Credit Portfolio Management, Provisioning and Stress Testing

Introduction

Credit risk is not managed only by deciding whether to approve individual borrowers. A bank can make apparently sound individual loans and still experience serious losses if its portfolio is excessively concentrated, exposed to correlated borrowers, or growing rapidly in a deteriorating economic environment. Credit portfolio management therefore examines the collective behaviour of exposures, the distribution of risk and return, the movement of accounts through risk grades, and the capital required to absorb unexpected loss.

Portfolio management links lending strategy with risk appetite, provisioning, capital allocation, pricing and governance. Its central question is not simply “Will this borrower repay?” but also “What does this exposure contribute to the risk of the entire institution?”

Learning Objectives

By the end of this chapter, a student should be able to:

• distinguish individual credit risk from portfolio credit risk;
• explain concentration, correlation, migration and diversification risk;
• design and interpret portfolio limits and risk dashboards;
• use vintage and cohort analysis to identify changes in underwriting quality;
• calculate expected credit loss and interpret staging under IFRS 9 or comparable frameworks;
• construct basic stress tests and explain their limitations;
• evaluate lending performance using risk-adjusted return and economic capital;
• discuss governance, ethics, regulation and technology in portfolio risk management.

1. The Portfolio Perspective

A credit portfolio is the collection of loans, guarantees, securities, leases and other credit exposures held by a financial institution. Portfolio risk arises from two components:

1. Systematic or common-factor risk results from macroeconomic conditions, interest rates, commodity prices, unemployment, property values or political events affecting many borrowers.
2. Idiosyncratic risk is specific to a borrower, industry, project or transaction.

Diversification can reduce idiosyncratic risk, but it cannot eliminate systematic risk. A portfolio of thousands of mortgages may be diversified by borrower while remaining highly sensitive to unemployment and house prices.

For an exposure i, a simple expected-loss relationship is:

Here, PD_i is probability of default over the selected horizon, LGD_i is loss given default as a proportion of exposure after recoveries, and EAD_i is exposure at default measured in currency units. The formula assumes that the three quantities are defined consistently over the same horizon and that LGD and EAD appropriately reflect collateral, guarantees, fees and future drawings. At portfolio level, expected loss is approximately the sum of individual expected losses, but portfolio variability depends on dependence among defaults and recoveries.

Expected loss is not the same as the actual loss that will occur. Actual outcomes fluctuate around expectations. Expected loss is normally covered through pricing and provisions; unexpected loss is addressed through capital, limits and risk transfer.

2. Concentration, Correlation and Diversification

2.1 Concentration risk

Concentration risk is the possibility that a relatively large share of portfolio loss arises from a single borrower, connected group, sector, geography, product, collateral type, maturity band or risk factor. Common measures include:

• Single-name exposure: exposure to one borrower or connected group as a percentage of capital or total exposure.
• Sector concentration: sector exposure divided by total portfolio exposure.
• Herfindahl–Hirschman Index (HHI):

where s_i is the proportion of exposure in segment i, expressed as a decimal. If four sectors each represent 25% of exposure, HHI = 4 × 0.25² = 0.25. A higher HHI indicates greater concentration. HHI is useful for comparison but does not reveal whether sectors are economically correlated.

• Top-N exposure ratio: the combined exposure of the largest N borrowers divided by total exposure or eligible capital.
• Granularity measures: statistics that assess how much risk is contributed by a small number of obligors.

Concentration limits should be risk-sensitive. A 10% exposure to government bonds, a speculative construction company and a highly collateralised mortgage pool do not necessarily have the same risk, even if their nominal proportions are equal.

2.2 Correlation and wrong-way risk

Correlation describes the tendency of defaults or losses to move together. If two borrowers operate independently, the failure of one gives limited information about the other. If they depend on the same commodity price or buyer, their defaults may be strongly related.

