RDP 2004-08: Housing Construction Cycles and Interest Rates 5. Discussion of Results

The responsiveness of demand-side variables to changes in interest rates lines up reasonably well with what would be expected from differences in mortgage finance institutions in these countries, as summarised in the first column of Table 2 for these four and a selection of other industrialised countries. As previously emphasised by Maclennan et al (1998) in the European context, the structure of the mortgage finance industry largely determines the speed of transmission of changes in interest rates to mortgage borrowing behaviour. Australia and the UK, where mortgages are predominantly offered at variable rates, display more overall interest sensitivity than does the US, where an array of institutional arrangements have been designed to support the existence of a long-term fixed-rate mortage market. Similarly, in the early period of regulated financial markets, housing demand in the US and UK was more interest-sensitive than more recently. This tends to confirm the findings of McCarthy and Peach (2002); that financial regulations affecting the supply side of the mortgage market resulted in the demand side of the housing market being more interest-sensitive than if those lending restrictions had not been in place. In Canada, five-year fixed-rate mortgages predominate, consistent with its intermediate degree of responsiveness to interest rates.

Another aspect of the mortgage market that may be relevant to a reconciliation of demand-side behaviour is the tax treatment of mortgage interest. In the consumption and asset demands framework outlined in Section 3.2, non-deductibility of mortgage interest implies that movements in interest rates have full force on user cost and the arbitrage conditions determining individuals' demand, Equations (1) and (2). Therefore it should be expected that housing markets in countries where mortgage interest is not deductible for owner-occupiers should experience larger effects of movements in interest rates. If this is true, then the recent completion of the abolition of mortgage interest deductibility in the UK should be expected to result in a shift in demand behaviour to become more interest-sensitive than is presently the case.

The patterns of interest sensitivity do not, however, completely line up with the differences in amplitudes of housing construction cycles identified in Section 2. Moreover, within this overall pattern of interest sensitivity, there are other clear differences, such as in the split between price and quantity responses. If the mortgage characteristics were the only consideration, the responses of Australia and the UK would be much more similar than they are. Two factors suggest themselves as likely additional considerations in a reconciliation of these cross-country differences evident in Figures 1 and 2. Firstly, even economies with otherwise identical structures will experience different housing cycles if they face different paths of interest rates, perhaps brought about for reasons other than developments in the housing sector. And secondly, the split of the total response between quantities and prices depends on the price elasticities of both demand and supply, as well as on the extent to which demand shifts in response to a change in interest rates.

As an illustration of this first point, we can use the models estimated in this paper to construct a counterfactual outcome for Australia that would have occurred had the United States' path of interest rates prevailed instead. This counterfactual series, shown in Figure 11 along with the actual series for both countries, is calculated by replacing the actual Australian interest rate series with the US equivalent, and generating the fitted values that would have resulted, given our estimated coefficients. Obviously this results in the level of housing investment being higher than the actual outcome, since nominal interest rates were lower in the US than in Australia for much of the period shown; we have not adjusted for the fact that inflation was lower in the US than Australia in the 1980s. The counterfactual series is also smoother than the actual series for Australia, particularly for the 1980s, although still containing more obvious cycles in the 1990s than the actual US series.

This exercise does not correspond to a true counterfactual: at the very least, income also responds to interest rates, and so the path of Australian income entering into housing demand should also have been changed to fit the US interest rate series. This may reduce the cyclicality of the counterfactual Australian series further, since our econometric results also showed that the Australian housing sector was relatively income-sensitive. However, this would have required a structural model of income that is beyond the scope of this paper. Even so, it is apparent from Figure 11 that a significant part of the visual impression of regularity in the housing construction cycle in Australia can be attributed to the greater cyclicality of interest rates in the 1980s; the amplitude of the cycle around 2000 is more a result of the introduction of the GST than changes in interest rates. This raises the possibility that, absent further changes to the tax system or other shocks, Australian housing construction cycles might be more muted in the future than was the case historically.

A more extensive illustration of the role of different paths of interest rates can be obtained by eliminating interest rates from the model entirely. Table 3 shows the changes in the variances of the endogenous variables that would ensue from holding interest rates constant at their sample means, given the estimated parameters and the paths of the other exogenous variables. For each country, the variances of the actual fitted values for the two quantity variables are greater than the variances of the counterfactual series where rates are held constant. This demonstrates that interest rates have an important role in explaining construction cycles: in particular, they appear to explain more of the variation in dwelling investment in Australia than in the US or UK, while housing starts are most interest-sensitive in the US. These results also imply that movements in interest rates have little exogenous effect on the variances of prices of housing and structures.

