stakeholders maximise return per unit of
risk while facing regulatory capital constraints. In
addition to recognising current regulatory capital
requirements, this article incorporates uncertainty
in the supply of and demand for regulatory capital
coming from changes in the credit environment.
Impact of IFRS 9 on regulatory
capital management
1. How IFRS 9 affects the dynamics of
regulatory capital at horizon
IFRS 9 affects the supply and demand for regulatory
capital in at least two ways. First, IFRS 9 generally
requires an institution to recognise 12-month
expected credit loss of a financial instrument as
soon as the instrument is originated or purchased.
Meanwhile, IFRS 9s predecessor, IAS 39, generally
requires material credit events to trigger loss
provision. Thus, IFRS 9 can cause an initial reduction
in the Tier 1 capital supply, driving required regulatory
capital to be more constraining for banks using
a standardised approach to compute regulatory
capital2. In addition, IFRS9 staging rules can result in
further reduction in the capital supply when lifetime
losses must be considered.
Second, IFRS 9 can increase the volatility in
 Model building

1 Loss allowance at horizon of diversified portfolio versus concentrated portfolio

capital supply. IAS 39 requires provisioning under
significantly negative credit triggers, which generally
dampens the impact of credit migration on capital
supply volatility. In contrast, under IFRS9, institutions
update loss allowance to reflect changes in credit
risk on every reporting date, resulting in credit
migration being accounted for in capital supply. An
important corollary to this observation is that more
concentrated portfolios will, in general, be more
impacted by the IFRS 9 volatility increase. Intuitively,
a perfectly diversified and granular portfolio exhibits
no volatility. Figure 1 compares the IFRS 9 loss
allowance at horizon of a well-diversified portfolio
and a portfolio with high concentration in the oil
industry. It can be seen that, for the simulated trials,
loss allowance for the well-diversified portfolio never
exceeds 10%. In contrast, the diversified portfolio
loss allowance at horizon exceeds 10% in 3.7%
of trials. All else being equal, organisations loss
provisions will be more extreme for concentrated
credit portfolios, driving a higher volatility in earnings
and likelihood of facing a regulatory capital shortfall.
2. Quantifying change in capital surplus
As discussed above, IFRS 9 loss provision affects
the supply of capital, potentially impinging on
an institutions ability to meet regulatory capital
requirements. The dynamics of loss allowance,
with each reporting date, can further constrain the
organisation, as it should consider buffering for a
1
 or example, Levy, Kaplin, Meng and Zhang (LKMZ) (2012) introduce a regulatory capital-adjusted Rorac measure by integrating economic capital
F
with regulatory capital under a capital asset pricing model framework. Xu and Levy (2015) extend LKMZs model and create a composite capital
measure that serves as a capital allocation measure accounting for both regulatory capital requirements and economic risk.
2
Under the Basel III rule for advance IRB banks.
3
Th e required regulatory capital for each instrument is computed based on the Basel III advanced IRB approach in all examples in this article.
deteriorated credit environment. Capital surplus
measures the gap between capital supply and
demand. In the context of Basel III and IFRS 9, the
change in capital surplus is driven by the change in
regulatory capital required by Basel III, and earnings
that are driven by interest income, default losses and
provisions either 12-month or lifetime, depending
on the assets stage.
Since the change in capital surplus captures the
dynamics of both required regulatory capital and
earnings, it provides a foundation for measuring
how much capital must be set aside. With that
said, the expected change in capital surplus
associated with each individual instrument does not
account for concentration and diversification risks.
Consequently, an institution should not use it as the
only measure when making investment decisions.
For example, all else being equal, an instrument
with a 4% expected increase in capital surplus may
be more attractive than one with a 6% expected
increase, if its credit risk is less correlated with other
instruments in the portfolio.
3. Leveraging an economic framework to
manage earnings dynamics and the demand
and supply of regulatory capital
As discussed in the introduction, the LKMZ
framework accounts for regulatory capital
constraints at the time of investment; in reality,
future credit deterioration results in changes to
regulatory capital requirements and its potential
violation. The likelihood of such a breach depends
greatly on the portfolio composition, the degree
to which it is diversified and its capital surplus.
Investment decision rules should account for
the likelihood and the cost of breaching future
regulatory capital requirements. Institutions have
addressed this issue by adopting buffers beyond
their stated required regulatory capital requirements.
The challenge is in quantifying the buffer, how
portfolio composition can improve managing that
buffer, and how all this should feed into investment
decision rules. Intuitively, institutions should set
aside capital buffers, so the likelihood of a capital
breach does not exceed a target probability. In
addition, institutions should assign an additional
capital buffer to each instrument according to the
expected change in capital surplus associated with
the instrument, and how that change contributes to
the overall likelihood of a breach. The distribution of
changes in capital surplus is depicted in figure 2.
The left-hand side of figure 2 shows the
distribution of changes in capital surplus over a one-
year horizon for a sample loan portfolio; the right-
hand side depicts the capital surplus distribution
for a more concentrated portfolio that does not
benefit from country and industry diversification.
In this case, the probability of a capital breach
more than doubles, to 26 basis points, if the
same 2.2% additional capital buffer is set aside3.
Limiting the probability of a capital breach to 10bp
requires a 2.2% additional capital buffer to be set
aside beyond what is needed to address current
regulatory capital requirements.

