The Central Bank of Egypt “CBE” has adopted IFRS in year 2008. In specific IAS 39 has a discussion about implementing a model that can derive the incurred credit losses for a pool of receivables/ loans, which was quite open for market development & practical initiatives.
From the part of the CBE, it has adopted same approach, which led to some wide different market practices, logic, and interpretations, which sometimes have been questionable on a wide scale basis!
So, I've thought to develop some sort of materials that can serve as a practical guidance for quantifying the credit risk, using different simple models, based on Basel II definitions of the risk components.
The intended users of this material are the credit risk professionals who conduct risk analysis, implement risk management policies, or/and are in charge of quantifying the credit risk for a loan portfolio (corporate & retail).
Also, other professionals or officers complying with IFRS, or CBE GAAP.
1. Credit losses modeling
„Deriving the incurred credit losses using Basel II
definitions for the risk components in line with the
CBE adopted IAS 39'
February 26th, 2012
Yahya M.Kamel
Financial Services Assurance
Banking & capital markets
2. Disclaimer
The information presented in here are
the explicit views of the author, and he is
held irresponsible for any loss or damage
caused by the use of these information.
Page 2 Market practice series “Credit losses modeling”
3. About this publication
The CBE has adopted IFRS in year 2008. In specific IAS 39 has a discussion about implementing a model that can derive the incurred credit losses
for a pool of receivables/ loans, which was quite open for market development & practical initiatives.
From the part of the CBE, it has adopted same approach, which led to some wide different market practices, logic, and interpretations, which
sometimes have been questionable on a wide scale basis!
So, I've thought to develop some sort of materials that can serve as a practical guidance for quantifying the credit risk, using different simple
models, based on Basel II definitions of the risk components.
The intended users of this material are the credit risk professionals who conduct risk analysis, implement risk management policies, or/and are in
charge of quantifying the credit risk for a loan portfolio (corporate & retail).
Also, other professionals or officers complying with IFRS, or CBE GAAP.
The paper includes interpretative guidance, illustrative cases, and analytical discussions to clarify the practical application of the accounting
standard IAS 39, which is mainly aiming at measuring the credit risk and help answer some of the market doubts & pre-assumptions, like the ones
below:
► Trade finance incurred losses should bear the same loss rates as for the direct exposure, you don‟t have to, even more credit losses might have
been overstated!
► Retail non-performing loans should be fully charged to P&L as a loss, you don‟t have to!
► The complex risk factors, like LCP, CCF, LGD, PD require highly developed systems, otherwise it can‟t be calculated, with few MS Excel skills
you can handle it!
► The documentation requirements are unspecific and time-consuming, important & can be short-listed!
Scope:
IFRS IAS 39, AG 84-92 „adopted by CBE‟; “Collective impairment assessment for a pool of receivables, bearing common credit risk”.
Page 3 Market practice series “Credit losses modeling”
4. About this publication
Acknowledgements:
This publication has been developed by Yahya Kamel of the Financial Services Assurance Office at Ernst & Young in Egypt “EY”, with no reference
or co-suggestions with other EY officials; that makes the paper the only explicit views of the author with no legal claims on EY.
Suggestions, comments, as well as inquiries regarding credit risk management and quantification from readers of the materials will be much
appreciated.
Please feel free to communicate directly with me, Linked-In: http://eg.linkedin.com/pub/yahya-kamel/4b/3b2/565
► Yahya Kamel
Financial Services Assurance Office, Audit senior, Ernst & Young
Other practice publications:
Through our market practice series, we have been working within the banking market, figuring out dilemmas, problem accounting matters, and
trying to provide practical guidance that can better assist resolve those problems and give some answers to problem matters.
Page 4 Market practice series “Credit losses modeling”
5. Market practice series
Credit losses modeling
Part I „Deriving the incurred credit losses using Basel II
definitions for the risk components in line with the CBE
adopted IAS 39‟
Credit losses modeling
Part II “Statistical migration analysis”
Credit losses modeling
Part III “Structured credit risk models”
Developing a valuation technique
„Deriving the FV for the inactive debt securities &
assessing impairment for the unquoted financial
securities in line with the CBE adopted IAS 39‟
Page 5 Market practice series “Credit losses modeling”
6. Content
1.List of abbreviations 7.Questionable market practices
2.Credit losses model objectives 8.Credit risk model documentation
3.Model adoption road map 9.FAQ
4.CBE new GAAP 'Credit losses assessment' 10.Data requirements
5.Expected Vs. Incurred credit losses
6.Credit risk measurement
1. Credit exposure segmentation
2. EAD
3. Measurement methods
A. Historical charge-off method
B. Migration analysis- simplex method
i. Credit and recovery event
ii. Probability of Default “PD”
iii. Loss Given Default “LGD”
4. Economic and market assessment
5. Model validation and back-testing
6. Reference data sets
Page 6 Market practice series “Credit losses modeling”
7. List of abbreviations
ALL: Allowance for Loan Losses Yr: Year
PD: Probability of Default IFRS: International Financial Reporting standards
LGD: Loss Given Default IRB: Internal ratings-based approach
EAD: Exposure At Default IASB: International Accounting Standard Board
CCF: Credit Conversion Factor LEQ: Loan Equivalent Exposure
LEQ: Loan Equivalent Exposure SD: Standard Deviation
LCP: Loss Confirmation Period GAAP: Generally Accepted Accounting Principles
RR: Risk Rating
NPL: Non Performing Loans
NPER: Number of Periods
EM: Effective Maturity
EIR: Effective Interest Rate
CBE: Central Bank of Egypt
PV: Present Value
Q: Quarter
Page 7 Market practice series “Credit losses modeling”
9. Credit losses model objectives
1. Crucial for investment decision making process that can be translated into:
a. Credit limits in a form of expansion or contraction loan investment policy
(geography limit, product, branch, sector limit,.. etc.).
b. Potential product opportunities.
c. Strengthening the underwriting procedures for certain segment or branch (branch,
product, sector, .. etc.)
e.g., deteriorated credit risk in certain branch portfolio might be due to fraudulent
underwriting, poor underwriting or weak monitoring procedures, which may require
higher level scrutiny and more strict underwriting procedures.
d. Product pricing to cope up with the increasing credit risk
e.g., a deteriorated credit worthiness, would require a decreasing portfolio credit
limit and higher interest rate to compensate for the expected higher credit losses.
Page 9 Market practice series “Credit losses modeling”
10. Credit losses model objectives
2. Compliance with financial reporting and regulatory bodies; in terms of the
credit loss reserves and the capital requirements.
It‟s note-worthy that soon or later the CBE will take time to review the basis
of calculation of the credit models for the banks operating in Egypt, by then
any credit model lacking proper rationale or reasonable risk studies,
supported by proper back-testing; might not be approved by the CBE, thus
leading to un-liquidation of the credit losses reserved in Equity; whether the
credit losses reserves created at the 1st time adoption of new CBE GAAP,
or the reserves created for the difference between the old and the new CBE
impairment standards.
Page 10 Market practice series “Credit losses modeling”
12. Model adoption road map
1. Determine the Credit losses Model objective.
e.g., to assess the incurred or expected credit losses.
2. Determine which credit loss measurement Model to be adopted.
e.g., Historical charge-off Model, Migration analysis Model (Statistical „EL‟, Non-statistical
„Incurred losses‟), or structured models (Merton‟s Valuation Model 'KMV', Moody‟s, KPMG‟s
Loan Analysis, Credit Metrics, Credit Risk Plus „Mortality rates‟, CPV-Macro, ..etc).
3. Determine the Model‟s parameters.
e.g., General parameters: Segmentation rule, Credit and recovery event, Periods assessed, Time-
horizon, Specific parameters (EAD, PD, LGD).
4. Determine the required data.
e.g., LGD using the simplex method= 1-Recovery rate,
Recovery rate= (Recovered amount or exposure – Costs)/Default exposure.
5. Determine the available raw data, and what‟s needed to be developed in the
future course of business.
e.g., For the LGD, recovered amounts not available, but rating recoveries available.
6. Determine the required job staffing, experience, and training.
7. Determine the time-table for the Model adoption plan.
Page 12 Market practice series “Credit losses modeling”
14. CBE new GAAP 'Credit losses assessment'
Retail collective ALL should be calculated based on the default rates; that is the
historical average recorded ALL „as per the balance sheet‟ divided by outstanding loans
per loan segment.
