I have received a bunch of questions from journalists about how to discriminate between the different house price measures that are available out there. I thought I would quickly summarise my responses in a user-friendly fashion for everyone’s benefit. There is no point being non-transparent about this stuff.
1. Why do we measure house price movements?
Exclusive of our human capital, or income-generating potential, the largest component of household wealth is residential real estate (around 65-85% according to the RBA). That is, the assets we tend to take for granted that nevertheless provide one of our most basic human needs to survive and contribute productively to society: well-located shelter. Our best guess is that the total value of privately-owned residential real estate in Australia is around $3.4 trillion (we recently revised this valuation using our ‘all-regions’, as opposed to capital cities, dwelling price estimate).
As an unrelated aside, there is a popular myth that investing in housing is ‘unproductive’. As I explained here, this is about as accurate as the old wives' tale that if you pick your nose your finger will fall off. Investments in new and established housing are highly productive because they supply critical accommodation for that most important foundation-stone of our economy: our 10.9m person labour force. This is no different to the shelter that commercial property affords businesses in order to enable them to operate. Furthermore, improvements in the quality of housing have been shown to positively influence labour productivity, which makes intuitive sense. Finally, new housing investment has a high economic ‘multiplier’. The ABS has found that for every dollar of new investment in residential real estate there are an additional 2.9 dollars of output generated throughout the rest of the economy.
2. Why is it hard to measure house prices?
In contrast to, say, shares listed on the ASX, where every individual unit is identical in legal form, and 'market prices' for shares in large companies are observed regularly during the trading day, houses are distinct in two fundamental respects: first, we only see sales transactions for individual homes on average every 6-8 years; and, secondly, each house tends to be different to the next. That is, it ordinarily has a unique location and physical characteristics, such as its land size, aspect, number of bedrooms/bathrooms, build quality, number of car spaces, presence or otherwise of a pool, air conditioning, tennis court, and so on.
Over the last 50 years the academic research literature has dedicated a great deal of time trying to come up with increasingly accurate house price measurement technologies that overcome these two fundamental problems: namely, the ‘illiquidity’ and ‘heterogeneity’ of housing. In this context, one can confidently conclude that the literature has made tremendous strides. And Australian researchers have been at the vanguard of these efforts.
Arguably the culmination of this work is the ‘hedonic’ method, which uses a statistical technique known as ‘regression analysis’ to assess the relationship between the prices of homes, which, simply put, is the ‘dependent variable’, and their individual attributes (ie, location, land size, number of beds/baths, etc), which are the so-called ‘independent variables’. The rationale underlying hedonic theory is that the value of a composite good, such as a house, is the sum of its individual characteristics. And so by decomposing a house according to its attributes one can control for the physical differences across all homes. Moving beyond this description the technical literature becomes rather complex, and is categorically not the subject of today’s note. (For completeness's sake there are, in fact, three alternative hedonic methods: the ‘pooled time dummy’; ‘adjacent period’; and ‘imputation’ techniques. RP Data-Rismark produce all three of them).
As I discuss below, the single biggest impediment to publishing hedonic indices has been the demanding data requirements: you need accurate information on the key characteristics of most homes, which, historically, few index providers have ever had. This is why hedonic indices are relatively rare, and today only produced widely in Australia and the United Kingdom. It is also why most countries have opted for much simpler and less exacting approaches, such as the ‘median price’ and ‘repeat-sales’ methods.
RP Data is in the enviable position of collecting very detailed property attribute data on 80-90% of all home sales across Australia (with this capture rate edging up every day). This has facilitated the launch of hedonic house price indices domestically for the first time.
For what it is worth, one of Australia’s leading academic economists, Professor Robert Hill, who is based at the University of Graz, is a global expert on price indices. And subject to getting access to the right data, Professor Hill’s preferred method is the hedonic approach.
3. What different house price indicators are there?
In Australia there are three major index providers—RP Data-Rismark, the ABS, and Australian Property Monitors (APM)—that are most frequently quoted by industry and the media. A fourth, Residex, also publishes house price data and is less regularly referenced. Occasionally, one also sees house price information supplied by different Real Estate Institutes (particularly in Victoria).
The RBA, which is Australia’s most well-regarded, knowledgeable and experienced economic analyst, follows the RP Data-Rismark, APM and ABS data (at least judging by the information reported in its Statement on Monetary Policy). Of these three sources, only RP Data-Rismark supply index data publicly on a monthly basis. This is because the hedonic measure that RP Data-Rismark publishes has significantly lower ‘noise’ and ‘revision bias’ attributable to its monthly estimates in comparison to the stratified median measures. Recent RBA disclosures in its Board minutes indicate that the RBA follows these preliminary monthly movements quite closely.
