Loan forbearance takeup in the Covid-era - The role of time preferences and locus of control.

During the COVID-19 pandemic, many countries eased the burden on borrowers through loan forbearance. Using a representative sample of the Hungarian adult population, we investigate whether time preferences and locus of control are associated with loan forbearance takeup. We find evidence that time discounting correlates with the resort to forbearance: ceteris paribus, more patient individuals are less likely to take up forbearance, even after controlling for their present/future bias, risk aversion, locus of control, demographic characteristics, educational level, financial status, and the effects of the pandemic. However, present bias and locus of control are not significantly associated with loan forbearance.

Introduction

Private and public loan forbearance programs have been widely used policy tools during the COVID-19 crisis around the world. According to the Oxford University database (Hale et al., 2020), 173 countries applied a nationwide temporary suspension of loan repayments or other contract reliefs for households until mid-2021. In the US, loans worth $2 trillion (a large share of mortgages, and almost all student loan debts) were in forbearance by the end of 2020 affecting 60 million individuals (Cherry et al., 2021). In Hungary, the forbearance program was the main economic policy tool to help households cope with the crisis, as no direct income support was provided (Drabancz et al., 2021). Participation was mandatory for banks. From an international perspective, it is one of the longest programs, the last prolongation ending on December 31, 2022.

Debt renegotiations, a dynamic bargaining between the lender and the borrower, is a well-researched area (Bergman and Callen, 1991; Chemmanur and Fulghieri, 1994; Hart and Moore, 1998; Moraux and Silaghi; 2014). However, the empirical literature focusing on the evaluation of public and private forbearance programs is mixed. On the one hand, it is a specific “emergency loan” which provides liquidity to distressed borrowers, helping them survive temporary crises. On the other hand, it can significantly increase the moral hazard, leading to a higher risk of default in the long run. Loan forbearance as a policy tool was found to be highly effective (Agarwal et al., 2017; Collins and Urban, 2018), ineffective in the long run (Dobbie and Song, 2020; Bergant, 2020), or dependent on the quality of institutions (Cherry et al., 2021; Godlewski, 2020; Mourad et al., 2020; Piskorski and Seru; 2021). Our study contributes to the empirical debate by investigating the borrowers’ side and focusing on the psychological characteristics of those participating in the moratorium.

A growing literature discusses how preferences and personality traits affect financial decisions. There are many preferences and personality traits, here we focus on two aspects of time preferences (time discounting and present bias) and locus of control due to their relevance indicated by the literature. There is ample evidence that individuals who discount the future less save more (Bradford et al., 2017; Falk et al., 2018) and have higher creditworthiness (Meier and Sprenger, 2012). Moreover, present-biased individuals are more likely to have credit card debt (Meier and Sprenger, 2010) and financial difficulty (Horn and Kiss, 2020). Locus of control has been shown to be relevant in saving decisions (Chatterjee et al., 2011; Cobb-Clark et al., 2016; Lunt and Livingstone, 1991) and in financial difficulty (Kuhnen and Melzer, 2018). Individuals who believe to a greater degree to have control over the outcomes of their life (exhibiting internal locus of control, as opposed to those with an external locus of control who attribute to external factors like luck or God what happens to them) have more savings and fewer financial problems. Moreover, there is growing evidence that time preferences (Meier and Sprenger, 2015) and locus of control (Cobb-Clark and Schurer, 2013) are stable in time at the individual level.

To the best of the authors’ knowledge, this study is the first to examine whether time preferences and locus of control are associated with the takeup of loan forbearance. Since taking up forbearance may be due to looming or (already present) financial hardship or the lack of savings, we focus on time preferences and locus of control that have been shown to associate with these factors. Our hypotheses are built on the findings of the literature that less time discounting, less present bias, and more internal locus of control are positively associated with having more savings and experiencing fewer financial difficulty. Therefore, we expect that individuals who discount the future less and are less present-biased are less likely to resort to loan forbearance, ceteris paribus. Similarly, we hypothesize that internal locus of control tendencies correlate with less participation in loan forbearance.1 Moreover, our study also sheds light on whether time preferences and locus of control are associated with loan forbearance takeup (if they are) only through the proposed channels (savings, financial difficulty, and the effects of COVID-19), or is there a genuine relationship beyond these channels.

