What is delayed may happen later. What do stats today tell us about tomorrow?

In this blog post I want to extend the analysis that I did using the numbers from Steve Saretsky’s website for the City of Vancouver to the numbers for the whole Real Estate Board of Greater Vancouver, which covers a lot more cities. (For your reference, areas covered by Real Estate Board of Greater Vancouver include: Whistler, Sunshine Coast, Squamish, West Vancouver, North Vancouver, Vancouver, Burnaby, New Westminster, Richmond, Port Moody, Port Coquitlam, Coquitlam, Pitt Meadows, Maple Ridge, and South Delta).

It was a lot of work to get this data. Johnny’s daughter went into the REBGV site to download the data entry by entry to get sales data. Thank you Aili! I also spent a few hours getting the inventory data. So now that we have the data, the questions I want to answer are the following:

  • There are two price indices available, the House Price Index posted by the CREA, and the Teranet Index. How do they compare?
  • What is the correlation between the Sales to Active Listing Ratio and the price indices?
  • What is the seasonality for sales, inventory and Sales to Active Listing ratios for the REBGV?

HPI vs Teranet

I discussed previously the HPI methodology. HPI data can be downloaded from here. The Teranet uses a different methodology explained in this video. It basically looks at sale prices of houses that have sold at least twice and intrapolates between them. Here’s a plot of how the two indices compare (since Jan 2008, as you’ll see below, that’s how far we could get SAL data so I limit the analysis for this period) :

It is clear that the two indices are highly correlated, here’s the correlation plot:

However, it does look like the Teranet Index lags the HPI. If you compute the Pearson correlation with the excel CORREL function for different delays you get the following:

Delay (months)Correlation
00.9935
10.9950
20.9946
30.9924
40.9886

A one month delay produces the best correlation. If you shift the Teranet index by one month you get the following plot:

Hence it appears that the Teranet index lags the HPI by one month. This in a way makes sense, as Teranet relies on land registries, which tend to happen weeks after a sale is reported in the MLS, which is what the HPI uses to compute its value.

Excel file with calculations and plots:

Correlation between SAL and the indices

In my previous blog post I discussed this correlation for the numbers of the city of Vancouver. Now let’s see if this correlation holds for the REBGV numbers.

Same as before, the monthly price changes are a lot noisier and suffer from seasonality than Y/Y price changes, but at the same time, Y/Y price changes suffer from an introduction of a lag due to the introduction of a rolling average. Below is the Pearson correlation between SAL and monthly/yearly price changes in the HPI:

Delay (months)Monthly Price ChangeY/Y Price change
00.8260.721
10.7720.824
20.6800.886
30.5280.884
40.3570.817

The correlation between SAL and monthly price changes was the greatest when no delay was introduced, which can be seen on this graph:

As seen, the price changes are very noisy, if we take the Y/Y price change, accounting for the 2-3 month delay we introduced, the plot looks this way

The best correlation with the Teranet index came with a 4 month delay (consistent with our discussion above):

The correlation between HPI price changes and SAL can be seen here:

This is in agreement with what we already discussed in the previous post. What I just realized is that the relationship appeared logarithmic at the left side. If you make the X-axis logarithmic and do a logarithmic fit, this is how it looks:

The values for the intercept are the following

FitIntercept95% confidence interval from the fit
Y/Y15.6%14.8-16.3%
Monthly15.3%15.4-17.4%

Since the confidence interval is ~14-18%. As the equation is in the form of y=ln(x) this means is that if the SAL is below 14% prices tend to decline in an exponential manner!, and above 18% prices tend to go up not exponentially. See this graph with the plot without a logarithmic scale on the x -axis:

In other words, small price differences are expected when SAL is close to 15%, but the farther it moves from it on the lower side, the greatest the price change will be, exponentially! This is illustrated in the following table (values within the 95% confidence interval are highlighted in bold, values computed using the best fit function from the graphs above):

SALExpected Y/Y change (2-month delayed)Expected monthly price change
5.00%-12.4%-2.4%
6.00%-10.4%-2.0%
7.00%-8.7%-1.7%
8.00%-7.3%-1.5%
9.00%-6.0%-1.2%
10.00%-4.9%-1.0%
11.00%-3.8%-0.8%
12.00%-2.9%-0.6%
13.00%-2.0%-0.5%
14.00%-1.2%-0.3%
15.00%-0.4%-0.2%
16.00%0.3%0.0%
17.00%0.9%0.1%
18.00%1.6%0.2%
19.00%2.2%0.3%
20.00%2.7%0.4%
21.00%3.2%0.5%
22.00%3.8%0.6%
23.00%4.2%0.7%
24.00%4.7%0.8%
25.00%5.1%0.9%
70.00%16.4%3.0%

This is not too far from what the REBGV states, which is that SAL of <12% is buyers’ market, 12-20% is balanced, and >20% is sellers’ market.

One thing to note is the data points for October-December 2016, where the SAL was about 24-27% but there were declines in the indices. Upon inspection, at that time the SAL for detached houses was very low (see the numbers here) after the foreign buyers taxes introduced in July 2016, but the SAL for strata was still high, so detached prices were falling bringing down the index.

Excel with calculations:

Seasonality

I also wanted to corroborate with the REBGV numbers what we discussed in a previous post. So let’s look at the heat maps of the sales, total inventory and SAL for the last 11 years:

Excel with the numbers:

A few observations are clear, the strongest months for the SAL are March-June:

But once again, there is a much larger distribution of values for a given month against different years:

Box plots showing 1.5 IQR regions, see https://en.wikipedia.org/wiki/Interquartile_range

than for the same year against different months:

Box plots showing 1.5 IQR regions, see https://en.wikipedia.org/wiki/Interquartile_range

Summary

I have shown that HPI and Teranet indices have a high level of correlation. Teranet tends to lag the HPI, and as I discussed before, HPI also lags average values.

Then I have shown that SAL is quite predictive for price changes. The data suggests that when the SAL is below ~14% prices drop, when above ~18%, prices go up. The price change seems to go up exponentially with respect to where the SAL is with respect to this threshold. Although there is seasonality in the data, where March-June are the strongest months of the year (and January is the weakest), there seems to be more correlation between months of the same year than between the same month across different years.

Given that the SAL for Vancouver has been <14% since October, and that there is a delay in the price changes in the indices, I expect the HPI and teranet to continue to decline for the next month. In my next post I’ll discuss the May 2019 SAL number in comparison to previous years. At the moment it appears that it will be ~18%, so prices are likely going to stabilize for a little while. I’ll be very interested to see what happens in June, and the months to come.

Correction: I corrected a few lines in the language SAL vs price change and put a new plot showing the y=ln(x) nature of the correlation. Thanks to Soma for pointing out the error in the original language.

Sep17, 2019: Corrected sales from Jan 2011 which was originally reported as 2219, but it was actually 1819. Corrected the SAL.xlsx excel file, the sales number and SAL in the charts. I did not correct all the other charts as this error had a very small effect in the numbers.

admin

One comment

  1. […] to April recorded among the slowest sales in decades. As we discussed in our previous posts (1,2). For price changes, what matters is the Sales to Active Listing ratio (SAL), a measure of supply […]

Leave a Reply