Nonetheless, one can assume that the RBA draws significant succour from having access to all three indices (and possibly others), which it can use to sanity-test anomalous outcomes. That is to say, it does not rely exclusively on any one dataset. This is, quite understandably, the principle the RBA applies to all of its analysis. Having said that, there is evidence to suggest that even when the ABS and APM indices agree with one another, which they often do given the similarities in their methodologies, but conflict with the RP Data-Rismark findings, such as in the first quarter of 2009, the RBA can still place greater weight on the latter due to the fact that the hedonic approach is better equipped to overcome extreme ‘compositional biases’ that sometimes afflict the ABS and APM proxies (such as with the unusual surge in first time buyers at the start of 2009). Generally though, there are reasonably strong commonalities between these three alternatives over the medium-term.
While I am, of course, a non-independent observer, many experts have concluded of late that RP Data-Rismark’s hedonic index is the most timely and accurate of the lot. This index also has the benefit of drawing on Australia’s largest property database, and by far the most sophisticated of all the house price measurement technologies (over which Rismark has been awarded patents by the Australian Government). In May 2009, CommSec’s chief economist, Craig James, commented, “The RP Data-Rismark index has emerged as Australia’s authoritative source on home price trends. The property database is Australia’s largest and, unlike the Bureau of Statistics, all properties are counted, not just free-standing homes.” Macquarie Bank’s highly regarded interest rate strategist, Rory Robertson, has also observed, “RP Data-Rismark’s monthly estimates are more timely and reliable than the ABS’s quarterly readings.”
4. What are the differences between these providers?
There are three key points of departure:
1) The data they collect;
2) The data they actually use; and
3) The accuracy/complexity of the index methodology they rely on.
All four index providers referenced above collect data from the Valuer Generals or Land Titles offices in each state and territory. In Australia, we have the important advantage that government agencies record data on pretty much all sales executed across the country. This is a function of our stamp duty system. These agencies then make this data available to a limited number of licenced contractors, such as RP Data, APM and the ABS. While some states report the data with up to a three month lag, the timeliness of the information is always improving with most agencies transmitting data within 1-2 months of the exchange (note, not settlement) of contracts. (The RBA is to be credited as a critical influence in driving these improvements.) This in turn means that most Australian house price indices benefit from the 'population' of all sales transactions—ie, there is little-to-no 'sample selectivity bias' wherein the index only employs a small subset of the overall population of information. By way of contrast, US and UK house price indices suffer from exactly this problem. For example, the Case-Shiller index in the US ignores around 40% of the US market, while the widely quoted Halifax measure in the UK captures less than 20% of all sales.
Differences in the various house price measurement approaches are best summarised according to the relevant provider:
1) The Real Estate Institute (REI) indices are based on simple ‘median prices’, which are crude and quite unreliable (the RBA and the Treasury have recommended against using this benchmark). A median price index ranks all sales from high to low and plucks out the middle or 50th percentile observation. The REI indies are normally reported quarterly;
2) The ABS reports a ‘stratified median price method’ that is broadly based on the methodology developed by two RBA economists, Richards and Prasad. The main author, Dr Anthony Richards, is head of economic analysis within the RBA and universally regarded as one of Australia’s most expert housing authorities. (While I am on this topic, there is a not a private sector or academic organisation in the country that can come close to competing with the RBA's peerless research resources. Including my old buddy Steve Keen. More on this later.)
The RBA has made the measurement of house prices a particular focus ever since the former Governor, Ian Macfarlane, correctly argued in 2004 that, “Housing…is an extremely important asset class for most people, yet…[I]t really is probably the weakest link in all the price data in the country so I think it is something that I would like to see resources put into.” The launch of RP Data-Rismark's hedonic indices were in part a response to the RBA’s call to arms.
Although the ABS measure is still a median price index, the stratification technique was created by Richards et al to help mitigate some of the severe 'compositional bias' shortcomings associated with simple medians such as those reported by REIs (I discuss these biases in more detail below).
The RBA nevertheless acknowledges that in a perfect world one would use more sophisticated regression-based techniques, such as the hedonic indices produced by RP Data-Rismark. Hedonic indices are, however, very complex to compute, and have intensive data requirements on the unique attributes of every individual property included in the index. In the absence of the necessary data, and the considerable statistical expertise needed to estimate hedonic measures, the stratified median price benchmark appears to be the RBA’s second-best preference (see here for the ABS’s methodological explanation). It is noteworthy that the Reserve Bank of New Zealand has recently decided to follow Richards and Prasad’s recommendations.
