Additional Credit for Liquidity-Constrained Individuals: High-Interest Consumer Credit in Korea

ABSTRACT

We find that the delinquency probability on formal sector debts of private loan borrowers in Korea increases from 2.4% to 20% in the first year after the borrowing and to 32% in the second year. This increase happens despite private loan borrowers trying to rebuild their financial health by reducing formal sector debts, credit card cash service balances, and credit card purchases during the post-borrowing period. This limits the possibility of moral hazard driving the results. Private loan amounts are positively associated with the delinquency probability after controlling other commonly used variables, suggesting that they contain additional information on the worsening financial situation of an individual.

KEY WORDS:

• delinquency probability
• household debt
• liquidity constraint
• payday loan
• private loan

Acknowledgments

We are grateful to Ali M. Kutan (editor-in-chief), three anonymous referees, and seminar participants at the Seoul National University Institute for Research in Finance and Economics for helpful comments and suggestions.

Funding

Jung-Wook Kim acknowledges the financial support from the SNU Institute for Research in Finance and Economics and the Institute of Management Research at Seoul National University. Seungyeon Won acknowledges the financial support from the SNU Institute for Research in Finance and Economics.

Notes

^1. We define private loan companies in Korea as those that provide expensive short-term consumer credit products. These companies are not formal financial institutions in Korea and, thus, are not regulated by the Financial Supervisory Service as much as other formal financial institutions such as commercial banks. A private loan company in Korea can be established by registering it with the local government.

^2. Typically, a loan application is made via the internet, telephone, or fax without visiting a branch of a private loan company. The applicant submits the loan application form along with supplementary documents regarding employment status and income. If the application package meets the requirements, the loan is made within 1 or 2 days without further investigation regarding applicants.

^3. The credit rating of an individual ranges from 1 to 10. Lower numbers represent better financial condition. Individuals with credit ratings of 7 or higher (out of 10) have little access to formal financial sector debt in Korea. The average credit rating of private loan borrowers in our sample is 7.

^4. However, the interest rate of private loan companies is much higher than the average interest rate on household loans provided by depository banks in Korea, which is about 5.2% in 2012.

^5. Formal financial sector debts include various loans with or without collateral from formal financial institutions. In the case of personal bankruptcy, borrowers in Korea are not exempted from repaying debts regardless of whether loans are secured by real estate or not.

^6. In our sample, all the private loan borrowers have at least one credit card. Unlike United States or other developed countries where strict screening is performed before a credit card is issued to an individual, credit card was relatively easily issued in Korea as a possible way to boost the economy in early 2000s. As a result, for example, in 2013, the average number of credit cards of economically active population is 4.

^7. The findings of Bhutta, Skiba, and Tobacman (2015) and our article are different from those of Agarwal, Skiba, and Tobacman (2009) who show that more than two-thirds of payday loan borrowers in their sample have credit card liquidity of more than $1000 at the time of payday borrowing. Since the interest on credit card cash services is much lower than that on payday loans, Agarwal, Skiba, and Tobacman (2009) discuss the possibility of payday loan borrowers being irrational. This seems not to be the case for both payday loan borrowers in Bhutta, Skiba, and Tobacman (2015) and Korean private loan borrowers in our sample, who exploited credit card liquidity.

^8. If private loan borrowers are fully rational and are not subject to any behavioral biases, private loan borrowing is the result of rational choice and thus welfare improving. However, there are many researches indicating that high-interest loan borrowers do not fully understand the cost that they have to pay, suggesting the possibility that private loan borrowers are not fully rational (Bertrand and Morse 2011; Stango and Zinman 2009) and may be subject to self-control problems (Elliehausen and Lawrence 2001; Laibson 1997) as mentioned in the Literature Review section. In this regard, it is possible that not fully rational borrowers may overestimate their future income prospects (overconfidence), which prompts borrowing of private loans with undesirable results. In fact, Skiba and Tobacman (2011) shows that serial payday loan borrowing is associated with higher bankruptcy in the United States.

^9. Iacoviello (2008) developed a dynamic general equilibrium model to explain the strong positive relationship between individual debt and income inequality in the United States. Individuals with negative earnings shocks should rely on debt financing. When formal sector debts are exploited, individuals need to rely on private sources. The increasing time trend in the private loan amounts in Korea may well reflect the same phenomenon.

^10. Original source data for the private loan company in our sample keep track of each individual over time. When KCB merges this dataset with their public debt-related data, KCB is able to identify each individual over time as well. However, when KCB provides us with data for this research, for security-related reason, they assign different numbers for each individual each year. However, they do this after they keep track of all the past and future delinquency history of each individual. Thus, we are able to track down the financing activity of each individual 12 months before and 24 months after the month of private loan borrowing.