risk.net 24
 Model building: Whitepaper
2 Probability of capital breach: diversified portfolio versus concentrated portfolio
3 Change in capital surplus versus change in capital risk
It is important to note that, even though the two portfolios shown in
figure 2 have different capital breach probabilities, both have the same
expected change to their capital surplus: 1.95%. Therefore, it is clear that the
expected change to the capital surplus by itself is not sufficient to describe
an instruments risk, as it does not account for portfolio concentration and
diversification effects. This trait is similar to expected loss measures not being
impacted by diversification and concentration.
Figure 3 provides an additional perspective to the dynamics of capital surplus
by comparing it with portfolio fair value loss. While the change in portfolio capital
surplus has a general inverse relationship with portfolio loss, there is a reasonable
amount of dispersion. One primary reason behind this observation is that fair
value portfolio loss, which includes both default loss and credit migration loss, is
entirely driven by the migration of point-in-time probability of default, while the
change in capital surplus is partly determined by the migration in through-the-
cycle probability of default, which feeds into regulatory capital calculations.
25 risk.net October 2016
To account for the full spectrum of economic risks and interactions with
regulatory and accounting rules, one can leverage the LKMZ framework
and associate an additional capital buffer charge to each instrument as the
organisation ensures capital solvency In the future. The resulting investment
decision rules account for capital surplus dynamics as well as the concentration
and diversification risks associated with each instrument.
Conclusion
The introduction of IFRS 9 changes the dynamics of capital supply and
demand and affects institutions investment decisions. In particular, the new
loss recognition rule under IFRS 9 can make regulatory capital requirements
more stringent and can increase the uncertainty of capital adequacy in the
future. IFRS 9 can also introduce significant concentration risk into capital
planning. These implicit costs should be accounted for in investment decisions
and capital allocation. An extended LKMZ model leverages an economic
framework and derives investment decision rules based on the full spectrum of
risk, and it accounts for regulatory capital as well as future dynamics in capital
supply and demand.
The authors
Amnon Levy (pictured, left)
Managing Director, Research
E amnon.levy@moodys.com
Jing Zhang, Managing Director, Head of Research
E jing.zhang@moodys.com
Andriy Protsyk, Associate Director, Research
E andriy.protsyk@moodys.com
Pierre Xu, Associate Director, Research
E pierre.xu@moodys.com
www.moodysanalytics.com
 Cutting edge: Credit risk
Cutting
Model edge:
building: Creditedge
Cutting risk

Loan classication under IFRS 9

IFRS 9 requires classifying non-defaulted loans in two stages depending on their credit quality evolution since initial
recognition by the bank. In this paper, Vivien Brunel proposes an optimal way to perform this classification. Target values of
some key performance indicators of the provisioning model emerge from the implementation of this process. In particular he
computes the target value of the stage 2 hit rate and the size of the stage 2 portfolio

S
coring and rating models have been used in the eld of the grant- a rating or scoring tool are based on the cumulative accuracy prole
ing of credit and in credit risk management for some time. In (CAP) or the receiver operating characteristic (ROC), and on their sum-
2001, the Basel Committee required the use of internal mod- mary statistics, namely the accuracy ratio (AR) and the area under the
els to be extended to capital charge measurement (Basel Committee ROC curve (AUC), respectively (see Sobehart, Keenan & Stein 2000).
on Banking Supervision 2001). Since then, banks and regulators have We mention that the AR and AUC are criticised for being awed, par-
both developed statistical tools to evaluate the quality of internal rating ticularly when expressed in terms of misclassication costs, and that
models because bad performance can lead to inefcient allocation of a more objective measure exists (Hand 2009).
capital. By considering that the two-stage classication process of perform-
In 2014, the International Accounting Standards Board (2014) pub- ing loans is based on a scoring model, we link the two-stage accuracy
lished the nal version of the IFRS 9 accounting standards, which aim ratio of a portfolio of loans to the underlying score accuracy ratio. We
to overcome the problems that arose during the nancial crisis because show that optimality in the two-stage classication can be reached by
of the previous IAS 39 incurred loss model. The new requirement is appropriately choosing the hit rate (called the stage 2 hit rate here-
to recognise loss allowances or provisions on all loans, including per- after) that we target, ie, the proportion of defaulted loans that come
forming loans. This is done in a two-stage process for non-defaulted from stage 2 loans. In many realistic cases the optimal target stage 2
loans. hit rate is in the range 7090%. Additionally, we derive a formula for
 Stage 1: if the credit risk of a nancial instrument has not increased the size of the stage 2 portfolio. We show that, for a given portfolio of
signicantly since initial recognition, the loss allowance is equal to loans, the main driver of the provision is the stage 2 hit rate and not
the 12-month expected credit loss (ECL). the size of the stage 2 portfolio.
 Stage 2: if the credit risk of a nancial instrument has increased To make our paper self-contained, we describe the main statistical
signicantly since initial recognition, the loss allowance is equal to tools used to assess the performance of a rating or scoring system in
the lifetime ECL. the next section. Later we show how these performance indicators are
In general, stage 1 loans are of better credit quality than stage 2 loans. shifted when loans are classied into only two buckets, and we com-
Paragraphs 5.5.10 and 5.5.11 of the norm (International Accounting pute the optimal target stage 2 hit rate. We go on to derive the formula
Standards Board 2014) provide some requirements about the transfer that links the size of the stage 2 portfolio to the other parameters, and
criteria. However, the norm is principle based and does not detail in the nal section we provide a simple proxy formula for the total
how to determine which instruments should be in stage 1 or in stage 2. provision.
Implementing the standards is subject to interpretation and to some
subjective choices in terms of credit risk quantication. The measures of discriminatory power
In this paper, we propose how assets should be assigned to the two A scoring model aims to rank the clients of a bank according to their
stages. Our proposal involves the following three assumptions. creditworthiness, ie, their ability to pay back the loan they have been
 We assume that the transfer criteria from stage 1 to stage 2 are based granted. Whatever it is based on either a mathematical model or an
on scoring and rating systems of the bank and that an instrument is expert-based judgment, or both the performance of a scoring model
transferred to stage 2 when its absolute level of risk has gone beyond is measured by the concordance of low scores with the occurrences of
a given threshold; this is a good approximation when the bank does defaults. When a scoring model is random, ie, contains no information
not originate any loan with a score under a given cutoff value. about the likelihood of a default, the conditional default probabilities of
 Loan classication aims to accurately predict defaults over a given the clients are uncorrelated with their scores. Conversely, for a perfect
time horizon, and its performance will be assessed accordingly. Even scoring model, the scores perfectly rank the risk of the clients: the
if the ECL is measured over the lifetime of an instrument, we assume clients that go to default are assigned the worst scores prior to default.
that the performance of rating and scoring models is measured over a We mention that the IFRS 9 norm focuses on the notion of credit risk
one-year horizon as this is usually the case in practice. deterioration and does not provide any denition of the default event.
 In the specic case of retail exposures, we assume that the scores Banks usually have only one denition of default, which coincides
and ratings under consideration incorporate some specic issues such with the Basel denition.
as multiple defaults, restructured loans or default contagion. We consider a homogeneous portfolio of loans, meaning that the
As we will emphasise, efcient provisioning in the IFRS 9 frame- loans have the same risk drivers. These loans are granted to the same
work is based on measuring credit risk and model performance accu- types of client in the same geographic area and belong to the same asset
rately. The main statistical tools used to assess the performance of class (for instance, prime residential mortgages in the UK originated

risk.net
risk.net
risk.net 73
77
26
 Cutting edge: Credit risk
Cutting edge: Credit
Model building: risk edge
Cutting

literature because it often leads to very good ts (Hanley 1996). We

1 The ROC curve (the hit rate as a function of the false alarm
rate)
will set b D 1 hereafter; as emphasised by Tasche (2012) and Cramer
(2003), this special case motivates the choice of modelling the prob-
ability of default curves with the inverse logit function. Additionally,
C
100 b D 1 is the only value of the parameter b for which the binormal
ROC curve is concave over the whole interval 0; 1.
80
Another t uses an exponential shape for the CAP curve. It was pro-
A D
Hit rate (%)

