The Four Numbers Everyone Quotes and Almost Nobody Underwrites

July 10, 2026 · Analysis

Real estate content in 2026 has a strange property: the louder the metric, the less anyone checks it. Cap rates, absorption, cash-on-cash, cost of capital. They get recited like liturgy on podcasts and pitch decks, and they quietly fail the people who deploy money based on them. Here is what the last month of data supports, and where the industry's favorite narratives fall apart under arithmetic.

Analyze any investment property in seconds - Sponsored

The rate story is true and useless at the same time

The 30-year fixed sits near 6.5% and Fannie Mae expects it to stay there through year-end. First-time buyer affordability remains below the qualification threshold on NAR's index. All true. None of it tells you where to put a dollar.

Here is the number that should reorganize your thinking: across 894 US metros, forecasts for the coming month split almost down the middle. 419 markets point up, 475 point down or flat, and the median move is close to zero. Call that what it is: two housing markets wearing one trench coat. A national affordability statistic averages them into a figure that describes neither. Rates tell you when borrowing gets cheaper. They will never tell you that Rochester and Savannah are heading opposite directions at the same coupon.

The discipline this demands is unglamorous: underwrite the metro, then the submarket, then the deal, and treat the national print as background radiation.

Your competition left two years ago. Your price floor left with it.

The most durable villain in housing content is the institutional buyer. The data stopped cooperating with that script. Institutional purchases are down more than 90% from their 2022 peak, representing roughly 1% of transactions. Total institutional holdings sit near a third of one percent of the single-family stock, per GAO. The January executive order restricting institutional purchases regulates a party that had already left the building.

The professional takeaway runs darker than "relax." Institutional capital concentrated in a specific footprint: Sun Belt, entry-level, high-yield submarkets. Its retreat removed a marginal bid from exactly those places, and you can watch the consequence in the listing data, where price-cut shares in Texas and the Deep South climb month after month. If your exit assumptions in those markets were formed between 2021 and 2022, they embedded a buyer who no longer shows up. Mark them down accordingly.

"94% accurate" is a mood, not a measurement

AI valuation marketing has settled on accuracy claims between 85 and 97%. Ask one question of any such claim: 94% of what? Within 10% of sale price? Median error of 6%? On which properties, in which regime? A 6% median miss on a $400,000 asset is $24,000. Levered at 75 LTV, that miss eats most of your equity cushion. The published numbers also score against homes that actually sold, a survivorship filter that flatters the model in the illiquid, mispriced, turning segments where a valuation error hurts most.

The models still earn a place. Peer-reviewed work shows multi-source and geospatial features genuinely improve valuation models. Treat them as ranking engines: use them to sort ten thousand candidates into the fifty worth human hours. The moment a single-number accuracy claim substitutes for deal-level diligence, you've bought speed with basis points. Demand published, horizon-specific error rates from any forecast or valuation vendor, misses included. Vendors who won't show you their scoreboard are asking for faith. (That standard is why every forecast on this site publishes its own error rate, and why we track where we disagree with Zillow's forecast in public.)

The metric canon needs a rate-regime audit

Every popular ratio rule in circulation was calibrated in a different cost-of-capital world. The 1% rule, the 70% rule, "buy above a 7 cap." Fixed ratios quietly change meaning across a 500-basis-point move in the discount rate. A 7 cap was a spread of 400 over financing in 2021. Today it can be a spread of 50, which leaves no margin after reserves. Cash-on-cash quoted without capex reserves, vacancy normalization, and refi risk is a brochure number. Absorption quoted metro-wide turns to noise when ZIP codes inside the same metro diverge, and they usually do.

The audit takes one afternoon: take every heuristic in your underwriting model and re-derive it at today's cost of capital, today's insurance costs, and today's exit-buyer composition. Most fail. The ones that survive are your actual edge.

What to do with all of this

Three moves. First, replace national narratives with market-level forecasts that publish their own error rates, and hold every data source you pay for to that standard. Second, re-underwrite Sun Belt exposure against a post-institutional exit environment; the price floor you remember has left. Third, demote AVMs to the screening layer and promote leading indicators, price cuts, days on market, and demand indexes to the top of your dashboard, because they move before the indexes everyone else watches.

Reading this market lazily costs real money. Reading it granularly is the edge, and the tools to do it are free: start with your metro.

Sources linked inline. Forecast counts from Sialim's June 2026 run across 894 metros. Analysis, not investment advice.