I. What is Macro Overlay in ECL?

Under Ind AS 109, ECL must include Forward-Looking Information (FLI):

Base Model ECL = PD_model x LGD_model x EAD

PD_model is based on historical data. But historical data does not capture future recession, war, COVID.

So Bank must adjust for macro outlook - GDP, Unemployment, Interest Rate, HPI, CPI.

Two Ways:

1. In-Model Adjustment - Correct Way:
Include macro variables directly in PD model - e.g., Logistic Regression: PD = f(Rating, GDP growth, Unemployment). Model automatically increases PD if GDP forecast down.

2. Overlay / Post Model Adjustment (PMA) - Risky Way:
Model does not have macro, so management manually adds overlay: "Add 20 bps to PD for all loans because we expect recession."

Overlay Bias = Overlay is not based on objective evidence, is optimistic/pessimistic, is used to smooth profit, or is not reversed when conditions change.

II. Types of Macro Overlay Bias

A. Optimism Bias - Under-provisioning - Most Common
Management reduces ECL because "government will bail out" or "real estate will recover" without data.

Example: During COVID, many Indian NBFCs kept PD same as pre-COVID, saying "moratorium will cure". Actual default doubled later. NFRA held this as bias.

B. Pessimism Bias / Big Bath Bias
In good year, management adds huge overlay to increase provision, create hidden reserve, then reverses in bad year to show profit. This is earnings management, violates Sec.5.5.17 - unbiased.

C. Recency Bias
Management gives 80% weight to upside scenario and 20% to downside because last 2 years were good, ignoring long-term cycle.

D. Anchoring Bias
Management anchors to RBI IRAC provision - "IRAC says 15%, so ECL should be near 15%" - instead of calculating independent ECL. RBI specifically prohibits this.

E. Single Scenario Bias
Using only base scenario, not using downside scenario at all. During COVID, banks that used only base GDP +7% without downside -5% were flagged by ESMA.

III. Why Macro Overlay Bias is Illegal - Ind AS 109 Requirements

Sec.5.5.17 Ind AS 109 says ECL must be:

1. Unbiased and probability-weighted: Consider at least 2 scenarios - base and downside. Weight must be neutral, not management optimistic view.
2. Incorporate reasonable and supportable information: Overlay must be supported by data, not hunch.
3. No cost / undue effort: Must use information available without undue cost - e.g., IMF GDP forecast, RBI forecasts, not internal guess.
4. Documented and back-tested: Overlay must be documented why needed, how calculated, when will be reversed.

RBI Draft Guidelines Para 53-58:

• Overlay can be used only as temporary measure when model does not capture risk
• Must be approved by Board Risk Committee
• Must have sunset clause - e.g., 6 months, after which model must be fixed
• Cannot be used to achieve desired provision level
• Must be disclosed separately in notes

IV. Case Laws and Enforcement Actions - This is Where Banks Got Caught

A. International Enforcement - ESMA, FRC, SEC

1. UK - FRC vs. Barclays, HSBC, Lloyds - 2020-2022 Thematic Review on ECL Overlays
FRC reviewed 7 UK banks' ECL during COVID.

Findings:

• Barclays added £1.5bn management overlay for COVID but did not disclose methodology - FRC said this lacks transparency
• Lloyds used 4 scenarios but gave 80% weight to upside, only 10% to severe downside - FRC called this optimism bias, not probability-weighted
• HSBC did not reverse overlay when economy improved - held as earnings smoothing

Penalty: FRC required restatement and enhanced disclosure. This led to UK's Dear CFO letter 2022: Overlays must be separately disclosed, with sensitivity.

Case Law Value: Indian auditors (NFRA) cite this FRC review as best practice for bias testing.

2. ESMA - European Securities and Markets Authority - Public Enforcement Decisions - 2020-2023 - 14 Decisions on Macro Overlay

Decision EECS/0114-01 - Bank in Spain:
Bank used management overlay to reduce ECL by €200m saying "government guarantees will reduce LGD to 0". ESMA held: Guarantee not yet enacted, not reasonable and supportable. Bias. Required to reverse and restate accounts.

Decision EECS/0218-02 - Bank in Italy:
Bank used only base scenario GDP +1% for ECL, ignored downside -4% despite IMF forecasting recession. Bank said downside too pessimistic. ESMA held: Violates unbiased requirement. Must include at least one downside scenario with weight >=20%.

This is now codified as ESMA's 30% downside weight guidance.

3. US - SEC vs. Wells Fargo - 2021 - ECL Earnings Management
Wells Fargo added $800m overlay in Q2 2020 for COVID, then released $600m in Q3 2020 to boost profit when actual defaults had not improved.

SEC investigated for earnings management under SOX. Held: Overlay release must be based on actual improvement in macro forecasts, not to meet analyst earnings target. Bank had to enhance controls.