Using representative data on the Hungarian adult population, we observe that time discounting correlates with loan forbearance while present bias and locus of control do not. Regression analysis confirms these findings. In successive specifications, we add more and more controls (demographic characteristics, educational level, financial status, and effects of the pandemic) to see if respondents’ time preferences and locus of control correlate with their resort to forbearance. We find that individual discount factors are negatively associated with the takeup of loan forbearance, even when considering all the control variables. Our findings suggest that time discounting may affect the takeup of forbearance through savings and financial difficulty: individuals who discount the future less have more savings and fewer financial difficulty and, as a consequence, are less likely to resort to loan forbearance. Interestingly, time discounting remains significantly associated with participation in the loan forbearance program even after controlling for these channels.

The rest of the study is structured as follows. Section 2 presents briefly the data, Section 3 reports the main findings, and Section 4 concludes.

Data

The research institute TÁRKI carried out the data collection in November 2020. They interviewed 809 respondents by telephone (due to COVID-19) who were representative of the Hungarian population in terms of gender, age, settlement type, and education level. The data collected provide information on gender, age, settlement type, region of residence, education level, family (marital status, household size, number of children), employment status, income, and perceived financial situation. Respondents also revealed how the COVID-19 pandemic affected their own health, the health of their family members, and their own financial situation (we refer to these effects as Effects of COVID-19 in the analysis below).2

The questions related to financial issues involved i) financial difficulty (a variable on the inability to pay utility bills, mortgage, or other debt in the last year, that is, in 2020), ii) the takeup of loan forbearance (binary variable on whether the respondent resorted to it, given that she had any type of loan), and iii) savings (the number of months that the respondent could live on their savings without problems).3 We focus on loan forbearance, but take into account the other financial variables as well since they may constitute potential channels.4

We followed the staircase (or unfolding brackets) method of Falk et al. (2018) when measuring time preferences. Respondents repeatedly had to choose between an earlier amount (fixed at 10,000 HUF, approximately equivalent to 35 USD at that time) and a larger later amount (X). The later amount changed in an adaptive manner so that we could approximate the amount that made the respondent indifferent between receiving the earlier or the later payment. There were three hypothetical interdependent questions. This task was carried out on two time horizons: now versus 1 month, and 12 months versus 13 months. In line with the literature, the discount factor (δ) that equalizes the 10,000 HUF in 12 months with X13 in 13 months (10,000 = δX13) is the proxy for time discounting (or patience). Using δ, based on the model of (β, δ)-preferences (Laibson, 1997), we could also calculate the parameter of time inconsistency (β) from the equation 10,000= βδX1, where X1 denotes the larger amount to be received in a month versus now. β < 1 indicates present bias, β > 1 indicates future bias, and β = 1 denotes time consistency.5

As time discounting involves risk by necessity (the future is inherently risky), we also measured risk attitudes with a hypothetical question to disentangle them. We followed Sutter et al. (2013) and proxied risk aversion by the amount placed as a bet in a gamble. The maximum amount that could be put at risk was 10,000 HUF.

Following Kuhnen and Melzer (2018), we used the Pearlin mastery scale (Pearlin and Schooler, 1978) to measure locus of control. This measure consists of seven statements, and the respondents could indicate to what degree they agreed with the statement on a 1–5 scale. We coded the answers so that higher scores denoted more internal tendencies. Locus of control does not correlate either with the discount factor (Pearson correlation = 0.044, p-value = 0.229), or the present bias (Pearson correlation = 0.053, p-value = 0.152).6

Findings

Our aim is to see whether two aspects of time preferences (time discounting and present bias) and locus of control are associated with the use of loan forbearance. Note that we only consider individuals who were eligible for forbearance, i.e. those who had a mortgage or other bank loan. First, we present some descriptive statistics to show how forbearance is associated with other financial variables (namely financial difficulty and savings), time discounting, present bias, and locus of control. Then, we proceed with the regression analysis to see if correlations persist when considering more and more control variables, paying special attention to potential channels.