Reference:
Default rates: CBE new GAAP, page 240-242, 244, “Collective ALL basis of calculation
for the retail loans”, last paragraph.
LCP: CBE new GAAP, page 257-258, “Collective ALL basis of calculation”.
e.g. a retail loan portfolio at a total value of $1200, as of 12/31/2012, assigned a risk
rating of 2 „Bucket 2‟, based on the repayment status, calculate the ALL;
ALL $60= ($1200 * Loss rate 10.2% * LCP 0.5).
Construction Yr.2009 Yr.2010 Yr.2011 Average balances Average loss rate
Bucket 3
Recorded ALL $100 $110 $105 $108
Recorded $1000 $1100 $1050 $1050 10.2%= $108/$1050
exposure
The estimated loss rate almost has the same rate of the historical recorded allowances, but with the LCP, the ALL should
be different, compared to the old CBE GAAP.
Page 14 Market practice series “Credit losses modeling”
15. CBE new GAAP 'Credit losses assessment'
Corporate collective ALL should be calculated based on the default rates; that is the
historical average recorded ALL „as per the balance sheet‟ divided by outstanding loans
per loan segment.
Reference:
Default rates: CBE new GAAP, page 242, “Collective ALL basis of calculation for the
corporate loans”.
LCP: CBE new GAAP, page 257-258, “Collective ALL basis of calculation”.
e.g. a corporate direct loan portfolio at a total value of $200 & other revolving loans
credit commitments of $20, as of 12/31/2012, assigned a risk rating of B „R.R.6‟,
calculate the ALL; ALL= ($200+$20) * Loss rate 10.5%= 11.
Construction Yr.2009 Yr.2010 Yr.2011 Average balances Average loss rate
Risk rating 6
Recorded ALL $100 $110 $105 $108
Recorded $1000 $1100 $1050 $1050 10.2%= $108/$1050
exposure
The estimated loss rate almost has the same rate of the historical recorded allowances.
Page 15 Market practice series “Credit losses modeling”
16. CBE new GAAP 'Credit losses assessment'
As per page 257 in the new CBE GAAP, the CBE opened the door for other credit risk
modeling approaches that may rely on algebraic or statistical equations.
► However, the CBE made it conditioned to the below restrictions to be considered with
any adopted approach:
1. The time value of money,
2. The credit lines different maturities,
3. The adopted approach should derive the incurred losses as per IAS 39.
► The wide difference in the market practices & confusion about the credit risk
measurement in line with the CBE new GAAP was due to the below:
1. The LCP was not clarified within the CBE guidelines,
2. The CBE opened the door for other approaches to be used which might deploy
statistical models, however most of the used statistical models primarily derive the
“expected value”, rather than the “incurred value” of losses,
3. The default rates as set by the CBE (average ALL/ Loans) will result in the same loss
rates as per the CBE old GAAP,
4. Finally, It wasn‟t crystal clear whether the credit commitments over the revolving loans
should be subject to assessment of impairment on gross basis.
Page 16 Market practice series “Credit losses modeling”
17. CBE new GAAP 'Credit losses assessment'
► The wide difference in the market practices and confusion about the credit risk
measurement in line with the CBE new GAAP is due to the incomplete guidance and
unclear instructions about the rationale and basis of calculation of the IRB-based
components, so we maintained to develop our rationale in this presentation from two
main references in addition to the CBE new GAAP guidelines „originally adoption to
IFRS‟; Basel II, US Federal reserve interpretations of the credit losses measurement; in
a way to derive the incurred losses rather than deriving the expected losses as per
CBE new GAAP
CBE new GAAP confused models
Incurred loss „IFRS‟:
IL= EAD*PD*LGD
Old GAAP default rates:
Retail: IL= (EAD * LCP * Average historical allowance rate)
Corporate: IL= (EAD * Average historical allowance rate)
Statistical models to derive the probability of default „EL‟:
EL= E(EAD)*E(PD)*E(LGD)
Page 17 Market practice series “Credit losses modeling”
19. Expected vs. Incurred credit losses
Incurred losses 'IFRS':
► 'Further, the IASB explains that the accounting model adopted is based on
'incurred losses' (rather than, say, expected losses and certainly not on future
losses). It believes that such a model, which does not take account of future
events or transactions, is more consistent with an amortized cost basis of
accounting
► The Board reasoned that it was inconsistent with an amortized cost model to
recognize impairment on the basis of expected future transactions and
events. The Board also decided that guidance should be provided about what
'incurred' means when assessing whether impairment exists in a group of
financial assets'
IFRS, IAS 39 'Impairment' BC 110.
Expected Losses 'Basel II':
► That‟s the future credit losses expected to be incurred in case of default of the
financial security‟s issuer, including and not limited to any contingent
obligations, accrued fees, accrued interest, and any potential payments to
collect the default loan
Page 19 Market practice series “Credit losses modeling”
21. Credit risk measurement
1. Credit exposure segmentation
Loan portfolio segmentation:
► Within the retail asset class category, banks are required to identify
separately three sub-classes of exposures: (a) exposures secured by
residential properties, (b) qualifying revolving retail exposures, and (c) all
other retail exposures
► Segmentation at a sub-portfolio level should be consistent with the bank‟s
segmentation of its retail activities generally. Segmentation at the national or
country level (or below) should be the general rule
► Data on loss rates for the sub-portfolio should be retained in order to allow
analysis of the volatility of loss rates
Source: Basel II
e.g., The secured credit cards‟ holders tend to maintain more frequent pastdues than the unsecured
c.c. holders, by mixing the two portfolios in the calculation of the PD & LGD, we maintain to
keep an over-estimated credit losses „inherent from secured cards probabilities‟.
Page 21 Market practice series “Credit losses modeling”
22. Credit risk measurement
1. Credit exposure segmentation
► The goal of segmentation is to provide meaningful differentiation of the risk,
with each pool composed of exposure with homogeneous credit risk,
accordingly banks should consider the risk drivers, while developing the risk
segmentation
► Segmentation should use relevant borrower risk characteristics that reliably
differentiate a segment‟s risk from the other segments and perform
consistently over time; such as (credit score, loan delinquency, debt-to-
income ratio, product, loan to value ratio, origination age, geography,
exposure amount, origination channel, ..etc.)
► A validation process should be in use to validate the manner upon which the
bank differentiated its loan portfolio into segments
Source: US Federal reserve system, Federal register Vol.69, 2004 notice'.
► For instance the project finance loans tend to bear higher risk than the
ordinary term loans, on the other hand the granted loans to Iraqi region tend
to bear higher risk than the other loans granted to other regions, also the
loans granted to the tourism sector tend to bear different level of risk,
compared to other loans granted to the food and beverage sector
Page 22 Market practice series “Credit losses modeling”
23. Credit risk measurement
1. Credit exposure segmentation
► Credit process and potential what can go wrong:
Credit assessment
Credit monitoring Provisioning
and approval
Credit policy 'Underwriting
Settlement monitoring
procedures'. Credit losses assessment
e.g., Increasing debt burden.
e.g., weak underwriting policies.
Branch compliance with the Portfolio analysis and obligor Corrective action
credit policy. follow-up
e.g., non-compliance with the e.g., poor industry . e.g., trend of losses might require
credit policy, or fraudulent credit reshaping the credit policy,
underwriting. approval process, and/or the
monitoring phase.
