Loss Given Default Estimation in Banks: LGD Explained (CAIIB RM)

Getting loss given default estimation in banks right is what separates a credit model that merely ranks borrowers from one that actually prices risk. Probability of default tells you how likely a borrower is to fail; LGD tells you how much money you will never see once that failure happens. For CAIIB Risk Management (Elective) candidates, this is a high-yield area because examiners can test it as a formula, as a concept, and as an Indian recovery-law question in the same paper.

This guide walks you through the definition, the workout and market methods, the drivers of recovery, the supervisory versus own-estimate distinction, and the validation expectations — in the sequence the syllabus follows.

📉 Where LGD Sits in the Expected Loss Identity

Every credit risk framework starts from the same identity: Expected Loss = PD × LGD × EAD. Probability of default (PD) is a likelihood over a one-year horizon. Exposure at default (EAD) is the amount outstanding when default occurs, including expected drawdowns on undrawn limits. LGD is the fraction of that exposure you finally lose.

LGD is expressed as a percentage of exposure at the time of default, and it is the economic loss on a defaulted facility — not the book write-off. Its complement is the recovery rate: LGD = 1 − recovery rate. A facility with 40% LGD means that for every ₹100 outstanding at default, ₹60 comes back in present-value terms after every cost is netted off.

Notice what this does to portfolio thinking. A borrower with a poor rating but rock-solid, quickly enforceable collateral can carry a lower expected loss than a better-rated borrower lending clean. That is precisely why sound loss given default estimation in banks matters as much as rating accuracy — the two multiply. Get one badly wrong and the product is wrong, whatever the sophistication of the other.

LGD is facility-specific, while PD is borrower-specific. The same company can have a 20% LGD on a mortgage-backed term loan and a 75% LGD on a subordinated instrument. Any bank building a risk management framework must therefore rate the obligor and grade the facility separately.

🧾 Economic Loss, Not Accounting Loss: The Workout Method

This is the distinction candidates most often fluff. Accounting loss is the write-off in the books — outstanding minus cash recovered. Economic loss is larger, because it must also carry the time value of money over the workout period and every direct and indirect cost of recovering.

The workout method builds LGD from your own history. You take each defaulted facility, collect every post-default cash flow — auction proceeds, one-time settlement instalments, insurance, guarantee invocation, IBC distributions — and discount them back to the date of default at a rate that reflects the risk of those uncertain recovery cash flows, not simply the contractual loan rate.

Then you subtract costs. Direct costs include legal fees, DRT filing costs, resolution professional and CIRP expenses, valuation and auction charges, and the cost of maintaining seized collateral. Indirect costs include the salaries and overheads of the recovery vertical, allocated sensibly. Skip them and your LGD is flattered.

⚠️ Common Mistake: Treating a ₹100 crore recovery received four years after default as a ₹100 crore recovery. Discounted at even 10%, that is roughly ₹68 crore in present value — the missing ₹32 crore is real economic loss, and ignoring it understates LGD badly.

The workout period is itself a modelling choice. You need a policy on when a case is treated as closed, and how to handle incomplete workouts that are still running. Using only completed cases introduces resolution bias, because quick recoveries close first and drag the average recovery upward. Robust loss given default estimation in banks therefore includes open workouts with conservative assumptions rather than quietly excluding them.

Cures matter too. Under the 90-days-past-due default definition, many accounts default and then regularise with near-zero loss. Whether you include cures changes the mean materially, so the treatment must be documented and applied consistently.

🧮 Three Ways to Estimate LGD — and When Each Works

Beyond the workout method, two market-based routes exist for loss given default estimation in banks. The market LGD method reads the price of defaulted debt shortly after default: if a defaulted bond trades at 35% of par, the market's implied recovery is 35% and LGD is 65%. It is fast and forward-looking, but it needs a liquid secondary market in distressed paper.

The implied market LGD method goes further and backs the number out of credit spreads on performing instruments — bonds or credit default swaps. Since a spread compensates for expected loss, if you assume a PD you can solve for LGD. The catch is that spreads also contain liquidity and risk premia, so the extraction is only as good as your decomposition. Candidates who have studied credit default swaps in Indian banks will recognise this as the same pricing logic run backwards.

For Indian banks the workout method is effectively the only complete answer, because the domestic market for defaulted debt is not deep enough to price a representative sample. Market-based readings are still useful as a sanity benchmark on large corporate names, in the same spirit that traders benchmark value at risk models against actual P&L swings.

🏦 What Actually Drives Recovery: Seniority, Collateral and Enforcement

Five drivers explain most of the variation. First, seniority of the claim: senior secured sits ahead of senior unsecured, which sits ahead of subordinated debt in the waterfall. Second, collateral type and quality, and the haircut you apply to its assessed value — a liquid financial security takes a modest haircut, while specialised plant and machinery or a half-built project takes a severe one.

