RDP 2004-08: Housing Construction Cycles and Interest Rates 2. Housing Investment Cycles in Developed Countries

The key facts we seek to explain are shown in Figures 1 and 2. Housing construction as a share of GDP has a negative relationship with interest rates.^[1] Construction prices are positively associated with construction activity, although there is a slight lag between the movement in activity and the change in prices. This lines up with the observations of Topel and Rosen (1988) for earlier US data, that the price-volume relationship is mainly due to shifts in the demand for new housing tracing out a largely unchanged supply curve. In this context, interest rates appear to serve as a demand shifter. A similar picture applies if growth rates are used rather than shares of GDP, or if housing starts are used instead of national accounts measures.

Within these relationships, however, there are clear differences both between countries and across time; these differences are the focus of this paper. In particular, the regularity of the housing cycle in Australia in recent decades is not matched by any of the other countries shown here. The early part of the US data is very cyclical, but as noted by McCarthy and Peach (2002), this pattern has been more muted since the early 1980s. However, this change cannot be wholly attributed to the milder cycle in interest rates more recently.

Although the share of nominal housing construction in nominal GDP for the UK displays some cyclicality, in volume terms the share has been virtually flat for more than a decade. There was a clear cycle in construction in the late 1980s, but this was almost certainly the result of the housing boom occurring at that time, mainly due to factors other than monetary policy and interest rates (Attanasio and Weber 1994; Ortalo-Magné and Rady 1999). The interest sensitivity of the UK housing sector seems to manifest primarily in structure prices, and in the land prices implied by the total price of existing housing.

In New Zealand and Canada, the housing construction share is more variable over time than in the UK or recent US data. It appears that the housing construction cycle has similar amplitude to that in Australia, although without the regularity seen in Australia's construction cycle. France, Italy and Japan have all seen substantial downward shifts in the share of housing construction in GDP, as population growth slowed and the post-war reconstruction phase came to a close. Abstracting from this by fitting these data to a cubic trendline, we can discern very mild cyclicality in France and Italy's housing construction share. In Japan, the detrended series displays noticeably more sensitivity to interest rates than the continental European countries, although without the regularity seen in Australia or the early part of the US data.

If the positive price-quantity relationship seen in Figures 1 and 2 captures the shifts of a demand curve tracing out the supply curve, then a regression of housing investment (quantity) on its deflator (price) should estimate the slope of this supply curve. The costs of building – such as labour and materials – would shift the supply curve, but there should not be any additional role for explanators of demand such as interest rates or income. Simple econometric estimates suggest that this is not the case. For all the countries shown in Table 1, non-price factors have a significant relationship with housing construction (and housing starts), even after controlling for structure prices in the form of the housing construction deflator. This is consistent with similar results cited in Egebo, Richardson and Lienert (1990). Reduced-form VARs constructed along the lines of those in Aoki et al (2002) and McCarthy and Peach (2002) were also consistent with this finding; these results are available from the authors.

In particular, the estimated coefficients on interest rates are negative, and significant for all countries shown except the United Kingdom and Japan. Part of the reason for this could be that interest rates are also a supply shifter, so that the reduced-form estimates are not tracing out an unchanged supply curve. Financing costs are one of the costs of constructing housing, so an increase in interest rates would shift the supply curve left, and reduce quantity supplied. The conclusion of earlier work, however, is that the extent of the estimated relationship between interest rates and construction is too great to be reconciled with financing costs over the duration of a construction project (Topel and Rosen 1988).

One initially plausible explanation for housing investment's interest sensitivity – even after controlling for prices – is that suppliers and demanders of housing are not paying the same price. Builders are affected by the cost and sale price of the structure. Home buyers, by contrast, pay the total price of the dwelling, including the cost of the land. As can be seen in Figure 3, the implied price of residential land generally swings around more than the price of the structure, driving significant, usually pro-cyclical, wedges between supply price and demand price. However, many of the causes of this wedge between the two prices, such as zoning laws (Glaeser and Gyourko 2002) and land shortages (Kenny 1999), do not appear to be the main driver of housing cycles. Chinloy (1996) found that cycles in housing construction can exist even when there is plenty of land for development and few zoning restrictions or rent controls to distort the supply decision. In English-speaking countries at least, the greater correlation of interest rates with total housing prices than with structure prices suggests that rates explain much of this pro-cyclical wedge.

Despite the apparent positive relationship between prices and construction visible in Figures 1 and 2, the results in Table 1 force us to conclude that this relationship is not capturing a stable supply function. This implies that a reduced-form estimation strategy is not appropriate. To understand the causes of interest-rate sensitivity in housing construction, we must turn instead to more structural modelling, and try to disentangle demand-side from supply-side influences. In the next section, we develop some variations on existing theoretical models of housing demand and supply, which we then translate into empirical models to be estimated in Section 4.

Footnote

Terminology in national data sources can differ substantially. To maintain a consistent set of terminology in this paper, we refer to the dwelling investment component of the national accounts as ‘housing construction’ or ‘dwelling investment’, its associated price deflator as ‘structure prices’, and building commencements as ‘housing starts’. ‘House prices’ denotes the prices including land or location value actually paid by home owners. Figures 1 and 2 show the share of housing construction in GDP detrended using a simple cubic time trend. [1]