Finally, the ABS reports on a quarterly basis and typically after the APM and RP Data-Rismark numbers have come out, which makes for the impression of rolling waves of housing information that ordinarily, but not always, coincides in a directional sense;
3) The Fairfax-owned APM also publishes a stratified median price method that is based even more closely on the Richards and Prasad method than its ABS cousin. APM report quarterly;
4) Residex use a ‘repeat sales’ method that is understood to be similar to the Case-Shiller technique published by S&P in the US. A repeat-sales index only examines purchases and sales of the same properties over time. It therefore has the strength of measuring buy-and-hold returns, but suffers from the deficiency that it excludes all sales transactions that do not have a previous purchase price (eg, new home sales). The repeat-sale index can also be biased towards homes that turnover more rapidly (eg, distressed sales). Finally, the repeat-sales proxy can be artificially inflated by renovations or capital improvements to the property, which are hard for this measure to control for (although there has been some work done at Yale on developing repeat-sales proxies that account for non-linear changes in return, which are thought to be triggered by renovations to the home). While Residex report monthly, they do so shortly after the end of the subject month and must, one can infer, be rather limited in the amount of data they actually include in their index. The RBA does not appear to focus on the Residex results, at least judging from the fact that it is not included in their Statement on Monetary Policy); and
5) RP Data-Rismark produce all of the above methodologies (including the Yale derivation) and our preferred benchmark, the hedonic index. In total, RP Data-Rismark privately compute up to 15 alternative index measures, including several median and stratified median price indices, four repeat-sales constructs, and a number of hedonic benchmarks. All of these are available to the public on request.
Over and above the contrasting methodologies, there are some material differences in the data used by these organisations:
* The ABS only examines detached house in capital cities and therefore excludes all ‘attached’ forms of accommodation such as apartments, terraces and semis (which account for around one quarter of the housing stock). However, I understand that the ABS is going to try to publish an apartments index in due course to remedy this problem;
* APM and RP Data-Rismark include all capital city data pertaining to all property types (ie, detached and attached housing);
* To the best of my knowledge, the only index provider that publishes an ‘all dwellings’ proxy (ie, an index that covers all property types) in addition to separate house and unit benchmarks, is RP Data-Rismark.
In order to report its monthly measure, RP Data substantially augments the sales data they receive from government agencies with ‘real-time’ information acquired directly from proprietary sources, such as Australia’s largest property portal, realestate.com.au and real estate agents. (Over 70% of Australia’s real estate agents use RP Data’s software to service their clients.) This real-time data has been tested to be highly accurate and furnishes RP Data-Rismark with up to 50% of the total population of final home sales within one month of the subject period. Over time this indicative data is then supplemented with the population of final sales transactions originated from government agencies.
5. Why are there different median prices? Which one is right?
There can only be one true median price as it is a strict mathematical definition. The median is actually very simple: it is the middle or 50th percentile observation. The median of a sample of homes sales is therefore the middle sales transaction if you lined up all those sales from low to high.
The median prices reported by RP Data-Rismark are based on close to 100% of all home sales executed across Australian and are believed to be absolutely accurate. However, these medians may differ from other data providers if they use smaller samples of sales, which they sometimes do, or if they are not actually calculating a true median.
For example, I understand that the medians reported by APM are not actually the 50th percentile (or middle) transaction of all sales, but rather the median deriving from their stratified index. The APM index divides all suburbs into 10 baskets (or deciles) ranked by their median price (from high to low). The median price sourced from this index is then presumably the median associated with the fifth decile. This can, of course, vary from the median of all sales in the absence of any stratification. (For the interested reader, the ‘index’ that is produced by APM is an average of the growth rates in the median prices associated with each of the ten baskets of suburbs referred to above.) To the extent APM estimate their medians in a manner inconsistent with this description, I would be delighted to know.
Another major point of distinction amongst median prices is what they relate to, which is not always obvious. For example:
* The ABS disseminates medians that cover detached houses in capital cities;
* APM reports medians relating to all property types in capital cities dissected according to houses and units;
* RP Data-Rismark publish medians that cover all property types in all regions (ie, not just capital cities). This is important since around 40% of all homes are not located in the capitals.
A final prospective difference is the time period during which the median price is measured.
RP Data-Rismark prefer to compute medians based on the previous 3 months worth of sales transactions. We do this because the medians can be very volatile on a month-to-month basis. This is precisely why when we measure house price changes over time RP Data-Rismark do not use a simple median price index.
As is well known, median price indices can be adversely affected by changes in the composition of buyers in the market, amongst other biases (such as capital improvements and variations in the type and quality of homes manufactured over time).