60 posed by Van der Burgt (2008) in the context of low-default portfolios

(sovereign credit risk in his paper). In this paper, we introduce the
40 exponential t of the ROC curve, which is similar to Van der Burgts
Perfect model
AR = 70% t:
20 Two-stages model 1 eku
O Random model RkE .u/ D (3)
z B E 1 ek
0
0 20 40 60 80 100 In what follows we will assume that the ROC function R.u/ is concave
False alarm rate (%)
and has either a binormal or an exponential shape. Other shapes are
of course possible and our results can be extended straightforwardly.
by entity X of the bank). We call p the one-year average uncondi- From (1), we derive the value of AR as a function of the parame-
tional probability of default within the loan portfolio. We consider a ters of the ROC function in the particular cases of the binormal (see
rating model that produces a continuous score over the set of debtors equation (3.14) in Tasche (2010)) and exponential approaches:
in the portfolio. The higher the score assigned to a loan, the lower its 9
a >
probability of default. We rank the debtors according to their credit- Binormal t .b D 1/ AR D 2N p 1 >
=
2
worthiness, starting with those that have the lowest scores and going (4)
1 1 1 > >
to those with the highest scores. Let us consider the fraction x of the Exponential t AR D 2 ;
1 ek k 2
debtors having the lowest scores. Among all the defaulters in the port-
folio, we call HR.x/ the hit rate function, which is the proportion
of defaulters that have been classied correctly regarding the score The stage 2 hit rate target value
value corresponding to x. Similarly, we call FAR.x/ the false alarm In contrast to the Basel framework, in which rating systems are
rate function, which is the proportion of non-defaulters that have been required to have at least seven grades for non-defaulted loans, the
classied incorrectly regarding the score value corresponding to x. IFRS 9 norms introduce an unusual classication for non-defaulted
The cumulative accuracy prole (the CAP curve) is obtained by loans with only two grades: stage 1 and stage 2. The resulting clas-
plotting HR.x/ when x ranges from 0% to 100%. The receiver operat- sication of loans performs less well than the original score because
ing characteristic (the ROC curve; see gure 1) is obtained by plotting of the loss of information in the bucketing process. Indeed, to assess
HR.x/ as a function of FAR.x/ when x ranges from 0% to 100%. In the two-stage classication rule, we assume that the ranking of the
what follows, we call the ROC function R.u/, where u D FAR.x/ is loans within each stage is random, but loans in stage 1 all have a better
the false alarm rate. For a random scoring model, the hit rate is equal ranking than loans in stage 2. The resulting ROC curve, named R2 .u/,
to the false alarm rate for all x, and the ROC curve is the diagonal of where u is the false alarm rate obtained with the new rankings, is an
the unit square; the area under the ROC curve, called AUC, is equal to afne function per interval (the purple line in gure 1).
1
. For a perfect model, the hit rate is always equal to 100% and AUC is Let us call AR2 the two-stage accuracy ratio and the stage 2 hit
2
equal to 1. The CAP and ROC curves are closely related to each other, rate, ie, the proportion of defaults captured by the stage 2 portfolio.
as well as to their associated summary statistics (see Engelmann et al We show in the appendix that AR2 is linked to the stage 2 hit rate :
2003):
Z 1 AR2 D R1 ./ (5)
AR D 2AUC 1 D 2 R.u/ du 1 (1)
0
We obtain a direct relationship between the proportion of defaults that
When considering real data, both the CAP and ROC curves are stage 2 catches and the performance of the two-stage model. We see
noisy. The most popular t of the ROC curve used in statistics is the geometrically that AR2 6 AR due to the loss of information in the
binormal approach, which is based on a two-parameter family of ROC bucketing process; we propose to undertake the bucketing in such a
functions (Hanley 1996): way as to minimise this loss of information. From (5), we show that
AR2 reaches a maximum for D , which is the solution of:
B
Ra;b .u/ D N.a C bN 1 .u// (2)

@R1 ./
where the function N./ is the cumulative normal distribution function. D1 (6)
@ D
This type of ROC function corresponds to a normal distribution of the
scores of both the defaulters and the survivors, which is a reasonable The selected value of depends on the calibration of the transfer
assumption in practice. The case b D 1, for which the score volatility criterion from stage 1 to stage 2. Transfer criteria for which >
is the same for defaulters and survivors, is often used in the statistical are not relevant because the two-stage accuracy ratio decreases for

74
78
27 risk.net
risk.net May
risk.net May 20162016
2016
October


 Cutting edge: Credit risk
Cutting
Model edge:
building: Creditedge
Cutting risk

this range of parameters. After some algebra, we obtain a one-to-one

2 Optimal hit rate and two-stage accuracy ratio as functions of
relationship between the underlying scoring model accuracy ratio and the score accuracy ratio (AR)
the maximum target hit rate from (6) and (3):
p 9 100%
2 1 1 C AR >
Binormal t .b D 1/  D N N >
=
2 2 (7) 80%
 1 1 1 C AR >>
;
Exponential t D D
1 ek k 2 60%

We note that in the binormal case we obtain the same result as that
40%
obtained by Tasche (2012) using a criterion based on misclassication
cost. We plot the values of the maximum target stage 2 hit rate  and Target hit rate (binormal b = 1)
20% AR2 (binormal b = 1)
of the associated two-stage accuracy ratio AR 2 as functions of AR in Target hit rate (exponential)
gure 2. AR2 (exponential)
0%
0 20 40 60 80 100
For realistic values of the accuracy ratio AR between 60% and
Score accuracy ratio (%)
80%, say the maximum value of AR2 is reached when the stage 2
hit rate is in the range 7080% in the binormal case and in the range
8090% in the exponential case. For the exponential ROC function, 3 Target size of the stage 2 portfolio as a function of the score
the link between AR 2 and AR is implicit, with intermediate variable k. accuracy ratio
This link is explicit in the case of the binormal approach, and we obtain
from (5) and (7):
50 p = 1%
p p = 5%
B 2 1 1 C AR
AR2 D 2N N 1 (8) 40 p = 10%
2 2 stage 2 portfolio (%)
Target size of the

