From real estate to consumption: the role of credit markets in the USA

Abstract

The aim of this article is to test whether the credit market conditions affect the strength of transmission of real estate wealth effects on household consumption in the US economy. Although many different works have dealt with the analysis of the existence of a real estate wealth effect, most of them as a reaction to the dramatic increase of housing prices in several OECD countries, there are only few papers analysing whether the consumption response depends on the positive or negative sign of the wealth shock and, as far as we know, none of them takes the effects of credit market conditions on that asymmetric response into account. This article tries to fill the existing gap in the literature on this matter. From an econometric perspective, we estimate the asymmetries in the consumption response within the momentum threshold autoregressive model (M-TAR) proposed by Enders and Siklos (2001), but following Stevans (2004), it is applied to a multivariate framework. The main results show that the credit market conditions play a significant role in the transmission of changes in real estate wealth to consumption. In addition, we find that there exists an asymmetric behaviour in the US aggregate consumption spending responses to real estate wealth and credit market shocks, which is only significant when a negative shock takes place.

Keywords:

• consumption
• real estate wealth effect
• credit conditions index
• housing liquidity index
• nonlinear cointegration
• M-TAR models
• asymmetric adjustment

JEL Classification:

• D12
• E21

Funding

E. Márquez and Ana R. Martínez-Cañete acknowledge financial support from the Spanish Ministry of Economy and Competitiveness [grant number ECO2012-31941].

Notes

¹ The construction of CCI and HLI is explained in detail in Duca et al. (2012) and summed up in the ‘Data’ section of this article. We are grateful to A. Murphy for kindly providing the data of CCI and HLI.

² Duca et al. (2012) point out that the credit crunch of the beginning of 1990 was due to the stricter regulation that implied the new requirements of Basil I.

³ Fig. 2 shows per capita data at 2005 constant prices of the following variables: total consumption excluding imputed rental of owner-occupied nonfarm housing, after-tax labour income and real estate wealth interacted with the HLI index. So, note that the variable associated with real estate wealth includes the HLI index. Details on the data source can be found in the ‘Data’ section.

⁴ The authors address the problem of alternative explanations of time series correlation between the variables assuming that consumption and housing wealth are endogenous and they are driven by movements in technology, preferences and monetary policy.

⁵ As Aron et al. (2012) point out, these two variables are crucial since it will be hard to find a cointegration relationship if credit market conditions are not taken into account. 

⁶ Lettau and Ludvigson (2004) emphasize the importance of the permanent or transitory character of the wealth shocks. In fact, their estimation methodology is based on the idea of considering only permanent wealth shocks in the estimation of the marginal propensity to consume out of wealth.

⁷ To be concrete, Brady et al. (2000) find that one-third of the loans were spent on home improvements, and just over one-fourth were used to pay off other debt. Roughly one-fifth went for consumer expenditures, and a similar amount was used to invest in real estate or business.

⁸ This indirect wealth effect channel would not operate in the opposite direction.

⁹ However, Donihue and Avramenko (2007) consider that the results in Stevans (2004) may be overstated because of how wealth and the threshold are measured. 

¹⁰ They also report a disproportionate impact on savings for those households with negative housing equity. 

¹¹ In fact, Márquez et al. (2013) show the opposite result for the UK economy.

¹² Aron et al. (2012) also use total consumption, excluding housing services, for the case of the United States of America.

¹³ For a more detailed explanation, see Lettau and Ludvigson (2001). We are grateful to Martin Lettau for kindly providing the data. 

¹⁴ Following the works of Aron et al. (2012) and Duca et al. (2012) for the United States, we use this specification to take into account the direct effect of credit on consumption and the housing wealth effect on consumption through the collateral channel.

¹⁵ In this respect, we follow the proposal by Aron et al. (2012).

¹⁶ Stevans (2004) considers the case of the financial wealth instead of real estate wealth. The indicator function in Enders and Siklos (2001) is different, since they define the value of the threshold as a function of the estimated residuals. The main advantage of the methodology proposed by Stevans (2004) is that it makes possible to distinguish where the wealth shock comes from, whereas this is not the case in the Enders and Siklos (2001) methodology.

¹⁷ Previously, different unit root tests to the consumption, income and wealth series were applied. Specifically, we consider the ADF tests and those proposed by Elliott et al. (1996) and Ng and Perron (2001); we are unable to reject the null hypothesis of a unit root. In all cases, the number of lags was chosen according to the Modified Akaike Information Criterion (MAIC). Although they are not included to save space, all the results are available from authors upon request.

¹⁸ To analyse whether the results are robust for the chosen number of lags , we have calculated the trace tests for , . In all cases, we cannot reject the null of no cointegration apart from , where cannot be ruled out. The results are the same when is treated as an exogenous variable.

¹⁹ Donihue and Avramenko (2007) do not obtain cointegration using total net wealth, but they do when breaking down wealth according to the asset liquidity. Stevans (2004) does not find a cointegration relationship when applying the Johansen methodology, but he does when an asymmetric adjustment through an M-TAR is considered. On the other hand, Ludvigson and Steindel (1999) and Lettau and Ludvigson (2001, 2004) do find a long-run relationship.

²⁰ The results provided do not include a time trend in the long-run relationship, since it was not significantly different from zero. 

²¹ See, for the US economy, Boone y Girouard (2002), Ludwig and Sløk (2004), Catte et al. (2004) and Carroll et al. (2011), among others.

²² The high value obtained for the marginal propensity to consume out of real estate wealth is due essentially to the effect of the HLI index. If it is not considered, the estimated value falls up to 0.1159. But in this case, this value is based on the estimated value of the elasticity of consumption in relation to the real estate wealth interacted with the HLI index. 

²³ According to the AIC, the ECM has been estimated with one lag. The results are quite similar when zero lags are considered following the BIC criterion. As the adjustment coefficients for were not significant, it has been considered as an exogenous variable.

²⁴ Unemployment and interest rates data come from the OECD Economic Outlook. The unit root tests applied to do not reject the null hypothesis of unit root, apart from the Ng and Perron (2001) tests, where both a constant and a time trend are considered. In this case, the unit root can be ruled out at a 10% significance level. With regard to , most tests reject the unit root assumption.

²⁵ Skinner (1996) also points out that housing gains may be anticipated by consumers, while that is not the case for losses.

²⁶ Brady et al. (2000) on the basis of 1999 survey of consumers, show that only around 20% of the US HEE was devoted to consumption.