Third, and uniquely important in India, enforceability. A security interest is only worth what you can realise from it. The SARFAESI Act, 2002 lets a secured creditor issue a demand notice under Section 13(2) and, on non-payment within sixty days, take possession and sell under Section 13(4) without court intervention — provided the security is not excluded from the Act. Where SARFAESI does not apply you fall back on the DRT route or, for corporate debtors, the Insolvency and Bankruptcy Code, 2016, whose corporate insolvency resolution process is designed for 180 days extendable by 90, with an outer limit of 330 days including litigation time. Actual elapsed time routinely exceeds that, and every extra quarter compounds into your discount factor.

Fourth, the industry and asset specificity: infrastructure and steel assets recovered very differently from retail-mortgage collateral across the last cycle. Fifth, the state of the economic cycle. Recoveries fall exactly when defaults rise, because distressed assets flood a shrinking pool of buyers.

📌 Remember: Recovery rates and default rates are negatively correlated. That single fact is the entire reason supervisors insist on a downturn LGD instead of a long-run average.

A sixth practical driver is the exit route. Selling the exposure rather than working it out changes the cash-flow profile completely, which is why banks model the sale of stressed assets to ARCs as an alternative recovery scenario. Counterparty exposures behave differently again, as students of derivatives and risk management know, because close-out netting and margin dramatically compress the amount at risk.

⚖️ Supervisory Values, Downturn LGD, Data and Validation

Under the foundation internal ratings based (F-IRB) approach a bank estimates PD only; the supervisor supplies LGD. The Basel framework's classic supervisory values are 45% for senior unsecured claims and 75% for subordinated claims, with the Basel III finalisation trimming the senior unsecured corporate figure to 40% and prescribing recognition rules for eligible collateral. Under the advanced (A-IRB) approach the bank uses its own LGD, EAD and maturity estimates — subject to far heavier data, governance and validation obligations.

In India this remains largely academic for capital purposes: banks compute credit risk capital under the Standardised Approach, where risk weights are prescribed and LGD is not a bank input. RBI issued IRB guidelines in December 2011 and permits banks to apply for migration, but the supervisory bar is high; the current rules are on the Reserve Bank of India website. LGD still matters enormously for internal pricing, provisioning and limit-setting even without IRB approval.

The data rules are exam-friendly. The observation period must cover a full economic cycle, with a minimum of seven years for corporate, sovereign and bank exposures and five years for retail under the advanced approach. And you must estimate a downturn LGD wherever recovery and default rates are cyclically linked — not simply the historical mean.

That mean is treacherous anyway. The empirical LGD distribution is bimodal: one cluster near total recovery (fully secured, quickly enforced) and another near total loss (clean exposures with nothing to attach). Very few observations sit near the average, so quoting a single portfolio mean describes almost no actual facility. Correct loss given default estimation in banks therefore segments the book — by product, collateral type, seniority and workout route — and models each pool separately.

💡 Exam Tip: If a question describes a distribution with peaks at both ends and asks what is wrong with using the simple average, the answer is that a bimodal distribution has no representative mean; you must segment before you average.

Downstream, the estimate feeds two engines. In provisioning it is the L in the expected credit loss calculation — NBFCs already apply Ind AS 109, and RBI has proposed moving banks from incurred-loss to an expected credit loss framework. In pricing, LGD drives the expected-loss component of the loan rate and the capital charge behind RAROC, which is why the same funding and liquidity assumptions used in liquidity risk management must be consistent with the workout horizon you assume.

Validation closes the loop. Expect at least annual independent review comparing realised against predicted LGD by grade and pool, tests of stability and discriminatory power, benchmarking against external data, and documented override policy — all owned outside the model-building unit under the three lines of defence. More CAIIB elective material sits on the Risk Management elective blog hub.

🧠 Practice MCQs: Loss Given Default Estimation

Want chapter-wise mock tests with 100+ MCQs? Start practising free →

🎯 Key Takeaways and Your Next Step

Reduce the topic to five sentences and you will answer almost any question on it. LGD is the economic loss on a defaulted facility as a percentage of exposure at default. Economic loss includes discounting over the workout period plus all direct and indirect costs. The workout method dominates in India because the market and implied-market methods need traded distressed debt. Seniority, collateral quality and the real enforcement timeline under SARFAESI or the IBC drive the number, and the distribution is bimodal, which is why a downturn, segmented estimate is required.

Do that consistently and loss given default estimation in banks stops being a formula and starts being a control — feeding expected credit loss provisions, risk-based pricing and capital planning from one validated set of numbers. Round out your revision with the related monetary-policy material on the cash reserve ratio in India, which the same CAIIB fleet of papers will test alongside this one.

Ready to test yourself under exam conditions? Work through the full CAIIB course with chapter-wise classes and mock tests and target the Risk Management elective paper with confidence.