Introducing the Kavanagh-Putland Index

Standing on the shoulders of giants, Gavin R. Putland describes a simple predictor of recessions.

Definition

The Kavanagh-Putland Index is an estimate of the ratio of property sales to GDP for Australia. Specifically, it is the ratio of the total property sales in those States and Territories for which figures are available (numerator), to the total estimated Gross State Product (GSP) in the same States and Territories (denominator).

Both the sales total and the GSP total are calculated in nominal dollars; when their ratio is taken, inflation cancels out. To allow for the limited range of years for which GSP figures are available from the Australian Bureau of Statistics (ABS) and for changes in methodology in the calculation of those figures, the GSP for each State or Territory is “estimated” as follows. For each year in which official GSP figures are available, (i) the GSP for that State or Territory is divided by the total GSP for all States and Territories, to obtain the GSP fraction, and (ii) this fraction is multiplied by the national GDP to obtain the “estimated” GSP for that State or Territory. (Obviously these two steps would cancel out if GDP were equal to the total GSP. However, due to an apparent change in ABS methodology, the GSP figures add up to GDP from 1989-90 onwards, but add up to somewhat less than GDP in earlier years. The two-step adjustment makes the “estimated” GSP figures add up to GDP in all years and thereby removes or minimizes the discontinuities in GSP.) For the years before GSP figures were compiled by the ABS, the GSP fractions are assumed to be the same as in the earliest year for which GSP figures are available (1984/5). For the years for which GSP figures are not yet available from the ABS, the GSP fractions are assumed to be the same as in the latest year for which GSP figures are available (2005-6 at the time of writing).

The GDP figures used are the quarterly, seasonally-adjusted, current-dollar figures from the ABS. When sales figures have been received for a quarter for which the GDP figure has not yet been released, the GDP for that quarter is assumed to be the same as in the preceding quarter.

Empirical significance

Since 1972, on every occasion on which the Index has declined more than 17.5% since the preceding year, and only on such occasions, a recession has started no later than the following year — that is, no later than the 2nd year after that from which the Index declined.

Theory: Bubbles, bursts, and recessions

A bubble in an asset market is a process in which prices rise out of proportion to earnings and are supported solely by the circular argument that prices will continue to rise (that is, by belief in the greater fool). When that argument loses credibility (that is, when the market runs out of greater fools), the alleged justification for present prices is taken away: the bubble “bursts”.

Every property includes a site — that is, a piece of ground or airspace. Because sites are inelastic in supply but essential to economic participation, the rental values of sites tend to absorb all income in excess of the minimum necessary returns to labour and capital, so that site rents, and hence prices of sites, tend to rise (expressed in percent per annum, the rise tends to be somewhat faster than that of GDP). The rational expectation of rising prices easily gives way to belief in the greater fool, with the result that the property market is susceptible to periodic bubbles and bursts. Since 1800, the U.S. property market has bubbled about once every 18 years, interrupted only by the Second World War. The 18-year cycle is now global.

The manner in which property bubbles burst is complicated because a property may include not only a site but also a building (or part of a building), which, unlike the ground or airspace that it occupies, can be produced by private agents who respond to market conditions. When buyers decide that prices have risen too high, sellers initially refuse to lower their price demands, so that properties take longer to sell. Thus the burst is initially manifested not as a fall in prices, but as a fall in turnover, hence an increase in the inventory of new unsold buildings. The fall in turnover is magnified because the earlier inflation of the bubble is characterized by rising turnover, due to flipping (buying for early resale into a rising market), builders' illusion (excessive construction as builders wrongly count uplifts in underlying site values as part of the return on construction), and loose credit (because lenders assume that in the event of default, they can sell the collateral into a rising market — and that taxpayers will bail them out if the market falls). When turnover falls from its peak, builders (or their clients) respond by reducing construction until the market clears.

In the best case (as seen by sellers), the market clears before there is any substantial fall in prices. But even that is enough to cause a recession, because:

• the decline in construction, and in the industries upstream and downstream of construction, is a direct deduction from GDP; and

• the willingness of buyers to borrow, and of banks to lend to them, has been based on rising prices, not stationary prices; when prices merely stop rising, flippers and builders have trouble repaying loans, and their creditors are forced to curtail lending to the rest of the economy.

Something close to the “best case” is occasionally seen in Australia, where the clearing of the market is assisted by rapid population growth. But more often (especially in countries with slower population growth), substantial numbers of sellers cannot wait until the market clears, so that prices actually fall. That makes the credit crunch and recession far worse, because falling collateral values both increase the rate of defaults and prevent lenders from recovering their principal.