Wrong-way risk occurs when exposure increases precisely when the counterparty’s credit quality worsens. For example, a bank may have a large derivative receivable from an energy company whose value rises when energy prices fall, while the company’s ability to repay also deteriorates. Correlation analysis must therefore consider both the borrower and the exposure amount.

A simplified portfolio-loss variance expression is:

where L_i is loss on exposure i, Var denotes variance and Cov denotes covariance. The first term represents individual loss variability; the second captures dependence. The equation is general, but estimating covariance requires a model of default dependence and recovery behaviour. It should not be interpreted as proof that diversification always reduces risk: adding a highly correlated exposure may increase portfolio risk materially.

2.3 Diversification

Diversification is effective when exposures have sufficiently different drivers of loss. Useful dimensions include borrower, industry, geography, currency, product, collateral, maturity and economic factor. However, superficial diversification can mislead. A portfolio spread across ten industries may still be concentrated if all industries depend on domestic property prices or consumer demand.

Portfolio construction should therefore use both exposure classifications and economic-factor analysis. A bank should ask:

• What common events could cause defaults?
• Which borrowers depend on the same supplier, buyer or government contract?
• Are collateral values driven by one market?
• Does a currency mismatch affect several borrowers simultaneously?
• Could a risk transfer instrument create concentration in a guarantor or insurer?

3. Credit Migration and Portfolio Dynamics

Credit migration is movement between risk grades over time. A borrower may migrate from “low risk” to “watch,” “substandard” and eventually default, or improve after restructuring and stronger financial performance. Migration analysis is important because default is a lagging indicator.

A transition matrix records the probability of moving from one grade to another during a specified period. For grades A, B, C and Default, a row may sum to 100%:

The matrix is valid only for its stated population, horizon, definition of default and economic period. A transition matrix estimated during stable conditions may understate migration in a recession.

Portfolio migration affects provisions, pricing and capital before defaults occur. A rise in loans moving from A to B may signal underwriting deterioration, sector stress or a change in risk appetite. Migration should be analysed by product, origination period, geography, relationship manager and borrower size.

4. Limits and Portfolio Governance

A limit system converts risk appetite into enforceable boundaries. Typical limits include:

• single-obligor and connected-counterparty limits;
• sector, country and geographic limits;
• product and collateral limits;
• rating-grade and probability-of-default limits;
• maturity, currency and fixed/floating-rate limits;
• non-performing loan and watch-list limits;
• concentration limits for vulnerable industries;
• limits on exceptions, overrides and unsecured lending.

A sound limit framework has four features. First, limits are linked to measurable risk and capital. Second, ownership is clear: business units manage within limits, while independent risk functions challenge assumptions. Third, breaches have escalation rules and deadlines. Fourth, temporary exceptions are documented rather than normalised.

Limits should not be treated as automatic approval thresholds. A transaction below a limit may still be unacceptable because of weak repayment capacity, fraud indicators or poor structure. Conversely, a limit breach may be justified temporarily during a controlled wind-down, but it requires formal approval and monitoring.

Portfolio risk dashboard

A useful dashboard combines exposure, quality, performance and forward-looking indicators. It may contain:

Dashboards should show trends, thresholds, variance from plan and data quality exceptions. A single aggregate risk score can conceal important deterioration, so management should be able to drill down from portfolio to segment and account level.

5. Vintage and Cohort Analysis

A vintage is a group of accounts originated during the same period, such as a quarter or year. A cohort is a group sharing a defined characteristic, such as product, branch, risk grade or industry. Vintage analysis compares outcomes at similar ages since origination.

For a vintage, a cumulative default rate can be expressed as:

The denominator must be selected consistently. Account-based rates answer a different question from exposure-weighted rates. If balances amortise, exposure-weighted analysis is often more informative for loss measurement.

A vintage curve that rises faster for recent originations may indicate weaker underwriting, aggressive pricing, fraud, poor documentation or a changed economic environment. Analysts should control for seasoning, prepayments, restructures and differences in borrower mix. Comparing a twelve-month-old vintage with a thirty-six-month-old vintage without aligning age can produce a false conclusion.

