The Same Loan, Five Business Models: How Mortgage Institutions Actually Differ
Two lenders can originate an identical loan to an identical borrower and be running fundamentally different businesses. The loan is the same. What differs is where the money comes from, how long the institution holds the asset, what it keeps afterwards, and therefore which risks it is actually managing.
Those differences explain most of what looks inconsistent about the industry, including why lenders that appear to be competitors are frequently each other's customers.
The four decisions that define a lending model
Every mortgage institution has answered four questions, and the answers together determine the business.
Where do the loans come from? Originated directly, taken from brokers, or purchased already closed.
Where does the funding come from? Deposits, or borrowed warehouse lines that must be repaid quickly.
Is the loan held or sold? Kept on the balance sheet for interest income, or sold for gain on sale.
Is the servicing kept or released? Retained as a long-lived asset and customer relationship, or sold with the loan.
Almost every structural difference between mortgage institutions reduces to a combination of these four.
The three origination channels
Retail. The institution originates directly through its own loan officers. Maximum control over file quality, highest cost per loan, and the quality problem is an internal training and process problem.
Wholesale. The institution funds loans originated by independent brokers. Lower origination cost and rapid scaling, but file quality is determined before the file arrives. The quality problem becomes a counterparty problem: which brokers submit clean files, and how is that measured.
Correspondent. The institution purchases loans already closed and funded by another lender. The least control of all: the loan is made, the borrower is unavailable, and the review window sits between purchase and downstream sale. The quality problem is a due diligence problem under time pressure.
Most substantial lenders run more than one channel, which creates the requirement to demonstrate that one standard was applied across all of them.
The institution types
Independent mortgage banks (IMBs). Non-depository lenders funding through warehouse lines and selling loans quickly. No deposit base, so the warehouse line has to be cleared, which makes saleability existential rather than merely desirable. A loan that cannot be sold because of a defect occupies a credit line that is needed for the next loan. IMBs are also the most exposed to volume cyclicality, since they have no other business to carry fixed costs.
Depository banks. Fund from deposits and can hold loans on the balance sheet. This permits portfolio lending outside agency guidelines, judged on the bank's own credit view. Subject to prudential supervision in addition to consumer compliance, and their retail cost to originate runs materially higher: averaging $16,320 per loan in 2025 against a broader industry figure of $11,109 in Q3. [1]
Credit unions. Member-owned and not-for-profit, regulated by the NCUA. Frequently portfolio lenders, competing on relationship and rate rather than scale, and typically operating without a large QC department.
State housing finance agencies (HFAs). Administer affordable housing programmes alongside lending: down payment assistance, income and purchase price limits, recapture provisions, compliance periods. Programme eligibility is a second body of requirements sitting on top of loan quality, and it is checked against documents gathered for underwriting rather than for programme compliance.
Non-bank aggregators. Purchase loans from smaller originators, pool them, and sell onward. Their entire risk position consists of loans they did not originate and cannot fix.
Agency, government and non-agency
Cutting across institution type is the question of where a loan can be sold.
Conforming / agency. Loans meeting Fannie Mae or Freddie Mac requirements, sold against representations and warranties about those requirements. The deepest and most liquid outlet.
Government. FHA, VA and USDA loans, insured or guaranteed by the relevant agency, securitised through Ginnie Mae. Additional programme requirements and layered documentation logic: particularly for 203(k), manual underwrites and streamline refinances.
Jumbo and portfolio. Loans exceeding conforming limits, or written to the institution's own credit standards and held.
Non-QM. Loans outside the Qualified Mortgage definition, generally securitised privately. Roughly one in ten US mortgages now falls outside QM. [3]
The outlet determines the rule set. A single institution running agency, government and non-QM production is operating three distinct requirement regimes simultaneously, which is why programme identification is a prerequisite for review rather than a detail of it.
Servicing: the decision that outlasts the loan
Servicing retained or released is the least visible decision and the longest-lived.
Retained creates an asset with value moving inversely to prepayment, and keeps the customer relationship. It also means the institution lives with the loan's performance and its documentation quality for years.
Released converts the servicing to immediate cash and ends the relationship at sale.
The consequence for quality is direct: an institution retaining servicing experiences its own origination defects as servicing problems later. One releasing servicing experiences them as repurchase demands. Same defect, different department, different lag, different visibility to whoever decided how much review to fund.
Why the model determines the quality problem
The useful conclusion is that "how should we do quality control" has no general answer, because the risk being controlled differs by model.
- A retail IMB controls file quality directly and is existentially exposed to saleability.
- A wholesale lender cannot control the file and must control counterparties instead, which
requires measurement by broker rather than in aggregate.
- A correspondent aggregator cannot control either, and is left with detection inside a short
window on loans it has already paid for.
- A portfolio depository carries credit risk rather than repurchase risk, so documentation
defects matter less and credit assessment quality matters more.
- An HFA carries programme compliance risk that has no equivalent elsewhere.
A quality control programme designed for one of these applied to another will be well run and aimed at the wrong thing.
Sources
- Mortgage Bankers Association, Quarterly Mortgage Bankers Performance Report Q3 2025, total loan production expenses per loan ($11,109); and MBA production-channel data for 2025, retail channel, depository institutions ($16,320). Both are total cost to produce one loan, not QC cost. Figures are quoted from MBA's published summaries; confirm against the current release before using them in a business case.
- Fannie Mae Selling Guide, D2-1-04, Identifying and Remedying Origination Defects Under the Remedies Framework. Sets out permitted corrections and the conditions under which additional documentation covering the underwriting period may resolve a defect.
- Stacker analysis, One in ten US mortgages now falls outside the Qualified Mortgage standard, August 2026. A secondary analysis; the underlying HMDA-derived figures have not been independently reproduced here.