Note: The Zillow Rent Index initially debuted in early 2012. Beginning with publication of July 2019 data, the methodology was updated to better reflect the true rental stock and to capture small market changes. For comparison’s sake, our original methodology can be found here.

The Zillow Rent Index (ZRI) is a dollar-valued index intended to capture typical market rent for a given segment (IE, multifamily or single-family units) and/or geography (IE for a given ZIP code, city, county, state or metro) by leveraging Rent Zestimates.

Historically, the ZRI was essentially calculated as the median Rent Zestimate of that segment. Recently, we revisited and updated this methodology to both better account for the stock of homes that actually rent and to capture small but important changes in rent levels.

These changes, summarized below, have been incorporated into a new Zillow Rent Index methodology that will apply to all published ZRI data series going forward. The largest changes include:

Reweighting to better reflect the rental housing stock
Taking the mean of the middle quintile instead of a median

Individual Rent Zestimates are computed for all homes, whether or not those homes are currently or ever were available for rent. Indeed, many of the homes for which we compute Rent Zestimates, especially single-family homes, are neither for-rent nor renter-occupied. Because single-family homes are, on average, higher-valued than apartments, including all of them in the ZRI and computing a median based on that full universe of homes creates an upward bias in the index. With our recent adjustments, we now reweight subsegments by the Census Bureau's American Community Survey (ACS) counts of units of different housing types so the sample is representative on the dimension the survey captures, namely number of units in the building and decade the building was built.

We also re-worked our methodology to calculate ZRI as the mean of the middle quintile of included Rent Zestimates, rather than the median. Doing so better captures small changes in the market, while also reducing noise. Rents are unevenly distributed and cluster around round numbers just below changes in the leading digit: there are many listings at $900/mo, for example, but few at $1013/mo or $842/mo. This is reflected in the algorithm that computes individual Rent Zestimates, but is somewhat problematic when creating a median-based index.

Imagine a region with 10,000 rentals where the median rent value was $900/month. If 1,000 of those rentals had exactly $900 as the median rent, the distribution could change quite a bit before the median would change. Conversely, if the median is near the $900/$1000 boundary, with no rents in between, index figures thrash back and forth between $900 and $1000 even if there is little change in the distribution. Taking the mean of the middle quintile alleviates this problem. Why not just take the mean? Rent price distributions are very skewed, so the mean is heavily influenced by a small number of expensive properties, and therefore does a poor job capturing changes in 'typical' rents.

We believe these changes will better reflect both the actual level and predominant trends in the rental market, while also reducing noise. Because of re-weighting, the ZRI level in most regions has shifted down by about 10% compared to prior index calculations. Otherwise, the effects are heterogenous and mostly small.

Methodology

Property Universe

Every month, a Rent Zestimate is created for more than 100 million U.S. housing units for which Zillow has sufficient data. The sources of this data include public records (property taxes, transactions), real estate listings and user-generated data. Real estate listing data comes from local Multiple Listings Services and/or direct feeds to Zillow from real estate brokers. User-generated data includes rental listings and for-sale listings posted directly on Zillow, and user corrections to incorrect and/or out-of-date data from listings and public records. Properties enter and leave this 'universe' due to many reasons, the most important of which are discussed below.

As the Zillow rental business and our data sources expand, we may become aware of more individual units within a building and add them to our “rental universe.” Periodically, these are retroactively given Rent Zestimate histories using attributes of the unit and the Rent Zestimate model. We take advantage of these retroactive histories in the first calculation of the Zillow Rent Index using our updated methodology.

Index creation

To create the index, we consider every home within a given region for which we have a Rental Zestimate for that period, then reweight and aggregate those estimates.

Weights are derived from the most recent 5-year ACS counts of renter-occupied housing units by decade built and number of units in the building available at the reported date (ACS file B25127, “TENURE BY YEAR STRUCTURE BUILT BY UNITS IN STRUCTURE”). These counts are compared to the number of units for which we have Rent Zestimates in the same region. For instance, if ACS states that 30% of rented units in a given county are in 2-4 unit buildings, but only 15% of our rent zestimates in that county are for 2-4 unit homes, the weight for each of those rent zestimates is 2. For regions where ACS data does not exist, we use county-level weights.For regions that overlap multiple counties, we sum over all counties overlapping with that region to compute weights. Therefore the weights are subject to:

And:

Where w is the weight and r is the rent zestimate, with i ranging over properties in the region. Using the weights, we can now compute weighted n-tile cutoffs:

Where t1 and t2 are tier thresholds (for low tier: 0.15, 0.35; high: 0.65, 0.85; middle and all others: 0.4, 0.6). For properties with weights satisfying the above, we compute a mean:

Which is our unsmoothed index for a given region and date.

After the index is produced, an exponentially weighted moving average is applied to smooth the series. Finally, we use a set of heuristic rules based on statistics of the time series in order to identify time periods within each series that likely have low signal-to-noise. We do not publish those time periods, which may cover the entire time series.

Tiers

The middle tier is the same as the headline number (mean of rent zestimates between the 40th to 60th percentiles). The low tier is the mean of the rent zestimates falling between 15th to 35th percentiles within a given region and the high tier is the mean between the 65th and 85th percentiles.

Available Segments

Currently, we compute indices at multiple geographic region levels: National, State, Metro, County, City, Zip, and Neighborhood.

At each region level, we compute for several home types:

Single-family
Condo/CoOp
Single-Family + Condo/CoOp (“all homes”)
Multi-Family
Single Family + Condo/CoOp + Multi-Family (“all homes plus multifamily”)

Additionally, for “all homes plus multifamily,” we compute tiers (“low,” “middle,” “high”) and cut by bedroom count.

The re-weighting procedure is only done for “all homes plus multifamily” and the multi-family series. Not all computed series are published, and all are subject to revision.

The post Methodology: Zillow Rent Index (2019 Update) appeared first on Zillow Research.

via Methodology: Zillow Rent Index (2019 Update)