What it means
A company estimated that some customer invoices would not be collected, and months later, actual write-offs and recoveries reveal how that estimate performed. A doubtful debt reserve backtest compares a dated allowance estimate with later outcomes for a matched receivables population.
The Journal of Accountancy discusses using historical data to assess allowance estimates, and Cornell explains the allowance and write-off accounting distinction, while exact impairment standards depend on the reporting framework and backtesting supports judgment rather than replacing it. Freeze the estimate date, keeping the allowance and supporting assumptions as they stood at period-end, and do not use later evidence to rewrite the old estimate.
Define the cohort as the receivables outstanding at that date by invoice or customer, because new invoices issued afterward do not belong in the same outcome population. Choose the observation window with care, since some customers pay months after the due date and a short window can call a delayed payment a loss while a long one postpones learning.
Track the outcomes separately. Cash collected against the cohort reduces the ultimate uncollected amount and remittances must be matched accurately, while a later price adjustment or billing correction recorded as a credit note may reduce the invoice for reasons unrelated to credit default.
An accounting write-off is evidence but not necessarily the final economic loss if a recovery later arrives, so cash received after a write-off changes the realised net loss and should be linked to the original invoice, and a genuinely disputed debt that resolves through a credit or collection should be classified consistently. Check opening balances and segment the risk.
Customer deposits, unapplied cash and offsets may already cover part of a nominal receivable, so reconcile before estimating loss, and age bands, customer type, geography and credit quality may each have different loss behaviour that a single portfolio average can hide. If the estimated loss was $100,000 and the realised net loss $130,000, the reserve was $30,000 below the later outcome under the chosen window, although even a sound method can miss a rare insolvency, so review systematic bias across several cohorts rather than one result alone.
Separate the model from the process and review changed conditions. A low reserve might reflect bad customer data or delayed dispute logging rather than a faulty percentage formula, and historical rates can be poor guides after a recession, a customer concentration change or altered credit terms.
Under some standards expected credit loss needs current and forward-looking information, so backtesting one historical rate does not prove compliance, and a new collection strategy or write-off threshold can alter observed losses, which should be explained before calling the estimate better. Watch for right censoring, concentration and ledger tie-out.
Some cohort balances remain open at the test date and should be treated as unresolved rather than as zero loss or definite loss without support, while one large customer failure can dominate the variance and should be shown separately. The starting cohort balance minus collections and approved credits should tie to open balances and realised losses under a documented bridge, and a reasonable period-end estimate cannot use facts learned later, so the question is whether the method was sound given the information then available.
In practice
Real-world examples.
Example
A distributor estimated a $100,000 loss on the invoices outstanding at 31 March and later compares it with a $130,000 net loss. Finance finds that a single customer failure explains most of the gap. It records the finding and then tests whether other cohorts show a similar pattern.
Example
A software reseller finds that a $40,000 credit note for incorrect pricing was initially counted as a bad debt in its backtest. It separates this from customer default and reruns the comparison. The corrected loss rate is lower, and the team documents the credit-note rule for future tests.
Example
A manufacturer wrote off a $50,000 balance last year, and the customer later paid $15,000 after a settlement. Under the stated method, the $15,000 is deducted from realised net loss, leaving $35,000. Linking the recovery to the original invoice lets the team see the true outcome rather than the write-off alone.
Formula
Calculation
Backtest difference = realised net credit loss on the original cohort - its dated estimated allowance. The convention here is that a positive result means the allowance was too low. Unresolved accounts require a stated treatment.
Worked example. At quarter-end, Pine Equipment has a cohort of $2,000,000 of invoices and holds an allowance of $100,000, which is 5% of the cohort. Over the test window the cohort produces $150,000 of write-offs and $20,000 of later recoveries, so realised net loss is $150,000 - $20,000 = $130,000, or 6.5% of the cohort. Credit notes of $40,000 for pricing corrections are excluded from the loss because they are not credit default, and $60,000 still unresolved is reported separately.
The backtest difference is $130,000 - $100,000 = +$30,000, so the allowance was $30,000 too low, or 1.5 percentage points of the cohort (6.5% - 5%). One cohort alone does not establish a bias, so the exercise is repeated for later quarters before the percentage is revised.Case study
Seen in the real world.
This entirely fictional example follows Pine Equipment. It reserved $100,000 against a quarter-end invoice cohort. Over the defined test window it recorded $150,000 of write-offs and $20,000 in later recoveries, a $130,000 net loss before other adjustments.
Finance investigated a large customer failure and found that most of the shortfall came from one account with unusually long payment terms. It revised its risk segmentation rather than applying a blanket increase to all customers. The case does not prescribe an IFRS or US GAAP impairment calculation, and finance repeated the test on later quarters before changing the allowance percentages.
Watch out
Common mistakes.
- Comparing one period's allowance with losses on invoices issued after that period.
- Treating every credit note or billing correction as a bad-debt loss.
- Calling open unresolved balances either fully collected or fully lost without evidence.
Questions
People also ask.
Is a write-off the final loss?
Not always. Later recoveries can reduce net realised loss.
What cohort should be tested?
Use receivables outstanding at the estimate date, with a documented observation window.
Does a successful backtest prove compliance?
No. The reporting framework and current forward-looking evidence still need review.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%