Philadelphia taxes buildings far more heavily than the land beneath them. In a rising neighborhood, a fenced parking lot costs little to hold, so waiting is often the smartest move an owner can make. This is the story of what that waiting costs the city, told through ten places, and of a different way to tax land.
Between Vine and Pine, from the Delaware to the Schuylkill, more than two dozen large lots in Center City are used for public parking. Many were paved during the urban renewal of the 1960s. Philadelphia's zoning code does not treat surface parking as a permanent use: owners need a special license, renewed every three years, and renewals have become routine.
The lots are already zoned for towers. What they lack is a reason to change. Some owners hold them with no mortgage, so their main costs are insurance and property tax. Because the tax falls mostly on buildings, an empty lot carries very little of it.
The pattern is not limited to downtown. Along commercial corridors in every council district, there are parking lots, shuttered drugstores, tire shops and vacant parcels on land zoned for apartments over shops. Each one has its own history. Together they describe how the tax system shapes the city we see.
On the 700 block of Chestnut Street, the owner of a parking lot tried four times to partner on an apartment tower. Each deal fell apart at the last minute. The lot is still parking.
Zoned for up to 258,120 sf of building. Four tower deals have collapsed. Parkway · Inquirer
The 2019 property tax on a 140-space lot zoned for up to 422,208 sf, against $91,346 in net parking income. Flynn Co. listing
A tower planned on this Parkway lot was dropped in 2023. The lot sold to a new owner in 2025, with no project confirmed. Philly Land Sites
The waiting land is concentrated where the market is hottest. District 1, which takes in Old City, Northern Liberties and Fishtown along the Delaware, holds 50 of the city's 120 CMX-3 vacant and parking parcels. Those are the places where land prices have climbed fastest, and where the gap between what land sells for and what the city assesses it at is widest.
The records also show how long land waits. Half of these parcels have no recorded transfer in nine years or more. A few have been held since the 1970s.
Every property tax bill in Philadelphia has two parts: the value of the land, and the value of whatever stands on it. Today both are taxed at the same 1.4%. Build a six-story building on a parking lot and its tax bill multiplies more than tenfold. A land value tax shifts the weight: a higher rate on land, a lower rate on buildings, the same total revenue for the city.
An owner holding a lot earns two things: whatever the parking brings in, and the land's rising value as the neighborhood around it improves. The cost of waiting is mostly the property tax. When that cost is low and the neighborhood is hot, the patient bet often pays better than building.
This is how speculation works its way into a market cycle. In good years, land trades on what it might be worth at the next peak, and prices run ahead of what an apartment building can afford to pay. When the cycle turns, projects stall and the lots go back to parking, waiting for the next run.
A land value tax raises the cost of waiting, every year, whether the land is used or not. It also lowers what land sells for, because buyers price in the higher tax. The buyer who builds keeps that difference as a lower cost of entry.
One vacant or parking lot zoned CMX-3 in each council district, drawn from the city's own records, with the story of its place. Districts 9 and 10 have none; their stories show why. Pick a district, or tap a site on the map.
Dots: every privately held vacant or parking lot of 4,000 sf or more in CMX-3 (orange) and CMX-4 or CMX-5 (gray). Dashed lines: council districts, approximate.
Philadelphia already pays developers to build. The 10-year abatement forgives most of a new building's tax for a decade, and a 20-year version is proposed for conversions. Both are paid for by everyone else, including the School District, which receives about 56 cents of every property tax dollar. A land value tax builds the incentive into the tax itself, permanently, and collects it from land that sits idle.
Pennsylvania is the only state that lets its cities tax land and buildings at different rates. Dozens of its towns have tried it over a century. Philadelphia never has, and it would need the state's permission to start.
Pennsylvania begins enabling split-rate taxes. Pittsburgh phases one in.
Harrisburg, near bankruptcy, adopts a split rate and later taxes land at four times the building rate.
Allentown writes a split-rate tax into its home rule charter.
Altoona adopts a split rate.
Detroit's land value tax plan stalls three times in the Michigan House.
Of 33 Pennsylvania places that have adopted or debated it, 16 still use a split rate and 5 have rescinded it.
Philadelphia cannot split its rate without authority from Harrisburg. A city solicitor told Council so at a hearing on the idea.
Shift the ratio a step at a year, as Pittsburgh and Harrisburg did, so owners and the market can adjust and assessments can be tested.
Layer the shift with the homestead exemption, longtime-owner and senior programs, so the change lands on idle land, not on families in rising neighborhoods.
Sources: Lincoln Institute · People's Land · Strong Towns · Crain's Detroit · WHYY
For planners, staff and anyone who wants to test the numbers: the full model behind every chart above. Pick a site and a building, change any assumption, and see how the developer, the landowner and the public fare under each policy.
Every number on this page can be traced, rerun and challenged. The data queries, the code that builds the districts and the model, the tests and the outputs are published as an open package, in the tradition of the City Controller's and the Lincoln Institute's published studies. Open any section below for the detail.
