What Does a Housing Bubble Look Like?

Many sources have reported that the housing bubble was spurred by a significant increase in investment in the housing market. The US housing market was seen as a safe place to invest money at a reasonable rate of return. When the financial wizards abused a method to diversify risk and applied it to home mortgages money flowed freely into the housing market. (The best description is on this podcast.)

The US government collects plenty of data on various aspects of the economy and it seems only natural to ask if there’s a smoking gun in the historical data. Is there a difference in a housing bubble vs genuine growth in the housing industry? Sure, growth wouldn’t suffer an immediate decline as bubbles do when they burst. But ideally we’d have some data, metrics, charts and graphs that can give us a warning sign ahead of time. So we’re looking for leading indicators of a bubble. What would those be?

One candidate may be found in the employment data. Genuine growth should happen organically. Slow increase in hiring, enough time to train qualified workers. Genuine growth in housing could occur because of an increase in population. Unless there was a burst of immigration, that sort of increase happens gradually. Families have babies, babies grow up, buy houses and repeat. A significant increase in the birth rate 20 to 30 years ago could foretell an increase in housing demand today.

Genuine growth in the housing market could also occur with wide spread prosperity across the country. If all sectors enjoyed increased wealth people may choose to spend some of that wealth buying a larger home or remodeling their current home. Growth in this way may be called “genuine” especially if the increase in wealth came from a diversity of sources and/or specifically from sources other than the housing market itself. However many believe that people borrowed money against the increased value of their homes to buy a larger home or fund a remodel. Housing itself begot more housing, not increased population, productivity or Gross Domestic Product.

This article is part one of an occasional series where we’ll to put the current economic situation in perspective with data graphics. In this article we’ll examine the employment data.

Once again we’ll turn to the Bureau of Labor Statistics to help us answer the question whether there was an increase in employment in the construction sector during the years leading up to the rise and fall of the housing market. It stands to reason that there should be an increase, so the real question is, is there anything unusual about the increase in employment in the housing industry?

The BLS collects data on employment from 2 main sources. For one, they query the employers. This provides a lot of data since a single employer employs tens or hundreds of people (or more). Data collected here is remarkably broad and detailed, in many cases breaking down the occupation to a level as detailed as “Plasterers and stucco masons”. But by querying employers, this source of data misses one key statistic, the unemployed. Thus, the 2nd source of data the BLS uses comes from directly querying the population. Through this mechanism they can determine who is working and who is not in addition to an occupational cross section similar to (but not as detailed as) the employer survey. Irritatingly these two surveys are not compatible with each other. So for reasons that will become apparent we are using the population based survey. The details of which are available here.

The BLS has data through the population survey that goes as far back as 1983. Graphing the raw data results in the chart shown below.


This graph is a stacked graph, meaning, for example, that bottom of the Service curve starts at the top of the Construction curve. If this still isn’t clear a portion of the raw data is shown in a table format below.

Year Construction
Service Sales
1983 11,463,000 16,039,000 27,763,000
1984 11,989,000 16,408,000 28,937,000
1985 12,174,000 16,599,000 29,648,000
2006 15,830,000 23,811,000 36,141,000
2007 15,740,000 4,137,000 36,212,000
2008 14,806,000 24,451,000 35,544,000

Thus, on the left side of the graph (1983), the size of the Construction sector runs from 0 to 11,463,000. The size of the Service sector runs from 11,463,000 to 27,502,000 (11,463 + 16,039). Adding in the size of the Sales & Office sector, another 28,000,000 (approx) we get something close to 60,000,000 as a cumulative total which is where the top of the yellow Sales/Office line in the graph above is.

Both the Population survey (CPS) and the Employer survey (CES) collect data at a detailed level, but the CPS combines many more of the lower level details into a single category. This is because there is less data (fewer people surveyed) in the CPS so the accuracy of the details may be off. But the CPS is a bit broader than you might expect. You may be wondering (for example) what category your job falls into. Teachers, for example are in the Professional side of the Management and Professional categorization. The full break down is available here.

