The author has been described by News Ltd as an "iconoclast", "Svengali", a pollie's "economist muse", and "pungently accurate". Fairfax says he is a "Renaissance man" and "one of Australia’s most respected analysts." Stephen Koukoulas concludes that he is "85% right", and "would make a great Opposition leader." Terry McCrann claims the author thinks "‘nuance’ is a trendy village in the south of France", but can be "scintillating" when he thinks "clearly". The ACTU reckons he’s "an enigma wrapped in a Bloomberg terminal, wrapped in some apparently well-honed abs."

Tuesday, August 3, 2010

Why the ABS house price index results will likely be artificially positive

What will the ABS print tomorrow?

The ABS ‘stratified median’ detached house price index should print at between +1% and +2% for the June quarter based on our internal analysis of a detached house, pooled quarterly (nb: not monthly), stratified median index model (one can, of course, never be certain!).

Using this method, RP Data-Rismark gets a +1.2% growth rate for the June quarter, although since the ABS collects fewer sales than RP Data the risk is that the ABS number is higher due to the relatively lower representation of the poorer June month in a ‘pooled quarterly’ model (see more below).

This is presumably the reason why RP Data-Rismark’s stratified median all dwellings June quarter estimate of +1.2% (nb: pooled quarterly not monthly) is significantly less than the +2.4% June quarter number estimated by Australian Property Monitors (APM).

RP Data-Rismark’s index likely has a greater weighting to the month of June due to the fact that RP Data, which is Australia’s largest real estate information provider and spends over $9 million each year collecting new data, has the biggest home sales database in the country. RP Data gets especially good ‘real-time’ sales data from the circa 80% of real estate agents across the nation who use RP Data’s software (this data is ultimately cross-referenced with the ‘final’ State government value general sales and is exceedingly accurate).

Why are the ABS (and APM) numbers likely biased?

Setting aside the fact that the ABS’s detached house capital city index excludes the roughly one-quarter of all homes that are not detached properties (they are apartments, semis, and terraces), the ABS’s (and APM’s) results are further biased by the ‘pooled quarterly’ (not monthly) approach that treats the three months in any quarter as one, and is therefore much more strongly affected by the earlier (April and May) months as opposed to the latest month (June) for which there is, definitionally, much less sales data.

As a rough heuristic, one month after the end of any given quarter the ABS and APM will probably have about 25% of the final population of home sales for the third (ie, most recent) month, around 50% of all sales for the second month, and circa 80% of all sales for the first month in the quarter. A solid, but nevertheless imperfect, sample.

Another problem for the ABS is that around 40% of all homes are not located in capital cities, which is why RP Data-Rismark recently developed a ‘rest of state’ hedonic house price index for the RBA. Unsurprisingly, growth rates in the rest of state markets across Australia have been much lower than the capital cities (ie, only +5.2% in the 12 months to end June cf. +10.5% for the capital cities).

The ABS and APM stratified median house price indices have also struggled in recent times with significant compositional biases attributable to the impact of the first time buyers boost. In Q1 2009 the ABS index initially reported a massive 2.2% decline in Australian house prices when in fact the RP Data-Rismark regression-based hedonic model showed that house prices were recovering over the quarter.

In short, the surge in first time buyers pulled down the ABS and, to a lesser extent, APM stratified median measures. Note that a stratified median index bins all suburbs into ten ‘deciles’ ranked by the historical median price of the suburb in question. The index averages these 10 medians every quarter and measures the change in the average estimate.

This means that if you have more first time buyers purchasing cheaper homes across all deciles, the medians (and thus the index) will experience composition-induced declines despite unchanged or even rising underlying house prices.

In the first quarter of 2010 we saw this same bias reverse out with the opposite effect. Recall that upgraders were buying more expensive homes, taking over from the rapidly fading first timers who had been acquiring cheaper ones. The ABS and APM consequently reported huge surges in house price growth in the first quarter of 2010 as the upgraders inflated their medians while RP Data-Rismark recorded a comparatively more modest (albeit still strong) outcome.

What does the RBA focus on?

Over time, the RBA has made it clear that it prefers the RP Data-Rismark hedonic and APM stratified median benchmarks over the ABS proxy. There have been a few other noteworthy developments in the RBA’s recent house price coverage:

* As mentioned above, the RBA now publishes ‘rest of state’ or ‘regional’ house price proxies from RP Data-Rismark and APM.

