Zillow’s dashboards are fine if you only need a headline number. I wanted something I could interrogate, remix, and extend, so I published my own repeat-sales toolkit for New York City. You can explore the live chart at ivoytov.github.io/manhattan and read the code at github.com/ivoytov/manhattan.

What the Code Does

1. Ingests & cleans two decades of NYC sales data, standardizing columns and filtering impossible prices or obvious duplicates.
2. Builds unique IDs for each property (block/lot/apartment) so repeat transactions line up across time.
3. Flags outliers using log price deltas adjusted for the time between sales.
4. Runs a repeat-sales regression with time based weights to estimate period over period price moves.
5. Slices the index by borough, neighborhood, property type, and price tier so you can zoom into the segments you actually trade.

How It Differs From Zillow’s ZHVI

• Data comes straight from NYC public records, not a black box AVM.
• Methodology is a transparent repeat-sales regression, similar to Case-Shiller.
• Segmentation is unlimited — if you can define the cohort, you can model it.
• Every transformation is auditable, so you can rerun the exact pipeline later.

Why It’s More Customizable

• Adjust filters (price bands, property buckets, time windows) with a quick edit.
• Swap in your own residual weighting or regression approach.
• Change the reporting cadence to weekly, monthly, or custom periods.
• Fork it for your favorite borough or roll it forward as new deeds drop.

Use Cases

• Policy analysis across boroughs or income bands.
• Hyper local investor updates for clients or LPs.
• Academic housing studies that require reproducible methods.
• Automated reporting in research decks or newsletters.

Grab the Code

Everything lives in a single Julia project. Clone it, tweak the thresholds, and send a PR if you find a smarter way to normalize the data. I am always up for ideas on segmentation, visualization, or alternative weighting schemes.