Every credit sale creates a receivable, and every receivable carries some chance it never becomes cash. Accounting handles that with the allowance, the contra-asset you met when building the balance sheet. It reduces gross receivables to the amount you actually expect to collect:
Net trade receivables = Gross trade receivables − Allowance for expected credit losses
Three entries run the whole machine. Learn them cold:
| Event | Entry | P&L effect? |
|---|---|---|
| Raise/adjust the allowance | Dr Impairment loss (P&L) · Cr Allowance | Yes — the provision charge |
| Write off a dead receivable | Dr Allowance · Cr Trade receivables | No (already expensed) |
| Recover a written-off amount | Dr Cash · Cr Impairment loss (or reinstate then collect) | Yes — credit/gain |
Analysts exploit two consequences. First, a write-off doesn't touch the P&L if it was provided for, so a company can bleed for years through provisions and then "clean up" quietly. Second, the provision charge is an estimate that management controls. Under-providing flatters today's profit at tomorrow's expense. That's the IL&FS/DHFL playbook, and you'll take it apart in Phase 2 and Phase 5.
Here's how it used to work. Until 2018-20, provisioning ran on incurred loss. You waited for objective evidence of default, then provided. The 2008 crisis exposed this as "too little, too late", because banks sat on performing-labelled books until the losses arrived all at once. Both regimes then moved to expected losses, by different roads:
Advertisement
How does the three-stage expected credit loss model work?
IFRS 9 / Ind AS 109: the three-stage ECL model (the "general approach"). Every in-scope financial asset (loans, debt investments, receivables, contract assets, lease receivables) falls into one of three stages:
| Stage 1 | Stage 2 | Stage 3 | |
|---|---|---|---|
| Status | Performing | Significant increase in credit risk (SICR) since origination | Credit-impaired (defaulted) |
| Allowance | 12-month ECL (lifetime loss × probability of default within 12 months) | Lifetime ECL | Lifetime ECL |
| Interest income accrues on | Gross carrying amount | Gross carrying amount | Net (gross − allowance) |
| Rebuttable presumption | — | > 30 days past due → SICR | > 90 days past due → default |
The measurement engine, whatever the stage:
ECL = Σ probability-weighted [ PD × LGD × EAD ], discounted at the original effective interest rate
where PD = probability of default (12-month or lifetime), LGD = loss given default (the % lost after collateral/recoveries), and EAD = exposure at default. The estimate has to be probability-weighted across scenarios, and it has to include forward-looking information. If the monsoon fails or rates spike, the model must say so before the defaults arrive.
Try a mini-example (₹ crore, synthetic lender). A Stage 1 book of ₹160 with a 12-month PD of 2% and LGD of 40% gives ECL = 160 × 0.02 × 0.40 = ₹1.28. Now a ₹30 slice slips 30+ days past due. It moves to Stage 2, with a lifetime PD of 15%, so 30 × 0.15 × 0.40 = ₹1.80. Look at what just happened. As a Stage 1 slice it carried only 30 × 0.02 × 0.40 = ₹0.24, so the same asset needs ~7.5× the allowance the day it crosses the SICR line. Then a ₹10 loan defaults with LGD of 60%, giving ₹6.00. The cliff between Stage 1 and Stage 2 explains why bank provisioning is lumpy. It also explains why "SICR criteria" are a fight between managements and auditors.
How do non-financial companies work out ECL?
The simplified approach: what non-financial companies actually do. For trade receivables and contract assets without a significant financing component, IFRS 9/Ind AS 109 skip the staging entirely. They require lifetime ECL, always. In practice that's done with a provision matrix. You bucket receivables by age, then apply loss rates drawn from historical experience and adjusted for the outlook. Here's a worked example (synthetic Sundaram Components Ltd, ₹ crore):
| Ageing bucket | Gross AR | Lifetime loss rate | ECL |
|---|---|---|---|
| Not yet due | 60 | 0.5% | 0.30 |
| 1–30 days past due | 20 | 2% | 0.40 |
| 31–60 days | 10 | 5% | 0.50 |
| 61–90 days | 6 | 15% | 0.90 |
| > 90 days | 4 | 40% | 1.60 |
| Total | 100 | 3.70 |
The required allowance is ₹3.70 crore, which is 3.7% coverage. If the opening allowance is ₹2.90 crore, the year's P&L charge is ₹0.80 crore. Schedule III P&Ls usually put it inside "other expenses"; US filers show it in SG&A/operating expenses. All the judgment is in the loss rates. Nudge the >90-day rate from 40% to 25% and profit rises with no improvement in the business at all. So compare the matrix year over year. Rates drifting down while ageing drifts out is a red flag.
How does US CECL differ from IFRS 9?
US GAAP: CECL (ASC 326, "current expected credit losses", effective 2020). CECL has one measure and no stages. You recognize lifetime expected losses on day one for every in-scope amortized-cost asset: loans, HTM debt, trade receivables, contract assets. Again the inputs are historical + current + reasonable and supportable forecasts. IFRS's simplified approach and CECL end up in nearly the same place for trade receivables. Both are lifetime, and a provision matrix works under both. For lenders they differ materially. A brand-new performing loan carries only 12-month ECL under IFRS 9 but full lifetime ECL under CECL, so CECL front-loads more pain. That front-loading is why US banks' provisions exploded in Q1 2020 the moment COVID forecasts turned. On impaired assets, IFRS moves interest to the net basis in Stage 3, while US practice uses nonaccrual status. Both sides have a live procyclicality debate. Do ECL models amplify cycles by demanding provisions just when capital is scarce? Know the argument, and hold it loosely.
What is different in India?
India specifics. (1) Ind AS 109 applies to corporates and NBFCs/HFCs, but Indian banks still report under Indian GAAP, with RBI's rule-based IRACP provisioning (90-day NPA recognition, prescribed provision percentages). Ind AS for banks has been deferred (as of mid-2026). RBI's directions of April 2026 move banks to expected-credit-loss provisioning from 1 April 2027, with a phased transition, while they keep reporting under Indian GAAP (as of October 2026). Expect a one-time step-up in provisions. So an HDFC Bank annual report and a Bajaj Finance annual report are written in different accounting languages, and you'll need to remember that in Phase 5. (2) NBFCs run ECL with an RBI prudential floor. If IRACP norms produce a higher provision than the ECL model, the shortfall is appropriated to an "Impairment Reserve." (3) Since FY22, Schedule III mandates a trade-receivables ageing table in every Indian annual report. Its buckets are < 6 months, 6 months to 1 year, 1 to 2 years, 2 to 3 years, > 3 years, each split disputed/undisputed and good/doubtful. That's a gift to analysts. Receivables ageing used to take guesswork, and now it's printed. US filers disclose the allowance roll-forward (provision, write-offs, recoveries) in the credit-losses note.
How can receivables data warn you early?
The early-warning habit (Phase 2 preview). Track receivables growth against revenue growth. Days sales outstanding is DSO = Trade receivables ÷ Revenue × 365. When it rises sharply, it can mean stretched customers, channel stuffing, or fictitious sales. Satyam carried fake receivables. Manpasand's receivables ballooned before it collapsed. You'll formalize this later with the liquidity-and-leverage ratios, and again with earnings games. From this week on, though, always glance at the ageing table.