Page 23 Market practice series “Credit losses modeling”
24. Credit risk measurement
1. Credit exposure segmentation
Retail portfolio Branch Geography Product Sector
Current - - - -
Bucket 1 - - - -
Bucket 2 - - New product 19% Tourism 10%, Aviation 5%
Bucket 3 - Aswan 15% - Tourism 7%
Bucket 4 Batal 4% Giza 4%, Cairo 8% Club 7%, Car 15% Tourism 3%
NPL „100% EL‟ Wadi Degla Br. 27%, Batal 8% Giza 3% Car loans 9% Tourism .5%
Credit risk concentration is calculated as below:
► Branch concentration: 2 branches had 'B4' of 10% in relation to the total branches
portfolio. Product concentration: (New product portfolio/Total loan portfolio) or (Bucket
balance/Total New product)
Basis of segmentation (credit risk pooling):
► Basis comes from the loan portfolio concentration, for instance a retail portfolio of
$10,000, might „ve two products, one accounting for $9,500 and a new product with
weak underwriting that accounts for $500, thus the pastdues concentration should be
based on two separate product portfolios rather than to the total retail portfolio
Page 24 Market practice series “Credit losses modeling”
25. Credit risk measurement
1. Credit exposure segmentation
► Ultimate retail segmentation could look as below:
Wadi Branch Mohandseen Other branches
'27%' '12%'
Tourism Aviation Other Tourism Aviation Other Tourism Aviation Other
'20.5%' 5% Sectors '20.5%' 5% Sectors '20.5%' 5% Sectors
New product New product New product New product New product New product New product New product New product
'19%' '19%' '19%' '19%' '19%' '19%' '19%' '19%' '19%'
Car loans Car loans Car loans Car loans Car loans Car loans Car loans Car loans Car loans
'24%' '24%' '24%' '24%' '24%' '24%' '24%' '24%' '24%'
Other loans Other loans Other loans Other loans Other loans Other loans Other loans Other loans Other loans
Segmentation analysis:
► In order to easily identify the loss making sub-portfolio for segmentation purposes, an
analysis of the volatility of the incurred losses can be made through calculating the
standard deviation 'SD' of the historical loss rates of the sub-portfolio under analysis
divided by the aggregate segment loss rate to figure out the segments with high
(SD/Avg. loss rate)
Page 25 Market practice series “Credit losses modeling”
26. Credit risk measurement
1. Credit exposure segmentation
► Ultimate corporate loan segmentation could look as below:
Tourism '10%' Construction '13%' Other sectors
Project Term loans Revolving Project Term loans Revolving Project Term loans Revolving
finance '5%' loans '20%' finance '2%' loans '5%' finance '1%' '50%' loans '3%'
'19%' '10%'
Page 26 Market practice series “Credit losses modeling”
28. Credit risk measurement
2. Exposure At Default “EAD”
Exposure At Default:
► For both the direct and indirect credit exposure; All exposures are measured
gross of specific provisions or partial write-offs that might be subject to credit
loss.
► For revolving exposures such as credit cards and overdrafts, each loan EAD
should include both; the outstanding exposure plus estimated net additions to
balances for loans defaulting over the following period.
► The net additions preceding a credit event are supposed to be a rate equal to
CCF, extended to the difference between the authorized credit limit & the
outstanding exposure.
► Changes in the underwriting policies, regarding the revolving loans utilization
might have a decreasing or increasing significant impact on the CCF, hence
the EAD, so banks should consider their policy changes, when developing its
CCF estimates.
► EAD= Outstanding (Principal + Accrued interest +or- deferred fees, premium,
discounts – collateral value) + (CCF * Unused credit commitment).
Page 28 Market practice series “Credit losses modeling”
29. Credit risk measurement
2. Exposure At Default “EAD”
Credit Conversion Factor “CCF”:
► The CCF should differ according to whether the exposure is being committed
or uncommitted.
► A credit line is considered uncommitted if it may be unconditionally cancelled
without prior notice, which in turn should bear less CCF rates.
► CCF: Credit conversion factor, alternatively known as Loan Equivalent
Exposure „LEQ‟
► For accounting purposes; the estimated allowances for the credit commitment
should be separately disclosed as credit commitments‟ provisions rather than
as being part of the allowance for loan losses.
► The collateral value should be once considered, whether as part of the EAD
or as part of the LGD calculation.
Page 29 Market practice series “Credit losses modeling”
30. Credit risk measurement
2. Exposure At Default “EAD”
Credit Conversion Factor “CCF” (cont‟d):
► Two main methods:
1. The Cohort method: under which the CCF is the average % of the additional
drawings for a defaulted credit exposure at a time period, compared to the
original exposure amount at time of default, in one exposure segment.
2. The fixed-horizon method: under which the CCF is the average % of the
additional drawings for a defaulted credit exposure at a time period, compared
to the exposure amount at certain date, regardless of the default date, in one
exposure segment.
► Regardless of the adopted method, the CCF can‟t be negative, thus only the
additional drawings in one exposure or loan segment should be assumed in
the CCF calculation.
► We have adopted the cohort method in this guidance.
Page 30 Market practice series “Credit losses modeling”
31. Credit risk measurement
2. Exposure At Default “EAD”
Illustrative case for the EAD, using cohort method:
► ABC construction Co. has been granted the below credit lines
► Overdraft $1000, Term loan $200
► Risk rating 3, CCF 15%
► At end of the FY2011, the outstanding exposure 'withdrawn principal +
accrued interest +/- deferred charges' has been as below
► Direct exposure: Overdraft $950, Term loan $198
► Indirect exposure: OD credit commitment $50 ($1000 - $950)
► EAD= $198 + $950*15%
► CCF: calculated based on historical conversion rates for similar (risk rated
and industry) obligors, for instance; it has been noted that the average
downgraded obligors from RR.2 to RR.3 had the below utilization history:
Available limit Period 1 Period 2 CCF%
For the downgraded portfolio
RR.2 $100
RR.3 „downgraded, originally from RR.2‟ $75 15%= ($100-$75)/$100
Note: the downgraded exposure should reflect the increased exposure alone, rather than considering the a whole balance that reflects the offset
of both the paid-off exposure „LGD‟ and the extra utilization „CCF‟.
Page 31 Market practice series “Credit losses modeling”
33. Credit risk measurement
3. Measurement methods (Historical charge-off method)
The graph below represents a loan portfolio over a time length of four years, showing the change in the
risk ratings.
► A historical charge-off analysis intends to derive the historical charge-off rate per loan segment,
extended to the period it takes to be a confirmed loss.
► For instance the project finance loan portfolio looks to bear historical CCC-rated loans of within an
average of 40% to 60% to the total portfolio, compared to the other commercial loans, which looks to
bear around 15% historical loss rate.
► A loss confirmation period would capture how long it takes a loan to be a confirmed loss, thus if the
other commercial portfolio borrowers take an average of two years to be a confirmed loss, then the
loss rate should be adjusted from 15% to 30%; meaning there are some other 15% incurred losses,
but still passive to the creditor.
Loan portfolio Other Commercial Project finance
100% 100%
100%
80% 80%
80% 60% 60%
AAA+BBB2
60% 40% AAA+BBB 40%
20% 20% CCC
40% CCC
AAA+BBB 0% 0%
20%
0% CCC
Page 33 Market practice series “Credit losses modeling”
34. Credit risk measurement
3. Measurement methods (Historical charge-off method)
Historical loss rate
Illustrative case for the historical loss rates:
► ALL= (EAD*Historical Loss Rate*Loss Confirmation Period)
► Example of a loan portfolio:
New product Period 1 Period 2 Average
„assumed risk pooling per product‟
EAD $1200 $2300 $2300 „Per.2‟
Current $1000 $2000 $1500
Net charge-offs $100 $50 $75
NPL „100% EL‟ $300 $400 $350
Historical loss rate 40% 23% 28%
= ($100+$300)/$1000 = ($50+$400)/$2000
Environ‟l adj.* 4%
Total Allowance for Loan Losses „ALL‟ $1,339 = (28%* 1.04)
*$2300* 2Yr.
For simplicity, the LCP is assumed to be 2 years „credit line tenor'.
Page 34 Market practice series “Credit losses modeling”
35. Credit risk measurement
3. Measurement methods (Historical charge-off method)
Loss Confirmation Period “LCP”
► Loss Confirmation Period “LCP”: that‟s through examining the past
defaults/charge-offs, the creditor determines that on average the borrower
takes certain time period before it defaults, for instance a retail loan would
take 6 months to default 'moving from current portfolio to NPL „100% EL‟ ',
however a corporate loan would take 2.5 year to default on average, since
the borrower rescheduling or pastdues tend to start after a weaken financial
strength has taken place.
Source: US GAAP “interpretation of the incurred losses Yr.2003”.
► The CBE temporarily set the Loss Confirmation Period to 1 for the 1st year of
adoption of the new CBE GAAP, however banks are required to develop their
own.
► Another definition can be found in Basel II, known as the EM, also known as
Macaulay duration.
