What Aggregate Loan-Tape Metrics Miss in Residential Pool Review

A residential loan tape is a good description of a set of loans and a poor description of a set of properties. Most pool review runs on aggregates computed from that tape — weighted-average leverage, credit bands, geographic mix, vintage spread. Those numbers are accurate. They are also structurally incapable of telling you how any individual property is positioned within its own market, because the tape represents each property with one value and an address.

What a loan tape actually contains

The tape is the loan-level file that accompanies a pool when it is offered for sale, pledged for financing, or contributed to a securitization. It is the shared reference document for everyone assessing the transaction: buyers, financing counterparties, third-party review firms and rating agencies all work from it. A standard definition describes it as the detailed data on the assets securing a facility, used by potential lenders or investors to analyze risk and determine pricing.

Typical fields cluster into four groups:

• Loan terms — original and current balance, rate, term, amortization, maturity, and for transitional products the rehab budget and draw structure.
• Leverage — loan-to-value, loan-to-cost, loan-to-after-repair value, and the valuations those ratios are computed against.
• Borrower attributes — credit score, liquidity, experience, entity structure, and an internal risk grade where the originator assigns one.
• Property — address, property type, occupancy, and a value. Sometimes square footage and year built. Frequently not much else.

Notice the asymmetry. Three of those groups carry many fields describing the obligation and the obligor. The fourth carries an address and a number.

What a candidate pool is, and what review has to answer

A candidate pool is a collection of loans still under evaluation — not yet acquired, financed or securitized. The composition is genuinely open: loans can be excluded, the price can move, the structure can change. As one practitioner description puts it, full visibility into a pool's strengths and weaknesses is what lets a buyer negotiate price, request exclusions, or structure the purchase to reflect the actual risk profile. The stage-by-stage mechanics are in candidate pool assessment.

Review has to answer three questions under time pressure:

• What is in this pool that the headline statistics do not show?
• Where should limited diligence hours go?
• How does this pool compare with the last one on a consistent basis?

How pools are traditionally assessed

Two workstreams run in parallel. Stratification summarizes the tape into distributions — the pool cut by leverage band, credit band, geography, vintage, property type, balance. Loan-level diligence samples or reviews the file itself. The established scope for residential transactions covers four streams: credit, property valuation, regulatory compliance, and data integrity, producing graded findings on the loans reviewed.

This machinery is mature and it works. Regulatory frameworks assume it — FHFA's securitization examination module treats collateral data quality and pool composition as core supervisory concerns. Nothing below suggests replacing any of it.

But both workstreams inherit the tape's asymmetry. Stratification can only cut on fields that exist, and the property fields are thin. Loan-level diligence verifies that the file is accurate and compliant — that the appraisal is supported, the documents are present, the data matches. Both are largely backward-looking and confirmatory. Neither asks whether this particular property is well or poorly positioned within its own local market going forward.

Where property-level differences hide

Geography is not a market

A pool cut by state or MSA looks diversified when its rows span many metros. But properties compete for buyers at a much smaller scale than an MSA. Two homes in the same ZIP code, at similar price points, can be positioned very differently relative to what buyers in that specific submarket want — and they can diverge accordingly over a hold period. When the only geographic field is a state or a metro, none of that is visible. The portfolio-level version of this problem is geographic concentration risk.

A value is a point, not a trajectory

Leverage ratios are computed against a valuation, so the entire property representation on the tape reduces to where the value sits today. Two properties with identical values and identical leverage can still have very different forward paths. That difference does not appear anywhere in a weighted-average LTV.

Averages hide their own tails

A weighted average is the wrong instrument for finding concentration. A pool can carry an unremarkable average while its weakest collateral clusters in one submarket or one origination vintage. Because the aggregate absorbs the cluster, the concentration only becomes visible when those loans resolve — which is the most expensive moment to discover it.

Count-weighted and balance-weighted views disagree

A group that is 15% of a pool by loan count can be a materially different share by balance. A composition view that reports only one basis will understate exposure roughly half the time.

