Uff da! Minneapolis and Changing Seasonality in House Prices

Tommorrow we get the latest data on nominal house price growth, both the Standard and Poor's Case-Shiller index and the Federal Housing Finance Agency. One series in particular to pay attention to is the Minneapolis series from Case-Shiller. What makes Minneapolis so interesting? In July it recorded its highest monthly growth rate ever, growing 4.8% from June to July. That was better than any month during the boom. See figure one. Even more interesting is that the growth in July came just four months after the worst month ever for Minneapolis, a negative 5.9% in March. To restate that, this year Minneapolis has seen both its worst and best month of nominal house price growth in the Case-Shiller index (the data for Minneapolis starts in 1989). Uff da!

Even more striking, is that dismal March for Minneapolis is the worst month ever for all of the cities for all of the months in the Case-Shiller index. Yes, March 2009 Minneapolis takes the title for worst Case-Shiller month. But the 4.8% in July 2009 is also one of the best. Only Phoenix June '05 (4.9%), a rogue July '88 in San Diego (5.3%) and Las Vegas from March '04 to July '04 (all above 5.0% with a peak of 6.0% in June '04). In other words, not only has Minneapolis gone from its worst month to its best in a span of 4 months, it has gone from the worst ever across all cities to one of the best ever for all cities! Uff da! For sake of comparison, figure two plots nominal house price growth for both Minneapolis and Las Vegas. The current seasonal cycle in Minneapolis is similar in magnitude to the boom and bust we saw in Las Vegas. Of course the Las Vegas cycle was longer so it has had a much larger impact on the house price level.

Why such a big change in house prices? Lars at the diner in Lindstom would shrug it off and say, `Oh, that's the snow, everyone knows that.' Lars is very likely correct. Inspection of figure one suggests that there has been an increase in the seasonality of house prices in Minneapolis. To get around the problem of seasonality in house prices most data is typically deseasonlized. Figure three plots the raw data (blue line) and the seasonally adjusted (black line) data for Minneapolis. Clearly, the seasonal adjustments are not fully accounting for changing seasonality.

The insufficiency of the seasonal adjustments leaves a problem. Due to increasing seasonality we cannot distinguish whether the high growth in Minneapolis house prices is due to snow or due to a recovery.

As I will show in the next post, the increasing seasonality is a problem for most cities in the Case-Shiller sample. (This appears to be due to most seasonal adjustments smoothing the seasonal factors, more on this in another post). This is making it very hard for us to distinguish whether prices are really rising, or we are just observing an artifact of increasing seasonality. This is why examining Minneapolis house price growth over the next few months is important. It will be the best data to disentangle the recovery from snow.

FHFA: House Prices for August 09

This morning the FHFA released their House Price Index for August 2009. They showed an decrease from July to August of 0.3%. The previously reported increase of 0.3% from June to July was unchanged. That means that the downward revision I had expected did not happen. As I showed in my earlier post, the initial estimate of monthly house price growth had been biased up by 0.22%, so i expected that the initial 0.3% growth from June to July could be revised down. This did not happen.

As I show in figure one, since December 2008, nominal house prices have risen 0.2%. Also shown in figure one, just using the headline initial monthly estimate we would believe that house prices have risen 2.6%, an overstatement of 2.4%. Using the August report, the initial monthly index for the house price index is on average biased up by 0.21%, a slight improvement over the previously found estimate of 0.22% using the July report.

Also included in the FHFA's August report, they now state that house prices are now 10.7% below their April 2007 peak. The latest fall in house prices is a bit surprising, but month to month numbers do not provide much information. Next month we will receive the report for the third quarter of 2009--that will contain a lot more accurate information.

FHFA Prices: Will they be revised down?

Tomorrow morning at 10 am est, the FHFA will announce their monthly house price index for August. As I reported earlier, their headline growth rate has been biased up by 0.22% on average. Last month, the FHFA reported that house prices increased 0.3% from June to July. If the bias holds true, then we can expect that the August report will include a revision to the July growth, revising the growth down from 0.3% to something between 0.0% and and 0.1%. I'll report back tomorrow with their release.

Monthly FHFA House Prices are Irrationally Exuberent: Headline Growth Rate Overstates Year-to-date Growth by 2.4%

On September 22nd, the Federal Housing Finance Agency (FHFA) released the July 2009 monthly housing price index with the headline “U.S. Monthly House Price Index Estimates 0.3 percent price increase from June to July.” Such headlines are fueling optimism that the housing market and the economy are turning a corner. However, in the text of the announcement we learned that the previously reported increase of 0.5% from May to June was revised down to 0.1%. Combining the two effects, house prices actually fell 0.1% relative to what was previously reported. The downward revision is not new for the FHFA monthly price index. In fact, if you only read the headline monthly growth rates, you would be under the impression that house prices have risen 2.9% so far this year through July. In fact, in the latest report, house prices had only risen 0.5%. In other words, the headline monthly house price growth gives the impression that house prices have grown 2.4% more this year than they actually have.

The FHFA monthly house price index is new, only having been reported for 20 months. In the FHFA’s original analysis of the potential revisions to their monthly index (reported in the second quarter 2007 release) they stated the initial monthly estimate `would not be reliable’ and did not bother to even report the magnitude of the revisions in their analysis. This unreliable estimate is now the headline of their monthly release. We should not overly criticize the FHFA. The initial monthly index does provide information, explaining roughly 84% of the movements in house prices (in an R-squared sense).


To put the revisions in context, suppose instead that the FHFA was a stock broker handling your million dollar portfolio, and he called you up to tell you the news on your portfolio for July. The conversation could have played out like this:

BROKER: Great news, your portfolio is up 0.3% June to July.

YOU: Great, combined with the 0.5% from May to June, I’ve earned $8,000 the past two months. Maybe I’ll start thinking about getting the deck redone on my house.

BROKER: Yes, about that 0.5%, we re-estimated the increase from May to June to be only 0.1%, so that you actually have only earned $4,000 the last two months. Sorry.