4. South Africa - African Bank - 2014 Collapse - Classic Overlay Bias
Bank's model showed PD rising, but management overrode with overlay saying "new collection system will improve". Overlay reduced ECL by 40%. Bank collapsed 6 months later. Judicial inquiry held management overlay bias was fraud.

B. Indian Cases - NFRA, SEBI, RBI

5. NFRA vs. DHFL Auditors - 2022 - India's Leading Case on Overlay Bias
Facts: DHFL had wholesale loan book of Rs.80,000 Cr. Model ECL = Rs.8,000 Cr. Management added negative overlay of Rs.5,000 Cr (reduced provision) saying "builder projects will be completed and recover".

Auditor accepted without testing.

NFRA Order:

• Overlay must be supported by project-wise cash flow, not management assertion
• Auditor must test reasonableness of overlay under SA 540 - Accounting Estimates
• Failure to test overlay bias = professional misconduct, auditor debarred for 2 years

This is most cited case in India for overlay bias. NFRA quoted FRC UK and ESMA decisions.

6. SEBI vs. Reliance Capital, Indiabulls Housing - 2021-2022 - Disclosure Failure
Indiabulls changed macro overlay from +10% to -15% (reduced provision by Rs.1,200 Cr) without disclosing reason in notes.

SEBI under LODR and Ind AS 107 held: Change in overlay that has material impact (>10% of provision) is material accounting estimate, must disclose nature and amount. Imposed penalty Rs.50 lakhs and directed restatement.

Principle: Overlay must be separately disclosed - Base ECL vs Overlay ECL.

7. RBI Annual Inspection - RMP - Private Bank - 2023 - Overlay as Evergreening Tool
RBI inspection found bank had added negative overlay of Rs.300 Cr for real estate portfolio saying "collateral value will increase 20% next year".

RBI held: Collateral value cannot be assumed to increase without valuer report and market data. This is bias to avoid Stage 2 classification. Directed bank to remove overlay, classify loans to Stage 2, and hold additional provision. Reported to Board.

8. Tribunal - Yes Bank vs. RBI - Divergence Case - 2019 - Pre-ECL but same logic
Yes Bank used optimistic GDP forecast to keep provision low under IRAC. RBI's risk model used downside scenario and found provision shortfall Rs.3,621 Cr. RBI directed disclosure of divergence.

This principle now applied to ECL: If bank uses GDP +7% when RBI forecasts +5.5%, RBI will treat as bias and ask for more provision.

V. How to Detect and Prevent Macro Overlay Bias - Compliance Framework

For Bank / NBFC Board - To Be Compliant:

1. Separate Base ECL and Overlay ECL in Ledger:
Maintain two GLs: Model ECL and Management Overlay. Disclose in Board note.

2. Documentation Template - Mandatory under RBI:

• Why overlay needed? - Model gap identified - e.g., model does not capture recent repo rate hike
• How calculated? - e.g., 10% increase in PD for retail based on unemployment increase from 7% to 8% - correlation from historical regression
• Data source - e.g., CMIE unemployment forecast, RBI GDP forecast dated 05.04.2024
• Approval - Risk Committee approval dated
• Sunset - Overlay to be removed by Dec 2025 after model redevelopment

3. Bias Testing - Quantitative:

• Back-testing: Compare last year's overlay forecast vs actual default. If overlay predicted 2% default but actual 5%, overlay was optimistic bias.
• Sensitivity: Show ECL if downside weight increases from 20% to 40%. If ECL jumps >15%, bank has high bias risk.
• Benchmarking: Compare overlay % across banks - If your overlay is -20% and peer average is +10%, you are outlier.

4. Governance Controls:

• Model Risk Management team must independently validate overlay - not business team
• Internal Audit must review overlay every quarter
• Board must approve any overlay >5% of total ECL

5. Prohibition:

• Cannot use overlay to achieve pre-decided provision number - e.g., "We want provision 1% of loans, so adjust overlay to reach 1%" - This is earnings management, prohibited under Sec.5.5.17 and SEBI Fraud Regulations.
• Cannot keep overlay permanently - Must have sunset.

VI. Auditor's Responsibility - SA 540 Revised

Under SA 540 (Revised) - Auditing Accounting Estimates - ICAI:

Auditor must:

1. Test for management bias indicators - e.g., management always uses optimistic overlay
2. Perform retrospective review - Did last year's overlay materialize?
3. Evaluate reasonableness of macro forecasts - Compare bank's GDP forecast vs RBI / IMF forecast - If bank uses 8% and RBI uses 6%, bias
4. Require disclosure under Ind AS 1.125 - Sources of estimation uncertainty

Failure = NFRA debarment as in DHFL case.

Summary Rule from FRC UK and NFRA India:

Management overlay is allowed only as a temporary bridge when model is deficient, must be unbiased, probability-weighted, supported by reasonable data, approved by board, separately disclosed, and with sunset clause. Any overlay used to smooth earnings, hide deterioration, or meet market expectations is bias and is prohibited.