Chart(s) of the Week: Housing Starts

    "Knowledge that is not quantifiable is of a meager and uninteresting kind."
    Lord Kelvin

    "Any figure that looks interesting is probably wrong."
    Sir Claus Mosley, Presidential Address to the Royal Statistical Society


Today we re-start a new feature, a popular category in the old Wisconsin Real Estate Viewpoint, namely the Chart (or charts!) of the Week.

The roughly 3,000 students I've taught over my career, and many colleagues can confirm that I'm a numbers freak. I also like to provide students and colleagues with my constructive -- some might say annoying -- suggestions for improving their data presentation.

As noted in the previous post, recently I've been spending some time working on a new edition of A Primer on U.S. Housing Markets and Housing Policy, coauthored with my friend Richard Green (also impresario of Richard's Urban Blog). For the second edition, we're pleased that Paul Carrillo joins us as the third coauthor.

What better place to start than updating the iconic chart of U.S. housing starts back to 1890! I never get tired of this chart.  Fascinating stuff:

Imagine jumping into the TARDIS and returning to look at these data in, say, 1960. Analysts of the time could have quite reasonably thought of the postwar boom up to a level approximating 1-2 million starts per year as a temporary phenomenon, while the country caught up to the backlog from the 30s and 40s. It would have been a farsighted thinker indeed who would have foreseen how broad postwar increases in income, changes in building technology (think Levittowns and other innovations in development and construction) as well as the expansion of the availability of housing finance, along with the baby boom and other demographics would have lead to the higher, if very volatile, levels of housing starts for another five decades.

Let's dig a little deeper. This first chart has three lines: the red line shows private housing starts; the green line shows public housing starts; and the blue line shows manufactured housing shipments. All three are in thousands of units started (or shipped). We'll focus mostly on private starts at first.

Around the turn of the century -- pardon me, I'm an old person, around the turn of two centuries ago, 1900 -- housing starts were bumping along at around 300,000 units per year; around 1905 they bumped up to around 500K per year. Students of The Panic of 1907 will be interested to find that this financial crisis had minimal impact on housing starts, maybe partly because at the time few households took out mortgages, and those were usually for perhaps a third of the purchase price. Housing starts did start to fall a few years before the U.S. entered WWI; and the 1918 trough in starts, 118K, remains the record low for the 120 years of data we examine.

Post WWI, starts boomed, hitting a quite substantial peak of 937K in 1925. They started to slide well in advance of the stock market crash of 1929, and fell further during the early years of the Great Depression, bottoming at 134K in 1932. Slowly they climbed back during the rest of the 1930s. GDP and unemployment data from that period are subject to larger-than-usual errors, but taking data in hand at face value, after declining by perhaps 30 percent between 1930 and 1933, with a concomitant rise in unemployment to perhaps 36 percent (!), overall GDP clawed back half that loss from 1934 to 1937, while unemployment fell to maybe 20 percent. The economy then took a second hit in the double dip of 1938, with a 4 percent decline in GDP and a return to rising unemployment. Things began to get better the following year, but with continued weakness in employment (Sound familiar? Well, it was, but much worse!)

During the war years, 1941 to 1945, GDP rocketed up by perhaps 70 percent, and unemployment fell to under 2 percent, while housing starts plummeted, as the nation shifted production from housing and consumption goods into military necessities as the U.S. economy became, in President Roosevelt's words, "the arsenal of democracy." Starts hit a trough of 142K in 1944. Then bounced back a bit in 1945 (the war ended in August), and shot up to an unprecedented 2.3 million in 1950.

After that boom, we settled down, but to a much higher plateau of around 1.5M units per year in the 50s and 60s, with substantial swings: peak-to-trough, housing starts varied by a factor of 2 to 1 or sometimes a little more! Housing, as our friend Richard Green has documented more carefully, became the leading edge of many business cycles. (Follow the Leader: How Changes in Residential and Non‐residential Investment Predict Changes in GDP, Real Estate Economics, 1997). Private starts hit their all-time high in 1972, with 2.4 million units underway.