* It has drawn further attention to the hedonic model by repeatedly publishing a new stratified version of the capital cities index broken up by the 20% of cheapest suburbs, middle 60% of suburbs, and 20% most expensive suburbs ranked by price (see here).

* It has started reporting seasonally-adjusted results for the first time with RP Data-Rismark being the only index provider to independently produce seasonally-adjusted numbers (see here).

* It has published and referenced monthly house price movements for the first time, which is a nontrivial innovation. More precisely, the RBA began highlighting monthly house price changes in the latest Statement on Monetary Policy whereas previously it only reported quarterly numbers (see here).

There are also specific examples of the RBA’s gradual shift away from the ABS’s numbers. Notwithstanding the ABS’s very large negative print in the first quarter of 2009, the RBA observed at the time that “nationwide housing prices were little changed in early 2009, although there is some variation in the range of available measures that use different techniques to control for changes in the composition of property transactions.”

In the February 2010 Statement on Monetary Policy, the RBA’s 10-12% range for 2009 house price growth correlated exactly with the RP Data-Rismark and APM results, yet was notably inconsistent with the circa 14% growth reported by the ABS.

In the May 2010 Statement on Monetary Policy, the RBA found that “over the past 15 months, capital city housing prices have risen by an average of around 1 per cent a month, to be 15 per cent above their trough in late 2008.”

For this same period, RP Data-Rismark reported 14.5% growth, APM were at 15.7%, while the ABS claimed 20%. The RBA therefore chose to ignore the ABS estimate (20%) opting for RP Data-Rismark’s outcome (15% growth rounded). See Table 8.

And in the July Board Minutes, the RBA commented, “Price data to May…provided some tentative evidence of a deceleration in growth relative to earlier in the year.” RP Data-Rismark were the only index supplier to report such an outcome.

It is useful to point out here that for longer time-series analysis, the RBA publishes the next-best stratified medians since RP Data-Rismark’s full hedonic index only begins in the early 2000s.

What is actually happening to house prices?

Based on RP Data-Rismark’s monthly (nb: not pooled quarterly) hedonic index results, we have documented a clear deceleration in monthly seasonally-adjusted growth rates in Australia’s capital cities over February (+1.0%), March (+0.9%), April (+0.6%) and May (+0.3%). The RBA has referenced this soft-landing in recent Board minutes. The trend continued in the month of June with a -0.7% preliminary print, which is consistent with the very weak credit growth and subdued auction clearance rates.

On a quarterly basis, RP Data-Rismark’s monthly hedonic model reported zero growth (-0.2%) for the June quarter (this is measured as the end June month over end March).

Interestingly, RP Data-Rismark have the benefit of internally producing a very wide range of other house price methodologies. For example, we compute simpler non-parametric stratified median price models that are near-identical to the ABS and APM approaches. We estimate our stratified median price house price indices on a monthly and a pooled quarterly basis to give us a feel for where APM and the ABS are going to land. APM and the ABS only publish their stratified median indices on a pooled quarterly basis due to the high volatility associated with this approach when computed on a monthly basis with only, say, 25% of the final sample of sales in the subject month, which in turn leads to high revision biases.

We also produce the Case-Shiller and FHFA regression-based repeat-sales indices, although the RBA has made it clear that they have little time for repeat-sales models, preferring to use hedonics and stratified medians.

What is an “hedonic” index?

The regression-based hedonic model provides an estimate of the true change in housing values over time by controlling for each individual property’s:

* Location
* Property type
* Exact land size
* Number of bedrooms
* Number of bathrooms
* Pool if applicable
* Car spaces if applicable
* View if applicable
* Waterfront if applicable

A very short summary of the hedonic index method is available here. Much more detailed working papers are available on request.

What are RP Data and Rismark?

RP Data (ASX: RPX with mkt cap ~$130m) is Australia’s largest real estate information company and spends over $9 million annually collecting new property information, amassing a database comprising >140 million property data records pertaining to 99% of all dwellings. Rismark International developed the underlying hedonic index intellectual property over which the Australian Government has awarded it a certified and sealed innovation patent. Rismark has a dedicated real estate research group comprising four PhD-qualified executives, including mathematicians, statisticians and financial economists, amongst others.