The maximum attainable values of AR2 are in the range 4565% when 30
the underlying score accuracy ratio is in the range 6080%, and they
20
are very similar for both approaches. We also observe the maximum
attainable values of AR2 depend only on AR and not on the portfolio 10
default probability.
From a practical point of view, we propose using the stage 2 hit 0
0 20 40 60 80 100
rate and the two-stage accuracy ratio to assess the calibration and per- Score accuracy ratio (%)
formance of the transfer criterion. People in conferences and working
groups sometimes refer to the value of 70% as a target for the stage 2
hit rate, which looks sound at rst sight. However, the above equa- We notice that this formula remains true whatever the transfer criterion,
tions show that the relevant targets for the stage 2 hit rate depend on even if it is not based on a score or a rating system. Let us focus on
the two-stage accuracy ratio or on the underlying score accuracy ratio, the binormal ROC function from now on in order to get some orders
ie, on the quality of the scoring model. In most cases, a stage 2 hit rate of magnitude. When the score accuracy ratio is equal to AR D 80%
of 70% is suboptimal. and the stage 2 hit rate is equal to D 70%, we get ARB 2 D 60:1%;
when p < 15%, the size of the stage 2 portfolio is then in the range
Stage 2 portfolio size formula 1020%. We plot the size of the stage 2 portfolio corresponding to the
Let us call B2 the size of the stage 2 portfolio expressed as a percentage optimal target stage 2 hit rate as a function of the score accuracy ratio
of the total portfolio exposure. The default probability within bucket 2 AR for several values of p in gure 3:
is then equal to p=B2 . The stage 2 hit rate is equal to the probability p
2 1 1 C AR
that a loan is in stage 2 prior to default; from the denition introduced B2 .AR/ D .2p 1/N N C1p (11)
earlier, we have HR D . The stage 2 false alarm rate is equal to the 2 2
amount of surviving loans in stage 2 divided by the total amount of We observe B2 is quite stable when the portfolio default probability
surviving loans of the portfolio: changes and remains considerably above the default probability for
B2 p realistic values of p.
FAR D (9)
1p
IFRS 9 provision proxy formula
These values of the stage 2 hit rate and stage 2 false alarm rate corre-
spond to the co-ordinates of point A in gure 1. By computing the area The amount of provision is equal to the ECL over one year for all
under the two-stage ROC curve (the area under the R2 .u/ function), loans in stage 1 and to the ECL at maturity for all loans in stage 2
we obtain the following relationship (see the appendix): (International Accounting Standards Board 2014). We assume that the
losses given default are the same within stage 1 and stage 2 (this
AR2 .1 p/ D B2 (10) assumption can easily be relaxed). As the stage 1 portfolio has a size

risk.net
risk.net
risk.net 75
79
28


 Cutting edge: Credit risk
Model
Cuttingbuilding: Cutting
edge: Credit riskedge

equal to B1 D 1 B2 , a proxy of the IFRS 9 provision is given by the
following formula:
 
.1 /p p
P D LGD B1 D1 C B2 D2
B1 B2
D p LGD.1 /D1 C D2 (12)
where D1 (respectively, D2 ) is the IFRS 9 risky duration within the
stage 1 (respectively, stage 2) portfolio, dened as the probability
weighted value of 1 invested in the stage 1 (respectively, stage 2)
portfolio over a one-year horizon (respectively, lifetime). Because of
the small probability of default in stage 1, we have D1 DF.1/, where
DF.T / is the discount factor associated with horizon T . Conversely,
defaults and maturity effects are no longer negligible in stage 2; we take
into account the survival rate of the loans thanks to the one-year default
probability within the stage 2 portfolio, which is equal to p=B2 , and
we get approximately:
 
p WAL
D2 WAL 1 DF.WAL/ (13)
B2
To obtain this formula we have computed the duration of a bullet port-
folio with maturity WAL (equal to the weighted average life of the
stage 2 portfolio). In general, banking book loans have a maturity
higher than one year on average and we have D2 > D1 . In such a
case, we see that the provision is not necessarily a decreasing function
of the stage 2 hit rate . A more careful study should then be made
to establish when an accurate model (with a high value of AR) gen-
erates lower provisions (this situation would correspond to negative
misclassication costs).
Conclusion
We obtain two important results in this paper. First, we derived a
quantitative criterion to determine loans that should go into either
stage 1 or stage 2. We obtain the optimal target stage 2 hit rate of the
two-stage classication at a one-year horizon, which is around 70
80% for realistic scoring models, corresponding to the binormal case.
Second, we obtained a formula that links the size of the stage 2 port-
folio with risk and performance indicators. We showed that the IFRS 9
provision is driven by the stage 2 hit rate and not by the size of the
stage 2 portfolio, whatever the transfer criterion between stage 1 and
stage 2.
The proposed approach sets some targets for the stage 2 hit rate
and the two-stage accuracy ratio that can be helpful for calibrating and
backtesting the transfer criteria that banks are using. The proxy formula
may also be used as a simplied approach to computing IFRS 9 pro-
visions when too little data are available for calibration. This formula
can, at least, be used as a benchmark for IFRS 9 provisions.
Appendix: proof of equations (5) and (10)
The area under the ROC curve of the two-stage model, called AUC2 , is
equal to the sum of the areas of the triangle OAB, the rectangle ADEB
and the triangle ADC in gure 1. The co-ordinates of point A are .z; /,
where, from (1), we have D R.z/. The geometry of gure 1 leads
to:
z .1 /.1 z/ 1Cz
AUC2 D C .1 z/ C D
2 2 2
We obtain AR2 D 2AUC2 1 D z D R1 ./. From (9),
the stage 2 false alarm rate links the size of the stage 2 portfolio and
the value of z:
B2 p
z D FAR D D R1 ./ D AR2
1p
This last equation leads to (10). 
Vivien Brunel is the head of risk and capital modelling at
Socit Gnrale and professor of nance at Lonard de Vinci
Ple Universitaire in Paris. The author is grateful to two anony-
mous referees who provided very valuable comments and sug-
gestions. He also thanks Benot Sureau and Yann Trguer from
Socit Gnrales risk department. This article reects the
authors opinions and not necessarily those of his employers.
Email: vivien.brunel@socgen.com. Previously published on Risk.net

