Secondary marketing, end to end

What secondary marketing does

Secondary marketing is the function inside a lender that converts closed loans into capital. The desk hedges interest-rate risk during the pipeline, executes loan sales to investors (delivery), and manages the economics of the trade — gain on sale, servicing disposition, and the ratio of locks that actually fund (pull-through).

From a software perspective, secondary marketing is mostly about data. Clean lock-time pricing, clean event timestamps, clean investor commitment data. When the substrate is good, hedging math and bid-tape execution become tractable; when it's noisy, every downstream model is.

Mandatory vs. best-effort delivery

The two execution modes:

• Best-effort delivery: the originator commits to attempt delivery; if the loan doesn't close, no penalty. The investor pays a tighter price (less risk for them) but the lender gives up the option premium.
• Mandatory delivery: the originator commits to deliver the closed loan. If they fail (the loan doesn't close), they pay a breakage fee or pair off the position. The investor pays a wider price (more risk for them) and the lender keeps more of the economics.

Mandatory delivery is more economic at scale but requires a hedging program to manage the risk that some locked loans won't close (fallout). Smaller lenders typically run best-effort because the hedging operations don't pay back at lower volumes.

TBA market basics

Most agency-conforming production hedges in the TBA (To-Be-Announced) market — forward contracts on agency MBS pools that aren't yet assembled. TBA mechanics:

• Pool characteristics (issuer, coupon, maturity) are specified at trade; the actual loans are assigned 48 hours before settlement.
• Settlement dates fall on six standard months per year per coupon.
• Liquidity is enormous; the bid-ask spread for actively-traded coupons is tight.
• Hedge ratios convert dollar volume of locks into TBA notional, with adjustments for expected pull-through and pool characteristics (coupon, weighted-average loan age).

Reference: SIFMA publishes good-delivery guidelines for TBA pools (see sifma.org).

Pipeline hedging in practice

The hedge desk's daily question: how much TBA notional should I be short to offset my pipeline rate risk? The math:

1. Aggregate the pipeline: locked-not-funded loans by program, rate, and lock period.
2. Assign a coupon to each loan's rate (rate-to-coupon mapping based on agency program rules).
3. Apply a pull-through estimate per loan or per cohort. Borderline loans have lower pull-through; recently-locked loans have higher.
4. Sum the pull-through-adjusted notional by coupon.
5. Compare to current TBA position; rebalance to target ratio.

The data substrate matters. If lock events arrive with delays, if cancellations don't flow through, if rate-to-coupon mappings depend on per-program logic that lives in a spreadsheet, the hedge is stale. Event-driven architectures with stable correlationIds and documented schemas remove most of the friction.

Pull-through analysis

Pull-through is the ratio of locked loans that actually fund. It varies by:

• Lock period (longer locks have lower pull-through).
• Borrower stage in the application (pre-approved vs. fully underwritten).
• Rate environment (rising rates lift pull-through; falling rates depress it as borrowers shop for better rates).
• Program (purchase vs. refinance; purchases pull through more reliably).
• Lender's operational throughput.

A pull-through model needs a clean event stream: every lock state transition, every rate change in the borrower's application, every funding event. The platform-side discipline is to emit those events with stable correlationIds so the hedge model can join them deterministically.

Bid-tape execution

For lenders that sell loans servicing-released to aggregator investors, bid-tape execution is the periodic (weekly or monthly) sale of a batch of closed loans. The flow:

1. Identify loans eligible for sale (closed, post-funding, not yet sold).
2. Build a tape: per-loan attributes that investors price against.
3. Distribute to investor pack; receive bids.
4. Choose execution: typically best-bid-wins for each loan.
5. Record settlement with each investor.

The hard part is reconciliation. The tape sent should match the loans sold should match the cash received. Software-side: every sell-side execution writes an audit row keyed by lockId + sellSideId + executingInvestor. Reconciliation against actual settlement is a deterministic join, not fuzzy matching on timestamps.

Sell-side pricing alongside loan-level

The integrated platform approach is to run sell-side pricing on the same engine as loan-level. Same loan facets, different investor pack, same trace mechanism. The desk sees the bid-tape outcome alongside the retail commitment without opening a second tool. When sell-side spreads tighten or widen, the desk adjusts loan-level pricing margins to reflect the change in execution economics.

Data infrastructure for secondary

What the desk needs from the platform:

• Event-driven primary feed. Pricing decisions, lock transitions, sell-side completions on a streaming bus with configured retention.
• Stable correlationId. Joins across pricing → lock → fund → sell-side become trivial.
• Webhook delivery semantics. HMAC-signed, retried, DLQ-replayable. Never silently dropped.
• Replay endpoint. For every disputed quote or settlement, reproduce the price as it was at lock.
• Hash-chained audit. The chain proves nothing has been altered post-write.
• Documented schemas. Stable, versioned, downstream- friendly. The hedge stack and the BI stack should not have to chase schema changes.

None of this is glamorous. All of it makes the desk's job tractable.