Table 1 shows the correlations between the variables of main interest. It indicates that time discounting and locus of control are associated with financial difficulty and savings in line with the literature.7 Furthermore, the correlations show that time discounting correlates negatively with forbearance, while locus of control seems to be unrelated to it. The association between present bias and the financial variables has the expected sign in all cases, but the correlation is never significant.

Table 1

Correlations between time discounting, present bias, locus of control, and the financial variables of interest.

	Forbearance	Financial difficulty	Savings
Time discounting	−0.141 ***	−0.096***	0.121 ***
Present bias	0.047	0.035	−0.055
Locus of control	−0.044	−0.167 ***	0.267 ***

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.

Pearson correlations and their significance. The discount factor δ represents time discounting from the (β, δ)-model, while present bias is a dummy variable (=1, if β < 1). Locus of control is the aggregated value from seven 5-point Likert scale questions, where higher scores correspond to a stronger internal locus of control. Forbearance is a dummy variable showing whether the respondent took up forbearance (conditional on having a loan). Savings is a continuous variable created from 4 categories. Financial difficulty is a dummy indicating whether the respondent had difficulty paying their loans or utility bills in the last year.

Table 2 shows the output of a logit regression where the dependent variable is if the respondent uses loan forbearance (in case she has any loan).8 We observe that – while controlling for present bias, future bias, internal locus of control, and risk preferences – δ, the discount factor correlates negatively with taking up loan forbearance, indicating that individuals who value the future more are less likely to use forbearance. This coefficient is both statistically and economically significant. The average marginal effects calculated from the model show that individuals whose discount factor is higher by one standard deviation are 9.26% less likely to use forbearance. The association between forbearance takeup and discount factor remains significant at the 1% significance level even if we take into account the respondents’ gender, age, settlement type, education level, financial status (that includes employment status and perceived financial situation), savings, and financial difficulty. Note that the coefficient barely changes in subsequent specifications, suggesting that there is a genuine and stable association between time discounting and resort to loan forbearance that is less likely to be affected by potential confounders. The association remains significant at the 5% significance level even after considering the effects of the COVID-19 pandemic.

Table 2

Forbearance, time discounting, and locus of control – logit regression.

	Dependent variable: takeup of loan forbearance (=1 if yes)
	(1)	(2)	(3)	(4)	(5)	(6)
Delta (discount factor)	−2.289***	−2.218***	−2.227***	−2.181***	−2.350***	−2.222**
	(0.694)	(0.726)	(0.749)	(0.775)	(0.833)	(0.865)
Present bias	0.385	0.339	0.302	0.214	0.124	0.088
	(0.298)	(0.307)	(0.313)	(0.324)	(0.344)	(0.357)
Internal locus of control	−0.023	−0.017	−0.013	−0.007	−0.013	−0.011
	(0.022)	(0.023)	(0.023)	(0.024)	(0.026)	(0.027)
Risk, future bias	✓	✓	✓	✓	✓	✓
Demographic controls		✓	✓	✓	✓	✓
Financial status			✓	✓	✓	✓
Financial difficulty				✓	✓	✓
Savings					✓	✓
Effects of COVID-19						✓
Constant	1.481*	1.228	1.036	0.931	2.032	2.031
	(0.848)	(1.146)	(1.201)	(1.232)	(1.353)	(1.654)
Observations	324	324	321	321	300	291
Log Likelihood	−206.098	−201.844	−199.813	−189.514	−173.757	−169.652
Akaike Inf. Crit.	424.195	429.687	433.626	417.028	391.514	389.303

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.

Weighted logit regressions on the takeup of loan forbearance using the stepwise regression method.

The delta discount factor represents time discounting from the (β, δ)-model, while present bias is a dummy variable (=1, if β < 1).

Locus of control is the aggregated value from seven 5-point Likert scale questions, where higher scores correspond to a stronger internal locus of control.

Higher values of risk preference represent more risk-tolerant respondents.