REFERENCES
Basel Committee on Banking
Supervision, 2001
The internal ratings-based
approach
Consultative Document, January
Cramer JS, 2003
Logit Models from Economics
and Other Fields
Cambridge University Press
Engelmann B, E Hayden and
D Tasche, 2003
Testing rating accuracy
Risk January, pages 8286
Hand DJ, 2009
Measuring classier
performance: a coherent
alternative to the area under the
ROC curve
Machine Learning 77,
pages 103123
Hanley JH, 1996
The use of the binormal model
for parametric ROC analysis of
quantitative diagnostic tests
Statistics in Medicine 15(14),
pages 15751585
International Accounting
Standards Board, 2014
IFRS 9 nancial instruments
July
Sobehart J, S Keenan and
R Stein, 2000
Benchmarking quantitative
default risk models: a validation
methodology
Moodys Rating Methodology
Tasche D, 2010
Estimating discriminatory power
and PD curves when the
number of defaults is small
Working Paper, available at http://
arxiv.org/pdf/0905.3928.pdf
Tasche D, 2012
Bounds for rating override rates
Journal of Credit Risk 8(4),
pages 329
Van der Burgt M, 2008
Calibrating low-default portfolios
using the cumulative accuracy
prole
Journal of Risk Model Validation
1(4), pages 1733

29
76
80 risk.net
risk.net
risk.net MayOctober
May 2016
2016 2016
 IT infrastructure

Implications for IT systems

Joined-up approaches are needed to couple risk and finance spaces in an effort at continuous default risk monitoring

1 Affected IT infrastructure for IFRS 9 projects

Scorecards and metrics Early warning system

Data mining
and analysis Reporting and dashboards
Taxonomies

Scenario and

Internal
Data marts OLAP cubes
Instruments

audit
stress testing

governance
Analytic
Messaging and business rules
Data model

IT
engines
Operational BI appliances
Metadata

Information security
Data warehouse Market risk
Reference and

Credit risk
master data

Financial
controls
Liquidity risk
RAROC

compliance

Source: Chartis, IFRS 9 Technology Solutions, 2016

Policy and
Operational data storage
Data quality

Finance
Prices

Operational
Trusted sources

risk
Data cleansing Changed data
and enrichment capture Others
Client &
account
mining
Data

Banking operations
Cards Treasury Auto loans Core banking

IFRS 9 applies further pressure to financial institutions already faced with Apart from the obvious changes to internal ratings-based (IRB) models
increasing regulatory reporting requirements. These cause particular stress to required for IFRS 9, workflow across the enterprise is a key area for analysis. The
legacy IT systems. extensive reworking of models is shadowed by the requirements for continuous
Impact analysis if IT design and architecture for IFRS 9 is required covers: monitoring of relative changes, since origination of loans and exposures in
Historic data analysis and retention probability of default, loss-given default and exposure at default. Early warning
Analytics systems and limits and exposure measurement systems will all require review to
Calibration ensure they are IFRS 9-enabled.
Monitoring The technologies required for IFRS 9 will need to bridge the gap between risk
and finance. As an accounting initiative, the controls and reconciliations that exist
Figure 1 summarises the affected IT infrastructure for a typical IFRS 9 project for other accounting processes will be a priority. Any repurposing of IRB and risk
those parts most affected are coloured yellow. technology will therefore have to address these emerging functional requirements.

risk.net 30
 IT infrastructure: Whitepaper
Preparing for the IFRS 9 game-changer
How to reconcile the demands of risk
and finance on a single, flexible platform
IFRS 9 is a new accounting standard for financial instruments, which firms around the world must implement by January 1, 2018 at the
latest. It has been designed to value assets and liabilities in a more risk-sensitive manner than the incumbent IAS 39. Despite being an
accounting standard and not a supervisory measure, it introduces a new approach to credit risk
IFRS 9 presents many challenges for financial firms,
including accommodating the differing demands
of risk and finance, and managing large volumes
of risk and financial data, which must be refreshed
at more regular intervals than ever before. This
whitepaper reveals how these challenges can be
overcome by using a single, integrated platform with
a flexible data model.
Overview of IFRS 9
implementation challenges
The major challenges of implementing IFRS 9
revolve around the requirement to accommodate
the perspectives of both risk and finance. Historically,
these two functions have operated in isolation from
each other, developing very different cultures. To
grasp the uncertainties of the future, the risk world is
fond of statistics, which it uses to analyse historical
patterns and spot recurring behaviour. In essence, it is
a world based on principles rather than formal rules.
In contrast, the world of finance and, more specifically,
accounting is characterised by many detailed rules. At
times, the excessive complexity of these rules results
in situations that defy common sense. These very
different cultures have deep roots in the bank, and
they translate into different types of organisations:
typically, risk management has a degree of liberty as
long as it is compliant with regulations; accounting, on
the other hand, has to strictly follow accounting rules
with no kind of freedom.
Accordingly, risk and accounting data is
organised differently, and the goal of combining
these different types of data is a real challenge. This
whitepaper analyses the technical challenges of
implementing IFRS9 and suggests effective ways
for firms to maximise their return on investment in
data management.
The common feature of the solutions proposed in
this whitepaper is the ability to feed processes and
31 risk.net October 2016
reports with data that is dynamically collected from
a number of internal and external sources, with no
duplication of data and no need to change the IT
infrastructure. This enables a rapid implementation
at limited cost, all the while maintaining strong data
lineage for mining and auditing. This is exactly what
AxiomSLs state-of-the-art platform offers. Proper
data management will produce significant benefits
by integrating risk and finance, but only if the
following challenges are addressed:
Data
 handling must be upgraded to support joint
risk and accounting compliance.
The nature of internal databases must be changed.
The divergent goals of chief risk officers (CROs) and
chief financial officers (CFOs) must be addressed.
Conflicting regulatory and accounting views of risk
must be reconciled.
Each of these issues is described below and
appropriate solutions are proposed.
1. Upgrading data handling for joint risk
and accounting compliance
The Basel Committee on Banking Supervisionsregula-
tion 239 (BCBS 239), Principles for effective risk data
aggregation and risk reporting, refers to risk data,
not accounting data, which is managed based on ac-
counting standards. Nevertheless, implementing IFRS 9
requires a large amount of data that will have to fulfil
both sets of requirements. So, what is the underlying
logic of each framework? How do they conflict? And
how can they be accommodated?
Accounting aims to provide accurate and
auditable figures. Billion-dollar balance sheets must
be published with figures that are precise down to
the last dollar, and it must be possible to trace these
financial figures back to their sources. Moreover, the
figures must be disclosed within tight timelines and
must be accurate and transparent.
All of this financial data is intended to
inform stakeholders how well the company is
performing, so they can make informed decisions
about its future. The financial figures are also
used to calculate taxes a serious matter. This
is why the disclosures bear the signatures of
both the chief executive officer (CEO) and CFO.
Inaccuracies, even if inadvertent, may result in
a prison sentence. Consequently, accounting is
performed in a very controlled, structured and
rigid production environment.
Risk works in a different, less structured
ecosystem because it is intended to support
decision-making in an uncertain environment. The
risk function tries to reduce the uncertainties of the
future by clarifying the alternatives. The resulting
insights are used to support decisions on which
risks to take and to what extent, and which risks to
hedge or avoid. Outputs are holistic by nature risk
appetite, risk tolerance, risk hedging, etc.
Despite the efforts of regulators to formalise and
industrialise risk assessment, practitioners need vision,
anticipation, openness and reactivity when measuring
risk and supporting risk/return decision-making. This
calls for flexible and easy-to-change decision-support
systems with an underlying IT environment that
should also share these characteristics.
Trying to combine the goals of accountants and
risk managers leads to the following question: can
data be managed in a controlled, structured and
rigid production environment while simultaneously
being available for open, flexible and evolving
decision support?
AxiomSLs platform provides both the controlled
environment required by accounting and the
flexibility required by risk. By imposing no constraints
on where the data is located and by avoiding
duplication and double storage, data can be used in
a rigid production structure as well as in a flexible,
 IT infrastructure

