The Wealthy Hand-to-Mouth Households in South Korea

Abstract

This study establishes the stylized facts on household balance sheets in South Korea and empirically investigates their macroeconomic implications based on the concept of ‘wealthy hand-to-mouth (HtM)’ households that hold little liquid wealth with owning large amount of illiquid assets. Using a household-level panel data for the period of 2000–2014, we find that (1) there are neither deleveraging of household debts nor a sharp decline in house price even during the financial crisis, (2) run-up in household debt in 2000s is led by high-income group, (3) regardless of net worth level, wealth is highly concentrated on illiquid assets such as housing and real estate, (4) the share of wealthy HtM households is very high compared to the cases of other advanced countries. We estimate the marginal propensity to consume out of a transitory shock and find that the consumption response of HtM households is larger compared to the non-credit-constrained group, posing a threat to macroeconomic stability. Using discrete choice models with fixed effects, we also find that a household that acquire more real estate assets is more likely to become wealthy HtM when its income is relatively lower or its indebtedness is relatively higher. We discuss the characteristics of HtM households and the role of macroprudential policy.

Key Words:

• Household balance sheet
• household debt
• wealthy hand-to-mouth households
• marginal propensity to consume

Jel Classification:

• D1
• D31
• E21

Acknowledgments

I thank David Berger, Kozo Ueda, Claudio Borio, Xuehui Han, and other participants at 2015 Bank of Korea-Yonsei University Joint Conference and Asian Development Bank Workshop for helpful suggestions and comments. All remaining errors are mine. The usual disclaimer applies.

Notes

1. BIS Statistical Bulletin, March 2016 (http://www.bis.org/statistics/bulletin1603.pdf)

2. From the experience of 2008 global financial crisis, it is relatively well known that what would happen when housing price collapsed or deleveraging process started because of external shocks. Mian et al. (2013) document how a shock on household net worth during the financial crisis leads to a drop in private consumption. And Mian and Sufi (2014b) clearly demonstrate how local shocks are transmitted to other areas, reducing the employment even in areas that were not subject to local shocks such as a sharp decline in housing prices.

3. Section 2.2.2 and Appendix below explain the concept and identification of these groups on the data. See Kaplan and Violante (2014) and Kaplan et al. (2014) for a model of wealthy hand-to-mouth household behaviour. The former provides a more full-fledged model in the context of two-asset incomplete-market life-cycle model and the latter briefly explains the concept using a two-asset two-period model.

4. Visit the website (http://www.kli.re.kr/klips_eng/index.do) for more information on KLIPS.

5. While it is not a main topic of this paper, it is important to examine whether the supply or demand factors of loans drives the patterns shown in Figure 2. Low-income groups, which are more likely to face severe borrowing constraints, would turn to non-bank financial institutions and pay higher interest rates.

6. We use the terms of illiquid assets, real estate assets (including housing), and non-financial assets interchangeably.

7. This number might seem too extreme. But, according to the national balance sheet provided by the Bank of Korea and Statistics Korea, Real estate assets, including land and buildings, accounted for almost 90% of national wealth. The statistics of national balance sheet is available at the Bank of Korea Economic Statistics System (http://ecos.bok.or.kr).

8. Jeonse is a term unique to South Korea that refers to the way houses are leased. Instead of paying monthly rent, a Jeonse resident will make Jeonse deposit, which is a lump-sum deposit, at anywhere from 50% to 90% of the market value. Typical contract lasts for two years and is renewable. Jeonse deposit is not liquid since it is in the hand of an owner during the contract period. This type of housing has been very popular, especially for the period of (expectation of) rising house prices, underdeveloped financial market, and high interest rates.

9. Appendix briefly explains the concept of wealthy and poor hand-to-mouth households and discusses how to identify them on our data. For more detail, refer to Kaplan and Violante (2014) and Kaplan et al. (2014).

10. Lusardi et al. (2011) use a similar concept, “financially fragile” households. Their definition is a household that would not be able to come up with $2000 in 30 days. Based on a survey, they document that a quarter of US households are reported as “financially fragile.”

11. As mentioned above, refer to Kaplan et al. (2014) and Kaplan and Violante (2014) for a more rigorous treatment.

12. We will treat this issue more rigorously using discrete choice models with FE below.

13. We will focus more on the wealthy HtM households below. It is because the share of poor HtM households is very small and, as evident in Table 1, the poor HtM households share the characteristics of poor people. The issue of poverty needs to be treated as a separate topic.

14. Blundell et al. (2008) use an MA(1) process instead of an i.i.d shock. Kaplan and Violante (2010) conjecture that this difference is quantitatively minor.

15. With an additional assumption of “short memory,” , one can show that the marginal propensity to consume out of a permanent shock, which is defined as , can be estimated byHowever, we do not delve into this because Kaplan and Violante (2010) show that the presence of tight borrowing constraints makes this estimator for permanent shocks biased.

16. We also use the age groups of 25–79 and 30–60 and the main result does not change.

17. For multinomial FE logit specification, the outcome variable, with j=1,2,3, is a polytomous categorical variable for all households and observation times. We set as the baseline outcome, which corresponds to wealthy HtM households. and correspond to non-HtM and poor HtM households, respectively. Our specification is as follows:

18. During the period of 2000–2013, the shares of homeowner, Jeonse, and paying rent are 59.3%, 25.2%, and 12.2%, respectively. We drop the category of “others,” which takes the share of 3.4%.

19. For our FE logit analysis, we use the explanatory variables at time t, not time . Suppose a situation when a household becomes wealthy HtM at time t by buying a house with incurring much debt between time and t. When we consider the factors that make the household wealthy HtM between and t, the amounts of real estate asset, liquid asset, and disposable income at time t matter, not the amount of those variables at time . For example, if a household has zero debt at time and hold much debt at time t to own a house, the appropriate variable to explain the wealthy HtM status at time t is the debt level at time t, not the debt level at time . Despite the possibility of spurious regressions when the lagged variables are used, we also estimate the same regression model with lagged-dependent variables and find that the main result does not change much.

20. Note that Table 4 reports the very low values of pseudo in column (1)–(4). There are three reasons. One is that fixed-effects estimation use only within-household differences, essentially discarding any information about differences between households. The other is that we do not consider the role of in estimating FE discrete choice models. Another is that we intentionally omit the variables of financial assets and net worth. When we include them in the regression and estimate, the values of pseudo 's in column (3) and (4) are 0.51 and 0.52.

21. While not reported here, we also find that a smaller family tends to become non-HtM, a household with older head tends to become poor HtM, and a household living in metropolitan area tends to become wealthy HtM.

22. The percentages need not add up to 100% because some households are double-counted.

23. See Cho (2012) for consumption behaviour after retirement. She uses the same data, KLIPS, and finds that consumption falls after retirement and the drop in consumption is related to the pre-retirement wealth level.

24. Refer to the following link for the Survey: http://www.bok.or.kr/contents/total/ko/boardView.action?boardBean.brdid=124918&boardBean.menuid=559&boardBean.rnum=1&menuNaviId=559&boardBean.cPage=1&boardBean.categorycd=0