The state of the Australian property market is generally measured by the “Corelogic Hedonic Price Index”. No index is 100% accurate in assessing the state of the market. I’ve done some relatively deep research into the index including a call with Corelogic’s Head of Research Tim Lawless and I can say that the Corelogic Hedonic index is generally a very well designed piece of statistics. However that doesn’t mean it’s perfect, the index has never been tested by market dynamics like what has transpired through Coronavirus, and due to a technicality in how the index is calculated, I am confident that the index is currently overstating the true values in the property market.
The below may be a little bit technical for some people, but for a TL;DR summary consider the following metaphor; the Corelogic index is a cruise ship and the destination is the true market. The true market (our destination) has just changed because now the cruise ship is riddled with Covid, so Corelogic is spinning the steering wheel to change our direction. However right now, even though we have spun the wheel, the ship is still not pointing toward our true destination (and it is taking some time to adjust course). Read more to understand exactly why this is.
A quick summary — how would the Hedonic value the entire market today?
- The index includes all sales conducted over the last 360 days as a starting point.
- The index then uses those sales to estimate how much buyers have been willing to pay for certain property features. For example, on average buyers in Manly have spent $30k incremental for a pool. This is done for all modelled property features to build a baseline estimate of each component of a property.
- Each property’s components are used to give a baseline estimate of the value of the property formed by the last 365 days sales (under the assumption that a property is the sum of its parts).
- To account for the market trend (up or down) of the current market relative to the last 365 days, a single additional trend component is added to adjust the baseline estimate from point 3, under the same approach as above (and based on all sales over the last 365 days).
Point 4 is incredibly important overall to the growth / decline of the index relative to prior months, since it will capture most of the change that is due to the overall market trend.
This methodology is solid under normal conditions, however importantly, point 4 has some serious limitations under extreme and sudden changes to the market (and exacerbated by low sales volumes) that I believe are compromising the integrity of the index right now.
To understand point 4, consider the diagram below of how this would be modelled under what could be considered normal conditions. In the diagram below, I represent each sale as a cross and the trend line as how the market is viewing whether the market is currently hot or cold. The fact the trend is going up means that the market is currently hot (apologies for scratchy sketches to illustrate these concepts):
Under these conditions, the straight trend line that the Hedonic fits to represent the market trend does a good job of representing that the market is generally ‘hot’ right now. I.e. buyers are generally willing to pay more today than the average sale in the last 360 days.
However, we know that after Covid-19 first hit, two factors immediately influenced the property market as were widely publicised:
- The volume of sales after the start of the forced lock-down dramatically dropped due to mandated online auctions.
- It is reasonable to expect that the price of those sales took a somewhat dramatic drop also.
Due to this, under these conditions, it could reasonably be expected that the state of the market could be represented to look something like the below (I’ve exaggerated some of the relevant defining characteristics to make this easier to visualise, however nonetheless the principle is the same):
From this we can observe a few important points:
- After the forced lockdown / crash date, there are relatively few sales when compared to before this date due to 2 factors. Forced online auctions and the recency with which the crash took place mean there are far more pre-crash sales than post crash sales.
- Due to how the ordinary least squares (OLS) regression algorithm works, this means the market “trend” dynamic would be overly biased toward the pre-crash sales instead of the post-crash sales.
- Because of this, my personal belief is that the market trend line is currently catching up to the true state of the market. This theory is supported by the continued month-on-month gradual decline we are observing in the index. The truth is the true market isn’t necessarily still falling now, it’s already fallen, and the Hedonic index is slowly adjusting course to catch up.
What can we expect moving forward?
Well, consider how the diagram above would look come September. Considering the start of the lockdown was at the end of March, at this point in time if it were true that there were essentially two different markets (pre-Covid and post-Covid) come September we would have 50% of the data points on either side of the “crash” line:
Shown above is what we can expect come September, with half the data points both before and after the crash line, we can expect that the trend line would do an accurate job of pointing toward the true state of prices (i.e. the end of the trend line is landing close to the true state of prices).
Due to this my personal belief is that we will observe the following in the coming months:
- The Corelogic Hedonic index will continue to fall until September. I am confident of this point. Come September the index will more accurately reflect the true market valuations.
- Beyond September, whether the index continues to fall further will depend on whether the market continues to weaken or bolster beyond this point. This is anyone’s guess, however I believe that the falling index will damage expectations at a time when the market will already be very vulnerable due to other September market changes. These factors combined will probably be more damaging for confidence and prices.
What do you think? I always value feedback from either a technical or an anecdotal perspective.
What can be done to improve the index?
If you’re not a stats / maths person then I’d suggest switching off at this point (thanks for coming!).
I want to stress in this section that these are just my ideas for how the index could be improved to make it more responsive, however I haven’t (obviously) had the opportunity to back-test the performance of this alternative across years of real data points to confirm that my proposal would outperform the Corelogic Hedonic index.
My idea for how to improve the index is to:
- Use a weighted OLS regression that gives increased weight to more recent sales over older sales. This would give a more current “baseline” estimate of the index.
- Retain the use of a linear market directionality factor in calculating the index, however calculate this factor using the same weighting of sales as proposed above. This would give a more current directionality of the market that would be expected to be more responsive to sudden changes.
Concretely, Corelogic’s Hedonic index uses an OLS regression with cost function below:
I would propose they switch to the below to improve the index’s responsiveness to sudden changes (while hopefully not compromising accuracy in more stable conditions):
So why haven’t Corelogic done this themselves? My understanding is that because market conditions like today’s have never tested the index, there has been limited appetite to adopt a more complex implementation of the index (since the existing index has performed well under normal market conditions). There is also the fact that the index’s current implementation is aligned to European standards for best-practice in how to calculate house price indices. However, having said this, I personally remain convinced that the above would outperform the current Corelogic index, and that the difference in the two methodologies would be especially obvious today.
