The New York rental market — for the people who rent, and the people who own.
A real-estate client wanted the market read for two audiences. I analysed 17,614 NYC listings to answer both: for renters, where the value is; for owners, what sets your price — and, separately, what actually gets you booked. The surprising part: those last two aren't the same lever.
The brief was open-ended: read the NYC rental market to inform renters and owners. Exploring the data first — price by room type, price by location, and a correlation check — surfaced one surprise that shaped everything: price and how often a listing books barely move together. That turned a vague brief into three sharp questions.
renters & owners — the pricing question
owners — the demand question
can you book more by pricing lower?
Two regression models answered Q1 and Q2; a clustering model turned the answers into market segments. Here's what each found — for renters first, then owners.
Median nightly price by neighbourhood
Darker = pricier. Manhattan and prime Brooklyn command the premium; the outer boroughs run less than half. Switch the metric, or hover any neighbourhood.
Median price. The premium is geographic and steep — West Village, Chelsea and the Financial District sit above $200, while much of the Bronx, Queens and outer Brooklyn fall below $80. For a renter, the value is one subway ride out.
Room type is the biggest price lever
Before location even enters, what you rent halves or doubles the price.
Cheapest neighbourhoods for value-hunters
Same city, a third of the price — the outer-borough value end.
What sets your price
A model predicting price (R² 0.42) leans almost entirely on two structural choices — location and room type.
Takeaway: price is set by where the listing is and what it is — set those right and you've priced correctly before touching anything else.
What gets you booked
A separate model for occupancy (R² 0.59) tells a completely different story — and price is barely in it.
Price and occupancy correlate at just −0.03 — effectively zero. What fills a calendar is availability and reviews, not a lower nightly rate. Owners chasing bookings with price cuts are pulling the wrong lever.
K-means on standardised behavioural features (price, occupancy, availability, reviews — identifiers excluded) splits the market into four groups that price alone can't separate. The revealing one is the last.
Three models did the work — a random forest predicting price, a second predicting occupancy, and K-means for the segments. The driver bars above are those models' output; these charts show the models themselves.
Price vs bookings — the proof behind −0.03
Every point is a listing, coloured by segment. The line is the fitted trend — nearly flat, because price and how often a listing books barely relate. Raising or cutting the rate does almost nothing to occupancy.
Choosing the number of segments
The "elbow" method: within-cluster spread as clusters increase. It bends at four — enough structure to be meaningful without over-splitting.
How well the price model predicts
Test-set listings: predicted vs actual price. On the dashed line = a perfect call. R² 0.42 — honest: it holds the mid-market and softens at the luxury tail, where price rides on things the data doesn't hold.
For renters: value sits one borough out, and a room instead of a whole home halves the price. For owners: your rate is set by location and room type, but your occupancy is won on availability and reviews — not price. Nearly a third of listings are "dormant" — always available, almost never booked — the clearest sign that price is not the thing holding them back.
Limitations
- Cross-year signals. Occupancy is measured in 2019 and availability in 2020 — a fair proxy for demand, but not the same window.
- Price is 42% explained. Location and room type dominate, but the rest sits in things this dataset doesn't hold — photos, amenities, host quality.
- No seasonality yet. A single snapshot can't show how demand shifts month to month.
- Neighbourhood coverage. The map shades the neighbourhoods with enough listings to read reliably.
Next steps
- Add the demand time-series. A monthly index would show when to raise or hold the rate, not just how to set it.
- Test the dormant fix. Run a reviews-and-photography push on a sample of dormant listings and measure the lift in bookings.
- Enrich the price model. Bring in amenities and photo quality to close the gap the current model leaves.