RDP 2015-08: Housing Wealth Effects: Cross-sectional Evidence from New Vehicle Registrations 1. Introduction

There is broad agreement among policymakers and academics on the positive correlation between house prices and consumption, but there is disagreement on the magnitude of the relationship. Some authors argue that changes in house prices are likely to have only a modest effect on consumption. If home ownership is merely equivalent to prepayment of expected future rents, then house price fluctuations have only a small effect on net wealth for households expecting to own their home for a long period of time (Sinai and Souleles 2005). But others argue for a sizeable effect of house prices on consumption (e.g. Case, Quigley and Shiller 2005). For credit-constrained households, increases in house prices may facilitate higher consumption by relaxing collateral constraints (Aoki, Proudman and Vlieghe 2004; Iacoviello 2004; Browning, Gørtz and Leth-Petersen 2013). Understanding the magnitude of the relationship between house prices and consumption is important, because it informs the extent to which developments in the housing sector can have broader macroeconomic effects.

The recent house price collapse in the United States lends weight to the view that fluctuations in house prices can have large macroeconomic effects. Mian, Rao and Sufi (2013) estimate that the decline in US house prices over the 2006 to 2009 period caused 40 per cent of the decline in US consumption over the same period, relative to trend. In accompanying work, Mian and Sufi (2014) estimate that during the 2002 to 2006 boom phase low-income households aggressively liquefied housing wealth to fund higher consumption. A key question is whether the large and heterogeneous effects of housing wealth on consumption estimated by Mian et al (2013) and Mian and Sufi (2014) are specific to that particular US boom-bust house price cycle. The 2002 to 2006 boom in US house prices was accompanied by a sharp increase in subprime mortgage lending and lax screening of borrowers (Keys et al 2010), while the subsequent collapse in housing wealth triggered a wave of mortgage defaults, bank failures, and a sharp tightening of lending standards. It is possible that these events amplified the usual effect of housing wealth on consumption, and differentially affected high- and low-income households.

We revisit the relationship between house prices and consumption using Australian data for the period 2006 to 2011. Average house prices in Australia rose at about 4 per cent per annum over our sample period. This contrasts with the collapse in US house prices over the 2006 to 2009 period studied by Mian et al (2013). Compared with the United States, non-conforming (subprime) mortgage lending remained a small share of the total lending in Australia. While the United States entered a deep economic slump in 2008, Australia experienced around average rates of economic growth over our sample period.

We use a cross-sectional identification strategy that exploits postcode-level variation in house prices and consumption for Australia's three largest cities, Sydney, Melbourne and Brisbane.^[1] We match postcode-level changes in house prices to administrative data from the Australian Bureau of Statistics (ABS) Motor Vehicle Census (an annual compilation of state motor vehicle registry data), from which we can infer the annual number of new (and near-new) cars registered by postcode. Disaggregated consumption data are difficult to come by, and new vehicle registrations are the only high-quality postcode-level consumption measure available in Australia. Official consumption data are available at no more than a state level of disaggregation, while biases in self-reported consumption data from household surveys can be large (see Koijen, Van Nieuwerburgh and Vestman (2015)). Survey data are likely to be particularly unreliable for durable goods consumption because sample sizes are typically too small to make reliable inference about infrequently purchased items. We show in the next section that new vehicle and total consumption growth tend to be synchronised, indicating that new vehicle registrations is a useful proxy for total consumption. Notably, Mian et al (2013) estimate motor vehicles to be the most responsive consumption good to a change in housing wealth. Under the assumption that this is also true for Australia, our estimates can be used to provide an upper bound on the relationship between total consumption and changes in housing wealth.

Our estimation period is aligned with the 2006 and 2011 Census of Population and Housing (the Census), allowing us to include a rich set of postcode-level control variables, including income, the unemployment rate, housing tenure, usual monthly mortgage repayments, and the level of education. This is a key strength of our analysis compared with earlier work, because it reduces the likelihood that the effect of changes in house prices on consumption that we estimate is caused by a third factor that simultaneously affects both house prices and consumption (see King (1990)). In our preferred specification, we divide Sydney and Melbourne into sub-city regions (e.g., north, south, east and west for Sydney), and identify housing wealth effects using only variation in house prices within each region. Each region is geographically small, so this controls for unobserved factors affecting consumption growth within localised areas. This contrasts with much of the existing literature, which uses cross-state or cross-city variation in house prices.