Roll-rate analysis complements vintage analysis. It measures movement from one delinquency bucket to another, for example current → 30 days past due → 60 days → 90 days. Roll rates support short-term forecasting, while vintage curves support origination-quality assessment.

6. Expected Credit Loss Provisioning

6.1 Conceptual framework

Expected credit loss (ECL) is a probability-weighted estimate of credit losses, discounted to the reporting date where required by the applicable accounting framework. Under IFRS 9, a simplified three-stage approach is commonly described as follows:

• Stage 1: 12-month ECL for exposures whose credit risk has not increased significantly since initial recognition.
• Stage 2: lifetime ECL when credit risk has increased significantly since initial recognition, but the asset is not credit-impaired.
• Stage 3: lifetime ECL for credit-impaired exposures; interest recognition rules differ from Stages 1 and 2.

The 12-month ECL is not the cash loss expected over only the next twelve months. It is the present value of losses arising from default events that may occur within the next twelve months.

A simplified discounted ECL model is:

Here, PD_t is the probability of default during period t, LGD_t is loss given default, EAD_t is exposure at default, r is the effective interest or appropriate discount rate per period, and T is the lifetime horizon. This representation assumes that the terms are defined consistently and that expected cash shortfalls are adequately represented by PD, LGD and EAD. More complete models use scenario-weighted cash-flow shortfalls and may account for competing defaults, prepayments, recoveries and timing explicitly.

If there are economic scenarios s with probabilities w_s, a scenario-weighted estimate may be written:

with ∑w_s = 1 and each w_s ≥ 0. Forward-looking information should be reasonable, supportable and consistently governed. Management overlays may be appropriate where models do not capture emerging risks, but overlays should be documented, independently challenged and removed or recalibrated when their cause is addressed.

6.2 Worked example

A performing loan has an exposure of $1,000,000. For the next year, the estimated PD is 2%, LGD is 40%, and the effective discount rate is 8%. A simple one-year ECL is:

The undiscounted expected loss is $8,000. Discounting reduces the present value because the expected loss is assumed to occur at the end of the year. This is an illustrative calculation; a production model would consider timing of default, recoveries, collateral costs, commitments and multiple economic scenarios.

6.3 Model risk and controls

ECL models can fail through poor data, unstable relationships, inappropriate segmentation, weak scenario design or incorrect treatment of cured and restructured loans. Controls should include data reconciliation, back-testing, validation, sensitivity analysis, independent review, approval of model changes and clear documentation of judgement.

Provisioning should not be manipulated to smooth earnings or conceal deteriorating underwriting. Excessive conservatism can also distort pricing and reported performance. Accounting estimates require disciplined judgement rather than mechanical use of historical averages.

7. Stress Testing

Stress testing examines how portfolio performance would change under adverse but plausible conditions. It is not a prediction of the most likely future. Its purposes include identifying vulnerabilities, assessing capital and liquidity consequences, testing limits, informing contingency plans and challenging strategic assumptions.

A stress test normally includes:

1. Scope: portfolios, legal entities, risk types and time horizon.
2. Baseline: current exposures, risk grades, provisions and capital.
3. Scenario: macroeconomic and market assumptions.
4. Transmission mechanism: how variables affect PD, LGD, EAD, income and capital.
5. Results: losses, provisions, capital ratios, limit breaches and management actions.
6. Challenge and governance: review by risk, finance, treasury, audit and senior management.

Scenarios may be historical, hypothetical or reverse. A historical scenario repeats an event such as a severe recession; a hypothetical scenario combines shocks, such as unemployment rising, property prices falling and the domestic currency depreciating. Reverse stress testing begins with an unacceptable outcome—such as a capital breach—and asks what combination of events could cause it.

A simple stressed-loss calculation is:

This is an analytical approximation, not a complete stress-testing system. It assumes the stressed parameters are coherent and does not automatically capture feedback effects, management actions, migration, income changes or correlations. A robust exercise models these elements where material.

