/ric-house-flip

Turning property transfer data into stats on house flipping developers

Primary LanguageJupyter Notebook

Richmond VA house flipping

Data processing

This data is mostly from the City of Richmond Assessor's Office property transfer reports, which go back to 2012.

  • get_all_transfers.ipynb - combines all of the Excel files into one dataframe and saves it
  • flip analysis.ipynb - does most of the work
  • data/ - this directory is mostly empty in this repo, but locally I store working data files here
  • data/flips.csv - the output from data processing

In the flip analysis notebook, I applied some data filters and transforms:

  • Filter to single-family homes (property classes 101, 110, 115, 120, 130, 150)
  • Group by parcel ID + sale to combine extension cards (e.g. see the extensions tab for this parcel )
  • To identify flips:
    • self-join on parcel id and grantee=grantor (ie, the buyer resold it)
    • where the property was held for between 0 and 2 years before reselling
    • removed most of the foreclosures / multiple parcel sales / special financing, etc.
    • removed new construction (<2 years old)
  • Added some fields - hold_time, price_diff, size_diff, year, age, council, roi

I also made some plots, like this one showing how the gross ROI varies by council district and year. If you're not familiar with the local geography, districts 1, 2, and 4 are statistically wealthier and whiter than the rest of the city.

Gross ROI % is calculated as the price difference (second sale - first sale) divided by the first sale. 100% means the second sale was twice as much as the first.

A messy line plot of the house flipping gross ROI in Richmond VA, by council district and year. Wealthy districts (1, 2, 4) give low ROI; districts 6, 7, 8, 3 can sell flips for twice what they paid.

Maps

To make this map, I joined the CSV to the Parcels shapefile from the city's GeoHub. The parcels are to-scale polygons, which isn't very convenient for heatmaps and also emphasizes the relative size of each parcel, so I converted the layer to geometry centroids, and applied an equal-count 5-class gray-red colormap based on the flips_roi column.

A map of RVA showing about 3500 flipped houses over 8 years. There are large red clusters in Northside and East End, with smaller clusters across Southside - they indicate large increases in the cost of houses.

Since several people asked about it, I also overlayed the 1923 HOLC "redlining" grades. As expected from my previous homeownership and redlining project, the clusters of high gross ROI are mostly in grade D areas with high "negro" populations, and in adjacent grade C areas. The gross ROI is the gross profit relative to the house price, and indicates the cost increase or "markup" that the flipper added to the house.

Similar to the previous map, but with an overlay of the 1923 HOLC map, showing the large red clusters mainly in historically black, devalued Grade D areas