Future bias is the opposite of present bias (=1, if β > 1).

Demographic controls: gender, age, settlement type (the capital, other towns, or village), and education level (primary education, no high-school graduation, high-school graduation, tertiary education).

Financial status: employment (employed, unemployed, inactive), perceived financial situation (good, okay, bad).

Financial difficulty: whether the respondent has problems paying their loans (if they had any) or utility bills.

Savings: measured by the number of months the respondent could live off of their savings.

Effects of COVID-19: see definitions in Section 2 (Data), lower values indicate that the pandemic had a worse effect on the respondent.

Appendix B contains a robustness check in which we include the self-reported income when considering financial status. We obtain qualitatively the same results as the coefficient of the discount factor is negative and significant at least at 5% in the specifications (1)–(4). However, once we control for savings and the effects of COVID-19, the significance vanishes. Our preferred model is the one without self-reported income, as about one third of the respondents do not report their income, so we lose many observations. Moreover, not reporting income correlates with the perceived financial situation, raising issues of self-selection.

We also consider potential channels through which time discounting may operate. More concretely, we assume that the association of time discounting with the takeup of loan forbearance may be mediated by savings (higher discount factor correlates with higher savings that, in turn, makes it less likely to resort to loan forbearance), financial difficulty (lower discount factor correlates with a higher probability of financial difficulty that, in turn, leads to the takeup of loan forbearance), or the effect of the COVID-19 pandemic (time discounting may be associated with suffering from COVID-19 which, in turn, may correlate with the use of loan forbearance). Table C1 in Appendix C contains all possible combinations of these potential channels, together with the control variables that we considered before. In line with natural expectations, savings correlate negatively with the takeup of loan forbearance, while financial difficulty is associated with a higher probability of resorting to loan forbearance. However, suffering from COVID-19 does not correlate with the takeup of loan forbearance. Importantly, the coefficient of the discount factor is remarkably stable and significant in all specifications, suggesting that time discounting plays an important role beyond these channels.

Table C1

Possible channels through which time discounting affects forbearance (logit regressions using financial difficulty, savings, effects of COVID-19).

	Dependent variable: takeup of loan forbearance (=1 if yes)
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Delta (discount factor)	−2.227***	−1.838**	−2.444***	−2.181***	−2.062**	−2.350***	−2.016**	−2.222**
	(0.749)	(0.775)	(0.803)	(0.775)	(0.841)	(0.833)	(0.792)	(0.865)
Present bias	0.302	0.307	0.211	0.214	0.210	0.124	0.193	0.088
	(0.313)	(0.322)	(0.332)	(0.324)	(0.346)	(0.344)	(0.333)	(0.357)
Internal locus of control	−0.013	−0.009	−0.015	−0.007	−0.011	−0.013	−0.004	−0.011
	(0.023)	(0.024)	(0.025)	(0.024)	(0.027)	(0.026)	(0.025)	(0.027)
Risk, future bias, demographic controls, financial status	✓	✓	✓	✓	✓	✓	✓	✓
Effects of the pandemic on health		0.095			0.273		0.141	0.302
		(0.280)			(0.291)		(0.294)	(0.305)
Effects of the pandemic on relatives' health		−0.120			−0.253		−0.187	−0.304
		(0.249)			(0.262)		(0.254)	(0.267)
Effects of the pandemic on the financial situation		−0.257			−0.241		−0.182	−0.168
		(0.164)			(0.178)		(0.170)	(0.186)
Savings: 1–2 months			−0.772**		−0.847**	−0.503		−0.634*
			(0.357)		(0.370)	(0.374)		(0.384)
Savings: 3–4 months			−0.363		−0.438	−0.063		−0.175
			(0.447)		(0.462)	(0.467)		(0.482)
Savings: 4< months			−1.515***		−1.498***	−1.208**		−1.241**
			(0.512)		(0.519)	(0.527)		(0.531)
Financial difficulty: no loan or problems paying utility bills				16.120		17.145	15.414	15.133
				(602.431)		(994.222)	(638.404)	(638.404)
Financial difficulty: could NOT pay loans or utility bills				1.238***		1.104***	1.167***	1.060***
				(0.357)		(0.379)	(0.369)	(0.393)
Constant	1.036	1.070	2.386*	0.931	2.260	2.032	0.995	2.031
	(1.201)	(1.466)	(1.328)	(1.232)	(1.602)	(1.353)	(1.518)	(1.654)
Observations	321	311	300	321	291	300	311	291
Log Likelihood	−199.813	−193.546	−182.711	−189.514	−176.437	−173.757	−185.414	−169.652
Akaike Inf. Crit.	433.626	427.093	405.422	417.028	398.875	391.514	414.828	389.303

