Expected Credit Loss (ECL) is a financial metric used by entities to estimate the potential losses from default on loans and other financial instruments. It incorporates factors such as Probability of Default (PD), Loss Given Default (LGD), and Exposure at Default (EAD) to calculate the expected loss over a specified time horizon. ECL is crucial for risk management and compliance with accounting standards, such as the International Financial Reporting Standards (IFRS 9), guiding how institutions provision for potential credit losses in their financial statements.
Under paragraph B5.5.28 of IFRS 9 – Financial Instruments:
- Expected Credit Loss: Expected credit loss (ECL) is the probability-weighted estimate of credit losses over the expected life of the financial instrument.
- Credit Loss: Credit loss is the present value of all cash shortfalls.
- Cash Shortfall: A cash shortfall is the difference between the cash flows that are due to an entity in accordance with the contract and the cash flows that the entity expects to receive.
Definition and formula of ECL
The definition of ECL under B5.5.28 of IFRS 9 (from above) gives us a wholistic idea on the linearity of calculating the ECL.
- First we determine expected cash shortfall for the instrument over its expected life
(This process involves both qualitative and quantitative estimations) - Then this cash shortfall are translated into present values
(This process involves determination of the discounting rate that is specific to the financial instrument which also involves both quantitative and qualitative estimation) - Then these present values are probability weighted to determine the ECL as at the reporting date
(This process involves determination of the probability of the occurrence of the particular cash shortfall which also involves both quantitative and qualitative estimation)
If we translate above definition into mathematical form, we come to the below expression:
This derivation using general mathematical identities also proves that the ECL formula of ECL=EAD×PD×LGD assumes that:
- Probability of Default (PD) is not weighted over the time period but rather indicates the probability of default at the point of calculation of ECL (i.e. at the point of reporting)
- Loss Given Default (LD) is not weighted over the time period but rather indicates the loss if default occurs at the point of calculation of ECL (i.e. at the point of reporting)
- Exposure at Default (EAD) is not the sum of present values of the amounts exposed to default over the term of the instrument (n) but rather the present value of the amount that is doubtful at the point of default (t)
So in essence, the above formula ECL = EAD × PD × LGD is an oversimplification of the actual definition of the ECL provided in paragraph B5.5.28 of IFRS 9 – Financial Instruments, for easier understanding. This is because it considers the single point of default (thereby negating the effect of time value losses due to delinquencies, renegotiation, restructuring and so on), single point of exposure amount (not an average amount) and also single ratio for loss at the point of default – which may not necessarily be the case for all instruments. This formula ECL = EAD × PD × LGD definitely simplifies our understanding but it doesn’t limit the BFIs from implementing a model that incorporates term (nth year) (i.e. time variable dependent ECL model) based estimation of probability, exposure and loss rate at default. A deep neural algorithm or random forest algorithm of machine learning models would provide more detailed Probability of Default (PD) models that would give not just the % of default but a % of default given the term/time of the instrument (i.e. time variable dependent). We will discuss this in more detail in other post.
Therefore a complete model would not be specific to a point of default but all the points in time (t) where shortfalls in the cash are observed which would bring our formula to:
This would enable us to prepare more complex and deterministic models capturing the data of time periods like origination date, term of instrument, trends of defaults or delinquencies etc. Basically, the entire foundation of the ECL depends on the availability of the financial data both qualitative and quantitative, which we will discuss in detail below.
Things we already know (as of 2/11/2024)
Originally, the updated standard NFRS 9: Financial Instruments (aligned with IFRSs 2018) was set to take effect from July 16, 2021, according to Institute of Chartered Accountants of Nepal (ICAN). However, due to various factors such as challenges posed by the emergence of COVID-19, limited time availability, and a lack of technical expertise, the full implementation of NFRS 9 was postponed until the fiscal year 2080/81 for banks and financial institutions. Consequently, the provisions of NFRS 9, which encompass expected credit loss, will become fully effective from the fiscal year 2081/82.
