For institutional desks

The rate your model applies to a whole market says nothing about the house under the loan or the bid.

Good Investment scores every home against its own market — and translates that score into a measured appreciation edge, in points per year, on top of whatever the market does.

Good Investment is the house-level appreciation layer for owners and lenders.

The problem

What does an area appreciation rate hide?

Every model that carries a growth assumption carries it for an area. A ZIP, a metro, a state: one rate, applied to every home inside the line. The homes do not behave that way. Bought and resold over the same years, homes in one ZIP land far apart from one another, on both sides of the area rate, and the distance between two of them is routinely wider than the distance between two metros. Most of what happens to a home's value is decided inside its own ZIP, at the level of the house, where no area rate reaches.

A per-property rate is the thing that captures it. Good Investment scores each home against the homes it competes with in its own market and reads that score as a rate: how far this house is positioned to run above or below its area, with the confidence behind the read.

Why doesn't the miss average out?

If the area rate were wrong evenly, it would cost nothing. It is wrong unevenly. A pool bid at one blended price pays the same for the homes that will run behind their market as for the ones that will run ahead, and the seller keeps the difference. A bid built on one growth cell overpays for the house that looks like its neighbours on paper and is positioned worse than them.

In a competitive bid the miss is worse than random, because the pools and the homes you win are disproportionately the ones you overvalued. The miss lands on the loans that go bad and the bids that overpay, which is why one column added to a strat table or an underwriting model earns a desk's attention.

Three products, one problem

Which product fits which desk?

Each one exists because the area rate is silent about the house, and each adds one column to a workflow you already run.

Collateral risk for loan pools

Whole-loan buyers, correspondent desks, non-QM desks, credit funds.

What you do today

The tape arrives, you stratify it into FICO and LTV bands, your credit model assigns an expected loss to each band, and the bands roll up to one blended price. The file describes the property in an address, a type and a value.

What changes with one column

A score band sits beside your FICO and LTV bands. The weakest band prices on its own instead of hiding inside the blend, or it goes on the kick list you already run. The loss differential flows through your own model and comes out as your price.

Appreciation input for underwriting

Acquisition desks, IC review, HEI residual holders.

What you do today

Your model has one growth cell per market and every home in that market gets it, so two homes on the same street at the same price get the same maximum bid.

What changes with one column

That cell becomes the market rate plus a per-property rate, quoted with the confidence behind it. Your house view on the market is untouched; we supply the deviation from it. Inside your own price mandate, the same score ranks the pipeline.

Hold and sell review

REIT and portfolio teams, asset management.

What you do today

The release list is chosen on operating grounds: lease expiry, turn cost, capex plan, submarket exit. Every criterion describes what the home costs now.

What changes with one column

Forward appreciation joins the release criteria. Same target count, same brokers, same timing, a different list, and the price mix of the book you keep stays where it was.

Built and sent on request.

Hold and sell review

Proof

How was it measured?

The figures behind each product are on the seat brief, the model card and the validation pack, and they go out on a call. What the page can say is how they were produced.

  • Blind historical tests

    Every product was measured on sales the model never saw, scored before the outcome and graded after it, with an interval on every figure.

  • Benchmarked against the column you already hold

    Each result is raced against the ranking a desk already runs, a price-based ranking included, and the model card reports where each one leads.

  • Point-in-time scoring

    A property is scored with only what existed before its decision date, and the test suite refuses a run where anything later leaks in.

  • A model card and a validation pack

    Written for a model risk team: data lineage, how the test was built, what it was benchmarked against and where the evidence stops. The external pack is free and goes out on a call; the full internal pack goes out under NDA as a paid engagement.

  • A pilot you grade

    You send a resolved tape with outcomes held back. We score it point in time, lock the ranks with a digest you hold, and you unblind and grade the result against the variables you already use.

Property Hub
Sample property

House-level appreciation

Neighborhood basis

Top tier of comparable homes in Charleston SC

One of the strongest relative appreciation screens in the local market.

88th percentile
Bottom0-20
Lower20-40
Mid40-60
Upper60-80
Top80-100
Below marketMedianAbove market
The within-market read on a sample home, rendered by the same component the signed-in product and the tape pipeline use. It places the home among the homes it competes with in its own market, and the chip names the basis it was scored on. The rate that read implies, and the interval around it, arrive on the brief and on your own report. Collateral risk analysis

The platform behind it

What is the engine behind the score?

The same engine scores one address for an individual buyer and a full tape for a desk. The institutional products are that analysis at scale, delivered as a scored file against your own tape, a written report, or an API. Behind them sits a complete residential analysis platform.

  • A national residential AVM with a confidence read
  • A rent estimate for the subject property
  • Comparable sales, plus sales and rental market reads
  • Neighborhood-level growth, mapped below the ZIP code
  • A full investor return model with financing, operations and exit
  • Sourcing tools that screen many candidates at once
  • A tape pipeline that returns every row keyed on your identifiers, with a score, a tier, a coverage status and a review route

An AVM is an estimated sale price for a property. It is not the same as the opinion of value in an appraisal developed by a licensed appraiser under the Uniform Standards of Professional Appraisal Practice.

Fifteen minutes, then a pilot on the seat that fits.

Tell us which desk this lands on. We reply within 24 hours with the brief for that seat, book fifteen minutes, and scope a pilot on your own tape with the output yours to keep. The debt brief, the underwriting brief, the one-pager, the model card and the validation pack are sent on the call.

NDA-ready. Replies within 24 hours. No mailing list.

Prefer email? harley@marbary.com

Questions a desk asks

Isn't this FICO and LTV under another name?

No. Removing the same share of a tape by LTV reproduced none of the loss reduction the score band finds, and FICO moved expected loss by almost nothing on the same test. Give every property an identical mortgage and the effect gets stronger, which is the signature of a property signal. The model card carries the test.

We already run a price-based ranking. What does this add?

Keep it. Across a whole market a price-based ranking is a strong column and we say so before you ask. Inside the price mandate a desk actually buys from, that column has almost no range left to rank on and ours still separates the homes, and on the underwriting side the figure we quote is the increment after price position is already accounted for. A desk that already varies its growth rate below the market level has captured part of this, and the pilot measures the rest on your own pipeline.

What happens in a downturn?

The spread between the top and bottom of the ranking held through the one historical downturn we can test, scored before the peak and observed through the recovery. It is one crash in a concentrated set of markets, so every figure is scoped to the regime it was measured in and the downturn test sits on the model card with its frame beside it. A pilot on your own history is the answer we trust more than ours.

How is a tape handled?

Your tape goes to a cloud project created for your pilot, separate from the project that runs our product, and it is scored there by a job that exists only while it runs. No copy touches an employee machine. Every delivered file carries a digest you hold, so a rank changed after you release outcomes would stop matching, and when the pilot ends we delete the objects, then the project, and send an attestation naming what was removed and when.

Do you have SOC 2?

No. We hold no audit report and no third-party penetration test, and we would rather you heard that from us. What a security reviewer gets instead is the architecture, the access list, configuration evidence, file hashes, execution logs and the deletion attestation, and we run the deletion and lock checks with you on the call.

What does a pilot involve?

You send one tape with the outcomes held back, keyed on your own identifiers. We score every row as of its original decision date and return a score, a tier and a coverage status per row, with the ranks locked and digested before you release anything. You join the outcomes and grade the result against the variables you already use. Two weeks, and a negative result is a result you keep.

Related reading

Looking at one property? Run a single address for the same analysis. The first one is free.