Credit Risk, Fraud and Collections

Introduction and context

Retail and rural banking convert uncertain household and livelihood cash flows into credit decisions. The same features that make these markets socially and commercially important—large customer numbers, small loan sizes, informal income, seasonal earnings, collateral constraints, distance and information asymmetry—also create distinctive credit, fraud and recovery risks. Effective risk management therefore cannot be reduced to classifying overdue accounts. It requires a continuous system linking underwriting, portfolio monitoring, early intervention, responsible collections, provisioning, restructuring, recovery and conduct governance.

Credit risk is the possibility that a borrower or counterparty fails to meet agreed obligations, causing financial loss. In retail banking, losses arise not only from default but also from fraud, operational errors, weak documentation, inappropriate lending, poor collateral enforcement and reputational or regulatory consequences. In rural banking, agricultural cycles, weather events, commodity prices, migration, local political conditions and informal indebtedness materially affect repayment capacity.

The central managerial principle is that risk should be managed over the entire credit life cycle rather than after an account becomes non-performing.

Learning objectives

After studying this chapter, the learner should be able to:

1. Explain retail and rural credit risk, delinquency and non-performing assets.
2. Design early-warning indicators and portfolio-monitoring routines.
3. Apply probability of default, loss-given-default and exposure concepts.
4. Explain provisioning, restructuring, stress testing and recovery strategies.
5. Distinguish common fraud schemes and design preventive and detective controls.
6. Evaluate collections strategies through financial, customer-protection and governance perspectives.
7. Recommend an integrated response to a deteriorating loan portfolio.

1. The architecture of credit risk

1.1 From origination to resolution

A sound credit-risk architecture contains seven linked stages:

Risk ownership is shared. The business originates and services accounts; risk functions establish independent oversight; operations maintain data and controls; collections manage arrears; internal audit tests the system; and the board approves risk appetite. Incentives must not reward loan volume while ignoring subsequent performance.

1.2 The components of expected credit loss

A widely used analytical decomposition of expected credit loss is:

Here, ECL is expected credit loss; PD is probability of default over the relevant horizon; LGD is loss given default as a proportion of exposure after recoveries and costs; and EAD is exposure at default, measured in currency. The simple expression assumes one exposure state and does not include discounting, multiple time periods or scenario weights. In a multi-period model, expected loss is normally estimated by summing probability-weighted, discounted cash shortfalls across periods:

where t denotes time, s denotes an economic scenario, wₛ is the probability weight of scenario s, and DFₜ is the discount factor. The formula is applicable only when the institution has defensible estimates, reliable exposure data and documented scenario assumptions. It is not a substitute for regulatory classification rules.

For a fully drawn loan of ₹100,000 with an estimated PD of 4%, LGD of 45% and EAD of ₹100,000, a one-period indicative loss is:

ECL = 0.04 × 0.45 × ₹100,000 = ₹1,800

This is an expected value, not a prediction that exactly ₹1,800 will be lost on that account. Actual outcomes are binary or otherwise uncertain; the estimate becomes meaningful when applied across a sufficiently appropriate portfolio.

1.3 Risk appetite and concentration

Risk appetite states the amount and type of risk the bank is willing to accept in pursuit of its strategy. Limits may cover product, geography, borrower segment, employer, crop, commodity, collateral type, channel and delinquency bucket. Concentration risk is especially important in rural portfolios: geographically dispersed accounts may still be economically correlated if borrowers depend on the same monsoon, crop or market.

A portfolio should therefore be examined by both borrower-level and common-driver dimensions. A low average delinquency ratio can conceal severe concentration in one district or product.

2. Delinquency, NPA and impairment

2.1 Delinquency measurement

Delinquency is failure to make a required payment by its due date. Useful measures include days past due (DPD), roll rates, cure rates, vintage curves and collection effectiveness.

The collection efficiency ratio can be expressed as:

The ratio should be interpreted carefully. Prepayments, settlements, reversals and the treatment of newly booked loans can distort comparisons. A high current-period collection ratio may coexist with a growing stock of old arrears.

A roll-rate matrix tracks movement between delinquency states—for example, current, 1–30 DPD, 31–60 DPD, 61–90 DPD and default. If 1,000 accounts in the 1–30 DPD bucket are observed at month-end and 120 move to 31–60 DPD next month, the observed transition rate is 12%, subject to treatment of cures, closures and new accounts.

