Methodology / Research
We Rejected Our First Two ZIP Scores. Here’s the One That Finally Passed.
Investors kept asking for one number: a single 0–100 grade for every U.S. ZIP Code, the way our flagship Markets tool grades cities. We tried twice and rejected both attempts: one leaned on ZIP “growth” the data cannot support, the other just floated the cheapest tiny towns to the top. Then we broadened the search to 68,256 transparent model configurations, and one finally passed. This is the story of what was wrong, what changed, and the new Rental Investor ZIP Score.
Why one ZIP grade is so tempting (and so hard)
Our city-level Balanced Market Score works because it balances thirteen factors across genuinely different axes: cash flow, but also population and job growth, crime, unemployment, price-to-income, months of supply. Growth and risk counterweight raw yield, so a cheap, shrinking town cannot ride a high rent-to-price ratio to the top. At the ZIP Code level, most of those counterweights are only measured at coarser geographies, which is exactly what sank our first two attempts.
Rejection #1: ZIP “growth” that was really noise
Our first design copied the city model’s Growth category. But a defensible ZIP growth number needs to compare Census survey vintages, and the only ZIP-boundary-compatible pair uses overlapping five-year windows, which the Census warns against for measuring change. The result was dominated by margins of error: population change was statistically indistinguishable from zero for the large majority of ZIP Codes. Ranking on that would have been ranking noise, so we rejected it.
Rejection #2: a “cheapest tiny town wins” machine
The second design dropped growth entirely and built a four-category score from a single Census vintage. It failed for a different, equally structural reason: at ZIP granularity gross-rent yield is almost the inverse of home value, so “cash flow” and “affordability” were measuring the same thing (cheapness), and nothing pushed back.
The overall score correlated −0.66 with home value; its top 100 had a $95k median value (vs $243k across all eligible ZIP Codes) and 61 of 100 had under 2,000 residents. It also correlated 0.64 with a plain affordability measure we already published; it added almost nothing. So we shipped only the two honest subscores and wrote up the rejection.
What changed for the third attempt
The mistake wasn’t rewarding affordability: a low price and a strong yield are legitimate advantages. The mistake was letting cheapness win unchecked. So the new model adds the constraints the ZIP data can actually support, and searches for the weighting that best balances them:
- Real ZIP-level demand and liquidity: renter-household depth and population scale, so a market has to be big enough to actually transact in, not just cheap.
- Low vacancy: a tight rental market, not empty units.
- A valid surrounding-market growth signal: annual county population estimates (non-overlapping, so the Z10 problem doesn’t apply) plus the surrounding place/county’s own market fundamentals, always labeled as context, never as a ZIP measurement.
- Data confidence: wider survey margins of error lower a ZIP’s confidence, so unreliable estimates can’t dominate.
We ran 68,256 deterministic configurations (different weights, factor sets, aggregation rules, floors) and scored each against hard safeguards (no single factor controls the ranks; the top can’t be tiny, high-vacancy, or in declining counties; ranks stay stable under weight changes). Cheapness is now a defect only when it comes with weak demand, thin liquidity, high vacancy, unreliable data, or a declining surrounding market. Affordable markets still rank well when the fundamentals back them up.
What finally passed: the Rental Investor ZIP Score
Each eligible ZIP Code now gets an overall Rental Investor ZIP Score (0–100), an A–F grade, and national and state ranks, from four categories: Cash Flow & Affordability (25%), Local Demand & Market Quality (30%), Surrounding-Market Growth (25%, context), and Stability & Risk (20%). ZIP-observed evidence carries about 70% of the weight; explicitly-labeled county/place context carries about 30%. It is a screening tool, not a return forecast.
The top of the list now reads like where rental investors actually look: Northwest Arkansas (Rogers, Bentonville, Springdale), Nashville suburbs (Spring Hill, Murfreesboro, Smyrna), Phoenix suburbs (Buckeye, Surprise), the Boise area (Nampa, Meridian), Huntsville and Madison, AL. And it handles the hard cases the way an investor would:
| ZIP Code | Value | Gross yield | Rank / grade | Why |
|---|---|---|---|---|
| 29445 Goose Creek, SC | $282k | 6.6% | #1 · A | growing, low vacancy, deep renter market |
| 72450 Paragould, AR | $178k | 6.2% | #1,122 · A | affordable and supported by demand + low vacancy |
| 48210 Detroit, MI | $68k | 17.3% | #8,608 · C | huge yield, but a declining county + weak stability |
| 60614 Lincoln Park, Chicago | $720k | 3.4% | #12,820 · D | expensive, low-yield, in a declining county |
Detroit’s 17% yield doesn’t buy it a top rank, because the surrounding market is shrinking and the stability signals are weak: exactly the check the rejected model was missing. Expensive, famous ZIP Codes aren’t forced up the list either; their placement follows rental economics, not name recognition.
How to use it
The ZIP Code explorer now opens on Best Overall. You can sort by the overall score or by any single category (Cash Flow, Market Quality, Growth Context, Stability), filter by state and data tier, add columns, compare 2–4 ZIP Codes, and export (all free, no login). Every ZIP profile shows the overall score, grade, national and state ranks, a meter for each category (labeled ZIP-data vs surrounding context), a plain-language “why it ranks here,” the cautions, and the two component subscores behind the overall. All 33,791 ZIP Codes stay searchable; about 19,970 clear the evidence gates for an overall score, and the rest show an honest reason.
ZCTAs are Census statistical areas that approximate USPS ZIP delivery geography. Every figure is an ACS 5-year survey estimate with a margin of error; the yield is a gross survey proxy, not a cap rate. Surrounding-market factors are county/place data, never presented as ZIP-level observation. ZIP scores and city scores use different inputs and universes and are not directly comparable. We use only public government data. We do not scrape Zillow, Homes.com, or listing portals. Not investment advice.
See the rankings. The Rental Investor ZIP Score is live: search, sort, filter, and compare U.S. ZIP Codes, no login, in the ZIP Code explorer. The full model, and the two designs we rejected first, are on the methodology page.