by NAR Research economists Danielle Hale and Hua Zhong
Home sales vary in (mostly) predictable patterns based on the month of the year and the days of the week in each month. Find out what data is the best estimate of real trends and not noise in the housing market in this article.
With the January Existing Home Sales data release, NAR Research released revised seasonal adjusted annual rate (SAAR) data for the last 3 years because we re-estimate and forecast new seasonal adjustment factors. Seasonal adjustment factors are used to try to extract a meaningful trend from the noisy home sales data that has mostly predictable big seasonal moves.
Imagine the headlines: “Home sales plummet 20 to 30 percent in January from December!” followed by “A recovery of 20 to 40 percent in home sales from February to March!” Those would have been the stories every year for the last decade if home sales data were not seasonally adjusted. But how helpful would that information have been for figuring out what was going on in the housing market?
In this article, we present answers to commonly asked questions (particularly from Wall Street analysts and journalists) on seasonality in housing data and a brief discussion of why seasonal factors don’t line up from year to year.
What is affected by the seasonal adjustment revision?
Only the monthly SAAR data change as a result of this revision. The monthly unadjusted data does not change and since the annual data is the sum of the monthly unadjusted data, that is also unaffected. Thus, the total sales for the year will not change but we may realize that June sales were a little worse than we thought before while November sales were a little better.
Why seasonally adjust the data?
In short, we seasonally adjust because the housing market has a fairly predictable pattern of many sales in the spring/summer and fewer sales in the fall/winter. See this commentary for a fuller discussion or click here to see several other articles discussing seasonality in home sales.
Why aren’t seasonal factors the same for each month every year?
Like many organizations, NAR employs a model for seasonal adjustment developed by the Census Bureau called X-12. Using this model, in addition to month to month seasonal effects discussed above, economists can estimate and adjust for trading day (weekend vs weekday or day of week effects) and holidays. To see why such an adjustment might be helpful in smoothing the data, let’s take a look at some closing data from January 2014 from a handful of MLSs that are part of the EHS program.
As shown in the chart above, there are 6 major spikes for closings clustered at: the middle of the month, the end of the month, and on Fridays. There are also considerably fewer closed sales on Saturdays and Sundays—less than 4 percent of the whole month’s closings even though they were more than a quarter of the days in the month. Without evaluating the reasons for those spikes (paychecks, moving and working schedules, amortization, etc.) we know that this pattern holds and the seasonal adjustment process adjusts for it in addition to the month-to-month seasonal trend.
How much of a difference could it make? While the days of the week don’t vary much year to year and month to month (all months have either 4 or 5 of each weekday per month) the total work days in a month can vary from 20 to 23 depending on the month, and holidays can eat even further into that variation. For an extreme example, February 2017, which begins on a Wednesday, has 20 working days one of which, Monday, February 20, 2017 is the Federal Holiday for Washington’s Birthday aka President’s Day. The following month, March 2017, which also begins on a Wednesday, has 23 working days, none of which is a Federal Holiday (perhaps to the dismay of the Irish). Given that the majority of transactions fall on weekdays, having 3 or 4 fewer of them in February vs. March can have a dramatic effect on the number of closings that happen.
Sometimes these day-of-week variations line up in funny ways year to year. For example, November 2013 had 21 working days (not counting holidays) 5 of which were Fridays. By comparison, November 2014 will have only 20 working days and only 4 Fridays. This affects the seasonal adjustment factor for November 2013 vs November 2014, by as much as 4 to 6 percent depending on the region.
So what’s the major take away?
Home sales vary in (mostly) predictable patterns based on the month of the year and the days of the week in each month. NAR releases seasonally adjusted home sales data because this is the best estimate of real trends and not noise in the housing market. This doesn’t mean that the data is completely free of noise, but the SAAR is the best picture of national and regional trends in the housing market.