I think most people with a million dollar portfolio would not like to frequently have such conversations with their stock broker. However, for the past twenty months that the FHFA (formerly OFHEO) has been reporting the monthly house price index, the FHFA has been regularly having just such conversations with Americans regarding their houses.

I do research using the quarterly house price index from FHFA. I know that this index states that house prices are down for the year, but it seemed to me that the monthly index had been reporting prices increases from month to month. I went back and checked the headline growth rates, here they are, stating the date reported along with the relevant period for the index:

• March 24: “1.7% increase from December to January”
• April 22: “0.7% increase from January to February”
• May 27: no headline, part of quarterly report. Reported change from February to March is 1.1% decrease
• June 23: “0.1% decline from March to April”
• July 22: “0.9% increase from April to May”
• August 25: no headline, part of quarterly report, in the text the FHFA states that the “monthly index for June rose 0.5%”
• September 22: “0.3% increase from June to July”

If you add up all of the growth rates, you arrive at a year-to-date growth of 2.9%. In fact, prices have only rise 0.5%. Furthermore, the 0.5% is most likely overstated as the 0.3% reported from June to July is likely to be revised down. Figure one plots the perceived series for house prices relative to December 2008 if one accepts the headline growth rates as the truth, showing an increase of 2.9%. Figure one also reports the current path for house prices relative to December 2008. Clearly the headline growth rate overstates the actual growth. So far, the overstatement has been 2.4% for 2009.

To put the FHFA’s revisions back into the context of a stock broker managing a million dollar portfolio, if you just looked at the headline growth (maybe in the subject line of a monthly e-mail) you would think that you had earned $29,000 so far this year. You could be thinking about making a large transaction such as renovating a bathroom. However, when you actually looked at your most recent statement, you’d see that you were in fact only up $5,000 so far this year.

The FHFA only started reporting the monthly house price index on February 26, 2008 as part of the report for the 4th quarter of 2007, giving us an observation for December 2007. Since then, we have received 20 observations in total (including Dec 07). Figure two plots the initial estimated price growth along with the current estimate for price growth. There is a clear bias down in the revision. Figure three plots the error in the initial estimate, given by the difference between the initial and the current. Once again there is a clear bias in the revision. Not including the latest observation, the initial estimate overstates the growth rate of house prices by 0.22% on average. This amounts to an annual overstatement of 2.68%.

Before the FHFA started reporting the monthly price index, in their Second Quarter 2007 report, released on August 30 2008, they performed an analysis of revisions. In their analysis they state (see page 11 of the report):

“A primary concern with the construction of monthly price indexes is that revisions will tend to be large. …the most recent monthly price measures would seem to be particularly susceptible to later revision. A June index estimate and the corresponding May-June appreciation rate estimate would be prone to the greatest revision, and a review of the evidence suggests that the estimates would not be reliable.”

To analyze the revisions, they compare their estimate in the second quarter 2007 release with an estimate only using date through the first quarter of 2007. When they do this analysis they do not show the error for March 2007! They leave it out, stating “the line only extends through February 2007 because the data submission only included data through March, and—as indicated above—the index point for the latest month (March) is not deemed to be reliable.”

In other words, from their analysis they deemed the first estimate unreliable and not worth even making an analysis of the size of the revisions. This estimate that `would not be reliable’ is the estimate that they currently report in their monthly release. But not only do they report it, they make it their headline!

Before we overly criticize the FHFA, they have put out many warnings in the text of their announcements. When the monthly house price index was first released as part of the fourth quarter 2007 report (release on Feb. 26, 2008) they warned “index users should recognize that, while OFHEO’s initial review suggests that revisions may be reasonably small, this is no guarantee that future revisions cannot be significant.” They also made a warning for the January 2009 report—which ended up being the largest revision to date—stating that due to a low sample size that “the estimation imprecision associated with the January estimate is relatively large and subsequent revisions … could be significant.”

The initial monthly estimate does provide information. As the then OFHEO director James B. Lockhart stated in the release with the first monthly index “Given the recent turmoil in the housing markets we thought it would be helpful to provide a greater amount of information about price trends.” The FHFA has succeeded along this dimension. Inspection of figure 2 shows that, although there is a bias in the initial estimate, it does provide information about the final estimate. The R-squared of the fit of initial to the final (through June 2009) is 0.84. Therefore, the initial does provide more information about the price trends in the housing market.

Last, there have only been 20 observations of the initial monthly estimate. In addition, theses twenty observations have covered a period of unprecedented falling house prices and very low level of sales, especially considering that the FHFA’s data only starts in 1991. With such a small sample over observations that are so unprecedented, it is difficult to say definitively that the initial estimated is biased. Furthermore, large revisions consistently in one direction are not surprising.

The main problem with the overstatement in the initial monthly estimate is that the FHFA is using the initial estimate as their headline. Given that the FHFA stated in their own analysis that this estimate ‘would not be reliable’, they may want to think about not using this observation as their headline. Instead they could go with the change in the year-over-year or use a more descriptive headline like they use for the quarterly report.

Another Article About Falling Rents

Marketwatch has an article providing anecdotal evidence that the high rental vacancy rates are putting downward pressure on rents:

Why you should ask for lower rent

As I have been saying, the downward pressure on rental rates will last for sometime, keeping downward pressure on all nominal prices.

Beveridge Curve and New Census Data: Owner-Occupied Market near Equilibrium, Rental Market Not

Recently, the census bureau released its latest figures on the housing stock, now updated to the second quarter of 2009. Interestingly, the vacancy rate for the owner-occupied market fell to 2.5% from 2.8%, but it rose to 10.7% in the rental market, a 40 year high. To shed some more light on the new census data, below, I redo the analysis in my "Beveridge Curve" paper which provides a quantitative measure of oversupply, or disequilibrium in the housing market. The main finding is that the owner-occupied market is almost back to equilibrium, in my opinion, a quite strong observation that the owner-occupied market has reached bottom. However, the overall housing market is still suffering from oversupply, with most of the oversupply residing in the rental market. This further supports what my research has been suggesting: a significant adjustment mechanism for the housing market is the shift of housing from the owner-occupied market to the rental market. I expect rents to continue to fall, putting downward pressure on the CPI. My hunch is that for the next several years multi-unit housing starts will remain depressed while the rental market corrects itself. Read on for the updated analysis.