ICAN, based on the recommendation of the Accounting Standards Board (ASB), Nepal, had introduced a non-optional carve-out for banks and financial institutions until the fiscal year 2080/81 in connection with NFRS 9. This carve-out allowed the utilization of the incurred loss model of impairment under NAS 39 – Financial Instruments: Recognition and Measurement, until the specified period. Additionally, it specified that financial institutions registered under the Bank and Financial Institution Act, 2073 must measure impairment loss on loans and advances as the higher of the amount determined according to the norms prescribed by Nepal Rastra Bank for loan loss provision and the amount calculated under the incurred loss model.
Regarding the applicability of NFRS 9, given that the carve-out/deferral is valid only until the fiscal year 2080/81 and NFRS 9 (2018) has already been declared applicable by the Accounting Standards Board, Nepal, from July 16, 2021, banks and financial institutions are mandated to adopt the expected credit loss impairment model of NFRS starting from the fiscal year 2081/82.
The present practice under Incurred Loss Model
Presently, the financial institutions registered under the Bank and Financial Institution Act, 2073 must measure impairment loss on loans and advances as the higher of the amount determined according to the norms prescribed by Nepal Rastra Bank for loan loss provision and the amount calculated under the incurred loss model.
In accordance with Unified Directives issued to Banks and Financial Institutions (A, B, and C class) by Nepal Rastra Bank, Loans and advances are broadly divided into performing and non-performing loans, primarily based on the past due period of interest or principal receivable. The performing loans include Pass and watchlist categories, while the non-performing loans encompass substandard, doubtful, loss, and restructured/rescheduled loans with respective loan loss provision rates specified for each class.
Whereas under standard NAS 39: Financial Instruments: Recognition and Measurement (equivalent to IAS 39 issued in 2013, before the update in 2018), utilized an ‘incurred loss model’ for recognizing the impairment of financial assets. Under this model, it is assumed that all loans will be repaid until evidence to the contrary, known as a loss or trigger event, is identified. Only at that point is the impaired loan (or portfolio of loans) written down to a lower value. At the end of each reporting period, entities are required to assess whether there is any objective evidence that a financial asset or group of financial assets is impaired. If evidence of impairment is found, an impairment loss is recognized.
For individually significant exposures and for all exposures not individually significant where a bank determines impairment on an individual basis, it is necessary to establish whether there is objective evidence that the exposure is impaired. Additionally, entities are required to make a collective assessment of impairment for all exposures not deemed individually significant and thus not assessed for impairment on an individual basis. Importantly, losses expected as a result of future events, regardless of their likelihood, are not recognized under this model.
Simplified and General Approach under IFRS 9
Under IFRS 9, entities have the option to use either the simplified approach or the general approach for calculating expected credit losses (ECL). The simplified approach is suitable for entities with trade receivables, contract assets, and lease receivables. Instead of separately calculating Probability of Default (PD) and Loss Given Default (LGD), entities employing the simplified approach utilize a loss rate method.
For trade receivables without a significant financing component, the loss allowance is measured as equivalent to lifetime Expected Credit Losses (ECLs). This is justified by the short-term nature of these receivables, usually due within 12 months, making the 12-month ECL and lifetime ECL the same. In cases where trade receivables or contract assets contain a significant financing component, or for lease receivables, entities can choose between the simplified approach and the general approach based on their preferences and requirements.
Simplified Approach may include simple models by preparing a Provision Matrix of Loss Categories using Aging Category, Credit Quality Category, Product Type Category as a basis. Even the simplified approach is expected to use the provision matrix that is adjusted for current conditions and future expectations, based on available forward-looking information. The default rates in the provision matrix should be calculated by segmenting the loan portfolio into appropriate groupings, based on shared credit characteristics. A provision matrix is simply a table that analyzes the trade receivables into groupings and applies a calculated loss rate to each one.
ECL under General Approach
Stages of Impairment of Financial Instrument under Expected Credit Loss computed under General Approach:
Particulars |
Stage 1 |
Stage 2 |
Stage 3 |
Criteria |
Credit risk is considered low |
Credit risk has increased significantly since initial recognition and is not considered low |
Credit risk increases to the point where it is considered credit-impaired |
Loss Allowance |
12-month ECL |
Lifetime ECL |
Lifetime ECL |
Interest rate applied on |
Gross Carrying Amount |
Gross Carrying Amount |
Net Carrying Amount |
On January 11th of 2024, NRB published a NFRS 9: Expected Credit Loss Related Guidelines, 2024 as an Exposure Draft for Consultation. The exposure draft published by NRB has been an important milestone for the process of transitioning into the ECL framework of loss allowance for Banks and Financial Institutions. NRB through this report has tried to provide both qualitative and quantitative guidelines for the implementation of the ECL model under NFRS 9: Financial Instruments. Among others the consultative paper goes in detail to delineate the qualitative distinction between the various stages of the financial instrument for the calculation of loss allowance under Expected Credit Loss Model.