Page 35 Market practice series “Credit losses modeling”
36. Credit risk measurement
3. Measurement methods (Historical charge-off method)
Loss Confirmation Period “LCP”
► The Effective Maturity “EM”: that‟s the maximum remaining time (in years)
that the borrower is permitted to take to fully discharge its contractual
obligation (principal, interest, and fees) under the terms of loan agreement
► One year floor doesn‟t apply to short-term exposures, this floor is only
available for short-term exposures with an original maturity of below one year,
► Effective Maturity (M) = Σ t* CF/ΣCF. „Basel II‟
Page 36 Market practice series “Credit losses modeling”
37. Credit risk measurement
3. Measurement methods (Historical charge-off method)
Loss Confirmation Period “LCP”
Illustrative case for the Loss Confirmation Period:
► Retail loan portfolio, based on the bank policy, we have the below:
► Credit event is to have a loan with pastdues for > one day,
► NPL is the loans with pastdues of > 60 days
► Buckets are Current „no pastdues‟, B1 „<30 days pastdue‟, B2 „<60 days pastdue‟,
NPL „>60 days‟
► Borrowers have been tracked through a history of one year, identifying the 1st time
of the loss trigger „credit event‟
NPL Historical data NPL EAD Period Weight Yr*W
„100%EL‟ 'Yr' 'W'
Q4.2010 Q3.2010 Q2.2010 Q1.2010 Q1.2011
Cust#1 NPL B2 B1 Current $100 0.6Yr= 51%= 3* 0.3Yr
(30+2*90) $100/$600
Cust#2 B2 B1 B2 B1 $200 1.1Yr= 99%= 3* 1 Yr
(30+4*90) $200/$600
Cust#3 B2 B1 Current B2, B1 in Q4.09 $300 1.6Yr= 150%= 3* 2.3Yr
(30+6*90) $300/$600
Total $600 LCP=
Max( 1Yr, Av.Period ) 1.2Yr
Page 37 Market practice series “Credit losses modeling”
38. Credit risk measurement
3. Measurement methods (Historical charge-off method)
Loss Confirmation Period “LCP”
Illustrative case for the EM or Macaulay duration:
► A loan of $300, with yearly repayment plan of $10 over 3 years, in addition to
one last payment of $300 at the maturity time, find the LCP at time of
inception:
► Macaulay duration can be simply the remaining life of the security= 3Yrs,
► or 2.9Yrs= [(1*$10/$330)+ (2*$10/$330)+ (3*$310/$330)]
Time 0 Yr.1 Yr.2 Yr.3
-$300 $10 $10 $10 + $300
Cash outflow Cash inflow Cash inflow Cash inflow
EM= 2.91 EM= 1.91 EM= 0.94
=(1*10/330)+ =(1*10/330)+ =(1*310/330)
(2*10/330)+ (2*310/330)
(3*310/330)
Page 38 Market practice series “Credit losses modeling”
40. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
The graph below represents a loan portfolio over a time length of four years, showing the
change in the risk ratings.
► A migration analysis intends to derive a loss rate, which is the probability that a AAA-
rated loan would become CCC-rated, less the probability that a due loan could be
recovered over a certain period of time.
► A credit event for a loan would be the loss trigger that kicks a loan from one risk rating
to another.
► The higher the degree of segmentation, the higher the accuracy of deriving the risk
components.
Loan portfolio
Other Commercial Project finance
100%
100% 100%
80% 80% 80%
60% CCC 60% CCC
60% 40% 40%
20% BBB 20% BBB
40% 0% 0%
CCC AAA AAA
20% BBB
0% AAA
Page 40 Market practice series “Credit losses modeling”
41. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
The risk components can be calculated per transaction or at the credit exposure segment
level.
For those who prefer to do their calculations on borrower/ transactional level, then a
conversion „if needed‟ to the segment level can be done as illustrated below:
Credit exposure segment: Project finance loans for the telecommunication sector
Customer Historical data Transaction Risk components Weighted Average rate for the selected
weight (PD, LGD, LCP) exposure segment
Yr.2010 Yr.2011 Yr.2012 W i1 W i2 W i3 Yr.2010 Yr.2011 Yr.2012
Cust#1 $1000 $800 $600 67% 50% 35% 3 1 2
Cust#2 $200 $300 $700 13% 19% 41% 20 15 10
Cust#3 $300 $500 $400 20% 31% 24% 30 5 15
Total $1500 $1600 $1700 100% 100%100% 10.6* 4.9 8.4 7.97= Av.(10.6, 4.9, 8.4)
*10.6: is the risk component, weighted by the transaction size, which is numerically derived from the [(3*67%)+ (20*13%)+ (30*20%)]= 10.6,
The input digits could be percentage or numbers, could be PD, LGD, or LCP.
Page 41 Market practice series “Credit losses modeling”
43. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Credit and recovery event
Credit (default) event:
► That‟s the loss trigger that indicates that a loss has been incurred, which may
lead to eventual loss „or default‟, for instance it could be the event of a
transaction or credit exposure to be downgraded.
► The definition of the credit event significantly impacts the calculation of the
risk components, thus the loan portfolio should be cross segmented based on
the credit event, for instance the retail loans would be segmented based on
the repayment status, and the corporate portfolio would be segmented based
on the risk ratings, thus deriving representative loss rates “PD*LGD”,
reflecting the sector, risk rating/ repayment status, product risk,…etc.
► It should be noted that the NPL with 100% of expected losses should be
defined in light of the regulatory requirements.
Retail 'based on Current Bucket 1 Bucket 2 Bucket 3 NPL „100% EL‟
repayment status' '30' „31-90‟ „91-180‟ „181-270‟ „>270‟
Corporate 'based Risk rating Risk rating Risk rating NPL
on risk rating' '1-3' '4-5' '6-7' '8-10'
Page 43 Market practice series “Credit losses modeling”
44. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Credit and recovery event
Illustrative case for the calculation of the allowance for loan losses:
► After defining the credit event and the portfolio segmentation has been made,
the Allowance for Loan Losses can be calculated for each credit exposure
segment as illustrated below:
Current Bucket 1 Bucket 2 Bucket 3 Bucket 4 NPL
New product $465 EAD $100 EAD $120 EAD $110 EAD $70 EAD $65 „100% EL‟ EAD
PD1 0.6% PD2 46% PD3 45% PD4 70% PD5 69%
LGD1 100% LGD2 63% LGD3 73.4% LGD4 73.5% LGD5 79%
'New product' EAD*PD*LGD EAD*PD*LGD EAD*PD*LGD EAD*PD*LGD EAD*PD*LGD
Allowance $143 $100*0.6%*100% $120*46%*63% $110*45%*73.4% $70*70%*73.5% $65*69%*79%
Page 44 Market practice series “Credit losses modeling”
45. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Credit and recovery event
Credit (default) event for a corporate obligor:
► A default for a corporate obligor is subject to the whole outstanding lines for
the borrower rather than a particular credit line as for the retail obligors.
► A default is considered to have occurred with regard to a particular obligor
when either one or more of the following events have taken place:
1. The bank considers that the obligor is unlikely to pay its credit obligations to
the banking group in full, without recourse by the bank to actions such as
realizing security (if held),
2. The obligor is past due for 3 installments or more on a material credit
obligation to the banking group. Overdrafts will be considered as being past
due once the customer has breached an advised limit or been advised of a
limit smaller than current outstanding,
3. The bank makes a charge-off or account-specific provision resulting from a
significant perceived decline in credit quality subsequent to the bank taking on
the exposure,
Page 45 Market practice series “Credit losses modeling”
46. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Credit and recovery event
Credit (default) event for a corporate obligor (cont‟d):
4. The bank consents to a distressed restructuring of the credit obligation where
this is likely to result in a diminished financial obligation caused by the
material forgiveness, or postponement, of principal, interest or (where
relevant) fees,
5. The bank has filed for the obligor‟s bankruptcy or a similar order in respect of
the obligor‟s credit obligation to the banking group, or the obligor has sought
or has been placed in bankruptcy or similar protection where this would avoid
or delay repayment of the credit obligation to the banking group,
6. The bank sells the credit obligation at a material credit-related economic loss,
7. Whether any of the above has resulted in a downgrade, a downgrade by itself
is considered as a default event.