The Waiting Lots: land, tax and time in Philadelphia. Pilot study, AIA Philadelphia Housing Committee. Draft for peer review, October 2026.
This appendix describes every data source, filter, formula and assumption behind the published page, so that any reader can check the work, rerun it and propose changes. Code and data are in the accompanying repository. Where a choice follows prior research, the source is named.
The study asks three questions:
The study is a set of transparent calculations, not an econometric forecast. Behavioral claims (that a split-rate tax increases construction) rest on the published research in section 9, not on new estimation here.
Parcels. Office of Property Assessment (OPA) records, table opa_properties_public, queried from the City's public Carto SQL API on 2026-10-28. The queries are in queries/ with their exact download URLs.
cmx_lots.csv: every parcel zoned CMX-1 through CMX-5 whose category is vacant land or whose building description is a parking lot, with lot area over 4,000 sf (787 parcels). Fields: parcel number, address, zoning, OPA category and building descriptions, lot area, market value, taxable and exempt land and building values, last recorded sale date and price, year built, and coordinates.citywide_totals.csv: citywide sums of taxable and exempt land and building value (583,718 parcels).Council districts. The 2022 redistricting assigns whole wards and listed divisions of split wards to each district; the ward and division lists were taken from the Committee of Seventy's published summary. Division boundaries are the 2019 election divisions from Azavea's open geo-data repository (Abaca Labs, CC BY 4.0). src/build_districts.py unions them into ten districts. Every 2019 division was assigned. Divisions have been redrawn since 2019, so parcels within a block or two of a district line may be misassigned. The City's official council district layer should replace this reconstruction.
Neighborhood outlines for the base map: Abaca Labs, CC BY 4.0, via Azavea.
Market inputs. Rents, retail rents, cap rates and land prices come from published listings and market reports cited in section 6. Construction costs are placeholders pending project bids.
From cmx_lots.csv, src/assign_lots.py keeps parcels that are:
CMX3);NON PD PKG LOT COMMERCIAL, PD PKG LOT COMMERCIAL, PKG LOT NON COMMERCIAL);Duplicate parcel numbers are removed. Each parcel is placed in a district by point-in-polygon on its OPA coordinates.
Result: 196 parcels, 203 acres, $695.8 million in assessed value, $9.74 million a year in tax at 1.3998%. Of these, 120 are CMX-3; 63 of those are "mid-sized" (8,000 to 60,000 sf), the population used for the apartment estimate.
Known limitations of OPA records. OPA's "vacant" category includes some parcels whose building description is not vacant (for example a shopping center recorded as vacant land); these were kept because OPA classifies them as vacant, and a reviewer may reasonably drop them. Recorded sale prices under $100 are transfers between related parties, not market sales (nominal_sale = true). One recorded price can cover several parcels sold together, so sale-to-assessment comparisons for individual parcels are indicative only.
Philadelphia taxes assessed land and buildings at one rate, t = 1.3998% (School District 0.7839%, City 0.6159%; the School District receives 56% of collections). A split-rate tax sets a land rate rL and a building rate rB with rL = k·rB, where k is the ratio. Holding citywide revenue constant:
rB·(1 − s) + rL·s = t, so rB = t ÷ ((1 − s) + k·s)
where s is land's share of taxable value. From the October 2026 records, taxable land is $44.50 billion and taxable buildings $121.30 billion, so s = 26.84%. (The 2020 Tax Shift Explorer used 24%.)
| Ratio k | Building rate | Land rate |
|---|---|---|
| 2 : 1 | 1.104% | 2.207% |
| 4 : 1 | 0.775% | 3.102% |
| 6 : 1 | 0.598% | 3.586% |
| Land only | 0 | 5.215% |
This is the same revenue-neutral construction used in Pennsylvania's split-rate cities and in the Lincoln Institute's Detroit study (Anderson et al., 2022), which modeled land taxed at five times the building rate.
Caution on s. The land share is only as good as OPA's land values. OPA's tax year 2025 methodology states that it usually allocates 20% of a property's value to land and 80% to the structure. The City Controller's 2019 review found that about half of single-family land values were set at a fixed 15% or 30% of total value, and that land values on the same block varied by as much as $22 per square foot. The split rates above should therefore be treated as provisional until land is valued directly (section 8).
For each parcel, src/assign_lots.py computes tax today, (land + building) × t, and under each split rate, land × rL + building × rB, using current assessed values. For the 196 study parcels at 4 : 1, tax would rise from $9.74 million to $21.40 million a year. This is a static first-round calculation: it ignores the fall in land prices that a land tax would cause (section 7.3) and any change in use.