Thus the observed correlation between property slumps and recessions is not a coincidence in search of an explanation, but empirical confirmation of a coherent theory of recession causation. That theory, in combination with the observed periodicity of property slumps, enabled two economists to predict the latest recessions in the UK and the USA more than a decade in advance [1,2,3]. These economists were not well known at the time; and as there is no greater crime in economics or politics than to be proven right when the establishment is wrong — or perhaps rather to be caught telling the truth when the establishment prefers to lie — they are still not well known.

Predecessor: Kavanagh's Barometer

Where the demand for new buildings is sufficiently high, a bursting bubble can be manifested as a fall in turnover and a temporary plateau in prices. But a price plateau by itself is not sufficient to identify a bursting bubble, because even a normal, rational upward trend in prices can be interrupted by an external event. In the absence of excessive prices and concomitant behaviours (flipping, builders' illusion, loose credit), the interruption will not cause a credit crunch, and the normal price trend will resume when the external event passes. For example, the near-plateau in Australian home prices through calendar year 1995 may be related to the raising of official interest rates by 2.75 percentage points through the second half of 1994.

The diagnosis of a national property bubble therefore requires a national measure of turnover. Accordingly, in October 1991, Bryan Kavanagh started assembling property-sales statistics from the Australian States and Territories. At first he expressed the estimated total sales in terms of value, adjusted for inflation [4, p.14]. This adjustment, although helpful, did not yield a long-term “normal” level of turnover, because property values tend to increase faster than the CPI. In about 1999, seeking a more effective cancellation of the long-term (non-cyclic) trend, Kavanagh divided the estimated total property sales by GDP, and made a crucial observation: every Australian recession since 1972 had been preceded by a fall in property sales to below 19% of GDP, from a peak above 19%.

The converse was not quite true; sales peaked above 19% in 1993-4 and fell below 19% in 1994-5, and no national recession followed. But the exception proved the rule in that the perception of a property “boom” was confined to Queensland, and was followed by a year of slow per-capita growth in Queensland relative to other States (1995-6). This “sharp economic decline”, which Kavanagh predicted while the bubble was still inflating [4, p.15], was occasionally described as a Queensland recession (although this is not confirmed by annual GSP figures) and was followed by a series of cuts in official interest rates (2.5 percentage points from July 1996 through July 1997). As a practising property valuer, Kavanagh also knew that the 1994 peak in turnover was partly due to delayed sales of repossessed collateral left over from the property crash of 1989 and the recession that followed [4, p.14]. The rise in interest rates in late 1994 may also have contributed to the fall in sales thereafter.

Another difficulty faced by Kavanagh was that sales figures were not available from all States and Territories for all years under study. Only Victoria's figures reach back to 1972. Qld, WA and NT join the picture in 1979, Tas and ACT in 1984, SA in 1985, and NSW in 1989. For each State or Territory other than Victoria, the sales figures used from 1972 until actual figures become available are estimates based on the assumption that the State or Territory has a fixed percentage of total sales, this percentage being calculated from later (actual) figures. A similar technique was needed for the most recent sales, because information from some states was more timely than that from others. The fixed-percentage assumption, although perhaps the best that could be managed in 1991, was less than ideal — as was emphasized when Queensland's sales exceeded those of Victoria from 1991-2 to 1994-5.

However, subject to those shortcomings in the data, and to the qualification that the rise in property sales must be nationwide and must not be distorted by known “distress sales”, it was empirically established that sales in excess of 19% of GDP indicated an economically damaging bubble. Kavanagh had constructed a “barometer of the economy”, on which the “bubble line” was marked at 19%. A graph of Kavanagh's ratio was widely published in 2001 [5] and 2007 [6]. The version below includes sales up to June 2008.

Kavanagh's Barometer:
Ratio of estimated total property sales to GDP

Alternative definition of a burst

The credibility of Kavanagh's Barometer would be enhanced if one could discover a bubble criterion, or at least a burst criterion, that correctly classified the 1994-5 event without the need for any “qualification”. The graph above contains a hint: the curve has an overall upward trend, suggesting that the “normal” ratio of property sales to GDP is gradually increasing. Gavin R. Putland noted two reasons why that should be the case. First, in the long term, property values grow not only faster than the CPI, but also slightly faster than GDP. Second, the recession of 1990-1 lowered expectations concerning inflation and nominal interest rates, so that property buyers could service larger loans from the same income. For these reasons the bubble line probably should trend upward, and probably should not be straight. Unfortunately, any attempt to quantify the slope or shape of the line would be no better than the underlying assumptions (the choice of which would be partly arbitrary), and no such calculation could filter out distress sales. But distress sales offered a second hint: when there is a “hump” in sales due to lenders selling repossessed assets at times convenient to themselves, the subsequent fall in sales might not be as sudden as that due to a bursting bubble.

