Case Study 2 — When the Pipeline Moves: Hedging and Pull-Through Under Stress

DataField.Dev

Case Study 2 — When the Pipeline Moves: Hedging and Pull-Through Under Stress

Case study 1 looked at pricing from the top down: what the agencies charge and who decides. This one looks at it from inside the lender, on the days when the assumptions in §29.9 stop holding.

It is the complementary angle because it is the failure side. Case study 1 ended with a proposed pricing input being withdrawn before it did any damage. This one is about a risk that is managed competently, continuously, by people who are good at it — and that still produces losses, still produces the extension fee your borrower is annoyed about, and twice in recent memory has produced genuine distress at real companies.

Background: the promise a lender cannot keep by itself

Recall the position a lender is in the moment a lock is taken. On day 12 of the Linden Street file, the lender promised 6.625% for thirty days on a loan that did not exist. It has sold something it does not own, at a price fixed today, deliverable at an uncertain date, in a quantity it can only estimate — because it does not know which of its locked loans will actually close.

Three separate unknowns sit in that sentence:

Price risk. If rates rise, the loan the lender is committed to make is worth less than the price it promised.
Timing risk. The loan might fund in three weeks or in seven, and the hedge has a settlement date.
Quantity risk — fallout. Some share of locked loans never funds at all.

Hedging addresses the first. It does not address the third, and the third is where the money is lost.

Episode one: March 2020, when a good hedge became a liquidity problem

A real, public event. In March 2020, as the pandemic disrupted financial markets, the Federal Reserve announced and then rapidly expanded large-scale purchases of agency mortgage-backed securities. MBS prices rose sharply and moved with unusual violence over a matter of days.

Consider what that does to a correctly hedged mortgage lender.

The lender is short TBA securities — that is the hedge. When MBS prices rise, a short position loses money, and those positions are margined: the counterparty calls for cash, promptly, in the amount of the mark-to-market move.

The offsetting gain is real. The lender's locked pipeline, full of loans at yesterday's higher rates, became more valuable at the same moment. But that gain is not cash. It is unrealized, sitting in loans that have not closed, and it converts to cash only when those loans fund and are sold — weeks later.

So the lender is economically hedged and simultaneously out of money. Cash goes out today against a gain that arrives next month.

THE MARCH 2020 SHAPE                            [structural; not a specific company]

   MBS PRICES RISE SHARPLY
        │
        ├── HEDGE (short TBA) loses ────► MARGIN CALL ────► CASH OUT, TODAY
        │
        └── PIPELINE (locked loans) gains ──► unrealized ──► CASH IN, IN WEEKS
                                                                  │
   ────────────────────────────────────────────────────────────────
   Economically flat. Operationally, a liquidity event.

This mismatch was raised publicly and urgently by the industry at the time — the Mortgage Bankers Association and others warned that margin calls on hedge positions posed a liquidity threat to independent mortgage banks. The period also produced separate but contemporaneous liquidity pressure from a different direction: servicers were obligated to advance payments on loans entering forbearance, and Ginnie Mae established a pass-through assistance facility for issuers while FHFA announced a four-month limit on the enterprises' servicer advance obligation for loans in forbearance. Those measures addressed servicing advances rather than hedge margin, but they are part of the same story about how quickly a nonbank lender's cash position can become the binding constraint. (Verify the specifics of each program and its dates at the agencies; the details matter and are on the record.)

The transferable lesson is one sentence: hedging converts market risk into liquidity risk. It does not make risk disappear. And a loan officer who understands that will never again be surprised that the lock desk cares intensely about the size and composition of the pipeline, not only its price.

Episode two: 2022, when pull-through stopped meaning what it used to

A real, public event. Over the course of 2022, mortgage rates rose faster than at any point in the modern history of Freddie Mac's Primary Mortgage Market Survey. The average 30-year fixed rate began the year near 3% and exceeded 7% by late October. Origination volume fell sharply, the refinance share of the market collapsed, and the industry went through a period of consolidation, capacity reduction, and closures. (Verify the series and the figures at Freddie Mac; the survey is public and weekly.)

For a secondary desk, that year did something specific and instructive to pull-through models. A model calibrated during the 2020–2021 refinance boom had learned a pattern:

refinance locks are plentiful,
borrowers renegotiate constantly because rates keep improving,
and pull-through is chronically low because there is always a better quote next week.

The 2022 market inverted every clause. Refinance locks nearly vanished. Rates rose almost monotonically, so a borrower with a lock held something worth more every week they held it — and pull-through on the remaining pipeline went up, not down. A model trained on the wrong regime under-hedged into a falling market, which is the expensive direction.

The structural point for a loan officer: pull-through is not a company-wide constant. It varies by purpose (purchase files pull through far better than refinances, which is a large part of why lenders prize purchase business — Chapter 37), by lock age, by how far the market has moved since the lock, by product, by branch, and by originator.

The arithmetic, on a composite

[Composite — constructed from documented industry patterns. Not a specific company. The dollar figures are illustrative.]