Stress tests should examine sensitivity to assumptions. If a small change in recovery rates produces a large capital impact, management should understand why and consider mitigation such as collateral controls, sector limits or hedging.

8. Risk-Adjusted Performance and Capital Allocation

Nominal interest income is an inadequate basis for judging lending performance. A loan with a high coupon may consume substantial capital and generate high expected loss. Risk-adjusted measures incorporate funding cost, operating cost, expected loss and capital.

A simplified risk-adjusted return on capital is:

All numerator components should refer to the same period, and economic capital should be measured consistently with the risk horizon and confidence standard. RAROC is useful for comparing products and segments, but it depends on transfer-pricing assumptions, model estimates and capital methodology. It should not replace judgement or statutory accounting measures.

Capital allocation assigns scarce risk-bearing capacity to businesses, products or transactions. A sound process considers marginal risk: the additional capital required when one exposure is added to the portfolio. Allocating capital solely by exposure size may reward low-margin growth or penalise well-diversified business.

9. Applied Case: Meridian Bank’s Commercial Portfolio

The following is an illustrative case.

Background

Meridian Bank’s commercial loan portfolio grew 28% in two years. Reported non-performing loans remained low at 1.4%, but 38% of exposure was linked directly or indirectly to commercial property, and the ten largest connected groups represented 24% of commercial exposure. Recent vintages showed increasing approval overrides.

Problem

Management focused on current default ratios and concluded that risk was stable. The portfolio team identified three warning signs: downgrades were increasing, collateral valuations depended on a buoyant property market, and several borrowers relied on the same construction contractors and property purchasers. Thus, measured concentration was understated by legal-entity reporting.

Analysis

The bank grouped exposures by ultimate economic dependency rather than borrower name alone. It calculated sector shares, top-group exposure, HHI, migration rates and vintage delinquency. A stress scenario assumed unemployment increased by three percentage points, commercial property prices declined by 25%, refinancing costs rose, and construction activity contracted.

The stress test showed that PD and LGD would rise together: weaker borrowers would default more often while collateral recoveries declined. Stage 2 exposures increased materially, producing a provision charge before most loans became non-performing. Several sector limits were breached on a look-through basis.

Approach and solution

The board approved a response consisting of:

• a temporary cap on new property-related lending;
• enhanced approval for exceptions and connected exposures;
• borrower-level cash-flow analysis rather than reliance on collateral;
• quarterly independent property valuations for vulnerable accounts;
• additional provisions supported by documented forward-looking information;
• active reduction of the largest correlated exposures;
• a dashboard combining migration, vintage, collateral and stress indicators;
• capital allocation that charged business units for marginal concentration risk.

Outcome

Growth slowed, but risk-adjusted returns improved. The bank avoided indiscriminate withdrawal of credit by distinguishing viable firms from speculative projects. The case demonstrates that portfolio management is anticipatory: waiting for defaults would have delayed action until losses were larger and options fewer.

10. Governance, Ethics and Regulation

Effective portfolio management requires a clear three-lines model. Business units own risks and operate within approved limits. Independent risk management develops frameworks, challenges decisions and reports breaches. Internal audit provides independent assurance over governance and controls.

Boards and senior managers should approve risk appetite, concentration tolerances, stress scenarios and capital planning. Credit committees should receive information that is timely, complete and understandable. Data lineage is essential: management must know how exposures, ratings, collateral values and provisions were produced.

Ethical issues include discriminatory credit allocation, opaque automated decisions, conflicts of interest in valuations, inappropriate restructuring to avoid recognition of impairment, and pressure to meet growth targets. Borrowers should not be denied fairly priced credit solely because an algorithm uses a prohibited proxy. Model explainability and human review are particularly important for small businesses and vulnerable consumers.

Regulatory frameworks commonly require capital for unexpected loss, large-exposure controls, governance, credit-risk data and supervisory stress testing. Exact requirements vary by jurisdiction; institutions must follow applicable national law and accounting standards rather than relying on generic textbook descriptions.

