by Calculated Risk on 8/07/2007 06:00:00 PM

This is an attempt to forecast how much further housing starts will decline.

Over the first half of 2007, housing starts averaged a 1.46 million unit annual pace. (data from Census Bureau: Housing Starts)

Over the same period, New Home sales have averaged a 0.87 million annual pace. (data: New Home sales)

This brings up a key point: New Home sales come from a subset of housing starts. Housing starts also include owner built units, rental apartments, and other units that would still not be included, if sold, in the New Home sales report. So when we try to forecast the decline in housing starts, based mostly on the dynamics of the new home market, we are assuming that starts for the other units will remain steady.

Click on graph for larger image.

This graph shows total starts, starts of single unit structures, and new home sales since 1970.

During some periods (early and late-70s, mid'80s) there was a surge of apartments being built (think baby boomers leaving the nest in the '70s). But in all periods, you can't compare starts (either total or one unit structures) directly to new home sales.

Definition: New Home sales

Another key point: the new home sales estimate reported by the Census Bureau includes only new single-family residential structures that include both the structure and the land. The Census Bureau defines single-family homes as either fully detached structures or certain attached homes with an unbroken ground-to-roof separating wall. This definition includes some condominiums (side by side units), but does not include condominium units with another unit above or below.

Although large multi-story condominium projects account for a small percentage of housing starts in the U.S., it is important to note that sales and inventory for these units are not included in the New Home sales report, but that the starts are included in the housing starts report. Based on anecdotal evidence, there is a large number of these units currently for sale - and it is probably reasonable to assume that starts will decline significantly for these large condominium projects.

For data junkies, the Census Bureau provides a quarterly report that breaks down the data by many of the above definitions: Quarterly Housing Starts by Purpose of Construction and Design Type

One more point: the National Association of Realtors (NAR) reports total units sold for existing homes, including all condominiums and co-ops. These different definitions can be confusing.

Inventory

This brings us to the inventory data reported by the Census Bureau. This inventory is for new single-family residential structures as defined above and therefore does not include some condominiums and co-ops. Also cancellations aren't included in the data and this can skew the inventory numbers during periods of significant changes in cancellations. See: How does the Census Bureau handle cancelled sales contracts in the published estimates of New Home Sales?

The Census Bureau breaks down the inventory as Completed, Under Construction, and Not Started. The following chart shows the inventory levels, and months of supply, excluding the "Not Started" category.


The median months of supply for hard inventory (competed or under construction) is just under 5 months since 1974. To reduce the inventory to the median (assuming sales stay steady) would mean an inventory decline of around 100K units. Some estimates suggest there might be another 100K+ completed units due to cancellations, so let's call it 200K units of excess inventory.

If the builders worked off the 200K excess units over 2 years, then starts would need to fall to 1.36 million units per year.

Other Excess Inventory

In addition to all the assumptions above, there are two additional problems with the above estimate: 1) sales will likely fall further due to tighter lending standards, and 2) there is substantial excess inventory that is coming back on the market from desperate sellers and foreclosures.

We can use the Residential Vacancies and Homeownership report from the Census Bureau to estimate the excess inventory.

Click on graph for larger image.

The first graph shows the vacancy rate of homeowner units for sale since 1956. From the Census Bureau: "The homeowner vacancy rate is the proportion of the homeowner inventory that is vacant for sale." A normal rate for recent years appears to be about 1.7%. The small decline in Q2 leaves the homeowner vacancy rate almost 1% above normal, or about 750 thousand excess homes.

Rental units are competing products for new homes too. The rental vacancy rate has been trending down for almost 3 years (with some noise). This was due to a decline in the total number of rental units in 2004, and more recently due to more households choosing renting over owning.

It's hard to define a "normal" rental vacancy rate based on the historical series, but we can probably expect the rate to trend back towards 8%. This would suggest there are about 600 thousand excess rental units in the U.S. that need to be absorbed.

This approach would suggest there are between 750K (homeowner units only) and 1.35 million excess housing units in the U.S. Perhaps the excess rental units will keep pressure on other areas of Starts, and we should just focus on the excess homeowner units. To work off 750K units over two years - assuming sales stay steady (unlikely) - housing starts would have to fall to about 1.1 million units per year. This housing is "permanent site" and isn't transportable, so some regions may have a shortage of units and other areas may have more excess inventory - so as a reminder, this is just a rough estimate for the aggregate national market.

There are several significant assumptions in this approach, but my expectation is that starts will fall to around the 1.1 million units per year level; a substantial decline from the current level.

UPDATE: BusinessWeek has some more on the excess inventory: Another Reason For Those Empty Houses (hat tip James)