LGD Is a Discounted Recovery Process, Not One Minus a Recovery Rate

Economic LGD begins with what remains unrecovered

The intuitive identity LGD = 1 − Recovery Rate is a useful starting point, but only if “recovery” already means consistently defined, discounted net cash flows. Otherwise it hides the questions that determine economic loss: what was owed at default, which recovery channels produced cash, what they cost and when they paid.

EAD_τ is exposure at the defined default date. Origination balance, current reporting balance and exposure at default are not interchangeable. A €4,000 discounted net recovery on €10,000 EAD implies 60% LGD; it does not imply the same result against a different balance date.

Identical nominal recovery can produce different economic LGD

Consider two fictional loans, each with €10,000 EAD and €6,000 nominal recovery. Using an illustrative 8% annual discount rate, Case A pays after six months; Case B pays after three years. The amount is identical. Its economic value is not.

Costs change the severity before time does

Suppose €5,000 gross recovery costs €700 to obtain and the €4,300 net cash arrives after 18 months. At 8%, its PV is €3,831.18 and LGD is 61.69%. Ignoring cost would produce 55.45%—a 6.24 percentage-point understatement in this example.

The recovery cash-flow curve is the mathematical core

R_t should be a net cash flow at a controlled time grain, not a final percentage detached from its path. The original profile below follows a €10,000 default through 24 months at 8%. Gross cash totals €6,300, cost totals €630, nominal net recovery is €5,670, but discounted net recovery is €5,386.03—an economic LGD of 46.14% rather than nominal LGD of 43.30%.

Recovery speed is not a cosmetic diagnostic. Even if ultimate nominal cash is unchanged, Time to Recovery ↑ makes PV(Recovery) ↓. A six-month delay sensitivity therefore belongs beside Recovery Amount ±10% and Collateral Value −15%, with ΔLGD reported rather than hidden inside a new average.

LGD is a mixture of cure and workout paths

A cure can reduce economic loss, but cure is not automatically permanent recovery. Define temporary and sustained cure, attach cash flows and costs to the correct default episode, and observe re-default. A borrower who cures, pays briefly and defaults again may lengthen the workout and change realised LGD. Cure ≠ Permanent Recovery.

Workout LGD reconstructs observed post-default cash flows: Default Date → Recoveries → Costs → Discounting → Realised Economic Loss. It provides strong empirical grounding when history is sufficiently mature and default, recovery and closure definitions are stable. Market LGD, implied LGD, regulatory downturn LGD and IFRS 9 expected LGD answer different questions; labels must not substitute for use-case alignment.

Accounting write-off can precede economic recovery. Post-write-off cash remains part of the recovery record under the chosen methodology. If a defaulted receivable is sold, sale proceeds may be a recovery cash flow, while the decision to sell remains a treatment that can influence observed LGD.

Collateral value is not recovery value

Collateral is one recovery channel, governed by value, priority and seniority, enforceability, liquidity, price volatility, time to sale and sale or legal costs. A haircut is not an explanation unless evidence supports how those components translate a valuation into cash.

LTV = Exposure / Collateral Value remains useful, but identical LTV can conceal different asset liquidity, lien position, borrower behaviour, enforcement and timing. Unsecured portfolios rely more on borrower payments, collections, cure and legal recovery; secured portfolios add collateral dynamics rather than replacing the rest of the workout.

The downside changes both amount and speed. It does not assert a market benchmark or proprietary haircut: it isolates why a €20,000 valuation shock, unchanged €8,000 sale cost and one additional year of delay can move LGD by 26.06 percentage points.

Recovery vintages reveal development, censoring and the tail

For default vintage v, track Recovery_v,t and cumulative recovery CR_v(m) by months since default. Curves expose collections effectiveness, legal change, macro conditions and portfolio mix. Compare like recovery ages: a six-month-old vintage cannot be judged against the final recovery of a four-year-old vintage.

Open cases, restructuring, legal enforcement and unresolved collateral are right-censored workouts—not final losses. Mature-cohort comparisons or survival-style methods can use partial information without pretending every case is closed. If T_obs < T_recovery, tail recovery must be estimated and its assumption disclosed.

Tail materiality is economic, not merely nominal. Discount distant cash and test whether it changes LGD enough to affect decisions. Very small distant recoveries may have limited PV; long legal or collateral tails may remain material and uncertain.

Segmentation should follow recovery economics, not available columns

Potential dimensions include secured status, product, collateral type, default path, exposure size, borrower type, geography and collections strategy. Variables measured at origination X_origination differ from recovery-relevant state at default X_default. For prospective ECL, post-default information unavailable at the reporting date must not leak into the estimate.