The fifth stream: collateral trajectory

The four established diligence streams — credit, valuation, compliance, data integrity — are all backward-looking or confirmatory. They establish what is true about the loan and the file today. None of them asks where a specific property sits within its own market going forward.

That question is collateral trajectory, and it is the natural fifth stream. Valuation tells you what a property is worth. Trajectory tells you where it sits in its market — and where it is headed relative to it. A trajectory read adds four things to a pool view. None of them is a credit opinion.

Local rank

Each property is scored relative to comparable properties in its own local market, not against a national or regional benchmark. The output is a relative position — where this property sits among its true competitors — rather than a price or a forecast. Ranking within market is what makes rows from different geographies comparable on a consistent basis.

Coverage

Some properties cannot be assessed with sufficient local evidence. Thin transaction history, an unusual property type, or a market with sparse data all produce genuine uncertainty. The honest response is to mark the row unsupported and route it to a person — not to assign a number with invisible error bars. Coverage should be reported as an output of the analysis, because a pool whose unsupported rows cluster in one market is telling you something.

Tails

Property-level signal is generally strongest at the extremes and weakest in the middle. A composition view should say so: a strong supported group, a weak supported group, a middle that stays in standard diligence, and an unsupported group. Presenting the middle as a smooth gradient implies precision that is not there.

Concentration

The operationally useful question is not the pool average but where the weak evidence sits — which markets, which vintages, which property types — and how that looks by balance as well as by count.

What this is not

Property-level collateral intelligence is a narrow input, and the distinctions matter more than the capability:

• It is not underwriting or valuation. It does not appraise a property, establish an after-repair value, or verify a file.
• It is not a default model. Credit risk depends on the borrower, the structure, the servicing, the project and the property. A property signal addresses one of those.
• It is not a security rating. Rating a transaction involves loss distributions, structural analysis, counterparty review and credit enhancement. A property read is an input to collateral analysis, several steps removed.
• It is not an inclusion recommendation. A composition view informs which loans deserve a closer look. Which loans belong in a pool is a decision that stays with the institution.
• Appreciation is not project profitability. For transitional loans especially, a property can appreciate while the project loses money on rehab overruns, carrying costs or timeline.

Where a property-level view is used

Acquisition. A buyer evaluating a tape they did not originate has the least context on sponsor and project, under a bid clock. An independent property read is most valuable exactly there.

Financing. Lenders against pledged residential collateral care about concentration and coverage across the borrowing base — the same composition question from the other side.

Securitization. Contributed pools are stratified and disclosed. A consistent property-level basis makes composition comparable across contributing originators. It is an input to how a pool is described and reviewed — not a substitute for rating analysis.

Portfolio monitoring. On a funded book the question becomes where to direct surveillance. Weak-tail and coverage views prioritize attention before problems surface in performance data.

How to find out whether it adds anything

Every claim above is a hypothesis about a specific portfolio until it is tested on that portfolio. The structure that settles it is a blind historical test, and its value comes from the order of operations — set out step by step in the institutional guide to residential property risk assessment:

• Agree first. Freeze the population, the exclusions, and the success measures before anyone sees a result. Choosing the measure after seeing outcomes is how retrospective analysis fools everyone involved.
• Separate. The institution keeps the outcomes. The analysis receives only information that existed as of the original decision date.
• Lock. Ranks and coverage statuses are delivered and locked before any outcome is released.
• Unblind together. Then join outcomes and evaluate against the pre-agreed measures.
• Test incremental value, not raw correlation. The question is never whether the property rank correlates with outcomes. It is whether it adds information after the variables already in use — leverage, valuation, borrower, project. Compare a baseline built from existing variables against the same baseline plus the property rank.

A negative result is a real result. If the signal does not transfer to a particular book, that is worth learning before a workflow is built on it.

The narrow version of the argument

Aggregate tape metrics are not wrong. They are answering a different question from the one they get used for. They describe loans well, geography roughly, and individual properties barely at all — and collateral outcomes are determined by individual properties. Adding a property-level read does not replace any part of pool review. It fills the one gap the tape was never designed to cover.