|Area||Full year 2011 Actual||Full year 2012 Proj||% chg from 2011||2012 Actual||Err % Diff to proj||Act 2012 % Diff to 2011||Forecast 2013||2013 F’cast to 2012 Act||Method Unemp or last Q|
Note: the error bars are at +/- 10%
The above table summarizes last year’s results and our expectations for 2013. These predictions assume no “major events”. As expected with any new trends model in its inaugural trial run, the 2012 predictions were mixed. Alabama residential saleswere up 5.9 percent in 2012.
In Huntsville, the forecast was within 2% of the full year 2012 results. In the aggregate, the results pointed in the right direction right, but with some error so this led the team to explore some alternative approaches for this year’s predictions to improve accuracy. In the markets with the greatest error we revised our methods to use the last quarter of 2012 sales via straight line linear regression instead of the unemployment rate. This method appears to be more accurate in most markets historically and hopefully going forward.
With these adjustments, above is the overview of what ABRE Analytics think might happen in 2013. As for the projection of a 12% increase for Tuscaloosa that is out of line with the other markets, ABRE’s opinion is that this may be a little too optimistic, but we have yet to figure out a consistent method to arrive at a better projection for Tuscaloosa in 2013. Of course, the local market’s near-term response to the tornado of April 27, 2011 certainly has a role with the difficulty in identifying a projection that could be presented with more confidence. Last year we experienced a similar issue with Montgomery, which prompted us to develop the alternative methodology of using last quarter sales for predictions.
The method ABRE used for 2012 was based entirely on the January unemployment rate for a market area. The assumption was that the January unemployment rate eliminated the effect of holiday temporary work and would reflect the mood of the populace towards buying and selling a new home in the upcoming year. A standard linear regression line yielded a better than 80% correlation since 2004 in all areas. The standard error however is somewhat high. This method also has the benefit of using two longer term trends, unemployment and home sales, and only at a single point per year, which eliminates a lot of “noise” in both series.
The seasonal regularity of sales is such that if you know the total sales for a year, dividing the total by the average proportion of the yearly sales attributable to a month has shown to be remarkably stable. Exceptions to the regularity do occur, such as fiscal cliff drama and the tax credit for first time home buyers in 2009. ABRE eliminates this data when calculating the monthly spreads.
ABRE originally chose unemployment data as it is one of the more timely pieces of data released by the government, as well as being released by geographies that generally correspond to the reported real estate market areas. Other data from various government agencies are released so late as to not be timely enough for meaningful projections.
So, what ABRE presents is our best estimate of next year’s sales which specifically excludes the possible impact of unpredictable governmental action or inaction, although even this seems to be having less impact as both the populace and markets begin to ignore political histrionics. For what it is worth, ABRE did test everyone’s favorite housing predictor, interest rates, and could not find any useful correlation.
In each case ABRE experiments with prior years data to see how well the prediction methodology would have worked. We tried using the last quarter sales of the prior year to “regress” against the full year sales of the projected (next) year. This is based on the same premise that recent data may be indicative of future results.
In each case, except Huntsville and Tuscaloosa, this methodology (last quarter of the year to the following year) resulted in greater historical accuracy in predicting and considerably better correlation numbers. We did exclude 2010 from the analysis since the last quarter of 2009 had abnormally low sales, (although the full year sales were as expected), due to the 1st time homeowners tax credit that pulled sales into earlier quarters and depressed the year end.
Complete spreadsheets with all data are available as public Google spreadsheets, which also include month by month projections, at http://goo.gl/jtJGW. ABRE presents these projections as a “work in progress” and as a tool for assessing how well current sales are performing against some level of “informed” expectation. You should not rely on them, but nonetheless ABRE hopes the projections are found useful. ABRE welcomes comments and suggestions for improvement.
About ABRE Analytics: Strategic collaboration is one of the keys to accelerating the flow of insights in the 21st century. The Alabama Center for Real Estate (ACRE) and Tom Brander has been successfully collaborating since 2009. The flow of ideas stemming from this relationship have led to solutions to better serve the Alabama real estate industry and consumers. ABRE (ACRE/Brander Real Estate) Analytics is designed to foster future creative thinking while also providing hands-on experience for student interns of ACRE.