Figure 1 plots the updated time series for the vacancy rates. In addition to the owner-occupied and rental vacancy rates, I provide a total vacancy rate, which provides the vacancy rate for the overall market, irrespective of ownership. We can see that the total vacancy rate has been flat of late, roughly at the all time high since the data started in 1965, while the owner-occupied rate has fallen and the rental vacancy rate has increased.


The census data also informs us about the rate of household formation, and how the rate of household formation is allocated across renters and owners. Figure 2 plots the rate of household formation for owners, renters, and total households irrespective of tenure status. Due to the noisiness of the data, I use the % increase in households over the past 8 quarters and report the rate of change at an annual frequency. In my opinion figure 2 is one of the most striking pictures of the housing bubble. We see quite clearly the shift in owning starting around 1995, and then the dramatic shift to renting starting around 2005. In fact, the first quarter of 2006 was the first observation where the previous two years had more growth in renters than owners since the last quarter of 1994! Even more striking, for the first time since Census began this series in 1965, the growth rate of owners has been roughly zero for the past two years.

My research suggests that there is a negative relationship between the rate of household formation and the vacancy rate. This is true in the owner-occupied market, the rental market, and the total market irrespective of ownership. I call this relationship the Beveridge Curve in the Housing Market, and it can be thought of as a long-run supply relationship. For a deeper analysis of the Beveridge curve see my earlier posts here and here or read the academic paper.


The most interesting aspect of the research is that the deviations from the Beveridge curve give us an estimate of the magnitude of the disequilibrium in the housing market. I can do this for the owner-occupied market, the rental market, or the total market irrespective of ownership. The result for the owner-occupied market is shown in figure 3, the units are in a % of the total housing stock. We can see quite clearly the oversupply that resulted from the bubble, peaking at 0.49% of the total housing stock in the fourth quarter of 2006. However, the market has been quickly correcting itself, now standing at 0.10% of the total housing stock. The correction in the owner-occupied market stems from two adjustments: (1) the growth rate of owners has fallen, now being consistent with the high vacancy rate; and (2) the number of units in the owner-occupied market has been falling, due to less construction and a shift of units to the rental market. I expect that these trends are still continuing, so that sometime in the next year the owner-occupied market will actually shift to a state of undersupply.

However, the picture is much different in the total housing market. Figure 4 shows the oversupply in all three markets: owner-occupied, rental, and total market irrespective of ownership status (the rental market reported here is simply the residual between the total market and the owner-occupied market—the independent estimate of the rental market implies an even larger oversupply in the rental market). Here, we can see that the total market still is far from equilibrium, with an oversupply of 0.89% of the total housing stock. However, this is down from the peak of 1.18% in the second quarter of 2006. Therefore, while the owner-occupied market is almost back to equilibrium, the overall market still has a large oversupply. Almost all of the oversupply is showing up in the rental market. Clearly, we are seeing the market respond by shifting resources from the owner-occupied market to the rental market. This will undoubtedly put pressure on rents to fall as has been recently reported. I expect rents to continue to fall for quite some time, putting downward pressure on the CPI.

Punchline: the market is adjusting, which means the owner-occupied market clears while the rental market holds the oversupply. The owner-occupied market has probably already moved back into a state of equilibrium, while the total market will slowly reach equilibrium as the production of primarily multi-unit rentals remains sluggish for several years.

More on Beveridge Curve: Disequilibrium and Oversupply in the Housing Market

Earlier I posted on some research that I was working on about a Beveridge curve in the housing market (the work that motivates the name of this blog). The working paper version of the paper is now finished and available for downloading.

The main results of the paper:

• The Beveridge Curve represents a long-run supply condition
• Short run deviations represent periods of disequilibrium, either over or under supply
• Using a years of oversupply metric, the observation of 2007-2008 was an all-time high of 0.995 years of oversupply, more than three times the previous peak of 0.285 in 1973-1974.

Results regarding the differences between the owner-occupied and rental markets:

• Generally, oversupply is a phenomena in the rental market
• Oversupply in the rental market is twice as volatile as in the owner-occupied market
• Oversupply first shows up in the rental market

Last, I find that

• Oversupply in the owner-occupied market is related to house prices, reinforcing the idea that short run deviations in house prices from fundamentals (such as bubbles) can lead to periods of oversupply

Below, I give a summary of these results, except for the last bullet on house prices which I leave to another post.

The primary change from the earlier results is that instead of doing the fairly ad-hoc HP filter, I have gone ahead and estimated the model using biannual data. In my opinion this gives a better estimation. In addition to the new estimation I have also found some interesting results relating oversupply of houses to house prices. I will go over those results in my next entry, for now I give an overview of my previous results under the new estimation.

The main idea of the Beveridge curve is that it represents the long-run equilibrium in the housing market. The Beveridge curve has its origins in labour economics, where Lord Beveridge found a negative relationship between the unemployment rate and the amount of job vacancies. For an example, see Rob Shimer's website. In the housing market I have found a negative relationship between the rate of household formation and the residential vacancy rate. I deem this negative relationship the Beveridge curve in the housing market. The Beveridge curve relationship exists in the owner-occupied market, the rental market, and the overall market irrsepective of ownership. Figure 4 from the paper, shown here below, shows the relationship for the overall house market. There is a clear and statistically significant negative relationship and the R-squared from a linear regression is 0.626.

As I stated, the curve represents a long-run relationship, so that short-run deviations (two to four years) represent periods of disequilibrium in the housing market, these are periods of under or over supply, see figure 7 below.
The next figure shows the estimated time-series of oversupply for the total housing market irrespective of home-ownership. The metric of oversupply is in a % of the total housing stock. Here we clearly see three periods of oversupply since the start of the data in 1968: (1) the 1974 crisis; (2) the mid to late 1980s housing boom; and (3) the current crisis. The oversupply in the current crisis is similar in magnitude to the 1974 crisis, with both having an oversupply of just under 1% of the total housing stock.
However, the rate of household formation is much lower now, so that it may take much longer to work off the oversupply. The next figure plots the estimate of oversupply in years of oversupply. This is the oversupply in years of household formation. For instance, if the rate of household formation was 1% and the oversupply of the housing stock was 1%, then it would take one year of no housing production for the oversupply to disappear. Using this metric, the amount of oversupply is staggering, being about one year of supply in 2007-2008 compared to under 0.3 years in the 1974 crisis.

