The Roebling Valuation

An informed estimate of your Manhattan apartment’s current market value.

The analysis begins with comparable sales in your building and expands to nearby properties only when necessary — the comparables and confidence range presented alongside the result.

About the methodology

Methodology and confidence

The research warehouse behind this tool holds 330,000+ recorded Manhattan transactions across 13,000+ buildings — co-op, condominium, and townhouse. Within it, 2,484 buildings carry public building profiles, and the comparable pools draw on the warehouse’s index-eligible sales. What these numbers mean →

The valuation methodology was tested against historical Manhattan sales. For each backtest, the model estimated a past transaction using only information that would have been available before the sale occurred. Inputs were retained only when they improved out-of-sample accuracy. Recorded sales are adjusted for changes in the surrounding market, and comparable selection gives priority to apartments of the same type within the subject building.

The result identifies:

the source of the comparable evidence;
the number and relevance of the underlying sales;
whether the analysis is building-specific or neighborhood-supported;
the historical error range associated with that evidence profile.

Valuation treatment	Comparable evidence	Historical median error
High confidence	Four or more comparable sales in the building	Approximately 12%
Moderate confidence	Two or three comparable sales in the building	Approximately 16%
Indicative range	Comparable apartments in nearby buildings	Approximately 21%
Individual review	Unusual property type, configuration, value range, or limited sales history	Reviewed separately

In historical testing, 85.8% of eligible co-op sales received an automated valuation across the first three evidence tiers.

How comparable sales are selected

The analysis begins with the most relevant available evidence:

Comparable apartments in the same building
Other relevant sales within the building
Comparable apartments in nearby buildings
Corridor and broader neighborhood evidence where appropriate

Non-arm’s-length transfers and co-op sponsor sell-outs are excluded where they can be identified from the public record. Ordinary resale transactions in converted buildings remain eligible.

Recorded prices are adjusted to current market conditions using local price indexes rather than treated as directly comparable without regard to sale date.

What recorded sales cannot fully capture

No automated valuation can observe every characteristic that affects the sale of an individual Manhattan apartment.

Renovation quality, condition, views, light, exposure, outdoor space, layout, ceiling height, and the specific desirability of an apartment line can materially affect value. These considerations are particularly important for highly renovated residences, combined apartments, penthouses, maisonettes, large co-ops, and other properties with limited direct comparables.

The Roebling Valuation is an analytical estimate based on recorded market evidence. It is not an appraisal or a substitute for an apartment-specific pricing analysis.

Methodology appendix — the numbers

Backtest universe. 222,691 historical Manhattan sales — 140,907 condo and 81,784 co-op — with recorded sales from 2004 through the current year. Model of record v1.5.1, frozen; any change to a threshold, band, or rule requires a new version and a re-run of the full backtest suite.

What “median error” means. The median absolute percentage difference between the model’s estimate and the price subsequently recorded for that same apartment, measured out of sample. Half of historical estimates in a cell landed closer than the stated figure, half further. Displayed ranges are the measured 50% and 80% bands of estimate-to-actual for the cell — asymmetric where the history is asymmetric.

Leakage controls. every historical estimate used only records dated strictly before the sale's closing; confidence variables are prediction-time only (never realized price or post-sale information).

Condos — median error by evidence tier and price band (out-of-sample)

Evidence tier	<$1M	$1-3M	$3-10M
Strong — 8+ size-matched building comps	7.8% (22,074 sales)	8.4% (31,875 sales)	9.3% (9,474 sales)
Strong-lite — 5–7 building comps	8.9% (4,758 sales)	9.5% (7,282 sales)	10.7% (2,397 sales)
Moderate — thinner building evidence	9.9% (6,439 sales)	11.8% (10,746 sales)	14.9% (5,203 sales)

Co-ops — median error by evidence tier and price band (out-of-sample, unanchored cases)

Evidence tier	<$500K	$500K-1.5M	$1.5-4M
High — same-bedroom building evidence	9.4% (10,715 sales)	10.7% (33,542 sales)	15.1% (10,514 sales)
Moderate — stale or thinner building evidence	12.8% (1,074 sales)	16.1% (1,990 sales)	19.4% (1,165 sales)

Interval-only evidence. Where building evidence is too thin for a point estimate (the B tier), the tool presents an indicative range keyed on bedroom type and nearby-building sales — never a single number.

Geography. Comparable pools begin inside the subject building and expand to the surrounding corridor or neighborhood only when the building’s own record is insufficient; recorded comps are indexed on the corridor’s measured price movement, not carried at their nominal prints.

Renovations and combinations. Condition and renovation quality are deliberately NOT modeled — a self-reported condition cannot be backtested, and a selector that does not change the number erodes trust. Combined, reconfigured, and otherwise unusual apartments (penthouses, maisonettes, townhouses, most $4M+ co-ops) route to a personally reviewed analysis instead of an automated figure.

Coverage. 85.8% of validated historical cases receive an on-screen estimate; the remainder route to review. Expected coverage for ordinary resale-owner requests is approximately 88% — an assumption about the production request mix, not a validated universal figure.

Every result shows its evidence — the source of the comparable sales (the building first, nearby buildings only when necessary), the number of sales behind the number, the confidence tier with its measured historical error, and the valuation date. When the record supports only a range, you get a range, not false precision.

What no model can see: condition, renovation quality, light and views, current competition, and the circumstances of the sale. Those routinely move a Manhattan price by more than the model’s error band — which is what the personally reviewed Private Pricing Opinion is for.

Run it on your apartment →