The model was wrong. Now what?
Updated: 11 hours ago
When an expected credit loss model is rebuilt, is the difference a prior-period error or a change in estimate?
Kilbride Distribution Ltd sells building supplies to contractors across Ireland. In 2026 its new financial controller rebuilds the expected credit loss model for trade receivables. The old model was a spreadsheet provision matrix that had grown, tab by tab, over several years. The new one is cleaner, uses more data and gives a bigger number: the loss allowance goes up by €416,000.
The board's first question is where that €416,000 goes. If it is a change in an accounting estimate, it goes through 2026 profit or loss. If it is the correction of a prior-period error, it goes back into 2025: the comparatives are restated and 2026 profit is left alone. Same model, same number, very different results for the year, and a very different conversation with the auditors and the bank.
IAS 8 draws the line, and it draws it on one question: what information was available at the time?
Two definitions, one question
IAS 8.5 defines prior-period errors as omissions from, and misstatements in, the financial statements for one or more prior periods arising from a failure to use, or a misuse of, reliable information that:
was available when the financial statements for those periods were authorised for issue; and
could reasonably be expected to have been obtained and taken into account in preparing them.
Errors include mathematical mistakes, mistakes in applying accounting policies, oversights or misinterpretations of facts, and fraud.
Accounting estimates are monetary amounts in the financial statements that are subject to measurement uncertainty (IAS 8.5). An expected credit loss allowance is a textbook example. An estimate may need to change because the circumstances it was based on change, or because of new information, new developments or more experience. By its nature, such a change does not relate to prior periods and is not the correction of an error (IAS 8.34).
So the test is not whether the old number was wrong. With hindsight, almost every estimate is wrong. The test is whether the old number was wrong given what the company knew, or should have known, when it signed the accounts.
Kilbride's rebuild
At 31 December 2025 Kilbride had trade receivables of €8,000,000 and reported a loss allowance of €164,000. While rebuilding the model, the controller finds three things.
First, a broken lookup. The old matrix was meant to apply a 60% loss rate to balances more than 90 days past due, the rate Kilbride's own history supported. When the spreadsheet was extended in early 2025, a lookup started pointing at the wrong row and applied 1% instead. The 90+ bucket was €400,000, so the allowance was understated by €236,000 (€400,000 × 59%). The right rate was sitting in the same spreadsheet. That is a misuse of reliable information that was available: an error.
Second, new data. During 2026 several contractors failed and Kilbride's loss experience worsened. The new model reflects these defaults and a weaker outlook for construction, which adds €180,000. None of this was known, or knowable, when the 2025 accounts were signed: a change in estimate.
Third, a new method. The new model estimates probability of default and loss given default for each customer segment rather than applying one matrix. A change in a measurement technique is a change in accounting estimate unless it results from the correction of a prior-period error (IAS 8.34A). Kilbride's switch is a refinement, not a fix, so the method creates no error of its own. Its effect is already inside the €180,000.
€236,000 plus €180,000 makes the €416,000. The rebuild is not one thing or the other. It is both, and IAS 8 requires the two parts to be accounted for separately.
What the restatement looks like
The €236,000 is corrected retrospectively: Kilbride restates the 2025 comparatives as if the error had never occurred (IAS 8.42). The example assumes a 12.5% tax rate and that the tax effect of the correction is recognised as a deferred tax asset.
2025 (€) | As reported | Restated |
Loss allowance on trade receivables | (164,000) | (400,000) |
Trade receivables, net | 7,836,000 | 7,600,000 |
Deferred tax asset on the correction (12.5%) | – | 29,500 |
Impairment loss on trade receivables | (61,000) | (297,000) |
Profit for the year | 1,450,000 | 1,243,500 |
Retained earnings at 31 December 2025 | 6,200,000 | 5,993,500 |
The correction stays out of 2026 profit or loss: IAS 8.46 excludes the correction of a prior-period error from the profit or loss of the period in which it is discovered. The €180,000 change in estimate goes through 2026 impairment losses, prospectively, as IAS 8.36 requires.
Because the lookup broke during 2025, the statement of financial position at 1 January 2025 is unaffected and no third balance sheet is needed. Had the lookup been wrong since 2023, the opening balances at 1 January 2025 would be restated too and, if the effect were material, IAS 1.40A would require a third statement of financial position at that date.