What is a significant increase in credit risk?
Paragraph 5.5.9 to 5.5.11 of the IFRS 9 details the procedure of determining the significant increase in the risk. Under IFRS 9, determining significant increases in credit risk involves assessing changes in the risk of default on a financial instrument since its initial recognition. Entities compare the risk at the reporting date with that at initial recognition, using reasonable information available without undue cost. If a financial instrument has low credit risk at the reporting date, it may be assumed that credit risk has not significantly increased. The use of forward-looking information is encouraged, but in its absence, past due information can be considered. A rebuttable presumption exists for instruments more than 30 days past due, but entities can challenge this presumption with reasonable evidence. The presumption doesn’t apply if significant increases in credit risk are identified before payments are overdue by more than 30 days.
Paragraph 9 of the Expected Credit Loss Related Guidelines, 2024 issued (as a draft for consultation) by NRB includes some additional indicators that involves quantitative tests and provides a non-exhaustive list that can been deemed as indicators of significant increase in credit risk. They are:
i) More than 30 days past due
ii) Absolute Lifetime PD is 5% or more
iii) Relative Lifetime PD is increased by 100% or more (i.e. relative to previous PD)
iv) Risk rating (internal or external) downgraded by 2 notches since initial recognition
v) Risk rating downgraded to non-investment grade by external credit rating agency (BB+ or below) or by bank’s internal credit rating system
vi) Deterioration of relevant determinants of credit risk (eg future cash flows) for an individual obligor (or pool of obligors)
vii) Expectation of forbearance or restructuring due to financial difficulties
viii) Deterioration of prospects for sector or industries within which a borrower operates
ix) Borrowers affected by macroeconomic conditions based on reasonable and supportable forecasts.
x) Modification of terms resulting in restructuring/rescheduling
xi) Credit Quality Indicators determined as per internal credit assessment of performing loans which are subject to individual monitoring and review, are weaker than that in the initial recognition
xii) Management decision to strengthen collateral and/or covenant requirements for credit exposures because of changes in the credit risk of those exposures since initial recognition
The guideline also recommends a holistic approach, considering both qualitative and quantitative factors, to assess significant increases in credit risk. Internal risk rating systems should have sufficient grades, acknowledging that credit risk changes may precede grade shifts. Financial institutions can group instruments based on shared risk characteristics for collective assessment. Examples include instrument type, credit ratings, collateral, initial recognition date, remaining term to maturity, industry, sector, geography, loan to value etc. Collective assessments may be needed even when evidence at the individual level is lacking. For retail or less-detailed exposures, collective assessment is advised, while individual assessments are suggested for Stage 3 and large exposures. Adjustments should be made if additional borrower-specific information becomes available for collectively assessed groups, with credit risk ratings included in collective Expected Credit Loss (ECL) measurements.
Stage One, Two and Three: Qualitative and Quantitative Criteria
Banks and financial institutions must categorize their financial instruments into three stages for measuring expected credit losses. Stage 1 involves recognizing 12-month expected credit loss, while Stage 2 and Stage 3 require recognition of lifetime expected credit loss.
Stage 1 classification includes initially recognized financial instruments, those without a significant increase in credit risk, those in which contractual payments are not overdue or are overdue for up to 30 days and those with low credit risk at the reporting date. Low credit risk criteria encompass exposures to the Nepal Government/Province/Local Level, fully guaranteed exposures by these entities, exposures to multilateral international institutions with risk weight 0% under CAF 2015, foreign sovereign exposures rated BBB- and above and bonds rated AA or above by external credit rating agencies.
