Skip to main content
Separate purchase, refinance and renewal decisions, and distinguish insured from uninsured mortgages. Record amortization, term, fixed or variable rate, lien position and owner-occupied or investment use.

​
Financial-risk measures

​
Public benchmark sources

CMHC reported a 0.24% national mortgage delinquency rate in Q4 2025 for its 90-day-plus measure. This is a dated portfolio observation, not a probability assigned to a new applicant. See the CMHC release. Use CMHC’s delinquency tables for matching geography and reporting period. The AMLOK dataset inventory also contains a US HMDA origination cohort. HMDA origination records do not establish later repayment performance, and a US cohort does not validate a Canadian mortgage model.

​
AMLOK evaluation

Evaluate document extraction for income, liabilities, mortgage terms and property evidence, then independently recompute affordability ratios. Test fixed-rate renewal exposure separately from variable-rate payment changes. Do not apply an immediate rate shock to a contract that cannot yet reprice. For outcome evaluation, prevent the same borrower or property from appearing across training and test partitions. Separate model performance from approval-policy changes, and report results by product, region and vintage. For Signal, score the evidence and lead time of renewal or liquidity signals rather than claiming that an alert prevented a loss. Apply the shared evaluation design and document benchmark protocol.