11. Technology and Contemporary Developments

Cloud platforms, application programming interfaces, alternative data and machine learning can improve monitoring frequency and early-warning detection. Models may identify unusual utilisation, declining deposits, payment delays or adverse sector news. However, technology introduces risks: data leakage, cyberattack, biased samples, unstable relationships, privacy violations and model opacity.

Good practice is to use AI as decision support within a controlled framework. Institutions should maintain data governance, validation, access controls, audit trails, drift monitoring, human override procedures and outcome testing. Machine learning can rank accounts for review, but responsibility for credit decisions remains with the institution.

Climate and nature-related risks are increasingly incorporated into portfolio analysis. Physical risks may damage collateral or business operations; transition risks may reduce the value of carbon-intensive assets. These risks often operate over longer horizons than conventional credit models, making scenario design and data quality especially important.

Challenges and Limitations

Portfolio models cannot eliminate uncertainty. Historical data may not contain comparable crises; default definitions may change; recoveries may be correlated with economic conditions; and exposures may be misclassified. Diversification measures can overlook common dependencies, while stress tests can create false confidence if scenarios are narrow or management actions are unrealistic. Provisioning models may produce precise-looking numbers that rest on subjective assumptions.

Accordingly, quantitative analysis should be combined with expert judgement, borrower contact, sector research, challenge processes and conservative governance. The objective is not to predict every loss, but to recognise vulnerability early and preserve the institution’s capacity to respond.

Best-Practice Checklist

A bank should:

1. define portfolio segments by genuine risk drivers;
2. monitor exposure on an ultimate-obligor and look-through basis;
3. combine concentration measures with correlation and scenario analysis;
4. track migration, vintage, roll-rate and early-warning indicators;
5. connect limits to risk appetite, capital and escalation procedures;
6. validate ECL models and document overlays;
7. run baseline, adverse and reverse stress tests;
8. allocate capital using marginal as well as average risk;
9. challenge data, models and incentives independently;
10. integrate ethics, customer outcomes, climate risk and technology governance.

Chapter Summary

Credit portfolio management views lending as a system of interdependent exposures rather than a collection of isolated accounts. Concentration and correlation can turn apparently diversified portfolios into vulnerable ones, while migration and vintage analysis provide early evidence of deterioration. Limit systems translate risk appetite into operational controls, and dashboards combine exposure, quality, performance, provisioning, capital and forward-looking information.

Expected credit loss provisioning estimates probability-weighted, discounted credit losses and requires consistent data, scenarios and judgement. Stress testing explores adverse but plausible conditions and supports capital planning and corrective action. Risk-adjusted performance measures such as RAROC connect revenue with funding cost, operating cost, expected loss and economic capital. Strong governance, ethical controls, model validation and responsible technology use are essential because portfolio metrics are estimates, not certainties.

Key Terms

Credit portfolio: The total collection of credit exposures held by an institution.

Concentration risk: Risk arising from excessive exposure to one borrower, sector, geography, product or risk factor.

Correlation: The degree to which two losses or defaults move together.

Diversification: Reduction of idiosyncratic risk through exposures with different risk drivers.

Wrong-way risk: Risk that exposure increases as counterparty credit quality worsens.

Expected loss: Average credit loss anticipated over a defined population and horizon.

Unexpected loss: Variability of loss above expected loss, normally covered by capital.

Migration: Movement of an exposure between credit-risk grades.

Transition matrix: Table showing probabilities of movement between grades.

Vintage: Accounts originated during the same period, analysed by age.

Cohort: Accounts grouped by a common characteristic.

Roll rate: Rate at which accounts move between delinquency categories.

Expected credit loss (ECL): Probability-weighted estimate of discounted credit loss.

Stage 1: IFRS 9 category generally using 12-month ECL.

Stage 2: Category generally using lifetime ECL after significant increase in credit risk.

Stage 3: Credit-impaired category generally using lifetime ECL.

