How the Investor Market Scorecard works

1. The Three Scores

Each market receives three scores on a 0–100 scale, where 100 = best in universe and 0 = worst in universe. Scores are computed as weighted percentile ranks across the universe of qualifying markets. They are not absolute measures — a score of 75 means "this market ranks in the 75th percentile for this dimension across all scored markets."

Cash Flow Score

Measures how well a market generates rental income relative to purchase price, weighted by evidence of active tenant demand. A high-yield market where homes sit unsold for 120+ days scores lower than a slightly lower-yield market where homes move in 20 days — because the latter is evidence that tenants and buyers are competing for the same space.

Components: Gross rent yield (60%) + Days to Pending inverse (40%).

Growth Score

Measures the appreciation trajectory of a market: where values have been and where Zillow's model expects them to go. A third leg — rent growth — cross-validates whether household demand supports the price trajectory.

Components: ZHVI 1-year change (40%) + ZHVF 12-month forecast (40%) + ZORI 1-year rent change (20%).

Stability Score

Measures how balanced and predictable a market is. An extremely hot market (homes selling in 5 days over ask) and an extremely cold market (homes sitting 200+ days) both score lower than a moderate market — because stability means predictable conditions, not maximum speed.

Components: Market Temperature Index (50%) + Absorption rate (25%) + Days to Pending bell curve (25%).

2. Components and Weights

Cash Flow Score — detail

Growth Score — detail

Stability Score — detail

3. The Math

Percentile rank

All components except the DTP bell curve use PostgreSQL window function percentile ranks:

pct_rank(c) = PERCENT_RANK() OVER (ORDER BY raw_value(c)) × 100

For "lower is better" components (Days to Pending in Cash Flow), the rank is inverted:

pct_rank_inv(c) = (1.0 − PERCENT_RANK() OVER (ORDER BY raw_value(c))) × 100

This means the market with the lowest DTP gets a percentile rank of ~100, and the market with the highest DTP gets ~0.

Missing-data re-weighting

When a component is null (e.g., no ZORI YoY data for a metro), the score is computed from the available components, with weights rescaled proportionally:

total_weight_present = SUM(weight(c) for c in components_present)
score(d) = SUM(weight(c) × pct_rank(c) for c in components_present) / total_weight_present

Example: If ZORI YoY is null for a metro, the Growth score is computed from ZHVI YoY (40%) and ZHVF (40%), re-scaled to 50/50.

Qualification rules

A market is excluded from scoring entirely (not just penalized) if:

• Cash Flow: No ZORI data — cannot compute yield without rent
• Growth: All three components are null
• Stability: All three components are null

4. Signal Logic and Thresholds

Signals are derived on every query — they are not stored and are recomputed from scores each time. This means methodology revisions don't require database migrations.

5. Coverage and Limitations

Metro coverage

The scorecard queries all MSAs in mv_metro_market_snapshot. As of 2026-04-30:

• 894 total MSAs in the snapshot
• 894 MSAs receive full Cash Flow + Growth + Stability scoring — limited by ZORI YoY coverage
• The remaining MSAs appear with Stability and partial Growth scoring only, marked "Unscored" on Cash Flow

ZIP coverage

ZIP-level scoring covers only ZIPs with native ZORI rental data. As of the current release:

• 8316 ZIPs receive full scoring (approximately 29.6% of ~26,297 US ZIPs)
• ZIPs without ZORI data are not surfaced in the ZIP table — they are not scored and not shown as "Avoid" or "Unscored"
• Stability components (DTP, Market Temperature, Absorption) are inherited from the parent MSA at the ZIP level — they are not ZIP-native

What this tool does NOT measure

• No vacancy data. Days to Pending and Absorption Rate serve as demand proxies. Actual vacancy rates are not in the warehouse.
• No population or income trends. Census ACS data is deferred to a future version. ZHVI trend serves as the demand proxy in v1.
• No neighborhood-level signals. All scores are market-level or ZIP-level. Block or street-level analysis is not possible with this data.
• No cap rate calculation. Gross yield is not net yield. Operating expenses, vacancy allowance, property management, and financing costs are excluded.
• No property tax adjustment. High-tax states (e.g., IL, NJ) may show attractive gross yields that erode significantly after taxes.
• No crime or school data. Out of scope for v1.

6. Data Sources

Refresh cadence

Scores recompute when materialized views refresh (currently monthly). The Redis cache has a 7-day TTL — worst-case staleness after an MV refresh is 7 days. Cache keys embed the latest observation_date, so a fresh MV refresh produces different cache keys and naturally invalidates prior entries.

7. Changelog

← Back to the Scorecard  ·  Report a data issue