The expected credit loss model under Ind AS 109 replaced the incurred loss model with a forward-looking approach: companies must recognise credit losses they expect to occur, not merely losses that have already occurred. The intention was to prevent delayed recognition of credit deterioration.
The consequence is that the most significant provision on many companies’ balance sheets, the ECL provision against receivables, loans and other financial assets, is now driven not by observable defaults but by management’s estimates of future defaults.
The estimate requires: a probability of default for each exposure (or portfolio of exposures), a loss given default assumption, an exposure at default calculation, forward-looking macroeconomic information and staging decisions that determine whether the exposure is in Stage 1 (12-month ECL), Stage 2 (lifetime ECL, not credit-impaired) or Stage 3 (lifetime ECL, credit-impaired).
Each of these is a judgment. And each judgment can produce a materially different provision.
Where the Model Creates Risk
PD estimation
The probability of default can be estimated from historical data (what percentage of similar exposures defaulted in prior periods?), from market data (what does the credit market imply about default probability?) or from statistical models (what does a regression against financial and macroeconomic variables predict?).
Each approach produces a different PD. The historical approach may understate PD if current conditions are worse than the historical period. The market approach may overstate PD during periods of market stress. The model approach depends entirely on which variables are included and how they are calibrated.
A management team that selects the approach producing the lowest PD produces the lowest ECL and the highest reported profit.
LGD assumption
Loss given default depends on collateral value, recovery process effectiveness and workout experience. A company that assumes 80% recovery on secured exposures produces a 20% LGD. A company that assumes 60% recovery produces a 40% LGD.
For a loan book of Rs 500 crore with a PD of 3%, the ECL at 20% LGD is Rs 3 crore. At 40% LGD, it is Rs 6 crore. The Rs 3 crore difference flows directly to reported profit.
Forward-looking information
Ind AS 109 requires the incorporation of forward-looking information: macroeconomic scenarios (base, optimistic, adverse) and their probability weights. The provision depends on which scenarios are modelled, what economic assumptions each scenario contains and what probability is assigned to each.
A company that assigns 60% probability to the base case, 25% to the optimistic case and 15% to the adverse case will produce a lower ECL than one that assigns 50/20/30. The probability weights are management’s judgment about the future economic environment, and they are not verifiable until the future arrives.
Staging
An exposure moves from Stage 1 (12-month ECL) to Stage 2 (lifetime ECL) when there has been a “significant increase in credit risk” since inception. The definition of “significant” is management’s judgment.
A company that defines significant increase narrowly (default must be imminent) keeps more exposures in Stage 1 and recognises lower ECL. A company that defines it broadly (any material deterioration in creditworthiness) moves exposures to Stage 2 earlier and recognises higher ECL.
The staging threshold, set by management, determines the provision. And the provision, which can represent the largest single estimate on the balance sheet, is therefore controlled by a threshold that management defined.
Management overlays
When the model output does not reflect management’s assessment of current conditions, management applies an overlay: an adjustment to the model output that increases or decreases the provision based on judgment.
Overlays are legitimate. They are also, by definition, departures from the model’s evidence-based output. An overlay that reduces the provision because management believes conditions are better than the model suggests is a judgment that should be documented, evidenced and scrutinised. In practice, overlays are frequently applied without the documentation rigour that the underlying model receives.
In Northrop Management Private Limited’s forensic and financial reporting work, ECL analysis examines the full chain of judgments: PD methodology, LGD assumptions, forward-looking scenarios, probability weights, staging thresholds and management overlays. The analysis tests whether each judgment is supported by evidence, whether the judgments are internally consistent and whether the pattern of judgments across all dimensions produces a provision that reflects economic reality or reporting preference.
Ashish Chaudhary, frames the analytical caution directly: “ECL models can create false precision when the underlying assumptions are weak, biased or deliberately optimistic. The model produces a number to two decimal places. The inputs are judgments that could credibly produce a range of outcomes spanning 50% to 200% of the reported figure. The board should know the range, not just the point estimate.”
Questions for the Boardroom
- What PD methodology do we use, and how does our estimated PD compare to historical default experience and to market-implied rates?
- What probability weights do we assign to our macroeconomic scenarios, and who determines those weights?
- What is our staging threshold for significant increase in credit risk, and how many exposures would move from Stage 1 to Stage 2 if we lowered the threshold by one notch?
- Have we applied any management overlays to the ECL model output, and if so, what is the documented justification?
- If we re-ran our ECL model using the most conservative credible assumption for each input, how would the provision change?
Closing Implication
The ECL model is a framework for estimating future credit losses. The estimate depends on judgments about default probability, recovery rates, economic scenarios and staging thresholds, every one of which management controls. The precision of the model’s output conceals the imprecision of its inputs. A board that accepts the ECL provision at face value is accepting management’s forecast of credit deterioration as a fact. It is not. It is an estimate, and the estimate should be tested with the same rigour the board applies to any other consequential management projection.