A mid-sized building allowed by right in CMX-3: a ground floor of retail and lobby covering 75% of the lot (the CMX-3 maximum occupied area for an interior lot), and five residential floors above. That is a floor area ratio of 4.5, under the 500% base allowed in CMX-3, so no mixed-income bonus payment is needed. A building of this height is typically built as wood frame over a concrete podium.
| Parameter | Value | Basis |
|---|---|---|
| Upper-floor efficiency | 82% | Typical double-loaded corridor; placeholder |
| Average apartment | 665 sf | 80% one-bedrooms at 650 sf, 10% studios, 10% two-bedrooms |
| Retail share of ground floor | 60% | Remainder lobby and services |
| Construction cost | $300 per gross sf | Placeholder, hard and soft, excluding land; to be replaced with bids |
| Residential vacancy | 5% | Metro vacancy 4.6% in Q2 2026 (Northmarq) |
| Operating costs | 30% of income | Before property tax; placeholder |
| Retail rent | $18–$28 per sf per year, net | Fishtown average about $27 (TenantBase); metro in-line $19.98 NNN (CBRE Q2 2026) |
| Retail vacancy | 10% | Metro retail availability 8.1% (CBRE Q2 2026) |
| Market rent | $2.20–$3.20 per rentable sf per month | Fishtown 23 West one-bedroom $1,895 for 646 sf (Inquirer, 2025); Rittenhouse one-bedroom average $2,329 (RentDataNow, Mar 2026); Class A metro asking $2,214 (Northmarq Q2 2026) |
| Exit cap rate | 5.5% | Philadelphia deals averaged mid-5% (Northmarq Q2 2026); 6.08% (Matthews Q2 2026) |
| Developer's required return | 10% profit on total cost | Common feasibility threshold; tested at 15% |
| Developer's discount rate | 8% | For valuing abatement savings |
| City's discount rate | 4% | Roughly a municipal borrowing rate |
| Construction period | 2 years | Land tax only during construction |
| Assessment level | 75% of market value | Simplification for new buildings and land; tested at 100% |
On the Laurel Street lot (21,858 sf, modeled as 22,000 sf) this gives 99,000 gross sf, 101 apartments and about 9,900 sf of retail.
All equations are implemented in src/model.js and tested in tests/test_model.js.
Stabilized net operating income before property tax:
NOI₀ = units × rent × 12 × (1 − vacancy) × (1 − operating ratio) + retail sf × retail rent × (1 − retail vacancy)
For each policy, NOI after full tax is NOI₀ − land tax − building tax. Value at stabilization is NOI ÷ cap rate, plus, for an abatement, the present value at the developer's discount rate of the building tax forgiven over its schedule. This follows the Controller's 2018 abatement analysis in treating the abatement as a capitalized benefit to the developer.
Return on cost = (value − land price − development cost) ÷ (land price + development cost). The rent needed is the rent per rentable square foot at which the return equals the required 10%, solved in closed form:
NOI needed = cap rate × ((1 + required return) × total cost − abatement value)
rent needed = (NOI needed + land tax + building tax − retail NOI) ÷ ((1 − operating ratio) × (1 − vacancy) × 12 × rentable sf)
The test suite confirms that plugging the rent needed back into the model returns exactly the required return.
A higher tax on land is capitalized into a lower land price, the standard result in the property tax literature and the reason a land tax is thought to have little deadweight loss. The model applies a one-time adjustment:
land price under the land tax = land price today × (land yield + t × assessment level) ÷ (land yield + rL × assessment level)
with a land yield (the pre-tax return buyers require on land) of 5%, tested from 4% to 7%. Yang and Hawley (2021) found that Pennsylvania's split rates lowered land values but not drastically, consistent with this adjustment.
Land is taxed in full under every abatement.
Tax from the site for 30 years under each policy, in constant dollars, discounted at 4%. The School District share is 56%. Two holding cases are included: the lot stays parking under today's tax, and under the land tax.
The landowner's yearly return from waiting = (parking income − property tax) ÷ land price + expected appreciation. It is compared with the developer's 8% discount rate as a benchmark for the return from building.
Built-in checks (tests/test_model.js). Split rates are revenue-neutral for every ratio and land share; a 1 : 1 ratio reproduces today's tax; abatement schedules match their definitions; the rent-needed solution returns the target return; the prototype envelope gives 6 stories, 99,000 sf and 101 apartments.
Sensitivity (src/sensitivity.js, data/derived/sensitivity.csv). Each main assumption was varied one at a time. In every case tested, a 4 : 1 land tax needs slightly less rent than the phased 20-year abatement and leaves the public more revenue from the site over 30 years. Construction cost moves the rent needed most: ±15% in cost moves it by about ±$390 a month, far more than any tax setting.
Recommended next validation steps.
It does not estimate how much new construction a land tax would cause citywide; it relies on the research above for that. It does not yet model homeowner protections, phase-in, or the effect of lower land prices on the tax base over time. It treats each site in isolation and holds rents and costs constant. Its district boundaries are approximate. Its land values are OPA's, which section 4 shows are partly rule-based.
Requirements: Python 3.10+ with shapely, pyproj and pyshp; Node.js 18+.
git clone --depth 1 https://github.com/azavea/geo-data vendor/geo-data
python3 src/build_districts.py # geo/council_districts_approx.geojson
python3 src/assign_lots.py # data/derived/*.csv
node src/sensitivity.js # data/derived/sensitivity.csv
node tests/test_model.js # model checks
To refresh the data, open the URLs in queries/README.md and save the CSVs to data/raw/.