A mid-size independent mortgage bank carries a \$400,000,000 locked pipeline. Its model says expected pull-through is 82%, so the desk sells \$328,000,000 of TBAs forward.

Scenario A — a sharp rally. Over three weeks rates fall and MBS prices rise 1.500. Borrowers who locked at the old rate call to renegotiate or move to a competitor, and pull-through comes in at 66% — only \$264,000,000 funds.

The desk is short \$64,000,000 more than it needed, and must buy it back after a 1.500 move:

$$\$328{,}000{,}000 - \$264{,}000{,}000 = \$64{,}000{,}000$$ $$\$64{,}000{,}000 \times 0.01500 = \$960{,}000 \text{ of loss on the over-hedge}$$

— before any renegotiation the lender grants to keep the borrowers it still has.

Scenario B — a sharp selloff. Same pipeline, same hedge. MBS prices fall 1.500 instead. Nobody leaves a lock they are winning on, and pull-through comes in at 94% — \$376,000,000 funds:

$$\$376{,}000{,}000 - \$328{,}000{,}000 = \$48{,}000{,}000 \text{ unhedged}$$ $$\$48{,}000{,}000 \times 0.01500 = \$720{,}000 \text{ of loss on the under-hedge}$$

Note the shape rather than the dollars. The desk was wrong in both directions and lost money both times, and in neither scenario did it make a bad call about rates. It made a call about people, and people changed their minds in exactly the way that hurt.

Why this lands on your desk

Everything a lock desk does that seems obstructive is a control on the quantity risk above.

Extension fees. The 15-day extension on the Linden Street file cost 0.250 point — \$914.38 — which the lender absorbed. That was a choice, and somebody in secondary marketing approved it. The budget it came out of is the same 0.250 the price waterfall reserves on every loan.

Worst-case repricing on an expired lock. If a borrower could let a lock lapse and then take the better of the old rate and the new one, they would hold a free option. Nobody sells free options.

"Is this file real?" Whether there is an executed contract, whether the appraisal is ordered, whether the borrower has actually decided — these are pull-through questions, and the desk asks them because the answer determines how much to hedge.

Your personal pull-through is measured. Here is the composite that makes it concrete. Suppose you lock 40 files a month at an average loan amount of \$365,750** — **\$14,630,000 of locked volume. Compare a 92% pull-through with a 70% pull-through:

$$\$14{,}630{,}000 \times (0.92 - 0.70) = \$3{,}218{,}600$$

of hedged-but-unfunded exposure every month. If the market moves half a point against that exposure:

$$\$3{,}218{,}600 \times 0.00500 = \$16{,}093.00$$

per month, from one originator, on a difference in habits rather than skill. (Composite; illustrative.)

Nobody will send you an invoice for it. It will arrive as slower exception approvals, less flexibility on extensions when your file is the one that needs it, and a manager who is unusually interested in whether your locks have contracts attached.

What this shows

One. A hedge is not a guarantee; it is a trade of one risk for another. Market risk becomes liquidity risk and model risk. Every mitigation in finance has this property, and it is worth recognizing early in a career.

Two. Fallout is a behavioral risk, and behavior is what models predict worst. Score and LTV are facts. Pull-through is a forecast about whether a stranger will change their mind, made in the middle of the market move most likely to change it.

Three. The loan officer is the sensor. You are the only person in the chain who knows whether the borrower is committed, whether the agent is real, whether the contract is solid. Locking a speculative file is not a neutral act; it puts a hedge on a loan that does not exist against a borrower who has not decided.

Four. Purchase business is worth more than refinance business at the same margin. It pulls through more reliably, so it costs the desk less to hedge. That is a capital-markets fact with a direct career consequence, and Chapter 37 turns it into a business strategy.

Discussion questions

Explain in your own words why a correctly hedged lender can face a cash crisis. Then say which of the two risks — market or liquidity — a small lender should be more afraid of, and why.
In the composite, the desk lost money in both scenarios without making a bad call on rates. Identify the single estimate that failed in each case, and propose two ways to make it more accurate that do not involve predicting the market.
The 2022 cycle broke pull-through models trained on 2020–2021. Name three other assumptions embedded in mortgage operations that a regime change of that size would invalidate, and say how you would notice.
Your borrower's lock expires in four days and the file is not clear to close. Using this case study, describe what the lock desk is actually weighing when it decides whether to grant an extension, and what you can put in the request that makes "yes" easier.
A colleague locks every pre-approval "so the borrower is protected," including files with no accepted offer. Write what you would say to them. Your answer must include a number.

Sources and verification. Federal Reserve announcements of agency MBS purchases, March 2020; Mortgage Bankers Association public statements on hedge margin calls and nonbank liquidity, spring 2020; Ginnie Mae Pass-Through Assistance Program and FHFA's announcement limiting the enterprises' servicer advance obligation for loans in forbearance, 2020; Freddie Mac Primary Mortgage Market Survey, 2021–2023. All figures in the composite are constructed. Verify every real-world figure at the source before quoting it.