Housing-analysis blog creater and owner, and financial planner, "mohican" noticed back in 2007 there was a (negative) correlation between so-called "months of inventory" (MOI), the ratio of reported for-sale inventory to monthly sales, to price changes. MOI tells us how long it will take to clear inventory at current sales rates. Other equivalent measures you may hear are months of supply, sell-list ratio (the inverse of MOI), and months to clear. This blog refers to MOI, and data can be found on REBGV news releases. (Note that Fraser Valley real estate board also published similar stats, and the correlation is good there as well; this analysis is only for REBGV data.)
Price changes are measured using the Teranet HPI, however similar correlations to MOI exist with the MLS-HPI and to a lesser extent median price changes. In order to account for month-month sales noise MOI is averaged over three months. I only have complete monthly inventory data for REBGV to 2005. There are likely data before this but I do not have access to them.
Two further innovations come by first recognizing that correlation changes depending upon the period over which price changes are measured. For example the correlation to quarter-on-quarter price changes is better than month-on-month price changes. Second the correlation becomes better if we time-shift the data (i.e. cross-correlate).
Below is a graph of the cross-correlations between price changes over different intervals and 3-month-moving-average MOI:
What this shows is that the highest correlation exists when correlating half-on-half price changes with a 3 month delay. The scatterplot showing this relationship is below:
Note MOI is displayed on as logarithmic scale. Other correlations are not as good but still not bad. Looking at year-on-year and quarter-on-quarter, both appropriately time-shifted, show the relationship is not as high but still obvious:
The most recent price weakness is shown on the QOQ graph above, indeed 2012 is behaving as previous years.
An important, and powerful, note on these graphs is that because of the time-shifted peak correlation, only MOI numbers from the past are displayed. For example the HOH graph only displays MOI to 3 months ago, the YOY graph only displays MOI to 6 months ago, and the QOQ graph only displays MOI to last month. That means that if we know MOI today (and we do thanks to Realtor Paul Boenisch posting daily sales, listings, and inventory on vancouvercondo.info) we can use this model to predict prices some months out.
Take the HOH graph above and apply current 3 month average MOI (about 10) to the graph. That implies that we should be seeing a -5% half-on-half change in the Teranet HPI 3 months from now, in January 2013. A weaker measure is to use the YOY graph and forecast prices 6 months from now; based on that measure we should see the Teranet HPI down -4% to -6% year-on-year in April 2013.
In summary a method of forecasting Teranet HPI for Vancouver has been presented. By finding peak negative cross-correlation between months of inventory and price changes over defined periods it is possible to calculate predictions with varying degrees of strength. Further work expanding this method into other cities is possible.