The Zillow Real Estate Research team has made use of our proprietary Zillow Mortgage Marketplace (ZMM) data in a few recent briefs (here and here). A natural question that a reader of these analyses may have is how representative this data is of the overall mortgage market as a whole. With this question in mind, we did what we do best—analyze our own data to show that we’re using good stuff.
First, before we get to the deeper analytics, a brief overview of ZMM data is in order. Since its launch in April 2008, over 2.3 million loan inquiries have been made on ZMM, generating over 33.5 million quotes to consumers from brokers or lenders. Each quote is a customized package that a mortgage broker offers for that particular consumer—location, size of loan, credit quality of the borrower, and loan-to-value ratio. A mortgage broker may offer multiple quotes to the same borrower, perhaps with different points associated with a different rate. Borrowers with a better credit profile typically get more quotes than those with a weaker one.
Ok, so ZMM has got plenty of volume, but is that volume representative of the overall market? How would one know? Probably the best way would be to compare aggregate ZMM rates to those produced by another commonly-accepted authority on mortgage rates. One of the most-referenced data series in the mortgage market is the weekly Freddie Mac Primary Mortgage Market Survey (PMMS) which is widely used to cite prevailing rates and trends in the mortgage sector. Freddie Mac’s PMMS is a weekly survey of 125 lenders, about 25 per region. According to Freddie Mac, this is a mix of lender types – thrifts, credit unions, commercial banks and mortgage lending companies – that is roughly proportional to the level of mortgage business that each type commands nationwide. The PMMS surveyors request the most popular rate-and-points combination offered to consumers, including the origination fee. The survey also standardizes on first-lien prime conventional conforming mortgages with a loan-to-value of 80%; there is no specification about the borrower’s credit profile.
Freddie Mac surveyors collect data from Monday through Wednesday and the results are posted on Thursdays. Average rates and points (and margin for ARMs) for each product are reported for the nation and each of the five Freddie Mac regions (Northeast, North Central, Southeast, Southwest, and West). A national average is calculated by using regional weights. The regional weights are based on Home Mortgage Disclosure Act (HMDA) data on origination volume and updated about once a year.
Freddie Mac includes both discount points and origination points in their “points” report. Discount points are points paid on the principal of the loan to get a lower rate. Origination points are the fees that the mortgage banks charge to cover the cost of making the loan. Both of these charges are expressed in terms of percentage of loan size.
Now that we fully understand what’s under the covers of the Freddie Mac data, let’s see how ZMM data compares to it. Using the Freddie Mac methodology, I created an analogue to the PMMS time series using ZMM data. Like Freddie Mac, I sought rates for conforming mortgages on a primary residence with an 80% loan-to-value ratio. In order to control for buyers with better credit (who tend to get more quotes), the simulation collapsed each loan inquiry into one observation by taking the mean APR, Rate, and Points of all quotes for a given inquiry. I then took an average across all inquiries for a region, thus giving a regional average. Weekly national averages are calculated by using the regional averages and the regional weights from HMDA data.
The first Freddie Mac data set we used for comparison with ZMM data was the 30-year fixed rate mortgage (“30yr fixed”), as that is widely considered the benchmark mortgage product (see the chart below).
The blue lines are Freddie’s data and the red are ZMM. ZMM rates track Freddie’s very closely, but our mortgage brokers consistently offer 30-year fixed rate mortgages with lower points (usually about 0.2-0.4 points lower) to customers. Statistical analysis of the rates show that Freddie’s 30yr fixed and ZMM’s 30yr fixed have a 98% correlation and the median absolute value of error is 7.9 basis points.
Why would brokers quote consistently lower points in ZMM versus in the PMMS survey? A plausible explanation is that the nature of our marketplace requires the brokers to be more competitive. When mortgage brokers are directly dealing with customers in a transparent marketplace alongside other brokers, they are more likely to offer their most attractive rates and fees. The evidence here suggests that a key element on which brokers compete in a transparent marketplace is the points charged to the consumer for either the rate or their services.
Turning now to 15-year fixed rate mortgages (“15yr fixed”), I used a similar approach as before for analyzing this product. The same specifications for eligible loans were used: conforming, first-lien purchase mortgages with an 80% loan-to-value ratio. For the 15yr fixed product, ZMM and Freddie Mac showed an even stronger correlation at 99% with a median absolute value of error also of 7.9 basis points (see the chart below). We see the same trends in the ZMM rates and points for 15yr fixed as we saw with 30yr fixed: there is high correlation between the rates, but the mean points and fees that are quoted on ZMM are about half what Freddie Mac reports as the national average.
Freddie Mac’s PMMS includes two additional mortgage products— 5/1 hybrid amortizing adjustable rate, and 1-year amortizing adjustable rate mortgage products. Unfortunately, these are not yet as popular products with our ZMM customers, and we do not yet have consistently large enough sample sizes to build a solid time series of data, so we will restrict ourselves to using the 30yr fixed and 15yr fixed for analysis projects until we get larger samples on the other products and can ensure data quality. Our datasets on 30yr fixed and 15yr fixed are large enough and the rates mirror the PMMS closely enough that we are very comfortable using our data for analytic purposes.
Additionally, one noteworthy benefit of our proprietary data is that it is nearly real-time since it is derived from an active, online marketplace as opposed to a weekly survey. As mentioned above, Freddie Mac’s PMMS is released each Thursday and covers rates that are quoted Monday through Wednesday. When there are rapid changes in the mortgage market, this one week lag in data may fail to give an accurate picture of where the market is and how it’s moving. To illustrate this difference, I compared the ZMM daily rate on a 30-year fixed mortgage with Freddie Mac’s weekly rate for the period from October to November 2010 (see chart below). Our sample size of mortgage requests on any given day ranged from 255 to 1123, with the mean number of queries per day being just over 750 (with a substantially higher number of actual quotes).
In early November, there was a run-up in Treasury yields generally thought to be fueled by concerns about future QE2 purchases by the Federal Reserve (yields on the 10-year Treasury note increased 37 basis points between November 4-18). These increases also drove a rise in 30yr fixed rates over roughly the same period. From November 4 to November 17, 2010, ZMM data showed a day-by-day, almost linear climb of 31 basis points (from 4.07% to 4.39%) in the 30yr fixed. PMMS, on the other hand, only has two data points for these 14 days, and their most recent data point on November 17 was 4.17%. The next day, Freddie Mac reported a 22 basis point jump to 4.39%. ZMM data shows that this jump happened over the course of two weeks, while the PMMS shows overnight. So, not only is ZMM data representative of the overall national mortgage market but it is available with much lower latency than the commonly used Freddie Mac data.
In the near future, we’ll be trying to integrate more mortgage data into the Zillow data portal but, in the meantime, if there are academic or industry analysts who would like to obtain ZMM time series data, please contact us at research@zillow.com.
Photo of calculator courtesy of Flickr.