Examining the data, we realize that spotting a unique growth trend will be harder than we expected because all these sectors as a whole grew over the 25 year time frame. One way to factor out this growth and dig deeper into the sector specific growth is to determine the growth of the labor force as a whole and then see if any sectors grew faster or slower than that. Notice, in the graph above, that the “Management & Professional” category grew significantly more than say “Sales and Office”. While both grew, Management grew more. So this approach may yield fruit. But there’s an easier way to get a similar restructuring. Since the population survey also includes the unemployed, we have a picture of the total labor force. So we can instead plot each sector as a percentage of the total labor force.

Additionally we’ll zoom in our scale a bit, using the full vertical range to graph a difference of 12%. Why 12? It turns out that the difference between each sector’s high and low fits within a 12% range. The largest is “Management and Professional” which had a low of 26% of the labor force to a high of 34%, a swing of 8%. The only oddity that this introduces is that on some of the graphs the bottom of the range will NOT be zero.

The graph below shows the Construction sector and if you click on it, you’ll be taken to an interactive application where you can click on the tabs at the top of the graph to see a close up of the other sectors. Note, for this graph we’ve also included an estimate of the 2009 data by merely averaging the monthly data available for 2009.


Here the full effect of a bubble in the Construction sector becomes visible. Firstly we notice a flat or somewhat declining influence in construction on the US labor force from 1984 till 2002 when it reached a low of 9.4%. It grew from there till 2006 when it reached a high of 10.5%. At the end of 2008 the annualized fraction of Construction workers in the labor force was 9.6% and the 2009 estimate appears to be 8.7%.

Taken all by itself it’s hard to tell if this is unusual behavior or not. Fortunately we have data from 4 other sectors to compare it to. When you do that the unusual nature of the construction sector becomes clear. First, except for Sales, all the other sectors follow their trend line for the entire 25+ year period studied. They are all either flat, growing or shrinking across the entire range. Secondly, all have bumps, but the bumps are significantly less than the 1% swing we see in the Construction sector1. Thirdly, Sales is the only other sector that experienced a rise and fall, but it seems to be what you’d expect from a slow systemic change, and not from a bubble. While it rose 1.4% during a 4 year period it took 14 more years before it fall back to its 1983 levels2.

We have found a hint of the impending doom for the housing market in the employment data. Interesting, but not enough. The construction employment bubble should betray additional details. Where did those workers come from? The construction sector added 1.5 million jobs between 2003 and 2006. It seems reasonable to suppose that these workers were not as highly trained as those who were already in the construction industry in 2002. If this is true, we’d expect to see signs of this untrained labor force. Were there more complaints filed against the construction industry during this time? Was the work below par? Did the work take longer to complete than expected? We will have to see.

It would also be interesting to learn what parts of construction these workers were engaged in. Were they building new homes? Remodeling existing homes? Building more commercial offices? While the answers to this question may not point to bubble vs growth, it will point to additional areas to probe. If an unusual amount of growth occurred in the residential housing market then probing residential statistics may yield fruit. Are there ways to determine if there was a genuine pent up demand for houses? Or was the only sign of demand the increase in home prices?

It remains a goal of this website to examine important news stories with data visualization tools in the hopes that one day we’ll be able to see the signs of situations like this before they turn into full fledged disasters.

1 A swing of 1% in Construction is 1% of the entire labor force. Since Construction represents about 10% of the labor force, this 1% “total” swing represents a 10% swing for construction workers.

2 There may be a larger story here, but it is also longer term. If Construction swung by 10% during the 4 year period around the bubble, Production decreased by 28% over the 25 year period while Management grew by 27% over the same interval. This long term shift in employment, while interesting, is not the topic of this article.

Leave a Reply

Your email address will not be published. Required fields are marked *