House Prices and Rents: Micro Evidence from a Matched Dataset in Central London by Philippe Bracke
1. House Prices and Rents
Micro Evidence from a Matched Dataset in Central London
Philippe Bracke
London School of Economics
PyData 2014, London (Feb 23)
2. About me
Studied economics
Wanted to become a theoretical
macroeconomist
PhD: discovered the joys of data analysis
Python (and R, Stata)
Current research focus Housing markets
Twitter @PhilippeBracke
10. Land Registry Price Paid data
All registered property sales in England and Wales, 1995–2013
→ 18.5m records, freely available!
full address
price paid
date of transfer
property type: Detached, Semi, Terraced or Flat/Maisonette
new build or not
freehold or leasehold
http://www.landregistry.gov.uk/market-trend-data/public-data/
price-paid-data
12. The problem: Data on private rents
Rental data are much less available than house price data
A gap exists in official private rental statistics with no
official private rental index currently available
The National Statistician’s Review of Official Housing Market
Statistics, September 2012
13. The problem: Data on private rents (cont’d)
The Office for National Statistics (ONS) released in 2013 an
experimental quarterly index of the private rental market
The index is based on individual rental data from the
Valuation Office Agency (VOA), who deploys rental officers to
collect the price paid for privately rented properties
This data is not publicly available
14. John D Wood & Co.
Rental Dataset
Real estate agency with 14 London offices and 6 offices in the
South-East of England
Focus on upper market: Central/South-West London and
countryside
15. John D Wood & Co. (cont’d)
Rental Dataset
new contracts, no
roll-overs
internal records +
exchange of data with
other agencies
18. Matching issues
Address format
Land Registry
Clean and easy:
postcode W2 3DB
paon 5
saon FLAT K
street WESTBOURNE CRESCENT
Ambiguous:
postcode UB4 8FJ
paon MARSH COURT, 561
saon 4
street UXBRIDGE ROAD
Agency data
Clean and easy:
hsename Flat K
hseno 5
address1 Westbourne Crescent
postcode W2
Ambiguous:
hsename
hseno 2
address1 Rupert House
address2 Nevern Square
19. Matched dataset
Construction
try as much as possible to harmonise the two datasets
all variables in upper case letters as in LR
rename “hseno” as “paon”, and “hsname” as “saon”
join together all transactions sharing the same “street”,
“paon” and “saon”
Rule 1 for each sale, keep the closest rent
Rule 2 for each rent, keep the closest sale
24. Matched dataset
Rent-price ratio over time
.02.04.06.08
01jul2006 01jan2008 01jul2009 01jan2011 01jul2012
R/P ratio 10−year UK Government Bond Yield
25. Matched dataset
Rent-price ratio vs. property value
0.02.04.06.08.1
0 1000 2000 3000 4000
Price (in £1,000)
Rent−price ratios vs Prices
0.02.04.06.08.1
0 500 1000 1500 2000 2500
Rent (in £ per week)
Rent−price ratios vs Rents
26. Matched dataset
Rent-price ratio vs. property type
.02.04.06.08.1
0 1000 2000 3000 4000
Price (in £1,000)
Rent−price ratios vs Prices (Apartm.)
0.02.04.06.08.1
0 1000 2000 3000 4000
Price (in £1,000)
Rent−price ratios vs Prices (Houses)
.02.04.06.08.1
0 1000 2000 3000 4000
Floor area (sqft)
Rent−price ratios vs Floor areas
NW1
NW3
NW8
SW1
SW10
SW11
SW3
SW5
SW6 SW7
SW8
W1W10
W11W14
W2
W8
W9
.046.048.05.052.054.056
400 600 800 1000 1200
Average Price (in £1,000)
Rent−price ratios vs Prices (by Postcode)
Patterns confirmed by multivariate regression:
Rent
Price
= α + Type β1 + Size β2 + Location β3 + Date β4 + ε
27. Depreciation/maintenance costs and rent-price ratios
Rent
Price
= rf + δ − g + m
House = land + structure
More expensive locations: higher land share ⇒ Rent
Price ↓
29. How to measure future appreciation and risk?
Rent
Price
= rf + δ − Eg + m
Need to find future sales and/or rentals of the same property
→ Match within-Land Registry or within-Agency data
easier
Repeat sales: not frequent
Repeat rentals: many
30. The effect of future appreciation and risk
Sales
Rentals
Matched Dataset Matched + Repeat Rentals Dataset
1,922 properties 859 properties
Max gap = 180 days
Average gap = 85 days
Max gap = 2,360 days
Average gap = 578 days
Regression results
One-standard deviation higher future rent appreciation
⇒ Rent
Price ↓ by 1.6%
Ambiguous results on rent volatility (one measure of risk)
32. Summary
Novel dataset on prices and rents in Central London
Measure rent-price ratios directly for matched properties
Find lower rent-price ratios for expensive properties
→ Effect of size
→ Effect of location
and other effects
Consistent with economic theory
33. Next steps
The Land Registry is a recent open data resource with huge
potential
Can be matched with many other datasets
private datasets
public housing-related websites
Let’s collaborate!
Github, philippebracke
Thank you!