2.2 Non-performing assets

An NPA is a credit exposure that meets the applicable regulatory or accounting definition of non-performance. Exact thresholds and classification rules differ by jurisdiction and product; institutions must follow current supervisory instructions rather than informal labels. In many frameworks, prolonged overdue status is a central criterion, but restructuring, borrower-specific evidence and agricultural lending rules may receive special treatment.

The distinction between delinquency and NPA is important. An account can be delinquent without yet meeting the NPA threshold, while a bank must not delay recognition merely because it expects eventual recovery. Evergreening—granting fresh or additional credit primarily to conceal distress—is a serious governance failure.

2.3 Provisioning and impairment

Provisioning recognises anticipated credit loss in financial statements and protects capital from delayed loss recognition. A bank may use collective models for homogeneous retail exposures and individual assessment for material or unusual accounts. Inputs commonly include historical default experience, current conditions, forward-looking economic information, collateral values, recovery costs and the time value of money.

Provisioning is not the same as cash placed in a separate account. It is an accounting recognition of expected loss. Write-off is an accounting action that removes an asset or reduces its carrying amount; it does not necessarily terminate legal recovery efforts. Recoveries after write-off should be tracked separately and governed transparently.

3. Early-warning systems and portfolio monitoring

3.1 Purpose and design

An early-warning system (EWS) identifies signals that repayment capacity, willingness to pay or data integrity may be deteriorating. It should produce actionable alerts, not merely large volumes of unexplained flags. Each alert needs an owner, severity level, response time, documented action and closure rule.

Indicators can be classified as follows:

A useful EWS combines rules and analytics. Rules are transparent and suitable for clear events, such as a missed instalment. Statistical or machine-learning models can detect combinations of weak signals, but they require representative data, validation, explainability, monitoring for drift and controls against discriminatory outcomes.

3.2 Monitoring dashboard

A senior-management dashboard should show stock and flow measures: gross and net NPA, delinquency by bucket, roll rates, cure rates, vintage performance, restructurings, write-offs, recoveries, provisioning coverage, fraud losses, complaints and collection productivity. Results should be segmented by product, branch, channel, geography, customer vulnerability and origination cohort.

The dashboard should distinguish correlation from causation. For example, a rise in missed payments may reflect a system debit failure rather than borrower distress. Operational reconciliation must precede aggressive collection.

3.3 Stress testing

Stress testing asks how a portfolio would perform under adverse but plausible conditions. A simple scenario may increase PD and LGD, reduce collateral values and raise EAD through unused commitments. If a portfolio has EAD of ₹50 crore, baseline PD of 3% and LGD of 40%, the indicative loss is ₹0.60 crore. Under a scenario of PD 6% and LGD 50%, it becomes ₹1.50 crore:

₹50 crore × 0.06 × 0.50 = ₹1.50 crore

The increase is ₹0.90 crore. This simplified calculation assumes one period, unchanged EAD and no discounting. A robust stress test also models migration, recoveries, management actions, capital effects and liquidity consequences. Scenarios should include macroeconomic shocks, climate events, cyber incidents, fraud spikes and operational disruption where relevant.

4. Fraud risk in retail and rural banking

Fraud is intentional deception for unlawful gain or to cause loss. It differs from error, although weak processes can enable both. Fraud risk management follows the logic of prevention, detection, investigation, response and learning.

4.1 Common schemes

Retail and rural institutions may face identity theft, synthetic identity, document alteration, impersonation, account takeover, mule accounts, collusion with agents, loan stacking, fictitious borrowers, diversion of group-loan funds, collateral fraud, insurance fraud, merchant or QR manipulation, phishing and social engineering. Digital channels add risks involving stolen credentials, SIM replacement, malicious applications and deepfake-enabled impersonation.

4.2 The fraud triangle and its limits

The fraud triangle identifies pressure or incentive, opportunity and rationalisation. It is a useful behavioural lens but not a complete predictive theory. Organisational culture, capability, collusion, technology and perceived punishment also matter. Controls should therefore address both individual behaviour and system design.

4.3 Control framework

Preventive controls include customer due diligence, segregation of duties, maker-checker approval, independent valuation, device and biometric controls where lawful, transaction limits and secure authentication. Detective controls include exception reports, velocity rules, duplicate-document analytics, device intelligence, post-disbursement verification and whistle-blower channels. Corrective controls include account freezing under due process, customer notification, investigation, recovery, disciplinary action, law-enforcement reporting and control remediation.

AI can support anomaly detection, but automated blocking requires human review, data-quality controls, appeal mechanisms and privacy protection. A model that flags rural customers disproportionately because of sparse digital histories may reproduce exclusion rather than identify fraud. Explainability and proportionality are essential.