We find a robust postcode-level relationship between growth in house prices and new passenger vehicle registrations. The relationship is robust to the inclusion of local-area fixed effects and the full set of control variables. In our preferred specification, we estimate an elasticity of new passenger vehicle registrations with respect to house prices of 0.4–0.5, with a relatively high degree of precision. To estimate a marginal propensity to consume (MPC) for new vehicles, we first scale the number of new passenger vehicles registered by the average price of a new passenger vehicle. We then regress the dollar change in new passenger vehicle consumption on the dollar change in postcode-level house prices and controls. We estimate an average MPC for new vehicles of about 0.06 cents per dollar increase in house prices. (In general, we refer to new vehicle consumption when referring to dollar values, and new vehicle registrations when referring to quantities.)

Using our results to infer the relationship between house prices and total consumption requires us to take a stand on the relationship between new passenger vehicle consumption and total consumption. If new passenger vehicle consumption has the same sensitivity to a change in house prices as total consumption, then we can get an aggregate MPC by scaling our estimate by the ratio of total consumption to new vehicle consumption. Doing so implies an aggregate MPC of 2 cents per dollar change in gross housing wealth. But if housing equity is more commonly used to finance durable than non-durable consumption, the aggregate MPC will be smaller. For example, if new vehicle consumption is twice as sensitive to a change in house prices as total consumption then our estimates imply an aggregate MPC of about 1 cent per dollar change in house prices. Mian et al (2013) estimate for the United States that new vehicle consumption is much more sensitive to a change in housing wealth than total consumption. Based on their estimates of the relative sensitivities of new vehicle and total consumption to a change in housing wealth, our estimates imply an aggregate MPC for Australia of 0.14 cents per dollar change in house prices. Thus, based on the US evidence that new vehicle consumption is particularly sensitive to changes in housing wealth, our results suggests that the MPC for total consumption is likely to be less than 0.25 cents. This is much smaller than the 5–7 cent range estimated by Mian et al (2013), and most Australian estimates, which fall in the 2–4 cent range. We discuss the Australian literature in detail in Section 7.

Notably, we identify heterogeneity in the response of households to changes in house prices: each $1,000 increase in annual postcode-level income reduces the estimated MPC for new passenger vehicle consumption by about 0.001 cents. At median household income, the estimated MPC for new vehicles is 0.1 cents, declining to about 0.025 cents at the 95th percentile of household income. Our finding that the MPC is larger for low- than for high-income households suggests that the heterogeneous effects identified by Mian et al (2013) are not specific to institutional features of the US housing finance system, or to the 2002 to 2009 boom-bust house price cycle. Thus, in general it is important for policymakers to monitor not only aggregate changes in housing wealth, but also the distribution of changes. From an academic perspective, our results lend support to the need for models that incorporate heterogeneity in MPCs across households. We believe our results represent the first non-US evidence that the MPC out of a change in house prices is decreasing in income.

Exploiting variation in housing-tenure type across postcodes, we attempt to estimate whether the effect of house prices on new vehicle registrations is via a standard wealth effect or a collateral constraints channel. A housing wealth effect is expected for those who own their home outright or with a mortgage, but only households with a mortgage are likely to be collateral constrained. In contrast to Windsor, Jääskelä and Finlay (2013), we find that the estimated effect of changes in house prices on new vehicle registrations is similar for households owning their home outright to those with a mortgage, although the difference between the effects is imprecisely estimated. We find no evidence of a positive relationship between new vehicle registrations and house prices for renters, indicating that our results are not spurious or caused by a third factor increasing both new vehicle registrations and house prices, such as income shocks.

Our finding of heterogeneity in MPCs by income is consistent with a collateral constraints channel. Low-income households are more likely than high-income households to have low wealth, and so be subject to binding collateral constraints. This implies that a dollar increase in house prices is more likely to reduce borrowing constraints and raise consumption for low- than for high-wealth households. But a precautionary saving motive can also explain a relatively larger MPC for low-wealth households: the MPC out of changes in wealth decreases in the level of wealth for a prudent consumer (Carroll 2001). Given that a range of theoretical models can explain a relationship between housing wealth and consumption, we remain agnostic about the relative importance of different channels from housing wealth to consumption.

Footnote

Like other studies using cross-sectional variation in house prices and consumption, some caution is warranted in using our estimates to make inference about aggregate behaviour. If general equilibrium considerations are important, the response of consumption to an economy-wide change in housing wealth may differ from relative changes in housing wealth. For example, monetary policy does not respond to relative changes in conditions across postcodes, but may respond to macroeconomic developments associated with changes in aggregate housing wealth. [1]