RP Data-Rismark’s hedonic index is not influenced by these changes and seeks to explicitly control for each property’s unique attributes, including, but not limited to, its longitude, latitude, landsize, type (ie, detached house or unit), and number of bedrooms/bathrooms using the regression procedure discussed earlier.
The problems associated with median prices were illustrated in the first quarter of 2009, when APM and the ABS reported that house prices were falling—by a record margin in the case of the ABS—when in fact they were rising rapidly. The medians were being dragged down by a surge in first time buyers purchasing cheap homes in the early months of 2009. RP Data-Rismark’s hedonic index, in contrast, reported strong growth during this period.
Since the first quarter, RP Data-Rismark’s index has shown relatively stable quarterly growth. In comparison, the median price indices have reported sometimes wild changes in value, which appears to be evident again in the fourth quarter. The latest median price estimates are likely being artificially boosted by the fading of first timers and the return of upgraders buying more expensive homes, which automatically bias the medians upwards (even if one uses the superior stratification technique pioneered by the RBA). At the current time, the true rates of capital gains across Australia are likely to be less than those reported by median price indices.
6. What's the value of looking at median prices when researching properties if there's no universal gauge?
There is a universal gauge: the median is simply the middle sales observation. Medians are not very useful for measuring house price growth rates because the median is affected by a range of biases:
* Different buyer types who happen to be dominating the market (first timers vs. upgraders);
* Changes in the types of homes built over time (if we build bigger (smaller) homes over time the median may rise (fall) suggesting house prices have appreciated (declined), when in fact they may have not);
* Renovations (if homes are renovated this can push the median up when capital growth rates have actually been unchanged); and
* The liquidity of different geographies (if more West Sydney homes trade than East Sydney homes, the median may fall when house prices could have been rising).
However, the median is useful if you want to simply know what the middle sales observation in, say, Melbourne was over the past, say, quarter. This gives you a quick and easy-to-understand (at least for most) guide for the price of the homes being purchased in the market. That is why RP Data-Rismark continue to report the simple medians alongside our hedonic index. It is really to satisfy the media’s needs. Median prices are also useful when seeking to address research questions that are targeted at identifying a ‘representative’ price at any particular point in time.
7. How should property investors treat these data?
If investors want to work out bona fide capital growth rates, they should not use unadjusted median price data. They should try and rely on more sophisticated index methods that overcome the simple median price biases, such as those that have been reviewed above. Our preferred approach is the hedonic method.
8. What's the risk of comparing data from RP Data-Rismark, APM, the ABS and Residex?
Since all the index techniques are different they cannot be directly compared. Having said that, they are all trying to quantify broadly the same thing: changes in the value of residential real estate over time. Accordingly, it is useful to be aware of how the different proxies behave, and to seek to understand what factors might be driving observed divergences. This is, I believe, exactly what the RBA does. As discussed, RP Data-Rismark privately produce all four index types: medians; stratified medians; repeat-sales; and hedonics. While we could, in theory, service everyone's needs, you would still be reliant on the one dataset. This is why it is helpful to keep an eye on the full cross-section of indicators.
A final observation
Let me leave you with one parting thought. I occasionally hear folks criticise the RBA’s analytical methods and/or its data. (Admittedly these protagonists tend to be of the certifiable type.) But I can assure you of one thing: there is no more thorough, and data-integrity focused organisation in the country.
I have worked at two incredibly anal places in my life: Goldman Sachs, where I pulled over 90 ‘all-nighters’ in my first 12 months, and for a brief time at the RBA. Now while, to be sure, the RBA folks don’t work as hard as their investment banking contemporaries (so much so that I had real trouble acclimatising to the 8-to-5 work day), their commitment to accuracy of input data and analytical excellence is typically far superior.
Whenever I hear or read somebody questioning the RBA’s technical methods, I normally chuckle to myself. If only they knew the extraordinary lengths our central bank goes to in order to validate the veracity of its information. Indeed, the RBA is uniquely responsible for gathering and reporting many vital economic statistics that we take for granted, such as the credit and financial system stability data. It has also had a decisive influence on the way in which many third-party statistics are measured (two obvious cases in point being the inflation and house price data).
As I have argued before, the RBA’s analytical efforts bequeath the community with many valuable ‘public goods’; that is, insights that are freely available to us all that would not have been accessible were it not for the RBA’s labours.
So while I am sometimes critical of the RBA’s governance structures, as a citizen of this nation I am very proud of the quality of the people sitting over at Martin Place, and their work product. When it comes to intellectual rigour, they should be the private sector’s gold standard.