One point that I cannot stress enough is that what we are currently facing is a huge oversupply of housing, irrespective of whether it is rental or owner-occupied housing. As I stated in the earlier post there is actually more oversupply in the rental market. To see this, below is figure 12 from the paper. The metric is years of oversupply in terms of the total rate of household formation, so that we are comparing apples to apples. There are several striking features in this figure:

• Generally, oversupply is a phenomena in the rental market
• Oversupply in the rental market is twice as volatile as in the owner-occupied market
• Oversupply first shows up in the rental market

These results suggest that the rental market is sponge the soaks up oversupply in periods of overall oversupply, and then lets it out once housing becomes relatively scarce. This suggests that research on the housing market, especially empirical, needs to include data from the rental market when doing structural work.

Distressed Sales and House Prices

In my last entry I argued that a shock was causing existing home sales to rise relative to new home sales, and that this same shock was causing house prices to fall more than would be suggested by the level of existing home sales. This `shock' is more than likely a supply side shock: quantity up, prices down, Econ 101 at work. A question is how much of this is being driven by distressed sales where the house is being sold because the current owner has stopped paying the mortgage.

Andrew Leventis, a researcher at the Federal Housing Finance Agency has attemped to answer this question is a recent working paper: "The Impact of Distressed Sales on Repeat-Transactions House Price Indexes".

In the paper, Leventis uses transactions data from California and breaks down transactions into distressed and non-distressed. A transaction is distressed if a Notice of Default was filed on the property up to a year before the transaction occurred and no other transactions occurred for the same property between the transaction and the Notice of Default. He performs the analysis for two different groupings of data. The first grouping he calls `Enterprise' data, and this consists of the transactions that would be included for calculating the FHFA House Price Index (HPI). The second grouping is `Recorder' data, which consists of the data that would be used to calculate the Case-Shiller index. His figure one shows that the share of distressed sales has increased from less than 5% before the fourth quarter of 2006, and has been rising steadily ever since, now over 45% for the first quarter of 2009. The rise in distressed sales is very similar to the gap that has appeared between new and existing home sales.

The main question of the paper is "How much are these distressed sales driving down house prices?" To paraphrase Leventis, we can breakdown the effect of distressed sales on house prices into two categories:

1. Direct Effect: distressed houses sell at a discount, therefore, as the share of distressed sales in the sample increases, then the reported house price index will fall.
2. Indirect Effects: the more distressed sales there are, the harder it is for a non-distressed seller to sell a house, lowering the prices for all houses.

In his paper, Leventis can only get at the direct effect. His figure 4 shows the discount for distressed sales. We can see that before 2006, that distressed sales sold at about a 4% discount. At the peak of the bubble in 2006, the discount almost went completely away--it didn't matter if a seller was distressed, seemed that buyers were willing to buy everything and anything in sight. However, after 2006, the discount gets significant, approaching almost 20% for the `Enterprise' data, and 15% for the `Recorder' data (there is an interesting divergence between the two types of data in the last couple of quarters).
Therefore, by the end of 2008, distressed sales were selling at a 20% discount and made up roughly 45% of sales. Relative to the peak in the housing bubble in 2006 (when the discount was essentially zero), if this sales pattern would maintain itself for a year, it would imply that the house price index is being pushed down an extra 9% due to the direct effect from distressed sales. However, the total effect so far has been smaller. His figure 2 shows the effect of the distressed sales on year-over-year price growth for the `Enterprise' data. The figure contains two plots: one showing price growth for the entire sample, another with the distressed sales removed. The effects are fairly small. The cumulative effect of the distressed sales is estimated to drive down prices an extra 5.3% from the peak, for a fall of 41.3% relative to only 36.0% when the distressed sales are excluded. The effect on the `Recorder' data is smaller, implying an extra house price fall of 1.9% from 44.8% to 46.7% (see his figure 3, not shown here).

To summarize, the divergence between new and existing homes seems related to the surge in distressed sales. The work by Leventis suggests that the direct effect of the distressed sales on the reported house price indexes is most likely small relative to the total decline we've seen in house prices. What we do not know is whether the distressed sales are directly responsible for the fall in house prices and the increase in existing sales relative to new sales, or if the distressed sales are simply the result of the large supply of housing, which is then responsible for distressed sales, housing price falls, and an increase in existing relative to new sales. My viewpoint is that we are just seeing the effects of supply at work.

House Prices and New versus Existing Homes Sales

This week we have received the May data on existing home sales (from the National Association of Realtors) and new home sales (from Census). Existing home sales have been flat or rising a bit for 2009. Such stabilization in the market for existing homes has been a sign to many observers that we may be reaching the bottom in house prices. However, at the same time new home sales have continued to fall relative to their 2008 levels. As people have noted (see CalculatedRisk), there is now a gap between new home sales and existing home sales that did not exist before 2006. As I show here, whatever is driving a gap between new and existing homes sales is also driving a gap between the relationship of existing home sales and real house price growth. The evidence suggests that a supply side shock is driving up existing homes sales relative to new home sales and at the same time driving down prices, just as standard econ 101 predicts. Furthermore, price growth is related to new home sales, not existing home sales. The punchline: if we are seeking stability in house prices we should look at new home sales, not existing sales.




Figure 1 plots annual single family existing homes sales and single family new homes sales as a percentage of the total housing stock (the total housing stock is taken from Census). We see that new home sales average about 1% of the total housing stock, while existing home sales average roughly 4%. To get a feel for how the two series move together, figure 2 plots the percentage deviation for each series from its mean from 1975-2008. We see clearly that from 1975 to 2006 (the solid lines) that new home sales and existing homes sales move around together, with a correlation of 0.944 over the the time period up to 2006. However, as shown by the dashed lines, a gap has developed post 2006, resulting in the correlation for the sample from 1975-2008 falling to 0.876. There seems to be some type of a shock that is driving existing homes sale up relative to new homes sales.


