The document discusses developing a housing market forecasting model that can be applied to over 20,000 zip codes. It describes using active market data like new listings, prices, and inventory as leading indicators to predict future housing transactions up to 3 months in advance. The model was developed using least angle regression to produce linear models with interpretable coefficients for each zip code. Future work may involve adding non-linear and macroeconomic variables to increase accuracy.
2. Research Question
• How to develop a housing market
forecasting model applicable to more
than 20,000 zip codes across property
types and price quartiles?
• How to enable regular model revision
and updates as new information and
data becomes available?
3. Active Market signals future transaction
price
Home listed
$429,000 Buyer financing fails,
Inventory 49 Neighbor home Property relisted
listed $394,000
$409,000
Deal closed
$389,000
Price reduced
$398,000
Inventory 69
Transaction
Offer made
Recorded
$391,000
March May July Sept Nov Jan
Closed transaction = 1 data point, months too late
Active Market = 9 months of pricing, price changes, supply and demand, leading
indicators
6. Housing Market News: Release &
Report Dates
Source Inflection Date Data Published
Altos Research Q3-2010 Webcast Jul-10
14-Jan-11 17-Jan-11
Altos 20-city Composite Price of New Listings (weekly)
Altos 20-city Composite Median Price (weekly) 4-Feb-11 7-Feb-11
Altos 20-city Composite Median Price 25-Mar-11 28-Mar-11
(90 Day Rolling Avg)
CoreLogic HPI Apr-11 1-Jun-11
FHFA national home price index Apr-11 22-Jun-11
Radar Logic RPX Apr-11 23-Jun-11
S&P Case Shiller 20 City Composite Apr-11 28-Jun-11
7. Importance of Active Market Indicators
• ANGLIN, RUTHERFORD & SPRINGER (2003), “The Trade-Off Between
the Selling Price of Residential Properties and Time-on-the-Market: The
Impact of Price Setting,” JJREFE
• MILLER & SKLARZ (1987), “Pricing Strategies and Residential Property
Selling Strategies,” JRER
• SPRINGER(1996), “Single-family housing transactions: Seller
motivations, price, and marketing time,” JREFE
• YAVAS & YANG (1995), “The Strategic Role of Listing Price in
Marketing Real Estate: Theory and Evidence,” REE
• KANG & GARDNER (1989), “Selling Price and Marketing Time in the
Residential Real Estate Market,” JRER
8. Published research & models limited by
local data sets & time series
• Boston: GENESOVE & MAYER (2001). “Loss Aversion and Seller
Behavior: Evidence from the Housing Market,” QJE
• Stockton, CA: KNIGHT (2002), “Listing Price, Time on Market, and
Ultimate Selling Price: Causes and Effects of Listing Price Changes,”
REE
• Arlington, TX: ANGLIN, RUTHERFORD, & SPRINGER (2003), “The Trade-
Off Between the Selling Price of Residential Properties and Time-on-
the-Market: The Impact of Price Setting,” JREFE
• Columbus, OH: HAURIN, et al (2006), “List Prices, Sale Prices, and
Marketing Time: An Application to U.S. Housing Markets,” Working
Paper
9. The Data
• 400 individual statistics & leading
indicators updated weekly for 20,000 zip
codes based on the active market
• Independently calculated by property
type (single-family & condo) and price
range quartile
• Primary data with uniform methodology
for all statistics
10. Model Development
Step 1: Traditional OLS
Objective: Build models by zip code for 20k zips
Process: Test set Limited OLS Models
Outcome: Variables & coefficients changed drastically from
market to market
Step 2: Regression Trees (CART)
Objective: Increase accuracy from OLS
Process: Test set Built models for 20k zips
Outcome: No coefficients, Trees randomly generated,
Interpretability problems
Step 3: Least Angle Regression (LARS)
Objective: Increased transparency
Process: Test set Build models for 20k zips
Result: Linear model with Coefficients, Transparent, Interpretable