In Depth Analysis: CalculatedRisk Newsletter on Real Estate (Ad Free) Read it here.

Friday, March 14, 2014

Kolko: Where Do Housing “Leading Indicators” Lead Us?

by Calculated Risk on 3/14/2014 07:31:00 PM

CR Note: This is from Trulia chief economist Jed Kolko:

How well do leading indicators predict housing activity? In theory they should if housing construction and home purchases follow a logical sequence. When homes are sold, contracts are signed (as measured by NAR’s pending home sales index) before sales close (NAR existing home sales); also, many buyers apply for mortgages (MBA purchase applications index) before a sale closes. Also, when new homes are built, they get permits (Census new home permits) before construction begins (Census new home starts); and roughly two-thirds of new home sales happen when homes are under construction or completed, rather than before they’re started.

But lots of factors can erode the link between leading indicators and the activities they foreshadow. Some homes inevitably fail to follow the standard paths: some sales under contract may fail to close, some permitted units might not get built, and – especially now – homes can be purchased with cash and therefore skip the mortgage-application step altogether. Unanticipated events can break the link, too: bad weather or a sudden crisis that hurts confidence could delay construction on already-permitted units.

Simple correlations and time-series regressions show empirically how well leading indicators actually predict key housing measures. Because the relationships between indicators can change over time, it’s helpful to focus on the most recent years of data. (The real test is whether leading indicators predict month-over-month changes, not year-over-year changes, since eleven months of a year-over-year change are already known before each monthly release of a year-over-year number.)

Based on national housing measures from 2008 to the present, the leading-indicator crystal ball is generally pretty cloudy, though better for existing home sales than for new home starts or sales.

1. Existing home sales

Existing home sales , which are closings, tend to follow pending home sales by one or two months. Pending sales turn out to be a reasonably good leading indicator of existing sales. The correlation between the month-over-month change (m/m) in existing sales and the m/m change in pending sales from one month earlier is 0.45; the correlation with the pending sales from two months earlier is also 0.45. The correlation between existing sales m/m and the average of the one and two month lags of pending sales m/m is 0.70.

How strong is this relationship? That’s weaker than, say, the very tight relationship between the change in the 30-year fixed rate and the same-month change in the MBA refinance index (correlation = -0.84). But the pending-existing sales relationship is stronger than any of the construction-related measures, as we’ll see next.

Putting that into a simple time-series regression shows that the best predictor of the existing sales m/m change is the simple average of pending sales m/m changes from one and two months earlier. Including the m/m change in the MBA purchase-application index from one and two months earlier improves the prediction of existing sales only minimally. The m/m changes in pending home sales for December and January suggest a 3% m/m drop in existing home sales in February.

2. New home starts

Building permits are a weaker predictor of new-home starts than pending sales are of existing home sales. The correlation between the m/m change in starts and the previous month’s m/m change in permits is 0.45, but permits from more than one month back bear little relation to starts. Rather, the m/m change in starts is partly correlated with the same month’s change in permits. That means that changes in new home starts tend to run, on average, half a month behind changes in new home permits. Since permits data are released in the same monthly Census report as new home starts, the value of permits as a leading indicator is limited.

The historical data suggests that the best guess of this month’s m/m change in starts is 0.6 times last month’s m/m change in permits. Permits fell m/m in January by 5%, pointing to a 3% drop in starts in February. But the relationship isn’t very tight, so even that best guess will often be way off.

3. New home sales

It’s even harder to predict new home sales based on other indicators. The m/m change in new home sales is most strongly correlated with the same month change in new single-family permits . Since the Census releases new home sales data one week after the release of starts and permits data for the same month, this month’s change in permits gives a few-day-ahead hint at the same month’s change in new home sales, but only because of the reporting lag.

Overall, the leading indicators don’t get us that far. Changes in pending home sales predict changes in existing home sales reasonably well, when the changes from one and two months earlier are averaged. But lagged pending home sales probably don’t beat out aggregated data from local realtor/MLS reports (regularly posted here on Calculated Risk) as an early look at existing home sales. On the construction side, building permits are halfway between a leading indicator and a concurrent indicator of starts, and neither is a good leading indicator of changes in new home sales. The modest value of leading indicators means that every month there’s plenty of room for housing-data surprises.