Turning to prices, figure 3 graphs expected annual real house price growth next to existing homes sales. Expected real house price growth is the FHFA (formerly OFHEO) house price index, made real by the rate of expected inflation from the Philly Fed survey of forecasters. The existing sales series is the same as in figure 2. Once again, we can see clearly that from 1975 to 2006 both series move around together, with a correlation of 0.921. However, post 2006, a gap develops just like the gap between new home sales and existing home sales. Using the whole sample up to 2008 the correlation falls to 0.812.


Both of the gaps suggest a shock hitting the housing market. To put it more clearly, figure 4 plots both a `price shock' and and `existing sales shock'. The price shock is the shock to price growth that is not explained by existing sales in figure 3. (This is actually the error term from a linear regression of price growth on existing homes sales relative to the housing stock). The existing sales shock is the shock to existing homes relative to new home sales, the difference in figure 2. (the existing sales shock has been normalized to be on the same graph as the price shock). From 1975 to 2006 these two shocks are essentially unrelated. However, in 2007 and 2008 both shocks are sizeable, moving in opposite directions. What we have is a classic supply side shock: quantities rising and prices falling. The glut of vacant houses on the market are doing what they do: drive down prices, and drive up sales.


The implication is that the supply shock is breaking down the standard relationship between sales and price growth. Existing home sales is not the place to be looking for stability in house prices. Instead, figure 5 plots new home sales relative to house price growth. Here we can see that new home sales are related to house price growth, and this relationship has maintained itself through the crisis. The correlation from 1975-2006 is 0.881, while for the whole sample from 1975 to 2008 it is 0.899. Therefore, if we are looking for price stability, we should look to new home sales, not existing home sales. The supply side shock that is hurting prices and raising existing home sales causes new home sales to fall. To stress the point a bit more, the stability in the housing market that we want for economic recovery is stable prices and new home construction. Stable prices are associated with new home sales not existing home sales. The current stability in existing home sales is most likely just the effects of a supply side shock that is driving down both prices and new home sales. In my next entry I will use search theory to guide us in understanding this behavior in the housing market.

Housing Market Recovery: Shadow Inventory?

We all want to know when the housing market will recover. A recent piece by David Leonhardt in the NY Times suggests that the housing market still has not reached bottom. Central to clearing the housing market is eliminating the oversupply of homes (see my earlier entry). There are several measures of oversupply: the National Association of Realtors provides a monthly measure of homes for sale; the Census Bureau provides a monthly measure of new homes for sale; and the Census Bureau also provides us with a quarterly measure of vacant homes for sale.

All of these measures rely upon the owner of the property putting the housing unit up for sale. As Calculated Risk has been saying (here, here, and here) there is most likely a large shadow inventory of sellers keeping their homes off the market. A recent piece in the San Franciso Chronicle (article here, Calc Risk Blog here) suggests that banks are also keeping foreclosed properties off the market.

There is no survey of banks and households that we can use to estimate the size of the possible shadow inventory. However, we can get a sense of it from the quarterly Housing Vacancies survey. This survey is the source for the data on owner-occupied and rental vacancy rates, and it provides a breakdown of vacant units beyond these headline vacancy rates. Table 1 provides the breakout of vacant housing units from Census. When census reports vacancy rates they exclude `Seasonal Vacant’ and `Held-off Market.’ The category Held-off market consists of three sub-categories: `Occasional Use’, `Usual Residence Elsewhere (URE)’ and `Other Reasons.’ These excluded categories are generally quite boring. They are generally increasing, following a trend, but display almost no cyclicality.


I say almost because the sub-category vacant held-off for other reasons has become quite interesting of late. Figure one plots the % of the housing stock that is vacant held-off the market for other reasons. My measure of the housing stock is the sum of total occupied, vacant for sale, vacant for rent, vacant sold or rented and vacant held-off market for other reasons. In the last two years this category of vacant units has surged from 2.3% to 2.9%. This suggests that there is a large shadow inventory equal to roughly 0.6% of the total housing stock, or 736,000 units.


Figure 2 plots the total residential vacancy rate with and without the shadow inventory (vacant units held off the market for other reasons). The total residential vacancy rate is the % of the housing stock that is vacant irrespective ownership status. The red line in figure 2 is the total vacancy rate solely from vacant for rent and vacant for sale, excluding the shadow inventory. The black line in figure 2 denotes the vacancy rate when the shadow inventory has been added to the vacant units for rent or sale. In the current crisis we see that the vacancy rate excluding the shadow inventory seems to be stabilizing while the vacancy rate including the shadow inventory seems to still be rising.


A rising vacancy rate is a potential bad omen for the recovery of the housing market. Ultimately, it is vacant units that are putting pressure on the cost of housing services, be it rents or house prices. Common sense suggests that stabilization of the vacancy rate is a necessary but not sufficient condition for stabilization of house prices. This suggests that inclusion of the shadow inventory implies that the housing market still has a long way to go before prices stop falling.

An interesting feature of the data is that the presence of a shadow inventory is not a new phenomenon. In figure 1 we see that during the crisis in 1974 there also was a surge in vacant held off market. In figure 2 we also see that the rise in the vacancy rate including the shadow inventory is more pronounced during the 1974 crisis. This suggests that movement in the shadow inventory is always part of the cyclical process of the housing market. The shadow inventory is just more visible during the severe downturns. Therefore, there may not be too large of an error in forecasts based upon historical estimates that ignore the shadow inventory.

The nature of the shadow inventory is not clear. The measure from the Census data is only vacant homes, so it does not consist of any owners who are waiting to put their house on the market. The vacant homes most likely consist of a large amount of foreclosed homes that are not listed as for sale. They could also be the former homes of the elderly who have moved to assisted living. When the housing market is in a downturn, the offspring of the elderly may take their time preparing a house before putting it on the market.