adaptable decision-support environment. Complex

1 Options Main benefits Main challenges
logic can be defined graphically and understood by
CFO owns the data First-hand IFRS 9 user Understanding and using risk data
both IT and business users. The platform enriches the Controls everything

data, but retains the links to all sources, providing Provisions managed dynamically

full data lineage. Thus, all data changes, whether due CRO owns the data Maintains coherence Complying with accounting requests
to human interaction or system logic, are tracked Regulatory compliance
Knows risk systems

and auditable. Full data lineage is also critical during

Duplicate sets of data one for CRO Both CRO and CFO own The possibility of discrepancies
testing and production to understand the results.

and one for CFO  Can implement different

The ability to drill down to the raw data sources, frequencies, perimeters and
which are already known to the users, is essential for content constraints

establishing trust in the system and for reassuring

management about the reliability of the final results.
2. Changing the nature of
internal databases
In order to implement IFRS 9, databases will need to
be larger, more dynamic and open to external data.
Under IFRS 9, the correct assessment of expected
credit loss (ECL) will require significant changes to
internal databases. The fact that almost all assets
will have to be impaired will dramatically increase
data granularity and data volumes; the point-in-time
(PIT) requirement will make high-frequency updates
compulsory; and the forward-looking approach will
require a significant inflow of external financial and
economic data.
Why will data need to be more granular? Assessing
ECL under the IFRS 9 framework means calculating
impairments for all assets in the amortised cost or
fair value through other comprehensive income
categories. These two accounting categories should
comprise the bulk of all assets. Therefore, the number
of assets to be impaired will increase dramatically
from thousands to millions. This is a whole new ball
game: in terms of data management, the processes of
purifying and enriching the data, and the operations
of extraction and transfer will have to be massively
enhanced and automated.
Why will data need to be updated more
frequently? The spirit of IFRS 9 is to identify credit
risk increases as early as possible. This is apparent in
the PIT requirement and the changes to the buckets
in which assets are grouped. PIT measures such as
PIT probability of default (PD), loss-given default
(LGD) and joint default correlations will probably
have to be updated at least once a month. This is
the update frequency for many macroeconomic
indicators affecting the markets. Higher frequencies
must also be considered: the supervisory guidance
specifies that ECL must capture all significant
increases in credit risk (Guidance #45), so ECL
assessment must be linked to market information
and must be ready to be updated at short notice.
Why will external data be needed? ECL is assessed
as a probability-weighted discounted cash shortfall.
The weighting refers to possible future scenarios.
These scenarios have to include information that is
available ... at the reporting date about past events,
current conditions and future economic conditions
(IFRS 9 5.5.17). As ECL changes will have to be
disclosed and explained, a significant mass of
external economic and market data, both qualitative
and quantitative, will have to be collected and stored
for future retrieval and analysis.
How can financial firms design an open solution
that can handle this granularity, these frequent
updates and large volumes of external data?
AxiomSLs platform can store, manage and control
data at different levels of granularity; and the
frequency at which this data is updated can be
tailored to meet the needs of individual parts of the
business. The AxiomSL platform is unified and can
be used to control all data and processes. A single
environment, team and architecture can be leveraged
to manage large volumes of historical, retail and
statistical data and precise balance-sheet information
for corporate clients. This unique approach allows
financial firms to reuse resources, including
hardware, software and teams, leading to significant
reductions in the total cost of ownership (TCO).
3. Addressing the goals of both
CROs and CFOs
IFRS 9 compliance will require risk data. Both the
CRO and CFO will want to own this data because it
is critical raw material they need to meet their own
current and emerging requirements. The CFO signs off
the financial accounts and is liable to the shareholders
and the auditors with regard to the transparency and
fairness of the disclosures. Consequently, s/he is used
to controlling accounting data from A to Z. The new
issues faced by the CFO include a massive increase in
the number of assets to impair and a higher volatility
of provisions, which s/he will have to dynamically
forecast and manage. Securing the quality and
availability of data will be a critical success factor.
The concern is the same for the CRO: s/he is
on the frontline managing regulatory pressure,
including compliance with BCBS 239, Basel III, the
Asset Quality Review (AQR) and European Banking
Authority (EBA) stress tests. The CRO must deliver
complex reports at high frequency. To that end, s/
he has an explicit responsibility from the board
to fulfil all upcoming supervisory requirements.
Moreover, s/he will undoubtedly be summoned
by finance to provide the credit risk data required
for IFRS 9 compliance. As a result, the new issues
facing CROs include managing more granular data
down to the transaction level, moving towards PIT
and introducing forward-looking components into
their analysis.
So, should the CFO or CRO own the credit risk
data used for ECL assessment? Should the data be
duplicated or can it be shared? Will CEO arbitrage
be necessary or should there be a chief data officer
(CDO) position to decouple data sourcing and usage
(figure 1)?
The current trend is to give ownership of the
data to the CFO because s/he owns the impairment
process and because the data used to assess PIT
measures and the forward-looking stance are mostly
related to the accounting standard. Nevertheless,
CRO supervision is usually granted. AxiomSLs state-
of-the-art dynamic data model makes it possible to
implement any of these options. For instance, giving
the CFO ownership of credit risk data can be done
without changing the risk IT infrastructure.
4. Reconciling conflicting regulatory
and accounting demands
IFRS 9 introduces a nave and refreshing view of
how to assess credit risk. This includes using past and
present data as well as available forecasts, assessing
risk based on the whole life of transactions and
properly including diversification effects.
In contrast, even the most sophisticated current
regulatory approach, the internal ratings-based (IRB)
approach, does not offer a proper way to assess
credit risk. This methodology, unchanged for the past
12 years, assesses credit risk using through-the-cycle
(i.e. backward-looking) measures. It only looks at a
one-year time horizon, regardless of the real maturity
of the transactions. Diversification and concentration
effects are still based on an oversimplistic asset
correlation methodology. Recovery risk assessment
is collapsed into one single figure, the downturn LGD.
Facing the tsunami of changes that are to come,
competent authorities are trying to stay in control.
The Guidance on credit risk and accounting for ECLs