Page 46 Market practice series “Credit losses modeling”
47. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Credit and recovery event
Credit (default) event for a retail obligor:
► The retail credit event is different from the corporate in a way that the retail is
applicable to a particular loan rather than the underlying outstanding
exposure of the borrower.
► A particular retail exposure is considered defaulted if one of the below events
have taken place:
1. A partial or full charge-off has been taken place against its exposure,
2. A retail obligor has filled for bankruptcy,
3. A retail borrower has missed one or more payments of the due principal,
interest , or fees.
Page 47 Market practice series “Credit losses modeling”
48. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Credit and recovery event
► A credit risk measurement policy should numerically define the credit and the
recovery event.
Credit event:
► For instance; a loan should be considered in default, regardless to the risk
rating, thus a 3 installment due loan, will be considered as an observation of
default, impacting the PD & LGD calculations.
► A sold-off credit exposure would be; that any sold-off loan with a market yield
at time of disposal greater than its original EIR due to a deteriorated credit
quality, then it should be considered as a default, impacting the PD & LGD
calculations. For instance a more than 30% increase in the sale yield should
be assessed for impairment whether it‟s been due to a credit deterioration.
Recovery event:
► For instance; a credit exposure would be considered a recovery if at least
90% of its defaulted due fees, interest, and principal have been settled.
► The recovery event is explained through the LGD section.
Page 48 Market practice series “Credit losses modeling”
49. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
50. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
Probability of Default 'PD':
► For corporate and bank exposures, the PD is the greater of the one-year PD
associated with the internal borrower grade to which that exposure is
assigned. For sovereign exposures, the PD is the one-year PD associated
with the internal borrower grade to which that exposure is assigned. The PD
of borrowers assigned to a default grade(s), consistent with the reference
definition of default, is 100%.
Source: Basel II
► The one-year default rate (or default frequency) is the number of accounts
that default at any time within the period divided by the number of accounts
open at the beginning of the year. A validation mechanism should be
deployed in case of using the $$ value in estimating the PD rather than the
number of accounts.
Source: US Federal reserve system, Federal register Vol.69, 2004 notice
► Segmenting the loan portfolio on (credit line size) basis to derive the PD,
using the $$ value approach should be an easy-smart alternative
Page 50 Market practice series “Credit losses modeling”
51. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
Probability of Default 'PD':
► PD should capture all the credit event observations for a credit exposure
segment over certain time-horizon.
► For instance; if the a credit event such as rescheduling, past-due default is
not being reflecting on the risk rating, then the PD calculation should consider
all such credit events as observations, as explained below:
Risk rating Period1 Period2
Downgrades Downgrades Rescheduled Defaulted on PD
to BBB to DDD “Not TDR” 3 installment
AAA $100 $5 $10 $15 $5 35%= ($5+$10+$15+$5)/ $100
Page 51 Market practice series “Credit losses modeling”
52. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
► Two main methods used to calculate the PD:
1. Unit/ account based PD “for retail”,
2. $$ value based PD “for corporate & retail”.
► The unit based PD supporters seem to view the PD from the number of
occurrences rather than from the exposure defaulted.
► For instance; an $800 retail loan portfolio, composed of 100 accounts, one
main account with a total value of $500, and the others make a total of $300,
spread equally. If that one account defaults, then:
► $$ value PD 63%= $500/$800,
► Unit PD 1%= 1/100.
► Alternative approach would be based on segmenting the loan portfolio over
two (one account making $500), and (99 accounts making $300), so the
difference between the unit & $$-value PD should reasonably come to a small
margin.
Page 52 Market practice series “Credit losses modeling”
53. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
► If the bank adopts $$ value PD, then it should consider the gross loan value
rather than net of the collateral value, thus deriving the default probability of
the risk ratings, industries in a portfolio that might have been incurred but not
observed in the fully or substantially covered credit lines
e.g., Listed below are the credit limits that were granted for the aviation industry:
► ABC Air Co., $1000, fully cash/bank guarantee covered, market share 70%,
► XYZ, $1000, 40% covered, market share 20%,
► ABC Co., $1000, 0% covered, market share 10%,
► Assuming same risk ratings at time of initiation, however at year end, ABC Air Co. alone has
been downgraded from RR.2 to RR.6; If we calculate the PD, based on gross loan balances,
then the derived PD will reflect the whole deterioration in the credit risk in the aviation industry,
as the downgraded credit lines will account for $1000, however if we calculate the PD based on
credit exposures net of the collateral value, then the derived PD won‟t reflect the deterioration in
the credit exposure with the aviation sector, as the downgraded credit lines will account for $0,
► The later mentioned PD is understated in light of the fact that the downgrading credit exposure
is being „hidden‟ by the cash cover, however in fact it represents a credit exposure to 70% of
the aviation sector.
Page 53 Market practice series “Credit losses modeling”
54. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
Illustrative case 1 for the PD ($$-value):
► Retail portfolio
New product Period 1 Period 2 PD1 PD2 Average PD Cumulative
„Outstanding dues‟ „given‟ calculated Av.(PD1, PD2) PD
Current $1000 $2000 40% 60%=$600/$1000 50%=Av.(PD1,PD2) 9%
Bucket 1 $800 $600 30% 88%= $700/$800 59%=Av.(PD1,PD2) 18%
Bucket 2 $600 $700 50% 83%= $500/$600 67%=Av.(PD1,PD2) 31%
Bucket 3 $400 $500 60% 75%= $300/$400 68%=Av.(PD1,PD2) 46%
Bucket 4 $300 $300 70% 67%= $200/$300 68%=Av.(PD1,PD2) 68%
NPL „100% EL‟ $100 $200
The Allowance for Loan Losses should then be calculated as EAD*
(PD „Col.#7‟+ Environ‟l adj.)*LGD.
* Environ‟l adj.: Environmental adjustment, standing for the incurred credit losses but not
yet observed in a form of default in the credit portfolio, derived from the change in
average historical PDs compared to the PDs at time of the crises.
Note: the environmental adjustment can be done to the total loss rate (PD*LGD) instead
of segregating it to the PD and to the LGD using the same rationale mentioned above.
Page 54 Market practice series “Credit losses modeling”
55. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
Illustrative case 2 for the PD ($$-value):
► Retail portfolio
Historical loss and delinquency data (simplified example)
March April May June
Current $2,500 $2,640 $2,600 $2,675
30 DPD $90 $100 $120 $140
60 DPD $42 $45 $47 $49
90 DPD $37 $36 $37 $39
Charge-off $29 $31 $32 $33
Roll rates
April May June $140/$2600 3 mo.avg.
Cur-30DPO 4.00% 4.55% 5.38% 4.64%
(4.00% + 4.55% + 5.38%
30DPO – 60DPO 50.00% 47.00% 40.83% 45.94%
60DPO – 90DPO 85.71% 82.22% 82.98% 83.64%
90DPO – Charge-off 83.78% 88.89% 89.19% 87.29%
Estimated credit losses:
July August Sept.
Current $2,641 $2,605 $2,558 $2605*
30 DPD $124 $123 $121
60 DPD $64 $57 $56
90 DPD $41 $54 $48 3 month loss
$34 + $36 + $47
Charge-off $34 $36 $47 $117
Page 55 Market practice series “Credit losses modeling”
56. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Probability of Default “PD”
Illustrative case for the PD ($$-value):
► Corporate portfolio
Period 1 Period 1 PD1 PD2 Average Cumulative PD
Jan.2011 Dec.2011 „given‟ „calculated‟ (PD1,PD2)
RR. 1 $7000 $10,000 40% 43%=$3000/$7000 41% 1%=
(41%*53%*76%*75%*68%*53%*25%)
RR. 2 $9000 $3000 50% 56%=$5000/$9000 53% 3%
RR. 3 $4000 $5000 76% 75%=$3000/$4000 76% 5%
RR. 4 $5000 $3000 70% 80%=$4000/$5000 75% 7%
RR. 5 $3000 $4000 70% 67%=$2000/$3000 68% 9%
RR. 6 $2000 $2000 55% 50%=$1000/$2000 53% 13%
RR. 7 $1000 $1000 30% 20%=$200/$1000 25% 25%
NPL 8:10 $100 $200 NA NA
The Allowance for Loan Losses should then be calculated as EAD*
(PD „Col.#6‟+ Environ‟l adj.)*LGD.