The vacant homes could also be abandoned homes. I was only just born during 1974 so I have no personal memory of that downturn, but it seems likely that it was during this period when there was an acceleration in the abandonment of vacant housing units in the cities that contributed to the inner city blight back then. Currently there is a same abandonment going on the midsize former industrial cities of the Midwest. For instance, the New York Times just had a recent piece about abandoned homes in Flint, see "An Effort to Save Flint, Mich, by Shriking It." Therefore, it is not clear whether the vacant units held off market for other reasons are a viable shadow inventory of future homes to be sold, or just an indication of a higher rate of depreciation via abandonment.

IU Econonics Forum

We had an Economics Forum at IU on Friday afternoon. There were several excellent talks put on by very bright economists with ties to IU. Here is a link to all presentations. George von Furstenberg's talk goes over the legal grounds upon which the growth in power of the Treasury, the FDIC and the Fed rests. Unfortunately his comments and turns of phrase cannot be completely gleaned from his slides.

Myself, I tried to explain how standard open-market operations were used to influence the overnight market for liquidity. I attempted to construct an example of IU undergrads lending excess cash to each other to have while they go on dates. The collateral they put for the overnight trades is video games. High quality video-games can be thought of as short-term Treasuries. A guy called Ben uses video-games to control the supply of excess cash--too much cash and the dating market overheats--too little cash and too few people go on dates.

We can think of the onset of the crisis by imagining there are a set of students at IU who go on dates, but don't get to trade directly with this guy Ben. Instead, they create their own video games that work as collateral in the overnight market. Particularly, a couple of guys started to issue `sub-prime' video games. Initially people liked these games, but in the summer 2007 people playing the games realized they were junk. These games were now lemons, and the overnight lending market froze up. Things got so bad in September 2008 that Ben's trading in money and high-quality games become impotent. Essentially, people preferred to keep their money, rather than lend it out or go out on dates. Money and high quality video games were now perfect substitutes, now serving as a store of value. The students at IU were afraid that if they lent, they would become insolvent and they would never be able to go on dates again. Ben doesn't like this situation, so he has now started to trade other types of video-games for money. For instance, he trades lower quality games and games that last longer--such as fantasy or war games that take longer to finish. This creates more liquidity for video-games, but if it will convince the students to go out on dates, we do not yet know. It's not clear how this liquidity solves the insolvency and lemons problems. It does however stem the liquidity problems that have arisen from the insolvency and lemons problems.

Intermediate Macro Blog

I've started my Spring teaching. As a byproduct I'm using a blog for the Intermediate Macro class (undergrad). I start off most classes by going over data relevant to the current crisis. While the data is not directly housing related, I'll post links to them since I think most people will find them interesting.

So far I've posted two entries.

The first covers the fall and subsequent (and quite recent) rapid increase in savings. See "The Trainwreck".

The second post covers the changes in Households' wealth and debt from the Fed's Flow of Funds. See "Household Debt Leading up to the Crisis." The data shows quite clearly the folly in the logic a lot of people (myself included) in thinking that the high levels of mortgage debt to income were justified by the high house values--we had seen that story before and high mortgage debt to income leads to high debt to assets, just with a lag.

The Beveridge Curve in the Housing Market: The Rental Market is more out of Equilibrium than the Owner Market

The long-run equilibrium in the Housing Market is characterized by a Beveridge Curve. In Labour Economics, the Beveridge Curve denotes the negative relationship between the unemployment rate and the number of job vacancies (Rob Shimer maintains an updated curve from the labour market). In my research (slides), I have discovered a Beveridge Curve in the Housing Market given by a negative relationship between the growth rate of households and the vacancy rate. The relationship holds in the owner-occupied market, the rental market, and the total market independent of ownership status.

The relationship for all three markets is plotted in figure 1. On the x-axis is the vacancy rate and on the y-axis is the growth rate of households for each type of market. The data source is the Housing Vacancies Survey put out by the Census Bureau (data in excel). The data have been smoothed slightly due to noise (details in slides). There is a clear negative relationship between the vacancy rate and the growth rate of households in all three markets. Currently all three markets are at their all-time highs for vacancies. (I also have a Beveridge Curve by decades).

What is most striking about the current housing market is that the rental market is significantly more out of equilibrium than the owner occupied market. This suggests that in addition to falling house prices, there is significant pressure for rents to be falling as well. In fact, there is evidence of rents falling (see CalculatedRisk, NYTimes, Guardian).What we also see is that the disequilibrium started back in 2003; the writing was on the wall even back then.

We clearly have an over-accumulation of houses. The current market imbalance is not an own versus rent problem, but a house versus household problem. The price that needs adjusting is not just the rent-price ratio. What needs adjusting is the overall cost of housing services, regardless of ownership status.

We can use the Beveridge Curve to get a deeper understanding of the long-run equilibrium in the housing market. The curve can best be thought of as a supply condition. The demand for new houses comes from household formation. Let’s assume that the cost to produce homes is increasing in the growth rate of the housing stock (as is assumed by Glaeser, Gyourko and Saiz in their papers here and here). When household formation is low, the cost to produce the homes to satisfy the market is low.

There are two adjustments to clear the market when the rate of household formation falls. The first is the standard adjustment that house prices fall. The second adjustment relates to the vacancy rate. When the cost to produce a home is low, a builder may want to produce more homes, potentially adding variety. This lowers the probability of an individual house selling, but it raises the probability that an individual builder may sell one house. In essence, builders respond to the lower costs by raising supply. The higher supply shows up in a higher vacancy rate. At some point in time the vacancy rate rises enough to stem more production and the market returns to equilibrium. Therefore, we get the following result: A lower growth rate of households lowers the costs of production to meet demand, leading to lower prices and higher vacancy rates. Thus, the Beveridge Curve.

Figure 2, illustrates how we should think of the long-run relationship in the Beveridge Curve. When the current market condition is to the northeast, there is an over-supply of homes—the vacancy rate is too high relative to the rate of household formation. High prices could drive the market into this area. When the current market condition is to the southwest there is an undersupply of housing units—the vacancy rate is too low relative to the rate of household formation.

