Atlanta TADs Did Not Cause Property Value Growth

Distance-controlled analysis of 104,597 Fulton County tax parcels, 2015 to 2025

Fulton County Board of Assessors data. City of Atlanta TAD boundaries.

Summary

8 of 8 TADs underperform their
matched controls
-14.6% TAD vs matched non-TAD
(same class, value, distance)
$905M Diverted from Atlanta
Public Schools

Using Fulton County's own parcel-level assessment data, TAD parcels did not produce excess property value growth. Non-TAD parcels at the same distance from downtown grew more, in both percentage and absolute dollar terms.

The City Auditor's comparison of TAD vs non-TAD property values does not control for location. Once you do, the TAD "advantage" disappears. In most cases, it reverses.

The Comparison Problem

What the Auditor Does

Compares TAD parcels (0-3 mi from downtown) to ALL non-TAD parcels (including suburban areas 5-10+ mi away).

Downtown grew faster than suburbs. That is not a TAD effect. That is urbanization.

What We Do

Compares TAD parcels to non-TAD parcels at the same distance from downtown.

Control for location, and the TAD "advantage" vanishes.

Why this matters: TAD parcels sit a median of 2.0 miles from downtown. Non-TAD parcels sit at 4.1 miles. Any comparison that ignores this difference is comparing apples to oranges.

Growth by Distance from Downtown

TAD vs non-TAD parcels within the same distance band from downtown. If TADs were causing growth, TAD bars would be taller. They are not.

Dollar Gains by Distance

Same comparison in raw dollar terms. Non-TAD parcels gained more at 0-3 miles from downtown.

Individual TAD Performance

Each TAD compared to non-TAD parcels at the same distance. Perry/Bolton at 5.1 miles from downtown shows the largest TAD growth, but even it underperforms when matched on property class and value.

Parcel Growth Map, 2015 to 2025

Each dot is a parcel. Brighter = higher growth. TAD boundaries in white, city boundary in gray. High-growth areas appear regardless of TAD boundaries, concentrated near downtown in all directions. Eastern Atlanta (DeKalb County) excluded due to data availability.

<50% 50-100% 100-200% 200-500% >500%

Map shows all TAD parcels + 15% random sample of non-TAD parcels (53,880 in city limits). Statistical analysis uses the full dataset of 91,787 parcels (after excluding stale assessments).

Matched Pairs Analysis

For each TAD parcel, we find the closest-matching non-TAD parcel: same property class, within 0.5 miles, within 50% of appraised value, within 75% of lot size. This is as close to an apples-to-apples comparison as the data allows.

7,283 Matched pairs
96.5% TAD median growth
111.1% Matched non-TAD
median growth

Property Class Breakdown

TAD areas have more condos and commercial properties. Non-TAD areas are mostly single-family residential. When you separate by class and distance band, TADs still underperform within each category.

Residential (82k parcels): TADs lag by 12-404% depending on distance band.
Condos (6.6k parcels): Roughly comparable at 3-4 miles.
Vacant Land (3.5k parcels): Non-TAD vacant lots appreciate faster at every distance.

The Border Test

The strongest causal test we can do with this data: compare parcels just inside a TAD boundary to parcels just outside. If TADs cause growth, there should be a jump at the boundary. There is one, but it goes the wrong direction.

108% Just inside TAD border
(within 0.25mi, n=7,890)
187% Just outside TAD border
(within 0.25mi, n=6,064)
-79% Difference
[95% CI: -93, -65]

Parcels immediately outside TAD boundaries grew 79 percentage points more than parcels immediately inside. This is the opposite of what a TAD effect would look like. The difference is statistically significant (p < 0.0001).

BeltLine: Proximity to the Actual Trail

The BeltLine TAD covers 37,000+ parcels, many of them miles from any trail investment. Critics say this dilutes the signal. So we split by distance from the trail itself:

Even parcels within a quarter mile of the trail (139% growth) underperform distance-matched non-TAD controls (155%). The BeltLine trail raised property values, but not more than the market was already delivering at the same distance from downtown.

Sensitivity Analysis

A finding is only meaningful if it holds up under different analytical choices. We tested four matching tolerance levels. TADs underperform at every single one.

Bootstrap CIs

Overall difference: -24.5% [95% CI: -27.5, -22.4]. Zero excluded from every confidence interval.

Tight Matching

+/-0.25mi, +/-25% value: -11.5% (1,483 pairs). Even strict controls show TAD underperformance.

Loose Matching

+/-1.0mi, +/-100% value: -16.8% (1,892 pairs). Wider tolerances make TADs look worse.

Property Class Changes

Only 1.8% of parcels changed class 2015 to 2025. Rezoning is not driving these results.

Known Biases and Limitations

Every known bias in this analysis works against our finding. We are understating any TAD effect, not overstating it.

Survivorship Bias

7,369 new parcels in 2025 not in 2015 are excluded. 76% of them are in TADs (new condo plats). Including them would help the TAD case.

Pre-2015 Growth

TADs created between 1998 and 2013 may have already produced growth before our window starts. If so, we are missing it.

Assessments Lag Market

Tax assessments are not market prices. But the same assessor applies the same methodology to TAD and non-TAD parcels. The bias is uniform.

Sample Size Imbalance

0-2mi: 22,773 TAD vs 285 non-TAD. But at 2-4mi where both sides have thousands of parcels, the same pattern holds.