In Stage 2 of credit risk classification, financial instruments encompass those experiencing a significant increase in credit risk since their initial recognition. This stage also includes instruments with contractual payments overdue for more than 30 days but not exceeding 90 days, loans listed under the ‘Watchlist’ in adherence to the NRB Directives on prudential provisioning, loans lacking an approved credit line or facing revocation by the bank, and loans that have undergone restructuring or rescheduling but do not meet the criteria for classification as non-performing, subject to specific exceptions outlined in the NRB directives. Additionally, Stage 2 involves claims on non-investment grade financial instruments, specifically those with a credit rating of BB+ or below.
In Stage 3, financial instruments are characterized by:
i) Contractual payments overdue for more than 90 days.
ii) BFIs determine that the borrower is unlikely to fully repay credit obligations to the bank, with indicators including placing credit obligations on non-accrued status, consenting to distressed restructuring resulting in material forgiveness or postponement of principal and interest, filing for debtor’s bankruptcy, selling part of the credit obligation at a substantial credit-related economic loss, debtor seeking bankruptcy protection to avoid or delay repayment, and evidence that full repayment based on contractual terms is unlikely without the bank’s realization of collateral, regardless of the exposure’s current or past due status by a few days.
iii) Loan classified as non-performing as per NRB prudential provisioning directive.
iv) Credit-impaired financial instruments with objective evidence of impairment, including events such as significant financial difficulty, a breach of contract, concessions granted by lenders due to financial difficulties, likelihood of borrower entering bankruptcy or financial reorganization, disappearance of an active market due to financial difficulties, or purchasing a financial instrument at a deep discount reflecting incurred credit losses. Credit-impaired financial instruments also include those defined as such by BFIs based on their risk management practices.
Transfer between Stages
In the context of transfer criteria between stages, paragraph 17 of the Expected Credit Loss Related Guidelines, 2024 (consultation draft) requires BFIs to follow specific procedures:
- Transfer from Stage 2 to Stage 1 requires evidence of a significant reduction in credit risk, and BFIs should monitor exposures for a minimum probationary period of 90 days before upgrading.
- Transfer out of Stage 3 involves monitoring for a minimum probationary period of 90 days after the conditions for Stage 3 no longer apply, with an obligatory upgrade to Stage 2 before reaching Stage 1.
- For restructured/rescheduled exposures, a minimum probationary period of 24 months in Stages 2 and 3 is required before potential upgradation to Stages 1 and 2, respectively.
Forward Looking Information
Paragraph 18 of the Expected Credit Loss Related Guidelines, 2024 (consultation draft) iterate that forward-looking information in the ECL model is a vital component, encompassing future credit developments alongside historical data. This inclusion, particularly considering economically stressed scenarios, is crucial for timely credit loss recognition. Key considerations include applying sound judgment in line with economic analysis principles, considering relevant, reasonable, and supportable information, and evaluating the impact of events on ECL measurement. Information should not be excluded based on low likelihood, and any exclusion requires well-documented justifications. The process involves identifying qualitative and quantitative factors, forecasting economic scenarios, and estimating unbiased and probability-weighted credit losses. Banking and financial institutions (BFIs) need board-approved policies for economic analysis and forecasting, considering various forward-looking information sources. BFIs should employ sound judgment, maintain internal controls, and conduct periodic sensitivity assessments. A minimum of three economic scenarios (normal, best and worst case scenarios) is recommended, with BFIs using recognized statistical methodologies for weightages. Utilizing forecasts from authentic sources and alternative credible sources is advised for adjusting ECL models based on economic conditions.
Breaking down ECL Function
Let’s break down each component of the ECL function in detail. From the discussion above we have agreed to the below mathematical identity of the ECL function:
1. Exposure at Default (EAD)
This is the amount to which the probability of default rate and the loss given default rate is applied. For any asset for which expected credit losses (ECL) is getting calculated, EAD represents the projected credit risk exposure at any given point of time.