The estimated long-run Beveridge Curve, rendered by the solid lines in figure 1, gives us a value for the current over/under supply of housing units. The best way to measure the amount of over/under supply is to find the current deviation of the vacancy rate from its long run value implied by the current growth rate of households. The deviation provides a measure of over-supply in terms of a percentage of the total housing stock. For the total market, the current over-supply is 1.04% of the total housing stock, or 1.22 million units.

A better way to measure the over-supply is to normalize it by the long-run growth rate of households. We then get a measure of over-supply in terms of years of household formation. That is, the measure is the numbers of years to erase the over-supply if household formation remains at its same rate and no more houses are constructed. This is a ‘years of supply’ measure equivalent to the ‘months of supply’ concept used in New and Existing home sales.

Figure 3 graphs the oversupply for all three markets. We clearly see a large increase in over-supply starting in 2003. Almost all of the over-supply originates from the rental market. Only recently has there been an over-supply in the owner-occupied market. Currently the total over-supply is at 1.03 years of household growth; by far the largest over-supply ever. Note that for this metric, the high rates of household formation in the 1970s kept the over-supply around 1974 low, at 0.29 years. Also, for the entire sample, the rental market is where most of the adjustment takes place—most of the over/under supply is there.

Finally, let me address foreclosures. A foreclosure does not affect the over-supply in the total market. When a household loses its house through foreclosure, the result is a vacant house and a renting household. Provided that the household remains an independent household and does not become homeless or move in with relatives, the household rents out a previously vacant house. The result is no change in the equation between houses and households. This is not to say that foreclosures are not a problem. Foreclosures are affecting the financial system, they create psychological problems for households being foreclosed upon, and a foreclosure can hurt a neighborhood just like any vacant house can hurt a neighborhood. However, foreclosures do not affect the imbalance in the overall housing market. As Glaeser and Gyourko state "the tyranny of housing supply suggests that no credit market intervention of any sort is likely to be able to stop housing price declines."

In summary, there is a Beveridge Curve in the Housing Market that defines a long-run equilibrium. Currently there is an over-supply of 1.22 million units, roughly one year of supply. What is striking is that most of the over-supply is in the rental market, not the owner-occupied market. This suggests that not only does the price-rent ratio need to adjust, but rents also need to adjust.

In a few days I will be posting about what we can expect for household formation. The data used for the Beveridge Curve suggest that this is also bleak.

Fooled by Search: The Great Housing Bubble

In the housing market, people .... just do not know how to judge the overall level of prices. Much more salient in their minds is the rate of increase of prices.

--Robert J. Shiller, Irrational Exuberance, 2nd Ed., 2005. p. 208


In the United States housing market, data suggest that temporary movements in demand are related to permanent movements in house prices, whereas theory predicts that temporary movements in demand should be related to temporary movements in prices. Alternatively, the data suggest that a permanent change in demand leads to a permanent change in the growth rate of prices, whereas theory would predict only a change in the price level of houses. The data and theory are in disagreement. The disagreement is especially striking if we think that market tightness, or the number of buyers relative to sellers, impacts prices via the bargaining process between buyers and sellers. The divergence of data from theory explains almost three-fourths of the rise of detrended housing prices in the bubble that starts in 1998.

The failure of theory can be explained by a behavioral inefficiency, where buyers and sellers who are currently in the market (irrationally) interpret prices of past transactions as the permanent value of a house. If buyers and sellers behave as such, then high demand for houses causes prices to be bid up relative to past prices. Future buyers and sellers then interpret the resulting price change as a change to the permanent value of a house. In this way temporary demand movements cause permanent movements in prices. If demand remains high, then prices continue to be bid up, so that there is an increase in the growth rate of prices. In this way, a permanent increase in demand results in a permanent increase in the growth rate of prices.

As I argue below, interpreting past prices as the permanent value of a house is irrational if the housing market suffers from `search frictions', which are the frictions in the decentralized trade of houses that cause market tightness to impact prices via the bargaining process. When these frictions are present, an increase in demand, even if there is no change to the fundamental value of house, will increase the price of a house by the interactions between sellers and buyers bidding up the price of a house since sellers know that they can easily find another buyer. However, when future market participants interpret that price change as a change in the fundamental value of a house, they are fooled. In essence they are fooled by `search frictions' into thinking that there has been an increase in the fundamental value of a house. As Shiller's quote states, households have a hard time understanding price levels, instead, they understand price increases.

In an estimated model, the irrational interpretation of past prices as the fundamental value of a house is responsible for over half of the rise in house prices from 1998 to 2006. To see this, figure Figure One plots the real price growth from the data along with an estimated counterfactual where the irrational assumption has been removed, marked `Rational Counterfactual.' The counterfactual is the model's prediction for the increase in prices due to high demand for homes over this period, if households had rationally interpreted that part of the past price increases were due to search frictions. Figure One also plots trend price growth. Removing the trend growth of prices, almost three-fourths of the rise in prices can be explained by households' ignorance of search frictions on past prices.

To see the relationship between prices and demand in the data, examine Figure Two. The most striking feature is the line marked `price level,' which is the real price of houses as reported by the Office of Federal Housing Enterprise Oversight (OFHEO), with inflation and the trend growth in prices removed (trend growth is assumed to be the average growth from 1976 to 2003). Clearly, the level of housing prices relative to trend is significantly greater than any that have been observed since the OFHEO series started. Also on the figure are turnover and price growth. Price growth is the percentage increase in prices from year to year, while turnover is the total sales of houses (new and existing, single family) over the total owner stock of homes (consists of owner-occupied houses plus vacant houses for sale as reported by the Census Bureau). Besides the bubble, we easily see in figure two that turnover moves with price growth, not the price level. Note that this is not an implication from the relative magnitudes of the curves, rather, it is the difference in the timing of the peaks and valleys of the curves: the price level lags turnover, while price growth and turnover move together. The correlation between price growth and turnover is 0.89 and turnover explains 79\% of the changes in price growth.

