House prices, (un)affordability and systemic risk

Abstract

This is the first paper to examine the role of the real estate sector and housing unaffordability in the determination of systemic risk. We measure the systemic risk of the UK by employing the $\mathrm{\Delta }\mathrm{CoVaR}$ method developed by Adrian and Brunnermeier [(2016). CoVaR. American Economic Review, 106(7), 1705–1741] and we explore both its cross-sectional and time series behaviour. Regarding the former, we show that when the real estate sector is under distress the tail risk of the entire financial system increases significantly. With respect to the latter, the findings of our dynamic model suggest that sustainable house prices positively contribute to the stability of the financial sector; whilst house price exuberance and rapid increases in housing unaffordability amplify systemic risk. Finally, we examine the conjecture that the banking sector comprises a transmission channel from the housing market to systemic risk. Our empirical results are in line with this argument and highlight the key role of housing unaffordability.

Keywords:

• Affordability
• real estate sector
• systemic risk

Acknowledgments

For comments and suggestions, we are grateful to the editors of the special issue and an anonymous referee, Simon van Norden, Enrique Martinez and participants of the 12th International Conference on Computational and Financial Econometrics (CFE 2018), and the 27th Annual Symposium of the Society for Nonlinear Dynamics and Econometrics (SNDE 2019).

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Ferrari, Pirovano, and Cornacchia (2015) consider the housing market as an important source of systemic risk and they present a novel graphical approach to identify early signs of real estate related crisis. They conclude that overvalued properties and increasing household debt are early indicators of a crisis.

2 The research on the real estate sector's systemic risk is limited. Li, Pan, and He (2016) use Contingent Claims Analysis to measure systemic risk in the RE sector in China. They use a Vector Autoregressive (VAR) model and obtain a negative but temporary effect on banking returns in response to a shock in the risk measure. Meng et al. (2014) use Random Matrix Theory to investigate the systemic risk and spatiotemporal dynamics of the US housing market. Their findings suggest that the increasing risk since early 1977 resulted to the 2007 bubble.

3 We note that the systemic risk measure CoVaR examines the system's stress conditional on an individual firm's (or portfolio of firms) stress. This implies that the conditioning set to compute the system's stress varies cross-sectionally. The focus is therefore on the behaviour of the financial system's returns assuming a RE firm (or portfolio of firms) is in distress and to measure their tail dependency employing $\mathrm{\Delta }\mathrm{CoVaR}$. This approach differs from alternative measures of systemic risk that examine a firm's stress conditional on systemic risk (or on a systemic event such as a financial crisis). This is the case of the Systemic Expected Shortfall (SES) measure developed by Acharya, Pedersen, Philippon, and Richardson (2017). The attractive feature of SES is that the conditioning set is held constant across firms (as a way of ranking the systemic risk of firms), while CoVaR has the advantage of keeping the analysis within the standard regulatory tool of VaR.

4 There are alternative methods to obtain the $\mathrm{\Delta }\mathrm{CoVaR}$. For instance, Girardi and Ergün (2013) employ a multivariate GARCH model. We have also used this method, but the results, presented in Appendix 1, suggest that it does not perform as well in capturing the build-up of systemic risk ahead of the crisis.

5 This data is obtained from Datastream and Morningstar.

6 The frequency and period of the analysis is restricted by some of the variables needed to implement the dynamic model described in the next section.

7 As a robustness exercise, we have also computed the systemic risk measures using the FTSE350 RE index instead of the portfolio of the 81 RE firms described above. The results are similar. We decided to employ the portfolio of 81 companies instead of only the ones in the FTSE350 RE index in our main analysis in order to have a more comprehensive and representative sample of the sector.

8 The banking sector's VaR increases by 47% conditional on the RE market being under distress. We have also reversed the analysis to check if in periods when banks are stressed, the RE firms' risk increases. The $\mathrm{\Delta }{\mathrm{CoVaR}}_{bank}^{RE}$ is 6.78%. Although the effect is sizeable, we note that it is lower than $\mathrm{\Delta }{\mathrm{CoVaR}}_{RE}^{bank}$, which may be indicative that the effect of the correlation runs from the real estate sector to the banking institutions.

9 All the regression analyses of this section are available upon request.

10 While there is no significant cross-section relationship between VaR and $\mathrm{\Delta }\mathrm{CoVaR}$, there is indeed a relationship between those two variables at the time series dimension as can be seen in Figure 3. However, these two variables provide different information. For most of the period $\mathrm{\Delta }\mathrm{CoVaR}$ is above VaR, and specially noteworthy is the difference in the build up of the financial crisis.

11 We have also considered an alternative measure of leverage, the Leverage Index provided by the Center of Risk Management Lausanne (data available at: http://www.crml.ch/index.php/systemic_risk/). The results with this alternative measure of financial leverage, available upon request, were similar to the ones reported below and we have therefore omitted them.

12 These two variables have been adjusted to monthly frequency using cubic spline interpolation.

13 We have also considered an alternative approach to the one of having three separate regressions for each of the variables in the set of housing market factors, H, as reported in Table 3. This approach involved having two separate regression specifications. The first specification included real house prices and the housing exuberance dummy in the same model and the second real house prices and housing affordability. In both cases, the macroeconomic risk factors, M, and the firm characteristic factors, C, remained the same as in Table 3. The results not reported here, but available upon request, were qualitatively similar to the ones in Table 3.

14 In Appendix 3 we provide information about the composition of the RE companies by type of business and their corresponding systemic risk measures.

15 The coefficients of the variables that control for firm characteristics, C, and macroeconomic environment, M, barely change and are therefore not reported here.

16 Similarly to the analysis reported in Table 3 and discussed in footnote 13, we have estimated two alternative regression model specifications. One with real house prices and the housing exuberance dummy, and another where we included real house prices and housing affordability. The results, available upon request, were similar to the ones in Table 5 for the former specification. For the latter, real house prices were not statistically significant.

17 Nationwide is the second largest mortgage lender (12.3%), but it is a building society hence not in the London Stock Exchange. We also leave out Santander because this company is not traded in the UK stock market either.