Stress testing: Analysis of portfolio outcomes under adverse scenarios.

Economic capital: Internal capital estimate for unexpected loss at a chosen confidence level.

RAROC: Risk-adjusted return on capital.

Risk appetite: The level and types of risk an institution is willing to accept.

Review Questions

Short Answer Questions

1. Distinguish expected loss from unexpected loss.
2. Why can a portfolio diversified by borrower remain concentrated by economic factor?
3. What information does vintage analysis provide?
4. Explain the difference between Stage 1 and Stage 2 ECL under IFRS 9.
5. Why are portfolio limits necessary even when individual credit approvals are strong?

Long Answer Questions

1. Explain how concentration, correlation and diversification interact in credit portfolio risk.
2. Describe the design of an effective credit risk dashboard and identify its principal indicators.
3. Discuss the assumptions, data requirements and limitations of an ECL model.
4. Explain how stress testing should influence capital planning and lending strategy.
5. Evaluate the usefulness and limitations of RAROC for performance management.

Analytical / Application Questions

1. A portfolio has four sectors with exposure shares of 40%, 30%, 20% and 10%. Calculate its HHI and interpret the result. What additional information would you require before concluding that risk is excessive?
2. A bank observes that recent vintages have higher early delinquency, rising approval overrides and increasing Stage 2 exposure. Develop a diagnostic and management action plan.
3. Design an adverse scenario for a property-linked commercial portfolio. Explain the transmission from macroeconomic assumptions to PD, LGD, EAD, provisions and capital.

Multiple Choice Questions with Answers

Question 1

Which risk is most directly associated with excessive exposure to one industry?

A. Concentration risk

B. Settlement risk

C. Interest-rate basis risk

D. Operational downtime

Correct Answer: A. Concentration risk

Question 2

A portfolio has four equally sized sectors. What is its HHI using decimal sector shares?

A. 0.04

B. 0.16

C. 0.25

D. 4.00

Correct Answer: C. 0.25

Question 3

In the simplified formula EL = PD × LGD × EAD, what does LGD represent?

A. Expected time to default

B. Loss given default

C. Loan growth duration

D. Liquidity gap deficit

Correct Answer: B. Loss given default

Question 4

Under the usual IFRS 9 staging description, which category generally uses lifetime ECL because credit risk has significantly increased but the asset is not credit-impaired?

A. Stage 1

B. Stage 2

C. Stage 3

D. Unrated stage

Correct Answer: B. Stage 2

Question 5

What is the main purpose of vintage analysis?

A. To compare origination groups at similar ages

B. To calculate statutory tax

C. To replace all borrower-level analysis

D. To measure branch staffing costs

Correct Answer: A. To compare origination groups at similar ages

Question 6

Wrong-way risk occurs when:

A. Exposure falls as credit quality worsens

B. Exposure and credit quality are unrelated

C. Exposure increases as counterparty credit quality worsens

D. Collateral value is always above exposure

Correct Answer: C. Exposure increases as counterparty credit quality worsens

Question 7

Which statement best describes a stress test?

A. A guaranteed forecast of losses

B. An analysis of outcomes under adverse but plausible conditions

C. A substitute for credit underwriting

D. A calculation limited to historical defaults

Correct Answer: B. An analysis of outcomes under adverse but plausible conditions

Question 8

A one-year exposure is $500,000, PD is 4% and LGD is 50%. Ignoring discounting, what is expected loss?

A. $1,000

B. $5,000

C. $10,000

D. $25,000

Correct Answer: C. $10,000

Question 9

Which item is subtracted in the simplified RAROC numerator?

A. Expected loss

B. Economic capital

C. Total assets

D. Share price

Correct Answer: A. Expected loss

Question 10

Why should a transition matrix be interpreted with its economic context?

A. Transition probabilities never depend on time

B. Stable-period data may understate migration in a recession

C. All grades must have equal exposure

D. Default definitions are irrelevant

Correct Answer: B. Stable-period data may understate migration in a recession