For simplicity; the identified downgrades are assumed to be just from the previous risk pool.
e.g., $3000 identified in RR.2 in period 2 is assumed to be a full downgrade from RR.1 that
had an exposure of $7000 in period 1.
Page 56 Market practice series “Credit losses modeling”
57. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
58. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default:
► That‟s the share of the defaulted exposure that will never be recovered by the
lending bank. The LGD of a transaction is more or less determined by “1
minus recovery rate”, in other words the LGD quantifies the portion of loss the
bank will really suffer in case of default. The LGD should be measured as a
percentage of the EAD. A bank should provide an estimate of the LGD for
each corporate, sovereign and bank exposure.
Source: Base II
► There are three main approaches as per Basel II, explaining the LGD
Standardized, Foundation, and Advanced approach „recommended by CBE‟
Page 58 Market practice series “Credit losses modeling”
59. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default:
► LGD is defined as the segment‟s credit-related economic losses net of
discounted recoveries divided by the segment‟s exposure at default, all
measured during a period of high credit losses for the particular loan, unlike
the PD, reference data sets for LGD contain only defaulted exposure.
► The concept of the economic loss is more broader than the accounting
measure of loss.
► Economic loss incorporates the mark-to-market loss of value of the defaulted
loan & collateral plus any direct & indirect costs to collect the loan, net of
recoveries, which all should be discounted to the time of default.
► The discount rate should be applied to the time period from the date of default
to the date of realized loss, or recovery on a pool basis.
► The discount rate should reflect the distressed rate of the credit line, in other
words the opportunity cost of the time value of money „mark-to-market‟.
Source: US Federal reserve system, Federal register Vol.69, 2004 notice”
Page 59 Market practice series “Credit losses modeling”
60. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default general note:
► EAD should be the aggregate value of the outstanding loan and any past
partial/full charge-off.
► In order to better present the incurred losses rather than the expected losses,
the exposure should only include the principal plus any accrued interest and
fees, same applies for the discount rate; as it „ll better assess the incurred
losses through discounting using the original discount rate of the loan rather
than being a market rate.
► The non-performing retail loans “100% provision” rule can be avoided by
supporting how much that portfolio recovers, thus it would be provided for
100% less the percentage of recovery; “100%-LGD%”.
Page 60 Market practice series “Credit losses modeling”
61. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default (Foundation approach):
► In the foundation approach, the “basic” loss-given default is fixed at 45% for
all senior, unsecured exposures. This value should be raised to 75% for
subordinated exposures, but can be adjusted downwards when some
recognized collateral is pledged against the loan. However, this reduction
can‟t be based on a bank‟s internal models or past experience. Instead, a set
of rules has been introduced that quantify the effect of financial and non-
financial collaterals.
► An adjusted formula for LGD* can be calculated as below:
LGD*= (45% or 75%).Max[0,1+HE-C/E(1-Hc-Hfx)]
C: Collateral value
E: Original exposure value
HE: Haircut rate to be added to the value of the exposure
Hc: Haircut rate the collateral, reflecting the risk of the collateral market value
Hfx: FOREX haircut, if a currency mismatch exists between the exposure and the collateral
Note: Haircut rates are the higher of regulatory or the internally developed rates
Page 61 Market practice series “Credit losses modeling”
62. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default (Foundation approach) :
► For banks applying its IRB approach “Internal Ratings Based”, haircuts are
replaced by a system of minimum and maximum haircuts, as below:
LGD*= (45% or 75%) less:
Max[0,{(Min(C/E, Tmax) – Tmin}/{Tmax-Tmin}].(45%-LGDmin)
Tmax: Maximum threshold for the C/E ratio, based on the collateral type
Tmin: Minimum threshold for the C/E ratio, based on the collateral type
LGDmin: Minimum ratio when C/E >= Tmax
Page 62 Market practice series “Credit losses modeling”
63. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default (Advanced IRB approach):
► Adopting this approach will permit banks to use their own estimates of LGDs
for the corporate portfolio, however for the retail portfolio there, only the
advanced approach should be adopted
► Moreover, the Basel Committee states that exposure risks on retail loans with
uncertain future drawdown (such as credit cards) may be incorporated into
LGD estimates, accounting for the expectation of additional drawings prior to
default
► In other words, when a bank does not reflect risk on undrawn lines in its EAD
estimates, it should reflect this in its LGD estimates. For example, if the bank
estimates that EAD on a retail pool will be 20% higher than current usage,
LGD can be increased accordingly (e.g., from 50% to 60%) to account for
exposure risks without having to establish a formal system of CCF on
undrawn revolving credit lines
Page 63 Market practice series “Credit losses modeling”
64. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Loss Given Default (Advanced IRB approach) :
► The basis of calculation is based on the ratio between the present value, at
the time of default, of all payments made on a defaulted debt instrument, and
the face value (plus any accrued interest) of this instrument, which can be
expressed as follows:
LGD = 1-Recovery rate
Recovery rate= [ {(FR-AC)/EAD}/(1+r)t ]
Alternatively, the LGD= Gross defaulted exposure/EAD
In order to derive the LGD, an observation should be witnessed, which is based on the Bank recovery
policy, for instance the recovery policy for a two-risk rated loan is at least 85% recovery rate, thus LGD
should be calculated as being the average rate for the observations of 85% or more as a recovery rate
for a B-risk rated loan.
FR: Face value of the Recoveries
AC: Amount of Costs associated with the recovery process.
r: The original effective interest rate of the credit line.
t: Work-out period or the recovery period, defined as the period from the date of default to a resolution
date. A resolution date should be defined whether it‟s the date of 100% settlement or 95% settlement of
the default exposure or otherwise based on the institution's policy.
Page 64 Market practice series “Credit losses modeling”
65. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Illustrative case for the recovery rates for a retail loan portfolio:
Assuming:
► The bank policy defines the credit event loan as being the downgrade from
bucket to another, with 30 days dues time length for the bucket, therefore the
recovery event is the reverse, which is a retail loan to get upgraded from a
bucket to another. Time-horizon is quarterly data, and the presented loan
exposure is for a 2-Yr, quarterly installment loan with a fixed interest rate of
10%. Initially the defaulted exposure was for $20 „value of 1st due installment‟
out of his original loan amount of $143.3, later he paid a total of $22.61;
Pastdues Period 1 Period 2 Period 3 Recovery rate LGD
Current - - $92.9 NA
Bucket 1 $147 (including $114.7 (including - 95%= [PV(r=10%/4, t=3 5%= 1-95%
$20 pastdues) $2 pastdues) periods to recover,,
FR=$22.61)/ EAD=$20+$2]
In case that the settled amount is not tracked on system, and can‟t be 93%= [PV(r=10%/4, t=3 7%= 1-93%
obtained the recovery rate would be based on the amount originally periods to recover,,
defaulted at. FR=$20+$2)/ EAD=$20+$2]
Page 65 Market practice series “Credit losses modeling”
66. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Detailed basis of calculation:
NPER EIR
4 10%
Period PMT Interest $143.4 Past-dues Collected Recovery Rate LGD
(Defaulted exposure)
