Please enable JavaScript to view the comments powered by Disqus.

We use cookies on our website to give you the best online experience. Please know that if you continue to browse on our site, you agree to this use. You can always block or disable cookies using your browser settings. To find out more, please review our privacy policy.


Real Estate Research provided analysis of topical research and current issues in the fields of housing and real estate economics. Authors for the blog included the Atlanta Fed's Jessica Dill, Kristopher Gerardi, Carl Hudson, and analysts, as well as the Boston Fed's Christopher Foote and Paul Willen.

In December 2020, content from Real Estate Research became part of Policy Hub. Future articles will be released in Policy Hub: Macroblog.

Comment Standards:
Comments are moderated and will not appear until the moderator has approved them.

Please submit appropriate comments. Inappropriate comments include content that is abusive, harassing, or threatening; obscene, vulgar, or profane; an attack of a personal nature; or overtly political.

In addition, no off-topic remarks or spam is permitted.

June 20, 2011

The housing wealth effect revisited

A lot of economic researchers have struggled to answer what seems like a simple question: how much does consumption rise when housing prices go up? Many economists believe that, by increasing the wealth of homeowners, rising house prices during the housing boom had a substantial positive effect on consumption (see this speech by former Fed Chairman Alan Greenspan). Along the same lines, the dramatic decline in home values during the Great Recession is thought by many to be a drag on consumption today.

Conventional wisdom among market participants, and to some extent academics, is that the marginal propensity to consume (MPC) out of housing wealth is somewhere between 3 and 5 cents per dollar for households in the United States (see this paper by Bostic, Gabriel, and Painter for a nice survey of the literature on housing wealth effects). In other words, on average, every $1 increase in housing wealth results in a 3–5 cent increase in consumption. However, many researchers are highly skeptical of these numbers for a number of reasons (see, for example, this 2009 critique). First, a basic life-cycle model that incorporates Friedman's permanent income hypothesis (PIH) would predict that modest changes in housing wealth (both anticipated and unanticipated) should result in changes in consumption that are smaller than the 3–5 cent estimate.

Second, most research that finds big MPCs out of housing wealth estimate these with aggregate consumption and wealth data (see, for example, this 2005 study by Case, Quigley, and Shiller). Aggregate regressions like these are highly susceptible to an omitted variables critique; it may be that both consumption and housing prices are driven by some other variable not accounted for in the regressions. One possibility is that as expected future income in a particular area goes up, the anticipated increase would cause residents in this area to consume more—because their future incomes are higher—at the same time house prices are rising—because more people would want to live in that particular area and enjoy these higher incomes. Ideally, one would use micro-level consumption and housing data to control for these confounding variables. Unfortunately, high-quality panel data on both housing values and household consumption have been hard to find.

An innovative new paper by Jie Gan of the Hong Kong University of Science and Technology recently published in the Review of Financial Studies may have partly solved these data issues. The paper uses property-transaction and credit-card data from Hong Kong to estimate the causal effect of changes in housing wealth on individual-level consumption behavior. For her mortgage and housing data, Gan uses a data set similar in quality and scope to those now being used in the United States (including some papers that we have written). Specifically, Gan obtained detailed mortgage data and some demographic information from a large Hong Kong bank for the period 1988–2004. She also obtained government data on the universe of all housing and mortgage transactions in Hong Kong from the early 1990s to the mid-2000s, which allowed her to construct district-level house price indices. For her consumption data, Gan used monthly credit card statements from the largest credit card issuers in Hong Kong, available for the early 2000s. While credit card spending is certainly not a perfect measure of consumption, Gan argues that credit

cards account for more than 20 percent of consumer spending in Hong Kong and are used to purchase a diverse, representative set of products.

 Merging these three data sets yields a panel of about 12,000 homeowners along with monthly information on credit expenditures and home values during the 2000–2 period. Gan then estimates a regression of quarterly consumption growth, measured with the individual-level credit-card data, on quarterly growth in lagged housing values, measured at the district level, and a set of individual fixed effects that control for time-invariant individual characteristics. The regression also includes a set of interaction variables between quarterly time dummies and occupational categories. These variables control for changes in income growth that may be correlated with both house price movements and consumption growth.