Turnover and price growth moving together suggests that movements in demand drive the housing market. When demand rises, more homes are sold and prices rise. Note that movements in supply would cause prices and turnover to move in opposite directions. The high correlation between turnover and price growth means that temporary movements in demand are related with a permanent change in prices. Alternatively, a permanent increase in demand is related with a permanent increase in the growth rate of prices. To see this in the data, in figure two, examine the period from 1998 to 2002. We see that turnover rose above its long-run level and stayed there for three years. In those same years we see an approximate permanent change in the growth of detrended prices from zero, being on trend, to something much larger.

Example of Turkish Bazaar

Turning to the theory, let's consider a Turkish bazaar where rugs are sold. There are a fixed amount of sellers who know that they can buy rugs for a certain price in the wholesale market. Each day buyers arrive and search for a rug to buy from a seller. Each rug is slightly differentiated and buyers have particular tastes in their rug preferences. Buyers keep searching from seller to seller looking for a rug to buy. Once a buyer finds a rug she likes, she bargains with the seller over a price. The price is influenced by how easily the buyer can find another seller from which to buy a rug. The sellers and buyers may not know all of the other buyers and sellers, but they can see the level of market tightness, given by the number of buyers to sellers, by observing how many buyers are stopping by each store. The tighter is the market, meaning the more buyers there are, the easier it is for a seller to sell the rug to someone else. In bargaining, both the buyer and seller know this, so increased market tightness leads to higher prices.

Now let's suppose that the market is in a steady-state where market tightness and thus the price is the same from day to day. Now suppose one week that there is an increase in the number of buyers. This would result in more sales, raising turnover. Prices would also increase from the increased market tightness. Now suppose the following week the number of buyers falls back down to the usual, or steady-state level. We would expect that the price of rugs would fall back down to the usual price. This analysis suggests that temporary movements in demand should only cause temporary movements in prices.

Instead we could examine the situation where the increase in buyers is permanent. In this situation, we would expect the price increase to also be permanent. This suggests that a permanent rise in demand should permanently raise the price of rugs, but it should not lead to an increase in the growth rate of the price of rugs.

To better understand the implications of the data from the housing market, place the results from figure two in the context of the rug market in the Turkish Bazaar. Consider first the temporary movement in demand. In this case, after the temporary increase in demand, if the price of rugs behaved like U.S. house prices, the price of rugs would remain high the week after the increase in demand, even though demand had fallen back to normal. In this sense, if you were shopping the week after the high demand you would end up paying too much for the rug. In fact, prices would remain high until there was a fall in demand below its usual level, driving prices back down to their steady-state level.

Next consider the permanent increase in demand. In the Turkish Bazaar this should only cause a permanent increase in the price level of rugs. However, the data on house prices would imply a permanent increase in the growth rate of prices. To put this context, if Turkish Rugs behaved like U.S. Houses, a permanent increase in the demand for Turkish Rugs would result in prices growing at a higher rate forever, or at least as long as demand remained high.

What is the difference between the Turkish Bazaar and the United States Housing Market? One explanation could be that the sellers in the bazaar do understand the price level. They know the cost to them to buy another rug in the wholesale market. Prices cannot deviate too much from the level, since some sellers would start to advertise cheaper rugs. The United States House Market does not behave this way. The equivalent of the wholesale market is new construction. But new construction is only a part of the housing market, and for certain types of houses, new construction is a very loose substitute. This would seem to be especially true in the coastal areas where bubbles have historically been more prevalent.

Instead of a wholesale market, there are new sellers (and buyers) entering the housing market each month. They probably have very little idea as to what the fundamental value of house is that would affect the price level. Efficient markets theory (Fama 1970 Journal of Finance) tells them that past prices should reflect all the relevant information, including fundamentals, for prices. But if households ignore that search frictions are part of those fundamentals they are making a mistake. This is especially true if they are then affected by search frictions in their own pricing of houses. In this sense, they recognize that search frictions affect pricing, but ignore that search frictions may have affected past prices, instead interpreting past prices as reflecting the fundamental value of a house in a frictionless world.

Another way to understand the difference, It's as if the Turkish Bazaar had new buyer and sellers each week and that the sellers had already made their own rugs. In addition, at the entrance of the market is a database with all of the transactions from the previous week. The buyers and sellers search the database looking for the prices of comparables to a rug that they are thinking about buying or selling. However, when doing this, the buyers and sellers ignore that the level of activity the previous week, which they can see, may have affected the price level of all of the rugs. When the buyers and sellers trade in the market, the current level of market tightness does affect their bargaining over prices. They then use the past prices as an anchor and bargain relative to them, not relative to a steady-state price level. Therefore, high demand causes increases in prices relative to the previous week. Each week the process repeats itself so that temporary changes in demand lead to permanent price changes, and permanent changes in demand lead to permanent changes in price growth.

Summary

To summarize, the data on the U.S. House Market is consistent with temporary movements in demand causing permanent movements in house prices, equivalently, permanent movements in demand cause permanent changes in the growth rate of prices. This contradicts rational theory that predicts that temporary movements in demand should only cause temporary movements in prices and permanent movements in demand should only reflect an increase in the price level, not price growth. This is especially true if there are search frictions in the housing market where market tightness affects the bargaining between buyers and sellers. Theory can be reconciled with the data by assuming that buyers and sellers interpret past prices as an indicator of the fundamental value of a house in a frictionless world, households are `Fooled by Search.' The past prices then serve as anchor for bargaining. The confusion between price levels and price growth is consistent with Shiller's hypothesis that households have Irrational Exuberance and interpret temporary movements in levels as permanent changes in growth rates.

In order for people to not be fooled by search, a separate housing price index could be started that adjusts the market price of houses to take into account the over and undervaluation of houses due to search frictions. This index would be a type of aggregate index that could be thought of as existing alongside a standard price index such as the OFHEO index or the Case-Shiller index. This index would give a better idea of the fundamental value of housing. Such an index could be useful for mortgage originators to use to think about valuing housing collateral. In addition, the index could help educate the public, to try and forestall the formation of Irrational Exuberance.