5. Collections and recovery strategy

5.1 Segmentation before action

Collections should be segmented by capacity and willingness to pay, not merely by arrears age. A customer who can pay but refuses may require firm, lawful escalation; a temporarily affected borrower may need a short moratorium or revised schedule; a deceased borrower’s family may require sensitive documentation and insurance assistance; and a disputed debit may require immediate operational correction.

A practical segmentation matrix considers:

• ability to pay;
• willingness and contactability;
• cause of delinquency;
• account value and exposure;
• vulnerability and hardship;
• fraud or dispute indicators; and
• legal and collateral position.

5.2 The collection sequence

A disciplined sequence is: validate the account and dues; send clear and respectful reminders; establish contact through permitted channels; understand the cause; obtain a realistic promise to pay; record it; monitor performance; offer an approved solution where appropriate; and escalate proportionately if default persists.

Promises to pay should be measurable. A promise-to-pay kept rate is:

Incentives based only on cash collected can encourage harassment, misallocation of payments or inappropriate settlements. Balanced scorecards should include cure quality, complaints, vulnerable-customer treatment, data accuracy and sustainable repayment.

5.3 Restructuring

Restructuring changes contractual terms—such as tenor, instalment size, interest treatment or repayment frequency—to restore viability. It is appropriate when the borrower has a credible repayment capacity under revised terms. It is not appropriate merely to postpone recognition of loss.

The assessment should cover verified cash flow, total indebtedness, revised debt-service capacity, cause of stress, collateral and legal implications, customer consent, affordability and exit criteria. Rural restructuring may need to reflect harvest cycles, but seasonal alignment must not disguise an economically unviable activity.

5.4 Recovery options

Recovery tools include negotiated settlement, collateral sale, guarantee or insurance claims, legal action, insolvency processes, debt counselling and portfolio sale, subject to law and policy. The decision should compare expected net recovery, time, legal cost, reputational risk and customer impact. Collateral is a secondary source of repayment; possession of collateral does not guarantee liquidity or value.

A net-recovery estimate can be stated as:

The items must be measured consistently in currency and over a defined horizon. The estimate is decision support, not a guaranteed outcome.

6. Conduct, ethics and regulation

Collections and fraud controls operate within consumer-protection, privacy, fair-lending, outsourcing, digital-payment and debt-recovery requirements. Rules vary by jurisdiction, but sound principles are broadly stable: communicate accurately; do not threaten, shame or disclose debt to unauthorised persons; respect contact hours and consent; protect personal data; provide complaint channels; accommodate reasonable vulnerability; and maintain evidence of decisions.

Conduct risk is not separate from credit risk. Mis-selling, hidden charges, unsuitable loan sizes and coercive collections can increase future defaults and legal exposure. Rural customers may have low literacy or limited connectivity, requiring local-language explanations, assisted channels and accessible statements.

Governance should include board oversight, a documented policy hierarchy, independent risk review, compliance testing, internal audit, incident escalation and regulatory reporting. Exceptions require written rationale and post-approval review. Vendor and agent controls must include due diligence, training, monitoring, audit rights, complaint handling and termination provisions.

7. Integrated applied case

Illustrative case: Sunrise Rural Bank

Background. Sunrise Rural Bank has a ₹240 crore microenterprise and agricultural portfolio. Its overall overdue ratio appears moderate, but the eastern region shows rising 31–90 DPD accounts. Most borrowers cultivate the same cash crop, and a local pest outbreak has reduced yields. At the same time, the bank detects duplicate identity documents in a small digital-loan channel.

Problem. Branch managers propose blanket restructuring to protect reported asset quality. Collections staff are contacting family members of borrowers without documented consent. The bank’s fraud dashboard generates many alerts but few investigations. Senior management must protect customers and capital without concealing deterioration.

Analysis. Portfolio segmentation reveals three groups. First, many small farmers have genuine temporary cash-flow stress and credible next-season repayment capacity. Second, some traders show persistent high leverage and repeated promise-to-pay failures. Third, digital accounts exhibit identity and device anomalies requiring fraud investigation. A system reconciliation also finds that several instalments were rejected because of a mandate-processing error.

Stress testing assumes crop-related PD doubles from 5% to 10%, LGD rises from 45% to 55% and EAD remains ₹80 crore for the affected segment. Baseline indicative loss is:

₹80 crore × 0.05 × 0.45 = ₹1.80 crore

Stressed indicative loss is:

₹80 crore × 0.10 × 0.55 = ₹4.40 crore

The incremental indicative loss is ₹2.60 crore, before considering scenario uncertainty, discounting or management action.

