RDP 2014-06: Is Housing Overvalued? 2. Previous Research

Perhaps the most common method of assessing whether house prices are overvalued is to compare the price-to-income ratio with its longer-term average. On this basis, The Economist (2013) and the OECD (2013) report that Australian house prices are 24 per cent and 21 per cent ‘overvalued’, respectively. A limitation of the price-to-income ratio is that its purpose is unclear. Being told that a house is expensive relative to incomes does not tell you whether the purchase is sensible. For that decision you need to know the cost of the alternative.

The price-to-income ratio could be used as a guide to future price movements if the series was mean-reverting. But in Australian data, it is not. Stapledon's (2012, Figure 3) measure of house prices has risen faster than incomes in each of the past six decades. A trending ratio means recent levels will be persistently higher than the average and that the reported overvaluation will grow over time. It is possible to find definitions of prices and incomes such that their ratio is flat over some periods, but these measures trend strongly at other times.

A trending price-to-income ratio is not surprising. An upward-trending ratio is to be expected when land is in limited supply. Then, as income (and hence the demand for housing) grows, both prices and rents increase. Because demand for housing is price-inelastic, prices need to rise faster than incomes to keep demand in line with supply. A persistent movement in the opposite direction would be expected when extra land becomes freely available, as in the first half of the twentieth century. Fox and Finlay (2012) discuss price-to-income ratios in greater detail.

Another popular approach is to compare price-to-rent ratios to their long-term averages. On this basis, The Economist (2013) and the OECD (2013) conclude that Australian house prices are 46 per cent and 37 per cent ‘overvalued’, respectively. These comparisons are incomplete. Potential home buyers look not just at the price of a house, but also at interest rates, running costs and other elements of the user cost of housing. The price-to-rent ratio is not stationary, but moves with these variables. Indeed, the price-to-rent ratio in Australia has increased over the past few decades, reflecting a decline in the user cost. Unless the user cost is expected to revert to its average, neither will the price-to-rent ratio. Gallin (2008) finds that the price-to-rent ratio, by itself, is useful for forecasting future house prices in the United States. However, we have not found that for Australia.

For these and other reasons, a large body of research compares the user cost of housing with rents. Often, but not always, this work is motivated by the desire to detect a ‘bubble’ in house prices. To give some illustrative examples, Baker (2002), McCarthy and Peach (2004), Himmelberg, Mayer and Sinai (2005) and Gallin (2005, from whom we copy our title) value houses using the user cost in the United States. Hatzvi and Otto (2008), Weeken (2004), Kivistö (2012) and Browne, Conefrey and Kennedy (2013) conduct similar exercises for Australia, the UK, Finland and Ireland respectively. The OECD (2005) provides international comparisons, including simple estimates for Australia. These papers provide citations to many others. A limitation of these papers is that prices and rents come from different samples with different characteristics. Owner-occupied houses tend to be larger and more expensive than rental dwellings. So a comparison of average prices with average rents reflects quality differences. This problem can be addressed by focusing on changes. Or, more commonly, the focus is on deviations from the average, on the questionable assumption that houses are fairly valued on average.

Our work differs in that we measure rents and prices for the same properties. This enables us to hold housing quality constant and hence assess the level of overvaluation. Several studies have also attempted to do this in the United States, including Smith and Smith (2006), Davis, Lehnert and Martin (2008), Campbell et al (2009) and Garner and Verbrugge (2009). We compare our results to this work in Section 5.2.

The paper that most resembles ours is Hill and Syed (2012), who examine Sydney house prices for 2001–2009, within a similar framework. Their paper focuses on technical issues relating to the use of hedonic regressions – specifically, imputation of price-to-rent ratios from incomplete data – and implications for the measurement of GDP, issues we do not address. We focus more on data issues, which leads to significant differences from some of their component estimates. For example, Hill and Syed assume no transaction costs or running costs, though we find these to be important. Offsetting these differences, Hill and Syed include land tax (not actually paid by owner-occupiers) and have high assumptions for depreciation and a risk premium. Some of their key results are similar to ours, as noted in Section 5.2. We hope that close attention to the relevant data increases the confidence that can be placed in these results. It also facilitates variations and extensions.

Other studies of the user cost of housing in Australia include Bourassa and Yin (2006), Stapledon (2007) and Brown et al (2011). Although these papers address somewhat different questions, we rely heavily on their discussion of the data, in particular the historical estimates compiled by Stapledon.