Repayment Rate Factor (RRF)
Exposure at default is the present value of the doubtful amount at the point of default. It is not the present value of outstanding balance of the instrument. But if there is a need to translate the EAD into the outstanding amount of the instrument, EAD can also be rewritten as follows using the repayment rate in the period to default (RRF):
EAD = Outstanding Balance × (1 – RRF)
Which essentially is the doubtful amount at the point of default. Using either of the mathematical identities of ECL is correct and its choice would depend on how your ECL model is trained. This will largely depend on whether the model would provide you the present value of the doubtful amounts at the point of default or the portion of repayment that can be expected from the current outstanding balance till the point of default. The use of repayment rate is preferred because repayment rate until the point of default is something that can be extracted from the historical data analysis of the repayments in the period to default, however this should not be viewed as negating the need to consider the use of both qualitative and quantitative forward looking information associated with the repayment rates. But in spite, using this alternative identity of EAD with repayment rate in the period to default, makes it possible to easily incorporate historical trends of repayments (also considering forward looking information) as a deterministic approach for computing the EAD.
Credit Conversion Factor (CCF)
In the context of Expected Credit Loss (ECL), the Credit Conversion Factor (CCF) is a factor used to convert off-balance-sheet credit exposures to on-balance-sheet credit equivalents. The purpose of applying the CCF is to capture the potential credit risk associated with certain off-balance-sheet items. The Credit Conversion Factor is used in the calculation of Exposure at Default (EAD), which is a key component in the computation of ECL. The formula for EAD is often expressed as:
EAD = CCF × Off Balance Sheet EAD
Paragraph 11(c)(viii) of the Expected Credit Loss Related Guidelines, 2024 (consultation draft) provides where the data is not available for off balance sheet exposures, BFIs may use CCF for the calculation of EAD for off balance sheet exposures:
- Unconditionally Cancelable Commitments (e.g., bills under collection): CCF 0%
- Forward Exchange Contracts: CCF 10%
- Short-Term Trade-related Contingencies (e.g., documentary letters of credit): CCF 20%
- Undertaking to Provide a Commitment on an Off-balance-sheet Item: CCF 20%
- Unsettled Securities and Foreign Exchange Transactions: CCF 20%
- Long-Term Irrevocable Credit Commitments: CCF 50%
- Short-Term Irrevocable Credit Commitments (excluding trade finance exposures): CCF 20%
- Repurchase Agreements, Securities Lending, Securities Borrowing, Reverse Repurchase Agreements: CCF 100%
- Direct Credit Substitutes (e.g., financial guarantees, credit derivatives): CCF 100%
- Unpaid Portion of Partly Paid Shares and Securities: CCF 100%
- Other Contingent Liabilities: CCF 100%
2. Probability of Default (PD)
The Probability of Default (PD) is a crucial metric representing the estimated likelihood of default within a specified time frame. In the estimation process, historical loss experience serves as a foundation, and this historical PD undergoes adjustments based on factors such as the historical correlation between key economic indicators (e.g., GDP, unemployment) and PD. Moreover, forward-looking macroeconomic information pertaining to future GDP and/or unemployment is considered to refine the calculated historical PD.
Paragraph 11(a) of the Expected Credit Loss Related Guidelines, 2024 (consultation draft) provides that for the Probability of Default (PD) calculation, BFIs should adhere to specific measures. They should derive PD by analyzing historical default migration rates, internal and external credit ratings, and other pertinent data. Additionally, BFIs should incorporate forward-looking PD information by adjusting for its sensitivity to changes in macroeconomic factors. To ensure robust calculations, BFIs are required to use at least five years of historical data for PD determination, with validation conducted by the Risk Management Department, avoiding undue smoothing of data or inputs. The integration of the internal rating scale with external credit ratings is employed for PD determination, while caution is advised against the use of proxies. Furthermore, the calculation of PDs for exposures denominated in foreign currencies issued by foreign sovereigns involves using a sovereign PD linked to the external credit rating scale.
In practical terms, even if a bank’s internal models or calculations yield a PD lower than 2.5%, the regulatory requirement mandates that the institution must use the higher of its calculated PD or the prescribed 2.5% floor by NRB. This regulatory floor helps prevent overly optimistic or aggressive PD estimates, promoting a more conservative and prudent approach to credit risk management. It provides a baseline level of risk sensitivity that institutions must adhere to, contributing to the overall stability and soundness of the financial system. It’s noteworthy that a regulatory prudential floor of 2.5% for credit exposures PD is prescribed, subject to review by the Nepal Rastra Bank (NRB) after a five-year post-implementation period of Expected Credit Loss Related Guidelines, 2024.
