Philadelphia Housing Price Prediction

Improving Property Tax Assessments

Team Name Consulting

Who We Are

Why Improve the Model?

Research Question

Identify which structural, spatial, and socio-economic predictors contribute to a more accurate Automated Valuation Model for the City of Philadelphia.

Motivation

Improving the accuracy of residential property tax assessment can mitigate inequity assessment methods, increase transparency in governmental processes, and make analysis more reliable and efficient

Data Sources

  • Property Sales: (Philadelphia, 2023-2024) City of Philadelphia - OpenDataPhilly
  • Socio-Economics: United States Census - American Community Survey
  • Spatial Features: City of Philadelphia - OpenDataPhilly

Sale Prices in Philadelphia

Higher sale prices are concentrated in Central and Northwest Philadelphia

Factors Impacting Sale Price

There is a notable relationship between sale price and total livable area

Model Comparison

Model Performance Metrics
Model RMSE MAE R²
Model 1 230,140.41 15,174.13 0.287
Model 2 215,430.39 90,217.97 0.380
Model 3 211,021.39 90,144.74 0.405
Model 4 196,081.17 71,582.45 0.490
  • Model 1 (Structural) and Model 2 (Structural + Census) had the worst performance.
  • Spatial features (Model 3) and interaction terms (Model 4) boosted model performance.

Top Predictors

  • In a wealthy neighborhood (\(\beta\) = 50,306.450, p < 0.01)
  • Number of bedrooms (\(\beta\) = 33,544.290, p < 0.01)
  • Number of bathrooms (\(\beta\) = 29,510.940, p < 0.01)

Interpretation: Higher home values in wealthier neighborhoods make sense, especially historically affluent areas with reputational appeal and historic housing stock. Higher numbers of bedrooms and bathrooms tend to indicate more total livable area, which aligns with our observation that total livable area impacts sale price.

Model Performance

Model Performance by Neighborhood (Part 1)

  • Neighborhoods in North Philadelphia and West Philadelphia (with the exception of University City), in addition to Chinatown are underpredicted.
  • Center City and parts of Northeast and Northwest Philadelphia are overpredicted.

Model Performance by Neighborhood (Part 2)

  • This suggests local factors not included in the model are having an impact on sale prices.
  • A particular area of concern is the underpredicting of sale prices in lower-income neighborhoods in North Philadelphia.

Recommendations

  • Policymakers should be aware that houses in lower-income neighborhoods will not be accurately predicted by this model.
  • Physical features of a structure such as the number of bedrooms/bathrooms serve as strong predictors of home sale price. Nearby attractive amenities such as proximity to transit stations as well as distance from crime also contribute.

Limitations & Next Steps

Limitations

  • Philadelphia sale prices don’t have a linear relationship, particularly among lower-priced homes
  • Variables used were aggregated and not weighted

Next Steps

  • Update the model to account for lower neighborhoods and tailor spatial features to add texture and depth to the model
  • Consider more layered data cleaning, given that we minimally cleaned the data in order to preserve complex property types (mixed-use etc.)

Questions?

Thank you! from Sujan, Henry, Ryan, Kavana, Chloe, and Nina :)

Contact us at: inquiries@tnc.com