risk.net 32
 IT infrastructure: Whitepaper

was published by the Basel Committee in December

2 Regulatory view IFRS 9 view
2015. Regarding loan portfolios, it adds to and
EAD expected value one year ahead (typically nominal + EAD = net present value (NPV) for two of the three
reinforces the IFRS 9 requirements. The guidance one year of interest) asset classes
was developed on the principle of non-objection by
PDs through-the-cycle PDs point-in-time
the International Accounting Standards Board (IASB)
One-year PD Multi-year PDs
and, consequently, it remains quite soft. In particular,
Downturn LGD LGD distribution
even though it mentions the discrepancies between
the approaches (...regulatory capital models Best worst case Weighted average scenarios

may not be directly usable in the measurement of One-year time horizon One-year and lifetime time horizon

accounting ECL due to differences between the

objectives of and inputs used for each of these
3 AxiomSL process flow for IFRS 9
purposes [Guidance#9]), it gives no indication of
how to overcome them.
Moreover, the same vocabulary is used across Collective impairment
Financial
the accounting and regulatory frameworks, and assets Measurement Portfolio Staging Through-the- Point-in-
Retail

classification segmentation cycle PD time PD

is certain to be a source of confusion. Commonly & risk drivers Definition of a
Multi-period Statistically or
Business On initial significant
selection
used parameters of credit risk assessment, such as model recognition change in
PD transition
matrices
expert base
adjustment
credit risk
exposure at default (EAD), PD, LGD, correlations and SPPI criteria Subsequent
measurement
expected loss, are used indiscriminately, despite their Amortised
Cost Individual impairment Expected credit
different concepts and realities (figure 2). FV through
loss (ECL)
Corporate

OCI Historical Monte Carlo simulation of balance sheet ECL 1 year for
It is therefore essential to have explicit definitions, FV through
Corporate (Algo Save) stage 1 and
ECL lifetime
Customers Produces correlated scenarios of default for stage 2 and 3
a well-defined data model and attributes that remain P&L
Financials Point-in-time PD and LGD assets and
Disclosures &
attached to the data wherever it is sent or used. reporting
Two features of AxiomSLs state-of-the-art, flexible
Bank
Market and
data model are particularly important for fulfilling exposures Effective interest rate
Exposures

macroeconomic data
Customer
the above requirement: on the one hand, every data Product
Amortisation and cashflow generator
Guarantees
source is clearly identified, named and mapped in Collaterals

the data model; on the other, even when delivered

within complex reporting, the data remains in its
source environment and is not duplicated. This data
lineage avoids the risk of losing track of the datas
nature and identity. It also facilitates data mining
and auditing, and guarantees the reliability of the
reporting, all the while allowing for quick checks
and controls at any stage of the reporting process.
Conclusion IFRS 9 calls for a
game-changer: AxiomSL
Fintech is a game-changer. Data has become
a critical asset for all financial institutions, and
fintech companies enable firms to manage it more
efficiently than ever.
IFRS 9 is premised on widespread adoption
of fintech its requirements would have been
unthinkable just a few years ago, before technology
played such a significant role at financial firms.
As demonstrated above, IFRS 9 will require:
an ever-increasing quantity of granular data;
more frequent data updates to support the PIT
approach;
the challenge of sharing the same data between
users with different objectives and concerns; and
the automation of complex processes, such
as ECL calculations for impairments, without
compromising data lineage and auditability.
AxiomSL, the leading global provider of regulatory
calculation and reporting solutions, has anticipated into the AxiomSL platform. AxiomSL has formed a
these challenges and offers a robust technology partnership with AlgoSave and has integrated its
platform that is fully equipped to implement IFRS 9. model, which assesses ECL using historical financial
The fully integrated platform is designed to enable data, current market data and economic forecasts
financial firms to make the best decisions in terms (Figure 3). The model has been field-tested in the
of organisation and modelling, while reducing the asset management industry and is now being made
time and effort needed to access and manage the available to other types of financial institutions. In
relevant data. this way, AxiomSL offers a compelling solution to the
External ECL models can be easily integrated challenges presented by IFRS 9.
The author
Jean-Bernard Caen is a subject matter expert at AxiomSL specialising in risk and
finance. He is responsible for supporting AxiomSLs client base in analytical and
reporting issues on IFRS 9 and providing direction and insights into the firms
professional services and product development teams.
Prior to joining AxiomSL, Jean-Bernard was head of economic capital and
strategy for Dexia Group for 12 years, where he was in charge of Basel II and Pillar
2 implementation, and risk/finance co-operation.
In 1990, Jean-Bernard founded Finance & Technology Management, a
management consulting firm he ran for 12 years as chief executive officer. As such,
he directed numerous assignments for European financial institutions in the areas
of shareholder value, risk management, capital allocation and asset-liability management.
Jean-Bernard is a member of the Professional Risk Managers International Association France executive
committee, of the Association Francaise de la Gestion Financire management board and teaches at the French
National School of Economics and Statistics. He is a French civil engineer and graduated from Massachusetts Institute
of Technology.

