Higher, higher and higher

How high can it go?  This chart, sent in by Michael McCracken and attributed to Yale economist Shiller of "Irrational Exuberance" fame, very effectively poses this question.  The "hockey stick" on the right side of the chart really hits us like a gigantic question mark.

When we have good data, or are looking at the data from the right angle, the charting task is that much easier.

Michael especially likes background shading to highlight specific periods.   I'm a bit perplexed by the "World War I" label as that period does not appear remarkable to my eyes; it is also the only shaded reason that is not a boom period.

The text explains the need to remove "new construction" in order to study housing as an "investment over time".  As an outsider to the real estate industry, I find this definition arbitrary.  The 2001 data presumably would include the sales price of any house that was constructed from 2000 and back.  Why exclude only current-year construction?  Could a sale of a one-year-old property be considered "investment" and not "speculation"?

Reference: New York Times, Aug 26 2006.

Comments

[Keith Anderson]

One of the steepest and most enduring declines appears to have started during World War I. The explanatory text ("Decline and run-up") also clarifies why that period is shaded.

[cscmike]

I'm always curious as to how the values have been adjusted for inflation. Somewheres on the interweb, I remember seeing how varied "inflation-adjusted" numbers could be whether you depended on consumer price index, core CPI (minus food and energy costs), etc.

[Chris Hibbert]

The reason that new construction is ignored is the heart of the Case-Shiller index. This index gives better comparisons of prices across time by looking at repeated sales of the same property. The index only looks at the same property when it has been transferred more than once, so new sales don't count. Basing an index on repeated sales of the same property seems to tell more about the increase in value of existing homes, even though it doesn't discount for remodels or other changes in fundamental value.

[Robert Kosara]

Interesting chart (and interesting website)! However, a problem I see here is that the scale does not start at 0, but at 60, thus overemphasizing the boom. Including the lower third of prices would make that look much less impressive, but would be a lot more accurate.

[derek]

Ironically, it would even let you see more structure. The graph as it is is too spiky. Like the example quoted by Tufte and Cleveland of sunspots, by shrinking the curve a bit in the vertical direction, more pattern would be discernable.

[Kaiser]

Keith - somehow Shiller decided to ignore the steep declines in the 1890s, early 1980s and early 1990s.

Mike - the inflation adjustment is dividing by a factor that equates 1 dollar at some past time to its value today. The potential for mischief is in defining the basket of goods used to estimate that factor.

Chris - I understand the need to differentiate speculation from investment. But it's unclear to me why 1 year is the correct delineation point.

[Robert]

I have written a SAS/Graph job that will graph this data almost exactly like this chart that's being discussed - if anyone is so inclined, feel free to download the sas code from my samples page, and experiment with changing the axes, etc :)

http://robslink.com/SAS/democd24/aaaindex.htm

[spencer]

The problem with the chart is that the scale does not appear to be a log scale.

Using this scale a move from 100 to 110 is the same size as a move from 200 to 210. But in percent terms a move from 100 to 110 is double a move from 200 to 210.

If you use a log scale the move from 100 to 110 takes double the vertical space as a move from 200 to 210. But in a log scale the same relative move at both the top and bottom of the scale takes the same space. Consequently, a log scale makes it much, much easier for a reader to compare reading at the top of the scale to moves at the bottom of the scale.

this chart makes people think the move in recent years is much larger proportionally than it really is.

This is the reason almost every long run chart of the stock market is scaled on a log scale.

[Mark]

Kaiser: Chris H didn't say that 1 year is the delineation point -- it isn't. The delineation point is after the house is sold for the first time by the developer. If the first buyer moves out after six months -- or indeed after a week -- and sells up again, that sale is included.

Having said that, I'm not sure I understand Chris's explanation of why this is the Right Thing. Every house is counted *for the first time* by the analysis at some point; I don't see why it matters whether this is at its first or its second sale. In other words, if you ignore the first sale, the second sale looks to the analysis like a sale of a new property (because it hasn't seen that house sold before); why not just bite the bullet and count the first sale too?

I could imagine that prices of new houses are thought to be "special" in some way -- unusually low because of high supply (the developer having just built a whole development of similar houses in the same place), or high (if it turns out that people put a premium on moving into a brand-new house). That might be a reason for excluding the original sale, but that doesn't seem to be what's being claimed.

[spencer]

The price index of resales is used to avoid measuring changes in composition to changes in price. The resale index compares the latest price of a home to its prior price --even if the earlier price was for a new house--so you are comparing exactly the same thing and determine exactly what the price increase is. But you can not do that for a new home
although there are price indices that attempt to adjust new home prices so you are always comparing the same 3 bedroom, 2 bath, 2,500 sq ft homes.