Benching the benchmark. REBGV House Price Index.

In this blog post, I want to discuss something that comes up every month: the benchmark price or each unit type for Greater Vancouver. This will become also relevant for a subsequent blog post where I’ll discuss the correlation between Sales to Active Listing Ratios and Price changes.

Every month when the REBGV releases its report (https://www.rebgv.org/monthly-reports), they discuss price changes using their benchmark House Price Index. On their website (https://www.rebgv.org/mls-home-price-index) they describe it as following:

The MLS® Home Price Index is modelled on the Consumer Price Index which measures the rate of price change for a basket of goods and services including food, clothing, shelter, and transportation.

Instead of measuring goods and services, the HPI measures the change in the price of housing features. Thus, the HPI measures typical, pure price change (inflation or deflation).

The HPI benchmarks represent the price of a typical property within each market. The HPI takes into consideration what averages and medians do not – items such as lot size, age, number of rooms, etc. These features become the composite of the ‘typical house’ in a given area.

Each month’s sales determine the current prices paid for bedrooms, bathrooms, fireplaces, etc. and apply those new values to the ‘typical’ house model.

In their methodology section (https://www.rebgv.org/mls%C2%AE-home-price-index-methodology), they discuss at a high level how they calculate the number, but they do not provide an example and/or disclose many of the parameters they used in their ‘typical’ house calculations. Moreover, they indicate the following:

In keeping with best practices, results are filtered to include records with values above 2.5% and below 97.5% of cumulative Normal distributions; other results are treated as outliers and automatically removed. To mitigate volatility, a moving five-year period is used, since the use of a shorter sample horizon may result in an insufficient number of sales over the period and cause index inaccuracies.

I assume that they tweak the weight of the data to emphasize recent numbers over those from 5 years ago, but I could not find a description of how they calculate this. Nevertheless, since the benchmark is a moving average, it is expected to have a lag. How much? To answer this, we can compare the HPI to the average price.

What is the lag of benchmark prices when compared to average prices?

I downloaded the Market Stat numbers that Steven Saretsky provides on his website: Market Stats. I then compared the average vs the benchmark (note that the axes are different)

Screen Shot 2018-11-04 at 12.51.07 PM.png

This visualization confirms that the detached benchmark appears lags the average values by a few months (shown in the figure). By shifting the average values by different number of months, the best correlations were obtained if the average prices were correlated to the benchmark prices 1,3 or 4 months later for the condo, townhome and detached units, respectively (top is correlation without shift, bottom is correlation with corresponding shift):

Screen Shot 2018-11-04 at 12.44.56 PM.png

Visualizing the effect of the shift by time shows the improvement:

Screen Shot 2018-11-04 at 12.52.20 PM.png

Since the benchmark uses a window of 5 years without a description of the weight of each point, and the average value contains data only for the month in question, it appears that the benchmark values are behind in showing recent changes in prices, particularly for the detached market. 

This is important as the mainstream media reports benchmark prices as the latest in price changes. This analysis shows that benchmark prices reflect changes in the market from a few months back, particularly for the detached house market. So either the average value is used (which is noisier and more susceptible to seasonal changes), or the benchmark price should be used to analyze how the market was a few months back.

It is interesting that the lag appears to be different for each unit type. Without knowing exactly how they calculate these values, it is difficult to figure out why the lags are different for each unit type. If you have some ideas, write them in the comment section below.

Excel with all the numbers: benchmark-vs-average

admin
Leave a Reply