33 risk.net October 2016


Oracle Financial Services Analytical Applications (OFSAA) provides financial
institutions solutions for risk, treasury, finance, financial crime, compliance
and the front office. The applications are built on the integrated and unified
Oracle Financial Services Data Foundation. This draws on a comprehensive
understanding of the industry over many years and brings together various
interlinked disciplines in a single source of truth cleansed, standardised,
reconciled and ensuring completeness, accuracy and financial integrity
of data. Shared data, metadata, computations and business rules enable
institutions to meet emerging business and regulatory requirements with
reduced expenses.
A common analytical infrastructure underpins the OFSAA suite with
metadata-driven R modelling capabilities and the industry-leading Oracle
Business Intelligence platform for management reporting. The OFSAA
infrastructure, built on open technologies, provides a set of tools to build
and maintain applications using a common business language across
the analytical applications suite and helps financial institutions leverage
existing investments.
OFSAA offers a single, unified IFRS solution that is compliant with IASB
Moodys Analytics helps capital markets and risk management professionals
worldwide respond to an evolving marketplace with confidence. Its expertise
and experience in credit analysis, economic research and financial risk
management enable the company to offer unique tools and best practices
for measuring and managing risk. By providing leading-edge software,
advisory services and research, including proprietary analyses from Moodys
Investors Service, Moodys Analytics integrates and customises its offerings
to address specific business challenges.
Moodys Analytics suite of credit risk models and data, economic forecasts,
advisory services and infrastructure solutions can assist with the granular
and dynamic methods required for implementing expected credit loss (ECL)
and impairment analysis under the new IFRS 9 accounting standard.
Moodys Analytics offers a wide range of solutions for implementing IFRS 9
requirements. Its advisory services support clients with IFRS 9 roadmap design,
model deployment, simulation of IFRS 9 provisions and implementation.
Credit modelling solutions allow firms the opportunity to implement internally
developed or off-the-shelf models and incorporate forward-looking information
into existing frameworks. Comprehensive and granular credit risk, economic and
financial data sets help capture and collect historical data required for building
Sponsor profiles
standards while using a modularised design approach. This allows for
easier integration and client-specific extension and allows banks to perform
granular calculations on a common set of data. The solution is integrated
with other Oracle applications related to regulatory and accounting
purposes, such as Basel reporting, regulatory capital reporting, Oracle
Financial Accounting Hub and Oracle General Ledger. Oracles approach
provides the following benefits:
Maximisation of existing investments in risk and finance systems, allowing
reutilisation of existing data, business rules and technology infrastructure,
reducing costs and time-to-market.
Consolidation of technology and data repositories for risk and finance.
The availability of a comprehensive, end-to-end solution to address IFRS 9,
from data management and computations to accounting and reporting.
OFSAAs IFRS 9 solution allows banks to actively incorporate risks into their
decision-making and to deliver actionable customer, business-line and
profitability insights. In addition, it helps the bank promote a transparent
risk management culture and pervasive intelligence across its departments.
www.oracle.com
a forward-looking impairment model, developing quantitative credit risk models
and benchmarking. Credit impairment analysis software facilitates the end-to-
end process of ECL calculations by centralising data from numerous sources,
co-ordinating and managing a wide variety of models, evaluating changes in
credit risk and calculating expected losses and provisions accordingly for export
to external accounting systems.
Moodys Analytics has more than 20 years of experience forecasting
challenges and helping financial institutions successfully address their credit
loss estimation. This experience, combined with deep domain expertise,
in-house economists, extensive data sets and modelling capabilities, and
award-winning regulatory and enterprise risk management software, is the
foundation for successful credit loss and impairment analysis.
Moodys Analytics has been recognised for its award-winning tools for
measuring and managing risk. Notably, Moodys Analytics was voted#1
in enterprise-wide credit risk management in the Risk Technology
Rankings 2015, is a Preferred Vendor in the IDC Financial Insights FinTech
Technology100 Rankings, and is recognised as a category leader in the
FinTech Quadrant for IFRS 9 technology solutions 2016.
www.moodysanalytics.com
risk.net 34
 Sponsor profile

AxiomSL is the leading global provider of regulatory reporting and risk
management solutions for financial services firms, including banks, broker
dealers, asset managers and insurance companies. Its unique enterprise
data management platform delivers data lineage, risk aggregation, analytics,
workflow automation, validation and audit functionality.
The AxiomSL platform seamlessly integrates clients source data from
disparate systems and geographic locations without forcing data conversion.
It enriches and validates the data and runs it through risk and regulatory
calculations to produce both internal and external reports. The platform
supports disclosures in multiple formats, including XBRL. The unparalleled
transparency offered by the high-performance platform enables users to drill
down on their data to any level of granularity.
AxiomSLs platform supports compliance with a wide range of global and
local regulations, including Basel III capital and liquidity requirements, the
Dodd-Frank Act, Fatca, AEI (CRS), Emir, Corep/Finrep, CCAR, FDSF, BCBS239,
Solvency II, AIFMD, IFRS, central bank disclosures and both market and
credit risk management requirements. The enterprise-wide approach
offered by AxiomSL enables clients to leverage their existing data and risk
management infrastructure, and reduces implementation costs, time-to-
market and complexity.
AxiomSL was voted best reporting system provider in the 2015 Waters
Rankings and was highlighted as a category leader by Chartis Research in
its 2015 sell-side risk management technology report. The companys work
has also been recognised through a number of other accolades, including
success in the best reporting initiative category of the American Financial
Technology Awards and the customer satisfaction category of the Chartis
RiskTech100 rankings.
www.axiomsl.com

About Chartis

Chartis is the leading provider of research and analysis for risk technology
on the global market.Part of Incisive Risk Information, which owns market-
leading brands such as Risk and WatersTechnology, Chartis goal is to
support enterprises as they drive business performance through better risk
management, corporate governance and compliance, and to help clients
make informed technology and business decisions by providing in-depth
analysis and actionable advice on virtually all aspects of risk technology.
Chartis is focused solely on risk and compliance technology, providing a
significant advantage over generic market analysts.
Chartis has brought together a leading team of analysts and advisers from
the risk management and financial services industries. This team has hands-
on experience of implementing and developing risk management systems
and programmes for Fortune 500 companies and leading consulting houses.

Our areas of expertise include: Visit www.chartis-research.com for more information, or join our global
Credit risk online community at www.risktech-forum.com
Operational risk and governance, and risk and compliance
Market risk
Asset-liability management and liquidity risk
Energy and commodity trading risk
Financial crime, including trader surveillance, anti-fraud and
anti-money laundering
Cyber-risk management
Insurance risk
Regulatory requirements including Basel II/III, Dodd-Frank,
Emir and Solvency II

35 risk.net April 2016