Where the line falls: common ECL findings
What the rebuild finds | Classification | Accounting |
A formula or lookup applied the wrong loss rate, or left a group of customers out | Prior-period error | Restate the comparatives (IAS 8.42) |
The old model ignored forward-looking information the company had at the time | Prior-period error: IFRS 9.5.5.17 was not applied | Restate the comparatives |
Default data or customer information that arrived after the prior-year accounts were signed | Change in estimate | Current-year profit or loss (IAS 8.36) |
A worse economic outlook, or a change in the forward-looking scenarios | Change in estimate | Current-year profit or loss |
Moving from a provision matrix to probability of default × loss given default | Change in estimate (IAS 8.34A), unless it corrects an error | Current-year profit or loss |
A customer that was current at year end failed the following spring | Neither: the outcome differed from the estimate (IAS 8.48) | Current-year profit or loss |
Switching between the simplified and the general approach for receivables | Change in accounting policy (IFRS 9.5.5.15) | Retrospective application (IAS 8.19–22) |
The last two rows are where people slip. A customer that was current at the year end and failed in March is not evidence that the allowance was wrong. Estimates are approximations that change as information becomes known, and an outcome that differs from the estimate is not an error (IAS 8.48). And the choice between the simplified approach and the general approach for trade receivables is an accounting policy choice under IFRS 9.5.5.15, so changing it is a change in accounting policy, not an estimate.
Five tests for the hard cases
Was the information available? Look at what existed when the prior-year financial statements were authorised for issue, not only at the reporting date. Information that arrived between the year end and the date of authorisation should have been reflected as an adjusting event under IAS 10. If it was missed, that is an error.
Could it reasonably have been obtained? IAS 8.5 does not demand perfection. Information buried in a third party's systems, or that would have taken undue cost or effort to collect, does not turn the old estimate into an error. IFRS 9.5.5.17 uses the same yardstick: reasonable and supportable information available without undue cost or effort.
Is it hindsight? IAS 8.53 prohibits hindsight when correcting a prior period: the restated number uses what was known then, not what is known now. The same discipline works the other way. An estimate does not become an error just because hindsight shows it was too low.
Is it a mix? Most rebuilds are. Split the difference into the part that corrects an error and the part that reflects new information, as Kilbride did, and document how the split was made. The auditor will ask.
Is it material? IAS 8 requires material errors to be corrected retrospectively. Immaterial errors can be corrected in the current period, but not errors made intentionally to achieve a particular presentation: financial statements containing them do not comply with IFRS (IAS 8.41). Judge materiality on the error itself, including qualitative factors such as covenant headroom or a reversal of a trend.
If materiality is the question, our Materiality Calculator sets the thresholds, and The Tipping Point walks through judgements like this one.
What to disclose
For the error (IAS 8.49): the nature of the error; for each prior period presented, to the extent practicable, the amount of the correction for each line item affected and for basic and diluted earnings per share; the amount of the correction at the beginning of the earliest prior period presented; and, if retrospective restatement is impracticable, the circumstances and how the error has been corrected. The EPS Calculator handles the restated earnings per share.
For the change in estimate (IAS 8.39–40): the nature and amount of the change that affects the current period or is expected to affect future periods, or the fact that the future effect cannot practicably be estimated. For credit losses, IFRS 7.35G(c) also requires an explanation of changes in estimation techniques or significant assumptions during the period and the reasons for them.
Columns headed "restated", a clear note and a reconciliation of the loss allowance usually tell the story. The Disclosure Review tool checks the note against the requirements.
The audit angle
ISA 540 (Revised) requires the auditor to review the outcome of previous accounting estimates (ISA 540.14). That review is not meant to second-guess judgements that were reasonable on the information available at the time. It tests the estimation process. When it uncovers something like Kilbride's lookup, the questions follow quickly: was the information available, why was it missed, is it material, and what does it say about the controls over the model? A broken model usually means a control deficiency to communicate under ISA 265.
Our Audit Risk Matrix and Audit Procedure Builder help plan the work on the expected credit loss estimate.
In short
Ask what was known, and knowable, when the prior-year accounts were signed.
A failure to use, or a misuse of, information that was available is an error: restate the comparatives and keep the correction out of current-year profit.
New information, new developments, more experience and new measurement techniques are changes in estimate: current-year profit or loss.
Split a rebuild into its error part and its estimate part, and document the split.
Switching between the simplified and general approaches is a change in accounting policy.
Tools and templates
Our Prior-Period Errors template does the restatement in Excel. It logs each error with its tax effect, produces the restated statements, the third statement of financial position when one is needed, the retained earnings and restated EPS, the IAS 8.49 note and the opening journal. Corravale is its worked example.
For the expected credit loss model itself:
ECL Calculator template: a provision matrix for trade receivables in Excel, with forward-looking adjustments.
IFRS 9 ECL Calculator: the same calculation online.
Intercompany ECL template: expected credit losses on loans to group companies.
Kilbride Distribution Ltd is fictional. References are to IAS 8 Accounting Policies, Changes in Accounting Estimates and Errors as amended by Definition of Accounting Estimates (effective 2023). From 2027 IFRS 18 replaces IAS 1 and renames IAS 8 Basis of Preparation of Financial Statements, without changing the requirements on errors and estimates discussed here.