Gan finds that the elasticity of consumption growth to house price growth is about 0.17, which implies that a 10 percent increase in house price growth leads to a 1.7 percent increase in consumption growth. While this is a large effect, it translates into a slightly lower MPC out of housing wealth than the 3–5 cent effect common in previous studies. The reason is that the estimated sensitivity of consumption growth to house price growth that Gan estimates must be multiplied by the consumption-to-housing wealth ratio in Hong Kong in order to construct an estimated MPC out of housing wealth. The consumption-to-housing-wealth ratio is relatively small in Hong Kong (11.5 percent) because housing is extremely expensive. Consequently, the MPC out of housing wealth in Hong Kong is about 2 cents per dollar. While 2 cents seems low, especially given previous empirical evidence, Gan argues that this seemingly low MPC should not be construed as evidence that changes in housing wealth have only a small effect on economic activity. To the contrary, because housing is so expensive in Hong Kong, a small MPC implies substantial impacts of changes in house prices on the real economy.1

Given these large and statistically significant results, an obvious question is whether they truly correspond to causal effects, or whether omitted variables might still be causing problems. The answer is probably mixed. On one hand, Gan's data is a significant improvement over previous studies on multiple dimensions. The micro-level credit-card panel data allow her to use individual-level fixed effects, something not possible when using aggregate consumption data. As Gan points out, a fixed-effects specification controls for unobserved household heterogeneity that could lead to biased estimates. Other studies have used micro-level consumption data from the Panel Study of Income Dynamics (PSID), but that data set only measures expenditures on food and has been found to contain substantial measurement error (see Runkle 1991 for evidence of substantial measurement error in the PSID measure of food consumption). Furthermore, the district-level house price indexes used to calculate estimates of home values are more disaggregated than the state-level and even MSA-level house price indexes that previous studies have used. As a result, Gan's price indexes probably suffer less from measurement error than the price indexes in other work.

But even with these significant improvements in data quality, Gan may not have solved the endogeneity issue. As Gan writes in her paper, "[s]ince housing prices are available only at the aggregate level or at the level of the metropolitan statistical areas (MSAs), the observed consumption sensitivities may be driven by economic- or MSA-wide shocks that simultaneously affect housing prices and consumption." In other words, from an econometric standpoint, a regression of consumption growth at the individual level on house price growth at the district level is only identified from time-series changes in average consumption growth at the district level and house price growth at the district level.2 That means that other, unobserved district-level variables could potentially be driving the correlation between consumption growth and house price growth. As noted above, one potential omitted variable is expectation of higher future income at the district level.

Gan downplays this issue by noting that many Hong Kong residents tend to work and reside in different districts. As a result, district-level shocks are less likely to simultaneously influence both housing prices and consumption. But as an empirical matter, it is not clear from the paper how common it is to work and reside in separate districts. And from a theoretical perspective, it is not clear whether this argument effectively rules out the simultaneity issue. If a significant fraction of individuals who work in one district reside in the same outside district, then a shock to employment in a district could cause prices and consumption to co-move in the outside district. In addition, there are other types of district-level shocks that could create simultaneity regardless of the commuting patterns of its residents. For example, a public works project that improved existing infrastructure or that developed new park space (or created some other desirable public good) in a district would be expected to increase the attractiveness of the district—thus increasing housing demand, which would raise prices and possibly also average district-wide consumption, by changing the income/wealth composition of its population. To completely solve the simultaneity issue, one would need time-series variation in home values at the individual level. Unfortunately, that data is simply not available at this point—in Hong Kong or anywhere else.