U.S. housing starts took a big hit during the post-S&L boom recession at the start of the 90s; starts hit a low point of about a million in 1991. They then started a long, fairly steady climb back to a peak of 2.1M in 2005. They started to fall in advance of the 2007 Great Recession, plummeting to below 600K in 2009 and 2010. Ouch! These are the lowest levels of housing starts since 1945.

In the past few years starts are getting back up a bit short of the million units range, still a bit below what we might expect from an average healthy market.  We'll take another look at this below, with a second chart that makes a rough demographic adjustment.

What about public housing starts? These have always been a small part of the market, albeit one that is an important concern of HUD, taxpayers, and of course the families that live in those units. Public housing starts rose during the later years of the Great Depression, maxing out at 87K in 1941. Along with other housing starts, they collapsed during WWII, bouncing back to 71K in 1951, bumping around at 50K or less for most of the 60s and some of the 70s. They declined to nearly nothing in the 70s as the U.S. shifted from "supply side" subsidies to "demand side" housing subsidies, with the creation of Section 8 Certificates, the precursor to today's housing vouchers. You can read more about those policy shifts in the Primer. We haven't built any public housing to speak of in over three decades, but of course we still have a stock that requires management. All in, public housing itself peaked at under 2 million units three decades ago, and now stands at about a million units, or roughly 1 percent of the U.S. housing stock.

Manufactured housing as an industry came into its own in the 60s, peaking at 576,000 shipments in 1972 (the same year as the peak year in housing starts; all in, about 3 million units started). For much of the next two decades shipments bumped along near 200K, not at all negligible; they hit their second peak of 354K in 1998, then slid; the slide accelerated with the collapse of the housing market in the Great Recession, to a low of around 50K in 2010.  Recently they are back up around 70K, but still shipments are a bit anemic.

Now, about that simple demographic adjustment.  When we are looking at the number of housing starts over more than a century, we might also want to make at least a rough adjustment for the fact that we are a much more populous country now than then.  In 1890 the U.S. had a population of about 63 million, in about 13 million households; now we have about 325 million in about 117 million households.  The next figure shows how private starts (the bulk of the supply) looks once we've normalized by the number of households:

It's no surprise that starts look a little different when we examine this simple transformation.  (A recurring theme of this blog will be the value of simple transformations of data: any series worth charting is worth charting several times!)

The dotted line is a regression line through all the points; roughly, a century or so ago starts averaged a bit over 20 per 1,000 households, each year; in recent years that average has dropped off below 20 per 1,000.  While starts are back up in the past few years, they are still at a very low level of around 9 starts per 1,000 households, perhaps half of recent averages.

This series actually splices older historical data to the "modern" starts data that you can readily find at Census or FRED.  (I've got a data source note below).  A lot of "modern" economic data begin in 1947, when national income accounting really took hold as a routine core activity of the Federal Government. (More on that history in a future post).  The monthly housing starts series that Census and FRED present begins in 1959.  Notice if you had started in 1947, or in 1959, you'd see a much faster decline in the trend -- and, if (as you shouldn't!) one used this trend to forecast, you'd expect starts to turn negative (!) sometime in the next several decades.  Not a sensible forecast!

Econometricians have a battery of tests to see if data are likely really "mean reverting," or revert to a trend (think of a trend as a moving mean!); or if they follow some other pattern, like a "random walk."  I haven't done the formal tests  here, but we don't need formal tests to see that we shouldn't put too much on trend reversion as a great forecasting tool for housing starts.  I put the regression line in just as a simple benchmark.

Quite a story, and still not all there is to say about housing starts. In future posts we'll examine monthly data, talk about seasonal adjustment, and relate starts to some basic demographics and other determinants. But for now, contemplate 120 years of housing starts.

Data note

Postwar starts are easy to find at Census; where did the early data come from?  I obtained them some time ago from Historical Statistics of the United States. This masterful work was originally published in (I think 2) editions by Census, and was later taken over by Cambridge University Press and put online. I checked the online version:

http://hsus.cambridge.org.ezproxy.library.wisc.edu/HSUSWeb/table/footnotes.do?id=Dc510-530&footnote=footnote#fn_1

Cambridge online Historical Statistics in turn cites:

U.S. Bureau of the Census, Housing Construction Statistics: 1889 to 1964 (1966).