^11. For brevity, we do not specify the year of the private loan borrowing, T, in defining most of our variables. The only exception is income as will be discussed later.

^12. While we have detailed information on different kinds of debts comprising the aggregate debt of each individual, we do not have information on the interest rate of each debt component for each borrower, making it impossible to calculate the risk-weighted aggregate debt amount for each individual. Thus, large formal financial sector debt of an individual in our dataset does not necessarily imply bad quality debt since even financially sound Korean households borrow money using their houses as collateral when they purchase their houses.

^13. Even though our sample period includes 2007, 2012, and 2013, these 3 years are excluded from the table as we calculate delinquency probability for the pre-borrowing 12 months and the post-borrowing 24 months.

^14. We calculate the average of each individual’s debt-to-income ratio to get this number. This number is slightly lower than the debt-to-income ratio calculated using two average numbers of debt and income, 53, 185 and 27,078, respectively, as reported in Table 2.

^15. With the maturity of 1 year, private loan borrowers do not need to repay the principal at least for the first 12 months. In addition, typically, interest of private loan is prepaid at the time of borrowing when borrowers receive principal net of interest payment.

^16. We calculate the average of each individual’s private loan to income ratio for each quintile to get these numbers. These numbers are about the same as the debt-to-income ratios calculated using two average numbers for each quintile reported in Table 3.

^17. It is interesting to note that the delinquency probability and delinquency density remain roughly at the same level after the third quintile. We discuss this issue in the Information Asymmetry and Loan Size section in relation with stricter requirement for private loan for the loan size greater than 5 million won. Note that the average private loan amount of the third quintile is about 4.6 in Table 3, which is very close to 5 million won cutoff.

^18. We calculate the average of each individual’s debt-to-income ratio to get these numbers. These numbers are slightly lower than the debt-to-income ratios calculated using two average numbers for each quintile reported in Table 3.

^19. We also run OLS regressions and logit regressions to check the robustness of our results. All the main results remain qualitatively similar in these specifications as well.

^20. Our dataset has information on whether borrowers are employed, self-employed, or unemployed. This allows us to estimate regressions separately for each group. However, the relationship between private loan amount and delinquency probability remains positive and significant in all subsamples. Our dataset does not have information on the maturity of various debts each individual has, and thus, we could not include the average debt maturity for each individual in our regressions.

^21. Skiba and Tobacman (2011) have data containing information on whether payday loan applicants are rejected or approved for a payday loan application. This information allows them to examine whether the access to payday loans causes the payday loan borrower’s Chapter 13 bankruptcy filing. They focus on the fact that the approval of payday loans depends on a fixed threshold of applicants’ credit scores in their sample. Under the assumption that the risk characteristics of payday loan borrowers around the threshold are randomly assigned, they find that access to payday loans causes higher levels of Chapter 13 bankruptcy over the following 2 years. We only have data for private loan borrowers who are approved for private loans and thus cannot use a similar methodology.

^22. We also use the standard error clustered by the yearmonth the private loan is made. The results are qualitatively similar.

^23. Since the estimated coefficient is very small, we report the coefficient after multiplying the original estimate by 1000 as written in the legend of the table.

^24. We also estimate regressions separately for each year to allow different coefficients for our variables in each year. The coefficients for private loan variables remain positive and significant in all years and in all specifications of Table 4.

^25. All control variables are evaluated at their respective sample means for each year. Appropriate year dummy estimates are used in evaluating probabilities.

^26. Note that regardless of whether we include $CC{L}_{it}$ (credit card cash service liquidity) or not in regression specifications, all the coefficients change little. We obtain qualitatively similar results when we include or exclude other variables. This suggests that our results are not sensitive to potential multicollinearity among independent variables. We are most grateful for an anonymous referee for pointing this out.

^27. Even though we include the year fixed effect in our regressions, it is possible that delinquency probability may covary with macro-level control variables and year fixed effect may not fully capture the heterogeneity in macro-level control variables affecting differently the delinquency probability. To address this issue, we construct macro-level control variables with annual frequency including the aggregate amount of household bank loans over GDP (to capture credit trend), stock index returns (to capture the impact of asset price on private loan demand), credit spread (to capture changes in risk premium), the volatility of stock index return measured by the standard deviation of daily index return for a given year. When we use these variables, year fixed effect term is removed but a constant term is added. Magnitudes and significances of the private loan variables are not affected much. We are grateful for an anonymous referee for suggesting this exercise as a robustness check.

Additional information

Funding

Jung-Wook Kim acknowledges the financial support from the SNU Institute for Research in Finance and Economics and the Institute of Management Research at Seoul National University. Seungyeon Won acknowledges the financial support from the SNU Institute for Research in Finance and Economics.