More excuses than George Costanza, Saul Goodman, and Tom Haverford combined

Today, I will make my excuses.

The report — 20 Years, 4 Co-ops, & 10002 — is good. But, it has its problems. (Who doesn’t? Even “reports” have them.)

This is the part where I reveal the known shortcomings of everything I put forth in the paper. (And none of these are aesthetic, like, “It needs more pictures!” Or, “No, really: It needs more pictures!”) This is simply to acknowledge that this data can only tell us so much. That it’s directional.

But, it’s the only one (that I know of) of its kind. So, it’s special.

You can download the complete report here

Here we go with our most glaring caveats and considerations , omitting some because this post is supposed to be interesting .

Data completeness varies — especially early on. The combined dataset includes 3,006 records, but its completeness is uneven across time. Listings from the early 2000s, particularly pre-2010, frequently omit key fields. Square footage is missing in 2,238 entries, and floor information is absent in 799 cases. List prices and sale prices are not always recorded together: 1,218 entries include both, while 231 include neither.

Missing data does not necessarily mean a transaction didn’t occur. In many cases, it reflects inconsistent reporting practices by real estate professionals over two decades—each working with different systems, norms, and incentives.

When sale prices are missing, list prices are used. In instances where a closed sale price is unavailable, the analysis substitutes list price as a proxy for valuation. This allows for continuity across the dataset but introduces a little noise — not quite to the volume of Quiet Riot — where asking prices can only reveal what a seller sought.

List prices are not 1:1 proxies for value. Agents may price above, below, or at perceived market value depending on strategy, timing, and market conditions. For that reason, the report’s narrative focuses on multi-year, cyclical patterns , not single-transaction precision. Long-horizon averages remain statistically informative even when individual data points are imperfect.

Also missing: Time on market, monthly carrying charges, number of price reductions (or increases), and a bunch of other things. While this information is out there — and super valuable — I am not able to access it at scale.

Physical unit data is inconsistent by nature. Square footage, bedroom counts, and bathroom counts vary widely in quality and availability. Older listings often omit bathrooms from listings entirely . Bedroom classifications sometimes reflect “marketing choices” rather than architectural reality — especially where alcoves were treated as second bedrooms or living rooms were partitioned.

As a result, a metric like price-per-square-foot is, inconveniently, less relevant. This is due directly to the properties being cooperatives, which — by virtue of their share-based structure — do not rely on square footage as a valuation metric. When available, price-per-square foot is treated as a directional indicator, not a primary valuation metric.

Repeated listings do not equate to repeated sales. This sounds obvious, but it may not be. Within the dataset, units appear multiple times as listings. This could be due to sales, yes, but also withdrawals and relisting at later points. Still, the analysis relies on 1,285 confirmed closed sales as the transactional foundation .

Human error is unavoidable in historical listing data. Over 20+ years, hundreds of professionals manually entered listings. Misspellings, truncated entries, incorrect addresses, omitted unit numbers—and even intentional relisting tweaks for marketing purposes—are inevitable. Obvious errors were standardized or removed .

For instance, one unit was listed at $27M . While that could be this Tribeca condo with the Staple Street skybridge , it’s certainly not one in the co-ops. (Can you imagine having a skybridge?)

20 Years, 4 Co-ops, & 10002 reflects the visible market — not the full ownership history. Private transfers, estate distributions, family inheritances, and internal transactions often do not appear in public records. As a result, turnover rates should be understood as indicators of visible liquidity , not total ownership movement.

Okay, that’s it for my excuses. But, still…

…despite these limitations, the dataset is sufficiently deep to model long-term appreciation, building behavior, line behavior, and cooperative identity with confidence.

Up next? We start to unpack the findings, and I promise there’s good stuff coming.