These considerations eventually led Putland to compute the year-on-year percentage change in Kavanagh's Barometer, hoping to tune out the long-term trend and reduce the sensitivity to distress sales. It worked: the three largest falls were followed by recessions, and the fall after 1994 was the 4th-largest, albeit only slightly smaller than the 3rd-largest (16.87% vs. 16.98%). One could therefore say that during the period under study, a recession followed if and only if the Barometer fell by more than a certain threshold; but the threshold had to be fixed within very narrow limits.

From the Barometer to the Index

On August 22, 2008, Putland suggested that because the statistic of interest was the ratio of property sales to GDP, incomplete sales data should be handled by using correspondingly incomplete GDP data — that is, dividing total available sales by total GSP for those States and Territories for which sales data existed. Unfortunately GSP data were not available for all years under study. But the use of GSP figures would still be an improvement because the fractions of total GSP represented by the respective States and Territories are far less volatile, and therefore far safer to extrapolate (as described in the above definition), than the corresponding fractions of property sales.

This proposed modification of Kavanagh's Barometer, dubbed the “Kavanagh-Putland Index” on Kavanagh's initiative, was first implemented on February 20, 2009, as Putland incorporated sales figures from SA and WA for July–December 2008. (Until then, the processing of new sales data had been entirely Kavanagh's work.) The result is shown in the following graph.

Kavanagh-Putland Index for Australia (blue)
with annual change (red),
based on data received to Feb.20, 2009

(The latest data point is for the half-year ended December 2008.)

The similarity between the Index and Kavanagh's Barometer is evident, except that the bubble line (as described by Kavanagh, with the usual caveats concerning the 1994 peak) must be drawn somewhere between 15.7% and 16.9%. A burst can be defined as a fall in the Index of more than 17.5% year-on-year.

The discrimination between a burst and a completion of distress sales is slightly better for the Index than for the Barometer; the fall in the Index to 1982 (which was followed by a national recession) was 17.9%, whereas the fall to 1995 (which was not) was 17.3%. Nevertheless, the closeness of these figures should be taken as a warning that the change in the Index may not function as a single-variable predictor of recessions in all future cycles. Other warnings, which may be gleaned from the history of Australian and American recessions, include the following:

• While the land market regularly bubbles about once every 18 years, its behaviour between those regular bubbles is highly variable.

• While the “18-year” bursts in the land market reliably lead to recessions, the overall frequency of recessions is about once every 9 years, not 18 years. The extra recessions, known as mid-cycle recessions, are preceded by bursting asset bubbles, but the asset class in question is not always land. The 1982-3 recession in Australia was indeed preceded by a land bubble (which, according to Kavanagh, was confined to residential land). But the bubble preceding the 2001 recession in the USA was in technology stocks.

But as long as the 18-year cycle continues — that is, as long as governments continue to allow the rent of sites to be capitalized into sale prices, inflated by speculation, and used as collateral for loans — the Index will be a useful tool for tracking the progress of the cycle. Such tools are needed because, of course, the “18-year” duration is only an approximation.

It will be noticed that the latest fall in the Index is about 30%, which is far in excess of the burst threshold. This fall is calculated on data from SA and WA only, by comparing the whole of 2007-8 with the first half of 2008-9. If, in an attempt to reduce seasonal influences, we compare the first halves of the two financial years, the calculated fall becomes 35%. Whether the Index can predict mid-cycle recessions is irrelevant to the interpretation of this massive fall, which is clearly an “18-year” event (or worse), not a “9-year” event. However, as the purpose of this article is simply to introduce the Index, we reserve further comment for a subsequent article.

Threats to this research

The States and Territories necessarily collect information on property sales for purposes including registration of titles, valuation, and imposition of stamp duty. It would be a simple matter to publish aggregate sales data on the internet, not only for the whole State or Territory but also for geographic divisions therein. But the States and Territories treat such information as a source of revenue. Worse, the NSW and NT governments no longer report aggregate sales by value. If these trends continue, they will at best require the Index to be computed from incomplete data, and at worst will make it impossible (or prohibitively expensive) for private non-profit entities to conduct research on the property market in the public interest.

That said, the aggregation of property sales by value is of such importance that it should not be left to the private sector. It should be done by the ABS. If the States and Territories are reluctant to supply the required information, the Commonwealth can compel them to do so under s.51(xi) of the Constitution (or, if all else fails, s.96!). After all, the Commonwealth has no scruples about compelling individuals to cooperate with the ABS.

Acknowledgments

References