Historical US Housing Data Availability

Question

Papers like Ng and Moench (2011) and Negro and Otrok (2007) use historical housing data to estimate latent variables that correspond to some "true" or "truer" measure of house prices or housing activity. I've put together a table of series one could use as inputs in such an exercise. However, as we can see in the table below, very little data is available before the mid-1970s. I would like to expand this list with U.S. historical data (free or to purchase) that help to characterize real and financial housing activity as far back as possible. What other series am I missing?

+------------------------------------------------------+--------------+--------------+------------+
|                        Series                        |    Source    |  Frequency   | Start Date |
+------------------------------------------------------+--------------+--------------+------------+
| Median Sales Price of Single-Family Existing Homes   |  NAR         |  Monthly     |  Jan1968   |
| Median Existing Home Prices                          |  NAR         |  Monthly     |  Jan1999   |
| Median Sales Price of Single-Family New Homes        |  CENSUS      |  Monthly     |  Jan1963   |
| Average New Home Prices                              |  CENSUS      |  Monthly     |  Jan1975   |
| S&P National Home Price Index                        | S&P          |  Quarterly   |  Q1-1987   |
| Freddie Mac House Price Index                        |  FHLMC       |  Quarterly   |  Q1-1975   |
| FHFA Purchase-Only Index                             |  FHFA        |  Monthly     |  Jan1991   |
| FHFA Home Prices                                     |  FHFA        |  Quarterly   |  Q1-1975   |
| CoreLogic National House Price Index                 |  CoreLogic   |  Monthly     |  Jan1976   |
| Housing Starts                                       |  CENSUS      |  Monthly     |  Jan1959   |
| Existing Home Sales                                  |  NAR         |  Monthly     |  Jan1999   |
| New One-Family Houses Sold                           |  CENSUS      |  Monthly     |  Jan1963   |
| Housing Units Authorized by Permit                   |  CENSUS      |  Monthly     |  Jan1960   |
| Housing Units Under Construction                     |  CENSUS      |  Monthly     |  Jan1970   |
| Housing Completions                                  |  CENSUS      |  Monthly     |  Jan1968   |
| Private Residential Fixed Investment (PRFI)          |  BEA         |  Quarterly   |  Q1-1947   |
| Wilshire REIT Total Return Index                     |  Wilshire    |  Monthly     | Jan1978    |
| Ziman-CRSP Residential REIT Value Weighted Index     |  CRSP-Ziman  |  Monthly     | Jan1980    |
| Ziman-CRSP Residential REIT Equally Weighted Index   |  CRSP-Ziman  |  Monthly     | Jan1980    |
+------------------------------------------------------+--------------+--------------+------------+

Commercial real estate data is not ideal, particularly data that is contaminated with mine and oil and water well construction. Regional data is fine but national data is preferred.

Answer 1

I feel this barely qualifies as an answer, but doesn't Shiller have house prices data on his webpage? Currently here.

Specifically:

Historical housing market data used in my book, Irrational Exuberance [Princeton University Press 2000, Broadway Books 2001, 2nd edition, 2005], showing home prices since 1890 are available for download and updated monthly: US Home Prices 1890-Present.

I don't have much experience with this data directly, and am only aware of it through Khandani, Merton, Lo (2013) ("Refinancing Ratchet Effect"), where they use this and many other data sources to construct a (very long) housing dataset for their simulation analysis.

Details of Case Shiller Series Construction methodology (from the end-notes of the 2006 edition for chapter 2):

Even though there were no regularly published home price indexes before the 1960s, some economists were constructing indexes of home prices that cover most of the years since 1890. We found home price indexes from 1890 to 1934 and from 1953 to the present that used in their construction some device to attempt to hold the quality of the home constant.

The nominal home price index 1890–1934 is from Leo Grebler, David M. Blank, and Louis Winnick, Capital Formation in Residential Real Estate: Trends and Prospects (Princeton, N.J.: National Bureau of Economic Research and Princeton University Press, 1956). It is a repeated-measures index based on a survey of homeowners in twenty-two U.S. cities, who were asked to give the value of their home in 1934 and the date and price of the purchase of that home. Since it is based on repeated measures of individual homes, the series is protected, in contrast to the simple median price, from any bias from changes in the mix of houses sold or of the increasing size and quality of newer homes. Its shortcoming is that it depends on memories of the surveyed homeowners for the earlier purchase price.

The nominal home price index that we constructed for 1934–53 is a simple average over five cities of median home prices advertised in newspapers. The cities are Chicago, Los Angeles, New Orleans, New York, and Washington, D.C. My students collected the data from microfilmed newspapers at the Yale University library, collecting approximately thirty prices for each city and year, except that for the fifth city, Washington, D.C., 1934–48, data came from a median price series from E. M. Fisher, Urban Real Estate Markets: Characteristics and Financing (New York: National Bureau of Economic Research, 1951). The median series for 1934–53 does not make any attempt to correct for home quality change, as do the indexes we use for the other subperiods. Improvement in home size and quality gives median home price an upward bias, and this is why I avoided using median price outside the 1934–53 interval.

The nominal home price index for 1953–75 is the home purchase component of the U.S. Consumer Price Index (CPI). The Bureau of Labor Statistics collected data on home prices for those years for homes that are held constant in age and square footage. In the 1980s they discontinued this index when they switched to a rental equivalence basis for housing in the CPI. They made this change to correct what was considered a conceptual flaw in the housing component of the CPI: the CPI is supposed to be a price of consumption goods and services, not of investment assets. For our purposes, however, the old home purchase component is acceptable. There are, however, some shortcomings in the home purchase component, notably that it is based only on homes with certain government-subsidized mortgages, and the procedure that the Bureau of Labor Statistics used to correct for changes in the ceiling on these mortgages was not optimal. See J. S. Greenlees, “An Empirical Evaluation of the CPI Home Purchase Index 1973–8,” American Real Estate and Urban Economics Association Journal, Carnegie-Rochester Conference Series on Public Policy, 17 (1982): 203–38.

The nominal home price index for 1975–87 is the U.S. home price index published by the U.S. Office of Housing Enterprise Oversight (OFHEO), which is available on their Web site. It is a repeat sales index, and thus controls for quality change. The nominal home price index for 1987–2004 is the repeat-sales U.S. home price index produced by Fiserv CSW, Inc., successor to Case Shiller Weiss, Inc.

Since 1987, the CSW and the OFHEO series have shown very similar patterns through time, though the sharpness of the increase is slightly more pronounced in the CSW series, an observation we attribute to the fact that our series is based only on actual sales, while the OFHEO series also uses both actual sales and appraised values as if they were sales. Appraised values are somewhat sluggish to respond to new market conditions.

Answer 2

I found an additional series.

There is a series Nonfarm Housing Starts (HOUSTNF on FRED) that runs monthly from 1946-01-01 to 1969-12-01. There is about 10 years of overlap with the new series Housing Starts: Total: New Privately Owned Housing Units Started (HOUST on FRED). These series have a correlation coefficient of over 95 percent in their overlapping period.