Innovative tests of the housing wealth channel
Still, while Gan's data may not completely solve the simultaneity issue, some of her additional empirical tests go a long way toward alleviating these concerns. The first is a check on whether the consumption–housing wealth relationship is stronger for people who own multiple homes. A pure housing wealth effect would predict that the consumption behavior of individuals with higher levels of housing wealth would be more sensitive to changes in wealth. This is exactly what Gan finds in her data.

In a series of other tests, Gan tries to distinguish between the role of credit constraints and a precautionary savings motive in generating the positive housing-wealth effect. The credit constraint story is that increasing housing wealth relaxes borrowing constraints for individuals, which results in higher consumption. This effect is expected to be relevant only for households that are borrowing-constrained. The precautionary savings motive refers to the tendency for risk-averse individuals to accumulate wealth in order to self-insure against negative future income or wealth shocks. If individuals consider housing equity to be a component of precautionary savings, then an increase in housing wealth might increase consumption by reducing other components of precautionary savings. For example, if a household is saving a certain percentage of each paycheck for precautionary motives, then an increase in housing equity might be considered to be a viable substitute, and the household might be expected to decrease the percentage saved of each paycheck, and thus increase consumption.

Gan focuses on households that refinance as a first test in distinguishing between these two effects. She argues that borrowing-constrained households would need to refinance in order to access any equity increases, while precautionary savers would not need to refinance, since they could increase consumption by decreasing other forms of saving. Thus, if the relaxation of borrowing constraints is driving the positive elasticity of consumption growth to changes in housing wealth, then the elasticity estimate should be larger among households that refinance.3 Gan finds evidence of both effects: the estimated elasticity is significantly higher for households that refinance but is still positive and statistically significant for households that do not.

In a second test, she identifies households that are likely borrowing-constrained based on their use of credit card lines and separately estimates regressions for households that are close to their credit limit and those that are far from their limit. If the credit-constraint channel is present, the elasticity should be higher for the households that are close to their limit, while the precautionary savings channel, in contrast, predicts that the elasticity should be higher for less-constrained households and thus households that are far from their credit limit. The results of this exercise provide support for the precautionary savings motive, as less-leveraged households have a stronger consumption elasticity than more-leveraged households.

Gan performs a few more clever tests to try to distinguish between the credit-constraint and precautionary-savings channels. She finds strong evidence in favor of the precautionary savings channel and little evidence of an important role for credit constraints. It appears that in Hong Kong, households use housing wealth as an important component of an overall self-insurance strategy and view increases in housing wealth as a substitute for other types of savings and thus an opportunity to increase consumption. This is a very interesting and important finding on its own, but we also view it as strong evidence that Gan has truly identified a causal relationship between housing prices and consumption. The reason is that if simultaneity bias is truly responsible for the positive estimate of the consumption elasticity, then we wouldn't expect the estimate to be sensitive to different samples. The fact that it is, and in ways that are consistent with theory, suggests that Gan has really identified the impact of housing wealth on consumption behavior.

One final caveat is that Gan focuses solely on homeowners, leaving renters out of her analysis. We would expect renters to be hurt by increases in housing values. Thus, while Gan finds a significant positive effect of housing wealth on consumption for the population of homeowners in Hong Kong, we interpret her results as an upper bound of the effect of housing wealth on aggregate consumption in Hong Kong.

Photo of Kris GerardiKris Gerardi
Research economist and assistant policy adviser at the Federal Reserve Bank of Atlanta

1 I believe that what Gan means here, although it's not completely clear, is that since the housing stock in Hong Kong is so highly valued, a 1 percent change in house prices translates into a lot of additional consumption, even with a marginal propensity to consume out of housing wealth of 2 cents.

2 The reason is that any time the variation in a right-hand side variable is at a more aggregated level than the dependent variable, the coefficient estimate associated with that right-hand side variable is identified from the variation in both variables at the more aggregated level.

3 Another potential way that a borrowing-constrained household could access increases in housing equity is through home equity lines of credit (HELOC). HELOCs do not require households to refinance their mortgage. This possibility is not discussed in the paper, perhaps because HELOCs are not quite as popular in Hong Kong as they are in the United States.