Approach. Sunrise should immediately correct the mandate failures and reverse inappropriate fees. It should offer transparent, individually assessed restructuring to viable affected farmers, aligned with harvest cash flows and supported by documented affordability analysis. Traders with persistent stress should receive intensified review rather than automatic relief. Fraudulent or suspicious digital applications should be placed under controlled investigation, with customer protection and lawful evidence preservation.

Collections should use local-language communication, prohibit third-party disclosure, and measure kept promises and complaints. The risk function should produce a weekly regional dashboard showing roll rates, restructures, cures, fraud cases, provisioning and customer outcomes. Internal audit should review whether restructuring is being used to hide NPAs. The board should approve a sector concentration limit and contingency plan.

Outcome. In this illustrative scenario, the bank’s reported delinquency initially rises because overdue accounts are recognised honestly. Over the next two quarters, viable farmers cure under revised schedules, mandate errors decline, suspicious digital disbursements fall, and collection complaints reduce. Capital planning uses the stress result rather than the optimistic baseline.

Learning points. Early recognition may worsen short-term reported metrics while improving long-term credibility. Portfolio correlation matters more than branch averages. A collection problem can originate in operations. Fraud analytics require investigation capacity. Restructuring is a risk decision, not a reporting device, and conduct quality is part of sustainable recovery.

8. Challenges and contemporary developments

Banks face incomplete bureau coverage, informal income, model drift, climate volatility, cyber-enabled fraud, third-party dependence and pressure for rapid digital growth. Alternative data may improve assessment but can create privacy, consent and discrimination risks. Generative AI can assist call summarisation, document review and analyst prioritisation, but it should not autonomously deny credit or intensify collections without validated controls, human accountability and secure data governance.

Climate-risk integration is increasingly important. Physical hazards affect repayment through income and collateral; transition policies may affect borrower sectors and asset values. Scenario design should avoid false precision and disclose limitations. Research and practice are moving toward more granular cash-flow underwriting, network-based fraud detection, explainable models, behavioural collections and outcome-based conduct metrics.

Best-practice framework

A mature institution should:

1. set portfolio risk appetite and concentration limits;
2. verify identity, purpose, affordability and repayment capacity;
3. maintain accurate, timely and reconciled data;
4. monitor behavioural, financial, sectoral and conduct indicators;
5. intervene early with proportionate and customer-sensitive actions;
6. recognise delinquency, impairment and fraud transparently;
7. validate models and stress scenarios independently;
8. restructure only viable exposures with documented consent;
9. use lawful, respectful and auditable collections;
10. investigate fraud quickly and learn from root causes; and
11. align incentives, governance and capital planning with long-term outcomes.

Chapter Summary

Credit risk management in retail and rural banking is a life-cycle discipline covering origination, monitoring, early warning, classification, provisioning, restructuring, recovery and learning. Expected loss analysis uses PD, LGD and EAD, but models depend on reliable data, appropriate assumptions and regulatory definitions. Delinquency measures such as DPD, roll rates, cure rates and vintage curves reveal portfolio migration more effectively than a single ratio.

Early-warning systems should combine behavioural, financial, external, rural-sectoral and operational indicators with clear ownership and intervention rules. Stress testing exposes sensitivity to correlated shocks such as crop failure, economic contraction, collateral decline and fraud. Fraud management requires preventive, detective and corrective controls, while AI must be governed for explainability, privacy and fairness.

Collections should segment borrowers by cause, capacity, willingness, vulnerability and legal position. Restructuring is appropriate only where revised terms restore viable repayment and must not conceal loss. Provisioning, write-offs and recoveries require transparent accounting and governance. Finally, lawful conduct, customer protection and data governance are not peripheral concerns: they directly influence recovery quality, reputation, regulatory standing and sustainable banking.

Key Terms

Credit risk: Risk of loss because a borrower or counterparty fails to meet obligations.

Delinquency: Failure to make a required payment by its due date.

Days past due (DPD): Number of days elapsed since a payment became overdue.

Non-performing asset (NPA): Exposure meeting the applicable regulatory or accounting non-performance criteria.

Probability of default (PD): Estimated likelihood that an obligor defaults over a specified horizon.

Loss given default (LGD): Proportion of exposure expected to be lost after recoveries and costs.

Exposure at default (EAD): Amount exposed when default occurs.