1 -20 $3.59 $146.99 20 0
2 -20 $3.67 $114.66 2 18 $0.00 0%
3 -20 $2.87 $92.92 0 4.61 $21.00 95% 5%
4 -20 $2.32 $75.24
5 -20 =PV(10%/NPer, date or period of
$1.88 $57.12 recovery, total collected amount)
1- R. rate
6 -20 $1.43 $38.55 5%= 1- 95%
7 -20 $0.96 $19.51 =MIN[100%, PV of the recovered amount
8 -20 $0.49 $0.00 $21/ Defaulted exposure ($20+$2)]
Page 66 Market practice series “Credit losses modeling”
67. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Illustrative case for the recovery rates for a revolving loan portfolio:
Assuming:
► Same as per the last illustrative case, but with a credit card exposure of $500
instead;
Pastdues Period 1 Period 2 Period 3 Recovery rate LGD
Current - - NA
Bucket 1 $500 - - NA
Bucket 3 - $200 $10 93%= [PV(10%, 9 months out of 1Yr, FR= 7% =1-93%
$500-$10) / EAD= Max(500, 200)]
Page 67 Market practice series “Credit losses modeling”
68. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Illustrative case for the recovery rates for a corporate loan portfolio:
Assuming:
► ABC has been downgraded in year 2, without being past-due, however his
industry perspective has been a bit speculative, meanwhile XYZ has been
unable to repay the last due 3 installments, finally JOE has been struggling to
pay off his dues with other banks, but before he comes due on installment
with our bank, he agreed to reschedule his debts,
► In later periods however, XYZ has been able to pay off the due installments
over one year; thus the recovery rate for risk rating 1 is:
Risk Yr 1 Yr 2 Past-dues Resched Recovery rate LGD
rating „> 3 uling
installments‟
1 ABC $500 - - - XYZ 91% =[PV(10%/4 periods, 9%=1-91%
XYZ $800 XYZ $800 XYZ $100 - Recovery period 4 quarters,, FR
JOE $300 JOE $300 - JOE $300 $100)/ EAD $100]
2 - ABC - - NA
$500
Page 68 Market practice series “Credit losses modeling”
69. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Types of LGD:
► Ordinary LGD:
LGD= 1-Recovery rate, Recovery rate= [ {(FR-AC)/EAD}/(1+r)t ]
► Collateral weighted LGD:
LGD= 1-Recovery rate, Recovery rate= [ {(FR-AC)/EAD*}/(1+r)t ]
EAD*= EAD x [C/E(1-Hc-Hfx)]
► Downturn LGD:
LGD= Average LGDs at time of a past crisis or to be adjusted by the
average change in PDs from the ordinary time to the time of the crisis
► Default weighted LGD:
LGD*= LGD x PDwi
Page 69 Market practice series “Credit losses modeling”
70. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Loss Given Default “LGD”
Illustrative case of the Default weighted LGD:
LGD*= LGD+(PDwi x LGD)
PD PD wi LGD Wi= (PD wi x LGD) LGD*= (LGD +Wi)
Bucket 1 30% 20% = 30%/130% 10% 2% 12% = 10%+2%
Bucket 2 40% 30% = 40%/130% 10% 3% 13% = 10%+3%
Bucket 3 60% 50% = 60%/130% 10% 5% 15% = 10%+5%
Total 130%
Page 70 Market practice series “Credit losses modeling”
72. Credit risk measurement
3. Measurement methods (Migration analysis- simplex method)
Trade finance
Basis of deriving the risk components:
Letter of Credit Letter of Guarantee Other products
Default definition: Based Default definition: Based Default definition: Should
on the frequency of on frequency of be based on the
liquidation* and/or liquidation and/or frequency of liquidation
stagnancy** stagnancy. or other technical
(Import LC, confirmed (Performance LG, Bid assessment that asserts
export LC, ..etc) LG) when a contract or
Same as direct exposure product has been
(Debt LG) defaulted.
* Liquidation frequency: Is stated to be the number of times the credit lines per certain risk rating are being converted
into direct loans.
e.g., 3 credit lines (LC, LG) at a value of 1mn each were made at the beginning of the year, however only one credit
was liquidated at the end of the year due to illiquidity of the obligor at a fee rate/interest rate of 5%, thus the PD for that
year is 33%=1/3mn.
By end of the following year, the liquidated lines „1mn‟ were fully collected, thus LGD= [1-(1mn/1mn)/(1+5%)^1]
* Stagnancy: Is stated to be the case of expiration of the credit line, however still unconverted into a direct loan and
couldn‟t be closed due to some technical problems between the line beneficiary and the obligor.
Page 72 Market practice series “Credit losses modeling”
74. Credit risk measurement
4. Economic and market assessment
The assessed credit losses should reflect the current economic
circumstances that might bear incurred losses, but not yet observed.
► Two main approaches:
► Credit risk stress testing,
► Past observation of historical loss rates at time of high loss severity.
► Risk components:
► EAD (CCF should reflect the change in draw down rates at time of high
loss rates),
► PD (Should be adjusted to match the slope in the PDs at time of high loss
severity),
► LGD (Downturn LGD, default weighted LGD, or an adjusted LGD to reflect
the loss rates observed at time of high loss severity).
Page 74 Market practice series “Credit losses modeling”
76. Credit risk measurement
5. Model validation and back-testing
The validation and back-testing process is mainly consisted of:
► Model methodology validation:
Intended to assure the logic & soundness of two processes; the exposure risk
segmentation, and the quantification of the risk parameters process.
► Operational process review:
Intended to assure the accuracy of the quantification process of the risk parameters;
that they are in line with the designed methodology and that any exceptional or unusual
circumstances have been reported to the upper management and properly addressed,
The quantification process should entail mapping the calculated risk parameters to the
data sets in addition to the mathematical calculations of the risk parameters.
► Model back-testing:
Intended to reassess the validity of the model through internal or external review,
mainly through default rates comparison over time to assess the adequacy of the
estimated allowances under the adopted methodology.
The validation & back-testing process should be conducted by an independent unit, on
periodic basis, on yearly basis at least.
Page 76 Market practice series “Credit losses modeling”
77. Credit risk measurement
5. Model validation and back-testing
Back-testing example
► The Model parameters should be subject to review and approval from the
management based on:
1. Internal review (correlation 29.9%!!)
10.0%
9.0%
8.0% Example of an internal review;
comparing the trend of the
7.0%
estimated Allowance for Loan
6.0% Losses 'ALL' to the trend of the
5.0% Non-performing Loans 'NPL' for
4.0% the retail loans.
Data extracted from the FS, thus
3.0% NPL should comply with financial
2.0% reporting definition.
1.0%
0.0%
Dec.10 Jun.11 Dec.11
R. ALL/ Retail loans R. NPL loans/ Retail loans
Page 77 Market practice series “Credit losses modeling”
78. Credit risk measurement
5. Model validation and back-testing
Back-testing example
► The Model parameters should be subject to review and approval from the
management based on:
2. Peer review (correlation 15%!!)
120.0%
Example of a peer review;
100.0% comparing the trend of the
estimated Allowance for Loan
80.0% Losses 'ALL' to the trend of the
Non-performing Loans 'NPL' for
60.0% the retail loans for my bank and a
peer bank. *ALL/NPL%:
40.0% Equals the ALL/Retail loans
%divided by the NPL/Retail
20.0% loans%.
Data extracted from the FS, thus
0.0% NPL should comply with financial
Dec.10 Jun.11 Dec.11 reporting definition.
ALL/ NPL% "My Bank" ALL/ NPL% "Peer Bank"
Page 78 Market practice series “Credit losses modeling”
80. Credit risk measurement
6. Reference data sets
► Data sets: Data that should be tracked and available for the calculation of the
risk components, and for segmentation purposes.
► Time horizon: That‟s the period of time by which the credit risk related data
sets are plotted in order to derive the risk components for the purpose of
calculating the credit losses.
For instance the corporate loans data sets are agreed to be on yearly basis,
however the retail loans data sets are argued to be on yearly or quarterly
basis, based on the bank credit risk policy.
► Data coverage period: For the IRB approach, three to five years is being
mandated as the minimum period to be covered in order for a bank to use an
IRB-based credit risk measurement model.
