Improving Property Tax Assessments
The Question: What data do we have on property, the Census, and built environment that can help predict housing prices?
The Motivation: Philadelphia is a city of more than 1.5 million people and growing, we need properties to be accurately assessed.
Improving the Office of Property Assessment’s Automated Valuation Model (AVM) can potentially:
More stable values aligned with market rates for properties being sold.
Alleviate burden from gentrification in previously disinvested neighborhoods.
Equitable taxation for residents.
Financial stability for Philadelphia.
Help set the future for more reliable open data.
Philadelphia Properties and Current Assessments from Open Data Philly (n = 559,322)
2023 5-Year American Community Survey (ACS) from the United States Census Bureau (n = 28,261; 2019-2023)
Higher Prices: Center City, University City, the riverfront, and affluent Northwest pockets.
Potentially due to easy access to transit and amenities.
Lower Prices: North of Broad Street into parts of West and North Philadelphia.
Potentially reflecting long-term disinvestment.
Sale price is place-dependent in Philadelphia, mostly due to neighborhood qualities.

RMSE = 138,279.40 → Predicted sale price differs by about ± $138,279 from actual market sale price.
R² = 0.746 → Explains 75% of variance in home prices.
Neighborhoods (e.g. Fitler Square $431,911.65 > East Falls)
Livable Square Footage ($187.32 increase)
Median Household Income ($0.63 increase)

Human review for over-valued neighborhoods that are being gentrified.
Limitations / Ethical Concerns:
Undervalues some disinvested neighborhoods and overvalues wealthier ones.
Areas with residents depend more on personal vehicles, this model is predicting on public transit, not highways.
Future Potential:
Incorporate additional categorical variables after consideration.
Separate rural modeling can be helpful too.