Jumbo Non-QM: Where the Banks' Pristine-File Machine Breaks Down

Key takeaways
- Every major Non-QM doc type — bank statement, DSCR, asset depletion — extends to jumbo and super-jumbo sizes.
- LTV steps down as loan size steps up: 80% at $1.5M can be 65% at $4M.
- Reserves are the silent killer: 6–24 months depending on size — run that math first.
- Above ~$2–3M expect a second appraisal; build it into the timeline, not the apology.
The bank's jumbo desk and the actual jumbo buyer have quietly diverged. Banks built their private-client machine for pristine W-2 files; the people buying $2M+ homes are increasingly founders, athletes, and investors whose income arrives in lumps and whose files carry a story. Non-QM jumbo exists for exactly that gap.
Same doc types, bigger grid
The Non-QM doc menu carries straight up the loan-size ladder: 24 months of business bank statements on a $2.8M purchase, DSCR on a $3M rental, asset depletion on a $4M primary for a post-exit founder. What changes is the grid around it — LTV tiers step down with size, reserve requirements climb, and collateral review deepens. A quote built off the $1.5M row of the matrix will be wrong at $4M every time.
The three checks to run before quoting
- Size-tier LTV: find the row for the actual loan amount, not the product's headline max.
- Reserve math: months of PITIA at this payment size is real money — retirement-account haircuts make the gap worse.
- Collateral story: unique properties (acreage, log construction, one-of-one architecture) need a comp narrative before appraisal, not after.
Why the premium is easy to defend
The Non-QM jumbo borrower isn't rate-shopping against the bank — the bank already said no, or said yes contingent on moving $2M in assets to their platform. Position the loan on total cost of the alternative: dead deals, asset-transfer strings, or ninety-day private-client timelines. Flexibility, at this size, is the product. The Altyverse desk holds current jumbo grids from multiple investors — run the scenario before the listing agent asks if your buyer is real.