Counterarguments

"You can't prove a counterfactual"

True. But we can show that comparable parcels grew faster without TADs. The burden of proof should be on the entity diverting $905M from schools, not on critics to prove a negative.

"The but-for test applies at time of creation, not now"

Fair point for early TADs. But the Auditor's report claims ongoing value using current increment as evidence of success. If the justification was in 1998, why is revenue still being diverted in 2025?

"TAD infrastructure benefits neighbors too"

If TADs make nearby non-TAD areas appreciate (spillover), our controls look artificially good. That would reduce the gap. Yet TADs still underperform.

"Perry/Bolton proves the model works"

One out of eight is a 12.5% success rate. Perry/Bolton is 5.1 miles from downtown, the only TAD far enough that market forces alone would not drive growth. That supports our point.

"TADs prevent blight, not just create growth"

Then growth rate is the wrong success metric. But it is the metric the Auditor uses. And non-TAD areas with similar blight concerns (South Atlanta, Pittsburgh) grew more.

"Property values are not the only metric"

The Auditor's report uses property value increment as the primary success measure. We are evaluating their claim on their own terms.

"TADs concentrate development, preventing sprawl"

Not the claim being evaluated. The question is whether $905M diverted from schools was justified by growth that would not have happened otherwise. There are cheaper tools for sprawl prevention.

Land Value vs Improvement Value

Land value reflects location premium. Improvement value reflects buildings and development. If TADs were driving growth through construction, improvement values should outperform. Instead, land values are nearly identical on both sides, confirming that the market prices location the same way regardless of TAD status.

Implications

$905M School Funding Diversion

Atlanta Public Schools lost $905M in property tax revenue to TADs that did not produce growth above what the market delivered on its own in nearby non-TAD neighborhoods.

Market-Driven Gentrification

Non-TAD neighborhoods at 2-3 miles from downtown (Grant Park, East Atlanta, Reynoldstown, Kirkwood) grew 217%, far more than BeltLine TAD parcels at 110%. No public subsidy required.

The "Halo" Illusion

The Auditor's claimed 6x "halo effect" is the distance-decay gradient from downtown. Everything near downtown appreciated, TAD or not.

Only One TAD "Worked"

Perry/Bolton at 5.1 miles from downtown is the only TAD where growth exceeded what the market would deliver. It is also the only one far enough to actually need a subsidy.

Methodology

  1. Downloaded 110,623 (2015) and 114,975 (2025) parcel records with geometry from Fulton County ArcGIS Feature Services
  2. Downloaded 10 official TAD boundaries from City of Atlanta Dept of City Planning GIS
  3. Computed centroid for each parcel polygon. Calculated haversine distance from CBD (33.749N, 84.388W)
  4. Point-in-polygon classification: 55,156 TAD parcels, 51,873 non-TAD parcels
  5. Excluded 2,432 stale assessments (parcels with identical 2015 and 2025 values, never reassessed by the county). 79% were in TADs, mostly vacant land.
  6. Computed percentage change in Total Appraised Value (TotAppr) from 2015 to 2025
  7. Distance-band analysis: compared median growth within 1-mile bands from CBD
  8. Class-stratified analysis: separated residential, condo, commercial, and vacant land. Compared within each class and distance band
  9. Multi-factor matched pairs: for each TAD parcel, found best-matching non-TAD parcel (same class, within 0.5mi, within 50% value, within 75% lot size). 7,267 pairs analyzed.
  10. Regression discontinuity: computed signed distance from each parcel to nearest TAD boundary. Compared parcels within 0.25mi on each side (n=13,904)
  11. Sensitivity analysis: re-ran matching at 4 tolerance levels (tight to loose). All show TAD underperformance.
  12. BeltLine sub-zone: split BeltLine TAD by distance to actual trail using 16-point approximation
  13. Bootstrap inference: 10,000-iteration percentile bootstrap for all medians and differences. Mann-Whitney U tests for significance.

Data Sources

  • Fulton County Tax Parcels 2015 and 2025, ArcGIS Feature Services (public open data)
  • TAD Boundaries, City of Atlanta Dept of City Planning GIS (10 zones)
  • Scope: City of Atlanta, Fulton County portion (~7mi from downtown)
  • DeKalb County portion of Atlanta excluded: no public 2015 parcel data available
  • All analysis code and output data available for independent replication

Known Limitations

  • Survivorship bias (conservative): 7,369 new parcels excluded, 76% in TADs. Including them would help the TAD case.
  • Stale assessments: 2,432 parcels excluded (identical 2015/2025 values, never reassessed). 79% in TADs, mostly vacant lots. Confirmed against live API: county data genuinely unchanged.
  • Pre-2015 window (conservative): earlier TADs may have already produced growth before our baseline starts.
  • BeltLine TAD covers 37k+ parcels, many far from investment. Sub-zone analysis shows even near-trail parcels underperform.
  • No pre-2015 data publicly available. Earlier data requires Open Records Act request.
  • Assessments lag market values. Bias is uniform (same assessor for all parcels).
  • Single CBD model. Growth radiates from multiple nodes but distance from downtown captures most of the variance.
  • Only 1.8% of parcels changed property class between years. Rezoning is not a confound.
  • DeKalb County (eastern Atlanta) not included. No comparable 2015 data available publicly.