Saturday, May 02, 2015

Lawler on Housing Vacancy Survey: More “Stunning” Results, But Tough to Interpret

by Bill McBride on 5/02/2015 03:06:00 PM

A technical note from housing economist Tom Lawler:

Earlier this week the Census Bureau released its “Residential Vacancies and Homeownership” Report (commonly referred to as the Housing Vacancy Survey, or HVS) for the first quarter of 2015, and while the results were not as “eye-popping” as those in the previous quarter, they were nevertheless – if reflective of actual housing trends – stunning.

The HVS estimate of the US homeownership rate continued its recent rapid descent last quarter, and the 1.2 percentage point decline over the last four quarters is the largest four-quarter decline since the HVS began. The YOY declines in the HVS HOR’s were especially steep for householders 44 years old or younger.

On the housing inventory front, the HVS estimate of the number of occupied homes (or households) declined by 407,000 in the first quarter following the all-time record 1.337 million jump in the fourth quarter of last year. After adjusting for normal seasonal patterns, the HVS household estimate fell by about 60,000 last quarter after jumping by a record 1.17 million in the previous quarter.

Here are some summary stats from the latest report.

Selected Statisitcs, Census Housing Vacancy Survey
  Q1/2014Q2/2014Q3/2014Q4/2014Q1/2015YOY Change
Rental Vacancy Rate8.3%7.5%7.4%7.0%7.1%-1.2%
Homeowner Vacancy Rate2.0%1.9%1.8%2.0%1.9%-0.1%
Gross Vacancy Rate13.8%13.6%13.5%12.6%13.0%-0.8%
Homeownership Rate64.8%64.7%64.4%64.0%63.7%-1.1%
Homeownership Rate SA65.0%64.7%64.3%64.0%63.8%-1.2%
Housing Stock (000')133,087133,209133,331133,453133,575488
Occupied Housing Stock114,762115,127115,310116,647116,2401,478
Owner74,40474,45874,24074,60674,018-386
Renter40,35740,66941,07042,04142,2221,865
Homeownership Rate by Age of Householder
< 3536.2%35.9%36.0%35.3%34.6%-1.6%
35-4460.7%60.2%59.1%58.8%58.4%-2.3%
45-5471.4%70.7%70.1%70.5%70.1%-1.3%
55-6476.4%76.4%76.6%75.8%75.8%-0.6%
65+79.9%80.1%80.0%79.5%79.0%-0.9%

The fourth quarter jump in the HVS estimate of total households was eye-popping, but there are several reasons to question the magnitude of the jump. First, in the fourth quarter of 2014 the HVS estimate of the number of “seasonal” housing units fell by a record 453,000 from the previous quarter, only to jump back up by 385,000 in the subsequent quarter. Since the HVS “controls” it housing units by category to independent estimates of the overall housing stock, this “strange” swing in the number of “seasonal” housing units may have artificially boosted the occupied unit category in the fourth quarter.

Second, beginning in April 2014 the HVS began “phasing in” in new sample which may have “artificially” increased household estimate gains since the early part of 2014. Here is an excerpt from the HVS “Sources and Accuracy” document.

“Beginning in April 2014, a new sample is being phased in over a 16-month period. The methods used to select the sample households for the survey are evaluated after each decennial census. Based on these evaluations, the design of the survey is modified and systems are put in place to provide the sample for the following decade. The previous decennial revision incorporated new information from Census 2000 and was complete as of July 2005. The design for the entire decade was selected from the 2000 based sample. The current revision incorporates new information from Census 2010 and will be complete in July 2015. The new sample is based on the Master Address File (MAF) compiled during the 2010 Census and will use annual selections from the MAF instead of the once a decade sample selection used previously.”

Comparisons between Decennial Census results and the HVS clearly indicate that the HVS estimates have not reliably reflected overall US housing trends in a number of important respects, including (but not limited to) overall vacancy rates (overstated), homeownership rates (overstated), and the distributed of households by age (overstated share of younger householders). These HVS errors widened from 2000 to 2010. It is not clear how much of the HVS estimation problem has been related to its “sampling frame” and how much is related to “non-sampling” errors. To the extent that the new sample better reflects the actual characteristics of the US housing stock, however, then one would expect that the phasing-in of the new sample would, among other things, result in (1) faster than “actual” household growth; (2) faster than “actual” growth in renter households relative to owner households; and (3) larger than “actual” declines in the homeownership rate.

My understanding is that Census “phases in” the new sample over such a long period of time so that there is not a “really big” discontinuous” shift it the CPS/HVS time series. If there are huge differences between the old and new samples, however, then even with such a “phasing in” there can be discontinuities in time series estimates. It would be nice, however, if Census would show a comparison of HVS results using the old sample and HVS results using the new sample so that analysts could assess the “sampling” vs. “non-sampling” errors of HVS results themselves.

While recognizing that the overall change in owner-occupied vs. renter-occupied housing units as estimated by the HVS over the last year may not reflect “actual” changes, I thought some folks might be interested in HVS-based estimates of owner- vs. renter-occupied housing units by units in structure. The HVS does not actually publish such estimates, but in detailed tables HVS shows estimates of (1) the distribution of rental and owner housing units (renter occupied plus for rent or owner occupied plus for sale only) by units in structure, and (2) vacancy rates by units in structure for renter and owner housing units. As such, one can produce the implied HVS occupied renter and owner housing units by units in structure. The HVS includes one-unit manufactured/mobile housing units in its one-unit structure category, so this category is not the same as what analysts generally consider “single-family” housing units. Here is a comparison of the first quarter of 2015 with the first quarter of 2014.

HVS-Based Estimates of Owner and Renter Occupied
Housing Units by Units in Structure (000's)
  Q1/2014Q1/2015Change
Renter-Occupied Total40,35742,2221,865
One-Unit15,83116,572741
2-4 Units7,7538,090337
5-9 Units5,0515,300249
10+ Units11,72212,260538
Owner-Occupied Total74,40474,018-386
One-Unit70,92070,518-402
2-4 Units1,3351,34712
5-9 Units574563-11
10+ Units1,5751,59015

HVS estimates suggest that the trend toward more single-family units being rented rather than owned continued over the last year. Compared to the fourth quarter of 2005, HVS estimates suggest that the number of owner-occupied one-unit homes fell by about 1.969 million units, while the number of renter-occupied one-unit housing units increased by about 4.345 million units.