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

Roll rate: Rate at which accounts migrate from one delinquency state to another.

Cure rate: Proportion of delinquent accounts returning to satisfactory status.

Early-warning system: Process for identifying and acting on emerging risk signals.

Stress testing: Assessment of portfolio performance under adverse plausible scenarios.

Restructuring: Modification of contractual terms to address borrower difficulty.

Evergreening: Extending new credit mainly to conceal an existing stressed exposure.

Fraud: Intentional deception for unlawful gain or to cause loss.

Collections: Activities undertaken to secure payment or resolve delinquent exposure.

Promise to pay: Customer commitment to make a specified payment by a specified date.

Provisioning: Recognition of expected loss in financial reporting.

Conduct risk: Risk of customer or market harm arising from inappropriate behaviour or processes.

Review Questions

Short Answer Questions

1. Distinguish delinquency from an NPA.
2. Define PD, LGD and EAD.
3. What is the purpose of an early-warning system?
4. State two preventive and two detective fraud controls.
5. Why is evergreening a governance concern?

Long Answer Questions

1. Explain the credit-risk life cycle in retail and rural banking.
2. Discuss how provisioning and stress testing support capital and risk management.
3. Evaluate the principles of responsible collections and restructuring.
4. Explain how rural-sector concentration can create correlated credit losses.
5. Design a governance framework for digital fraud risk.

Analytical / Application Questions

1. A portfolio has EAD of ₹30 crore, PD of 8% and LGD of 50%. Calculate its one-period indicative expected loss and explain the limitations of the estimate.
2. A region shows rising 1–30 DPD accounts, falling deposits, a crop-price decline and increased customer complaints. Construct an EWS response plan, including data validation and management actions.
3. A bank’s collection ratio improves after aggressive third-party calling, but complaints and repeat defaults rise. Analyse the likely causes and recommend a balanced performance framework.

Multiple Choice Questions with Answers

Question 1

Which expression represents the simplified one-period expected credit loss model?

A. PD + LGD + EAD

B. PD × LGD × EAD

C. PD ÷ LGD × EAD

D. EAD − PD − LGD

Correct Answer: B. PD × LGD × EAD

Question 2

Which indicator most directly measures movement from one delinquency bucket to another?

A. Roll rate

B. Capital adequacy ratio

C. Net interest margin

D. Loan-to-deposit ratio

Correct Answer: A. Roll rate

Question 3

Why should restructuring not be used automatically for every overdue account?

A. It always increases collateral value

B. It eliminates the need for monitoring

C. It may conceal non-performance when repayment is not viable

D. It guarantees full recovery

Correct Answer: C. It may conceal non-performance when repayment is not viable

Question 4

Which is primarily a detective fraud control?

A. Segregation of duties

B. Maker-checker approval

C. Post-disbursement exception analytics

D. Customer consent at onboarding

Correct Answer: C. Post-disbursement exception analytics

Question 5

A loan has EAD of ₹10,000, PD of 5% and LGD of 40%. What is the simplified expected loss?

A. ₹20

B. ₹200

C. ₹400

D. ₹2,000

Correct Answer: B. ₹200

Question 6

Which factor is particularly relevant to correlated rural credit risk?

A. A common crop and weather exposure

B. The colour of branch signage

C. The number of ATM screens

D. The age of the accounting software

Correct Answer: A. A common crop and weather exposure

Question 7

What should a bank do first when a borrower disputes an overdue amount caused by a failed debit mandate?

A. Immediately commence coercive recovery

B. Validate and reconcile the account records

C. Sell the loan without investigation

D. Classify every related account as fraudulent

Correct Answer: B. Validate and reconcile the account records

Question 8

Which statement about write-off is correct?

A. It necessarily ends legal recovery

B. It is identical to a customer refund

C. It is an accounting action and may not end recovery efforts

D. It guarantees that no loss occurred

Correct Answer: C. It is an accounting action and may not end recovery efforts

Question 9

A stress test is mainly used to:

A. Guarantee the exact future loss

B. Assess performance under adverse plausible conditions

C. Replace all underwriting controls

D. Remove the need for provisioning

Correct Answer: B. Assess performance under adverse plausible conditions

Question 10

Which collection practice is most consistent with responsible conduct?

A. Disclosing debt to neighbours to create pressure

B. Using clear, respectful and proportionate communication

C. Ignoring customer complaints

D. Rewarding agents only for any cash collected

Correct Answer: B. Using clear, respectful and proportionate communication