Page 80 Market practice series “Credit losses modeling”
82. Questionable market practices
Loan segmentation PD/historical charge-off LGD/LCP EAD
rate
► Risk ratings alone (misstating ► Average historical recorded ► LGD: Average (recorded ► The direct exposure alone
the credit losses due to the credit losses in the P&L to the allowances based on old (understating the credit losses
fact that product type, region, credit exposure (ignoring the GAAP/exposure), leading to by an amount = CCF *
sector „ve their own loss and fact that the recorded losses (estimating losses in adverse revolving loans commitment)
recovery rates) are based on old GAAP, and relationship with the trend of
► All the direct and indirect
understating the credit losses NPL)
► Certain product level without a exposure (overstating the
due to the calculating a
proper analysis of high credit ► LGD: Old GAAP loss rates credit losses
proportion of the loss instead
risk concentrated or (understates the losses, as the
of calculating the whole
deteriorating segment old GAAP loss rates
exposure being under default)
(misstating the credit losses; if compensate for the „PD*LGD‟)
the most of the portfolio quality ► Rate of migration between the
► LCP: set at the maximum of 1
is clean, then the incurred risk ratings for all the total
(understates the losses as it‟s
losses over the high loss exposure „direct and indirect‟,
usually floored to 1, and
making segments will be (very conservative approach
termed into years)
understated and vise versa) as the indirect exposure
losses aren‟t expected to be
as large as the direct)
► Assigning the same PD rates
originally driven from the direct
exposure; to the indirect
exposure (overstating the
credit losses by as the indirect
exposure isn‟t supposed to
incur as much as the direct
exposure)
Page 82 Market practice series “Credit losses modeling”
84. Credit risk documentation
Minimum requirements:
1. Credit exposure segmentation 4. Reference data sets
A. Definition of a credit exposure A. Time-horizon
B. Credit exposure types B. Data coverage period
C. Segmentation basis C. Data sets
D. Rationale of the segmentation basis D. Mapping the risk components to the data sets
2. Measurement method 5. Exceptional & unusual circumstances
A. Model scope & purpose A. Basis of treatment
B. Adopted measurement method B. Rationale of the treatment
i. Historical charge-off 6. Model validation and back-testing
ii. Migration analysis A. Review of the model methodology
iii. Other structured models B. Review of the operational process
C. Basis of calculation of the risk components C. Back-testing
D. Definition of the loss trigger & recovery event D. History of the model amendments
E. Rationale of selection E. Oversight BOD and management approval
3. Economic and market assessment
A. Stress testing (objective & scenario basis)
B. Other approach
Page 84 Market practice series “Credit losses modeling”
86. FAQ
IRB risk components (EAD, PD, LGD) calculation:
1. How should the sold off loan portfolio impact the risk components?
The IRB risk components should be adjusted to recognize the risk
characteristics of the exposures that removed reference data sets through
sales or securitization
It becomes substantially important for banks that usually sells off primarily
credits that are poorly performing
Source: US Federal reserve system, Federal register Vol.69, 2004 notice”
2. Should the history data cover a time period of recession?
The PD covered period should entail at least one period of recession,
furthermore the LGD is the loss severity observed during periods of high
credit losses „distressed periods‟
Source: US Federal reserve system, Federal register Vol.69, 2004 notice”
However the above mentioned practice is a US GAAP requirement, but isn‟t
according to the CBE GAAP, rather it would be considered as a conservative
approach
Page 86 Market practice series “Credit losses modeling”
87. FAQ
PD calculation:
1. How should the withdrawn ratings be treated?
The „withdrawn ratings‟ is observed when an obligor has a risk rating at the
beginning of the period but eventually no risk rating by period-end „due to
settlement‟ of the credit exposure
An approach being adopted by S&P is to adjust for the withdrawn ratings by
subtracting all their exposure from the denominator
Note that the withdrawn accounts are treated in adverse to the sold exposures.
The difference in the treatment can be reasoned by the fact that the withdrawn account, proved to be able to settle its
exposure, and the risk model‟s objective is to measure the risk of „loss severity‟.
Page 87 Market practice series “Credit losses modeling”
88. FAQ
PD calculation … continued:
2. How should the new credit exposure that arrive in the middle of the period be
treated?
There are two approaches:
A. Consider the mid-period credit line as an observation,
That‟s to embed in the calculation of the PD, the balance of that observation in the
nominator, and the balance of the credit line at time of initiation in the numerator.
B. Consider the mid-period credit line “not” as an observation.
That‟s to ignore the value of the credit line in the calculation of the PD, thus the PD
shouldn‟t get impacted by the change in the mid-period credit line initiations.
Page 88 Market practice series “Credit losses modeling”
89. FAQ
CCF calculation:
► There are instances when the borrower have settled a portion of the
outstanding loan, resulting in a negative CCF%, how should it be treated?
e.g., a borrower has been granted a credit card with a limit of $150, as of period 1 the
total due balance is $100, however in period 2 his due balance has been $75 due
to settlement, then CCF would be:
Available limit $50= ($150-$100) period 1,
Available limit $75= ($150-$75) period 2,
CCF -50%= ($50-$75)/$50.
The negative CCF% should be eliminated from the calculation of the average
CCF%.
An alternative solution is to calculate the CCF only for the increased credit lines,
rather than for a total portfolio with an offset impact of both draw-downs and
settlements; means negative CCF per borrower should be eliminated.
Page 89 Market practice series “Credit losses modeling”
90. FAQ
LGD calculation:
► There are instances when the LGD is negative or some other instances when
it‟s very highly positive, how should it be treated?
A negative LGD (1-R.Rate) usually comes from the fact that the recovery rate is over
100%, which is mainly attributable to higher collateral value or more cash settlement for
the due loans.
However the highly positive LGD „being above 100%‟ comes from the fact that there
were additional lending to the default loans whether in form of support to help the
borrower meet its short term dues or in a form of agreement to postpone the loan
settlement, thus accruing more fees and interest.
Whatever the cause is, there has not been specific guidelines in this regard, however
the market practice has been; flooring the negative LGD to -10% and capping the
positive to 175%, and another market practice has been; flooring the negative LGD to
0%, and capping the positive LGD to 105%.
Page 90 Market practice series “Credit losses modeling”
91. FAQ
Securitized loan calculation:
► How should the credit losses of the securitized loan portfolio be measured?
It should be noted that a securitized loan portfolio should be subject to the
same basis of calculation of the risk components (EAD „CCF‟, PD, LGD) to
the extent the originator “seller” has retained an interest in the securitized loan
portfolio; thus for banks with a regular history of securitization or sell-off,
especially securitizing loans of particular type „mainly poor performing loans‟,
reference data sets should be available from the trustee, or loan servicer.
Alternatively, refer to the data sets for the retained pool of loans.
Source: US Federal reserve system, Federal register Vol.69, 2004 notice”.
Page 91 Market practice series “Credit losses modeling”
93. Data requirements
Example of the data requirements
► Hereby we list an example of the data requirements, subject for use under
any of the previously mentioned methodologies, whether under the historical
charge-off method or under the migration analysis methodologies
► Specific requirements should be customized to each methodology by its own,
based on its risk components
Page 93 Market practice series “Credit losses modeling”
94. Data requirements
Corporate loan portfolio:
1. Direct exposure: Performing & Non-performing (customer ID, name, total
outstanding, deferred fees, accrued interest, risk rating, loan type (Term,
revolving), credit limit for the revolving lines, tenor for the term loans,
collateral type, collateral value, interest rate, sector, branch #, Pastdues in
value, pastdues in days, pastdues in number of installments),
► If any; (restructuring date, restructured value, tenor before restructuring, modified
tenor), especially for the customers who either are not identified as a default loan,
or as a restructured loan, or as a pastdue loan.
► Additional data for the non-performing loans; (time of default, recoveries made in
value, source of recovery „guarantee/collateral/asset liquidation‟, recoveries in
dates, charge-offs in value, charge-offs in dates)
► Obligor pricing model as of the date of assessment „interest rate that compensate
for the credit risk‟ to determine the yield spread & discount rate
Page 94 Market practice series “Credit losses modeling”
95. Data requirements
Corporate loan portfolio:
2. Indirect exposure: Performing & Non-performing (customer ID, name, total
outstanding, deferred fees, risk rating, credit line type (Term, revolving), credit
limit for the revolving loans, collateral type, collateral value, interest rate,
sector, branch #, Pastdue fees, pastdue fees in days, liquidation date, expiry
date, reason of default if any)
3. Covered Period: Data for at least 5 years backward, with an appropriate
time-horizon.
Page 95 Market practice series “Credit losses modeling”
96. Data requirements
Retail loan portfolio:
1. Direct exposure: Performing & Non-performing (customer ID, name, total
outstanding, deferred fees, accrued interest, product type, credit limit for the
revolving lines, collateral type, collateral value, interest rate, sector, corporate
employer, branch #, Pastdues in value, pastdues in days, pastdues in number
of installments, geographical location)
► Additional data for the non-performing loans; (time of default, recoveries made in
value, source of recovery „guarantee/collateral/asset liquidation‟, recoveries in
dates, charge-offs in value, charge-offs in dates)
► Obligor pricing model as of the date of assessment „interest rate that compensates
for the credit risk‟ to determine the yield spread and discount rate
2. Covered Period: Data for at least 5 years backward, on quarterly basis or
semi-annual basis.
Page 96 Market practice series “Credit losses modeling”