PIT PD and TTC PD
Point-in-Time PD (PIT PD) refers to the probability of default at a specific point in time. This measure captures the probability of default at the exact moment when the assessment is made. PIT PD considers the current conditions, economic factors, and other relevant information available at a particular date. It provides a snapshot of the credit risk at that specific point.
Through-the-Cycle PD (TTC PD) represents the average probability of default over the entire economic cycle. Unlike PIT PD, TTC PD is not time-specific and is a more stable measure, often used for long-term risk assessment. TTC PD smoothens out short-term fluctuations and provides a broader view of credit risk that is less sensitive to immediate economic conditions.
While PIT PD is specific to a particular point in time and reflects current conditions. TTC PD, on the other hand, is an average measure over the economic cycle and is less influenced by short-term fluctuations. PIT PD is specific to a particular assessment date, and TTC PD provides a more stable, cycle-averaged measure of credit risk. The mathematical difference between PIT PD and TTC PD lies in the calculation methodology. PIT PD incorporates adjustments for current economic conditions, making it more dynamic and responsive to short-term variations. In contrast, TTC PD is based on the long-term average, providing a stable estimate that is less influenced by immediate economic changes.
Hazard Function for 12-Month PD
Hazard rate is the initial probability of default. For instruments in stage 1 the expected credit loss is calculated over the 12 month’s period. Twelve month expected credit losses is the portion of lifetime expected credit losses that represent the expected credit losses that result from default events on a financial instrument that are possible within 12 months after the reporting date. An amount equal to 12 month ECL is not only losses expected in next 12 months rather, it is the expected cash shortfalls over the life of the lending exposure or group of lending exposures due to loss events that could occur in the next 12 months. Twelve month expected credit losses are to be recognized for financial instruments with low credit risk or no significant change in credit risk since initial recognition, at the reporting date. Probability of default (PD) is an index that increases over the time function, so the 12 month’s probability of default is lower than the probability of the default over lifetime.
When estimating simply the lifetime probability of default using the PD model, it is also necessary that we obtain the point in time where the default occurs, as the point in time is an essential index to compute all other factors of ECL like EAD, LGD as well as PD. When we have the point in time associated with probability of default, it will enable us to translate the lifetime probability of default into 12-month’s probability of default using the Hazard Function.
The Hazard Rate function is a concept commonly used in survival analysis and reliability engineering. It represents the instantaneous rate of an event happening at a specific point in time, given that the event has not occurred before that time. In the context of credit risk modeling, the event of interest is often default. The hazard rate is a measure of how the probability of an event (such as default) changes over time. If we have the probability of default (PD) for a specific time horizon and you want to estimate the hazard rate, you can use the formula below.
Let’s assume a scenario where the probability of default in 4 years is 25%, we can use the above formula to estimate the hazard rate. Once we have the hazard rate, we can then use it to estimate the probability of default in 1 year. Please note that this assumes a constant hazard rate over time.
So, the estimated probability of default in 1 year using the hazard rate is approximately 6.89%.
3. Loss Given Default (LGD)
LGD is an adjustment to the ECL calculation for post-default recoveries. These can be in the form of cash repayments, proceeds from the realization of security or sale of the debt to a third party. In an alternative mathematical identity, LGD can also be written as:
LGD = 1 – Post default recovery rate.
Paragraph 11(a) of the Expected Credit Loss Related Guidelines, 2024 (consultation draft) provides guidelines on Loss Given Default (LGD) and collateral valuation for credit exposures. The guideline includes factors to consider for LGD, advising financial institutions to develop models based on historical data, cash recovery experience, and relevant forward looking information. If unable to compute LGDs due to data limitations, a minimum LGD of 45% is recommended. LGD percentage is 0% for same currency cash-backed loans, 20% for foreign currency exposures with government guarantees. All subordinated claims on corporates, banks and foreign sovereigns will be assigned a minimum of 75% LGD.
Thanks for reading !!
Next up – How to train a time variable independent Credit Default Classifier Model on AI/ML using Random Forest Algorithm?
Leave a Reply