In unsecured consumer lending, delinquency stage, payment behaviour, balance, cure history, legal state, collections strategy and time since default can be central. Frequent defaults may create larger empirical samples and faster learning, but only when account-level cash flows and episode logic are reliable. Short contractual tenor does not imply a short LGD horizon: workout can continue long after maturity.

Collections strategy makes LGD partly operational

Early settlement, restructuring, legal action, receivable sale, internal teams and agencies can change amount, timing and cost. If difficult cases systematically receive aggressive treatment, outcomes reflect Borrower Risk + Treatment. Naïve strategy comparisons therefore suffer selection bias. This is a monitoring and experimental-design warning, not an invitation to automate collection decisions.

Default risk and loss severity can deteriorate together

The same conditions that increase default frequency can reduce recoveries: unemployment can weaken borrower payments, house or used-car prices can fall, markets and courts can slow, and collection capacity can become constrained. PD ↑ may coincide with LGD ↑. Assuming independence can understate tail loss.

For IFRS 9, the objective is expected, forward-looking LGD under relevant scenario conditions—not a mechanical import of regulatory downturn LGD. Economically justified drivers may include unemployment, GDP, rates, property or vehicle prices and legal recovery conditions. Scenario-specific ECL followed by probability weighting may preserve joint and nonlinear PD–LGD–EAD behaviour better than averaging each parameter first.

LGD_t,s can vary with expected default timing as balances amortise, collateral values evolve, workout horizons change and macro paths unfold. One static LGD across all future periods is an approximation that needs evidence.

SICR remains primarily a change-in-credit-risk question, not a severity trigger. Yet LGD changes materially affect ECL. In Stage 3 or other credit-impaired exposures, cash-flow and recovery expectations become especially central to measurement; staging and severity should remain conceptually distinct.

Validate recovery level and timing separately

Calibration-in-the-large compares Mean Predicted LGD with Mean Realised LGD on sufficiently mature defaults, but portfolio averages can hide offsetting errors. Analyse segments, default vintages, recovery age and collateral type. If predictions differentiate severity, test whether higher predicted LGD corresponds to worse realised economic loss, with care for censored outcomes.

Backtest CR^pred_t against CR^obs_t, not only final recovery. Drift can arise from borrower mix, macro conditions, collateral markets, legal process, recovery cost or collections operations—even if the PD model remains stable. An improved agency contract may reduce observed LGD because the process changed, not because borrowers became safer.

This is a management decomposition, not a universal accounting identity. It connects LGD to ECL attribution: management should know whether impairment moved because collateral fell, cash slowed, cost rose, mix changed or scenario weights shifted.

Reliable LGD starts with an account-level recovery ledger

The practical data chain is Default Event → EAD → Recovery Cash Flow → Recovery Type → Cost → Collateral → Cure / Re-default → Legal Status → Closure Date. Each ledger event should carry account or facility ID, default date, cash-flow date, amount, type, cost and source.

Borrower, facility and default episode are distinct keys. Explicit episode logic prevents a cured exposure and later re-default from being merged automatically. Ledger versioning, late cash, reversals, post-write-off recovery and sale proceeds should remain reproducible.

LGD feeds EL = PD × LGD × EAD and can change cut-off economics, pricing, collateral policy and collection prioritisation. See Cut-Off Strategy for the decision bridge; governance should keep measurement evidence separate from autonomous action.

Common failure modes are usually architecture failures

An LGD & Recovery Analytics Agent should build evidence—not approve estimates

A future LGD & Recovery Analytics Agent could reconstruct default episodes; ingest recovery cash flows; identify cure and re-default; calculate nominal and discounted recovery; construct curves; analyse collateral and workout cost; estimate segment LGD; test timing; compare forecast with realised cash; detect vintage drift; run approved sensitivities; attribute movement; and prepare validation evidence.

Its bounded role is recovery analytics + LGD monitoring + attribution + validation support. It must not autonomously determine accounting estimates, approve macro assumptions or select collections action. Recurring value comes from continuously evolving cash flows, costs, open workouts and default vintages.

The practitioner logic is: Define Default Episode → Measure EAD → Reconstruct Recoveries → Net Costs → Discount Cash Flows → Segment → Stress / Scenario → Validate Level & Timing → Monitor.

Entimema's Credit Risk capability connects LGD development, validation and portfolio evidence. The CFO Function bridge connects recovery assumptions with allowance, impairment movement and management explanation. Continue through IFRS 9 Expected Credit Loss, Lifetime PD Term Structures, Significant Increase in Credit Risk, Default Definition, Credit Vintage Analysis, Model Calibration Drift, Model Validation and Roll Rate Analysis.