It is also apparent from Table 3 that cross-country differences in housing outcomes persist even when the influences of the path of rates, and consequently the differences in interest sensitivity, are removed. Because Table 3 compares variances of fitted values, these remaining cross-country differences are not due to the contemporaneous effects of different shocks. They may partly relate to the differences in the strength of the indirect influences of rates via the income channel, which as mentioned earlier, have not been removed in these results, as well as variation in other exogenous variables. However, it is also likely that some part of the cross-country differences can be attributed to different dynamic responses of the endogenous variables once a shock has already occurred.

This underlines the importance of the second factor mentioned earlier – the role of differences in price elasticities. This can be seen by comparing this paper's results for the UK and Australia. In both countries, housing demand is quite interest-sensitive, as well as being quite price-elastic. However, as shown in Figure 1, housing construction is almost completely acyclical in the UK, in contrast to the clear cycles seen in the Australian data, and the impulse response of quantities to interest rate shocks shown in Figures 7 and 8 is likewise much smaller than in either Australia or the US. What appears to be happening is that in the UK, transmission of changes in interest rates to credit markets is similar to that in Australia, but the onward transmission from that market is via housing prices to other aspects of consumer behaviour, rather than directly to physical investment in housing. This suggests the presence of a steep supply curve, or some rigidity causing the adjustment in the UK to come through price rather than quantity. This is confirmed by the impulse response analysis: as mentioned in Section 4, the econometric results imply that the notional supply curve is quite steep.

The last two columns in Table 2 provide a hint of the underlying reason for this inelasticity in supply. The UK – and indeed most of continental Europe – is characterised by slow population growth and high population density, in contrast to the situation in Australia, Canada and the US. A given-sized cyclical shift in demand represents a proportionately larger shift compared to the lower average levels of construction activity that might be expected in a country with low population growth. This might entail greater costs of adjustment than if the shift in production was only small relative to the size of the industry. In other words, adjustment costs might provide reasons for construction supply curves to be fairly steep in countries where construction is small relative to its importance in other countries. Given the likely constraints imposed by population density and thus land scarcity, the preponderance of price adjustment rather than quantity adjustment in the UK seems entirely consistent with the characteristics of that market. If countries in continental Europe with a similar supply environment had mortgage markets like that of the UK, one might also expect them to experience similar variability in housing prices.

Another possible contributor to the greater amplitude of housing construction cycles in Australia may be the apparent intrinsic cyclicality identified in the impulse response analysis for Australia, but not elsewhere. Since some of the estimated parameters in these models are imprecisely estimated, it would be a mistake to make much of the finding that only Australia's housing sector exhibits intrinsic cyclicality in the face of a simple demand shock. This finding could, however, be another factor explaining the differences shown in Figures 1 and 2. It is also important to note that intrinsic cyclicality does not imply that supply is inelastic. Rather, intrinsic cyclicality is likely to occur when both demand and supply are quite elastic, but supply is sluggish; that is, the short-run elasticity is much lower than the supply elasticity in the medium to long run. By comparison, the sticky supply evident in the results for the UK imply that construction does not respond much to a shock over any horizon. In that case, there is not enough variation in quantity for an intrinsic cycle to get started.

Although mortgage finance conventions are not the only factors driving differences in the interest sensitivity of housing construction, they have an important role. This naturally raises the question of why the deregulated outcomes for finance differ so much. Presumably households in all four countries would value the reduced volatility of payments inherent in a fully fixed-rate loan contract, but only in the US do a significant proportion of households pay for this reduced volatility. A full investigation of the reasons for this difference is beyond the scope of this paper. However, one of the key contributors to this difference appears to be the substantial involvement of government-sponsored agencies in the US mortgage market. The financing advantages of these institutions reduce the term premium that US households must pay for a fixed-rate loan over a variable-rate loan over the whole term. Additionally, the differences in tax treatment mentioned earlier may also be relevant. The incentive to pay off their mortgage ahead of schedule is much stronger in a country where home mortgage interest is not tax-deductible, such as Australia, since alternative uses of funds earn a post-tax, not pre-tax, return. This makes variable-rate loans relatively more attractive, since unlike fixed-rate loans, they do not generally involve penalties for early repayment.