The 12-Month Window: What a Year of Bank Transactions Reveals That a Credit Pull Can't
Income stability, negative balance events and unreported obligations sit in a year of transactions. What the 12-month window holds, and what isn't proven.
Alfred BEditorial Reviews
A credit pull tells you how someone has handled formal credit obligations. A year of bank transactions tells you what their money actually did. The second view contains things the first structurally cannot hold: how stable income is month to month, how often the account runs dry, what obligations are being paid that no lender ever reported.
Twelve months is the window lenders doing cash-flow underwriting typically use. FinRegLab's July 2019 research found participants evaluating cash-flow data extending back by as much as 12 months. Worth saying immediately: that is documented practice, not a tested optimum. No published research establishes that twelve months outperforms six or three.
What follows is what sits inside that window, and why each signal resists being seen any other way.
What does a transaction history actually contain?
FinRegLab's July 2019 work described the specific metrics participants derived from cash-flow data: income-to-expense ratios, differences in flows of fixed and variable income, minimum balances, and the frequency of negative balance events.
Those four categories cover most of the practical value, and each answers a question the credit file does not ask.
Income stability, not income
A credit application captures income as a number. A transaction history captures it as a pattern, and the pattern is often the more decision-relevant fact.
The JPMorganChase Institute, analyzing de-identified account data in September 2025, found that hourly workers' typical month-to-month earnings change was 9%, and that in one month out of four the swing was 21% or greater. It also observed that those swings were frequently larger than the worker's checking account balance, meaning the volatility was not absorbable from savings.
Two applicants can report the same annual income while one receives it in 26 near-identical deposits and the other in irregular amounts from three sources. A stated figure cannot distinguish them. Deposit history does, without asking anyone to explain themselves.
This matters more in Canada each year. Statistics Canada reported in March 2024 that 871,000 Canadians did gig work as their main job, and that 468,000 earned income through a digital platform in the previous 12 months. Paystub-shaped verification does not describe those workers well.
Negative balance events
Overdraft and NSF activity is among the most direct behavioural signals in an account, and it is invisible to a credit pull unless it escalates into something reportable.
FinRegLab's June 2025 research on small business lending found NSF transactions and low or negative ending balances associated with higher default risk. That study used a three to six month window on business accounts, so it should not be blended with the twelve-month consumer practice above; the two describe different populations.
On the consumer side, the US Consumer Financial Protection Bureau reported in December 2023 that 26.5% of consumers lived in a household charged an overdraft or NSF fee in the prior year, and that households paying frequent fees had lower credit scores and more delinquent debt. That is US data. No equivalent Canadian regulator figure has been published.
The useful property of an NSF string is timing. It appears in an account immediately, while it appears in a credit file only after something defaults, if at all.
Obligations nobody reported
Not every recurring payment reaches a bureau. Rent frequently does not. Neither do many private arrangements, some subscription financing, and a range of informal obligations.
In a transaction history these appear as what they are: regular outflows on a schedule. For debt-service calculations built on reported obligations only, that gap is a systematic underestimate, and it runs in the direction that flatters the applicant.
What each source can see, on the dimensions that decide a file:
| Signal | Credit file | 12-month transaction history |
|---|---|---|
| Formal repayment history | Yes | Only where paid from the account |
| Income amount | As stated on the application | As deposited |
| Income stability | No | Yes |
| Negative balance events | Only if escalated | Immediately |
| Unreported obligations | No | Yes, as recurring outflows |
| Recency | Last reporting cycle | Through yesterday |
Why twelve months rather than three
The honest answer is that the practice is better established than the evidence for it.
Twelve months captures a full seasonal cycle, which matters for anyone whose income moves with weather, school terms or holiday trade. It also gives enough observations to distinguish a bad month from a pattern, which three months cannot reliably do. Both arguments are sound reasoning rather than measured results.
What has not been published, as far as we can find, is any controlled comparison of predictive lift at twelve months against six or three. If that research exists we have not seen it, and anyone claiming a specific improvement from a longer window should be asked where the number came from.
What we still don't know
Three gaps worth naming, since this area attracts confident numbers. There is no Canadian regulator figure for consumer NSF or overdraft incidence. There is no published measure of how often transaction review surfaces obligations absent from a credit file. And there is no established optimum for window length.
Those are all measurable. They have not been measured publicly.
Common questions
Why do lenders look at 12 months of bank transactions?
FinRegLab's 2019 research found participants evaluating cash-flow data extending back as much as 12 months. It is established practice; no published research establishes twelve months as an optimum.
What can bank transactions show that a credit check cannot?
Income stability rather than stated income, negative balance events as they happen, and recurring obligations that were never reported to a bureau.
How volatile is income for hourly workers?
The JPMorganChase Institute found in September 2025 that typical month-to-month earnings change was 9%, with swings of 21% or more in one month out of four.
Are NSF events predictive of default?
FinRegLab's June 2025 small business research associated NSF transactions and low or negative ending balances with higher default risk. That finding is from business accounts over a three to six month window.
How many Canadians do gig work?
Statistics Canada reported in March 2024 that 871,000 Canadians did gig work as their main job.
Carousel's intake flows return categorized transaction data ready for income recognition and obligation detection. See how verification fits your flow