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.

Weighted logit regressions on financial difficulty using the stepwise regression method.

The delta discount factor represents time discounting from the (β, δ)-model, while present bias is a dummy variable (=1, if β < 1).

Locus of control is the aggregated value from seven 5-point Likert scale questions, where higher scores correspond to a stronger internal locus of control.

Higher values of risk preference represent more risk-tolerant respondents.

Future bias is the opposite of present bias (=1, if β > 1).

Demographic controls: gender, age, settlement type (the capital, other towns, or village), and education level (primary education, no high-school graduation, high-school graduation, tertiary education). Financial status: employment (employed, unemployed, inactive), income, perceived financial situation (good, okay, bad).

Effects of COVID-19: see definition in Section 2 (Data), lower values indicate that the pandemic had a worse effect on the respondent.

Savings: measured by the number of months the respondent could live off of their savings.

Financial difficulty: baseline category consists of those who had no difficulty paying their loans or utility bills in the last year.

Even though the coefficient of present bias has the expected sign in all specifications (indicating that present-biased respondents are more likely to take up loan forbearance), it is never significant. Turning to the locus of control, even though the sign of the coefficient is consistently negative (suggesting that individuals with more internal tendencies are less likely to resort to forbearance), in no specification do we observe a significant relationship between locus of control and loan forbearance.

Conclusions

Based on the existing literature, time preferences and locus of control seem to be important non-cognitive determinants of financial decisions and outcomes. Our aim in this study was to see if they are also associated with the takeup of loan forbearance, a widely used policy tool during the pandemic to ease the financial burden on households. We find that present bias and locus of control are not associated with the takeup of forbearance, but we document a negative relationship between the discount factor and the use of forbearance. This relationship is stable, even if we control for present/future bias, risk preferences, locus of control, several demographic and socioeconomic variables, savings, financial difficulty, and the effects of the COVID-19 pandemic.

Our study has several limitations. We use cross-sectional data, so we are able only to document associations. Our analysis may suffer from omitted variable bias. Notably, we do not have data on the interest rates that the respondents face. Interest rates may be a confounder if they are associated both with loan forbearance takeup and personality traits. Suppose that a group of respondents faces high interest rates (for instance, on their personal loan), while another group enjoys lower rates. If interest rates correlate with personality traits (higher rates being associated with higher discount factors and lower internal locus of control) and also affect loan forbearance takeup (individuals with higher interest rates being more likely to participate in loan forbearance), then we would observe the same associations between time discounting and loan forbearance that we see in our data. More research is needed to find out if such confounders are behind our results.

Contrary to the predictions of our theoretical model (presented in Appendix A), locus of control does not significantly correlate with loan forbearance (though, as we show in Supplementary material C, it is significantly associated with savings and financial difficulty). This nil result may be due to several factors. First, the set of eligible respondents to participate in loan forbearance is restricted relative to the sample as only those with a loan could take advantage of it. Within this smaller set of respondents, locus of control does not correlate with the takeup, while in the whole sample locus of control is associated with savings and financial difficulty. Second, to have savings or to avoid financial difficulty may require more effort than participating in a highly regulated and widely promoted loan forbearance program. If locus of control exerts its effect through effort, then the difference in the required effort may explain why it has a reduced or no role in the takeup of loan forbearance. Future research will tell whether this nil result is specific to our sample, or is a general finding.

Understanding how borrowers’ personality traits are associated with their financial decisions has become a burgeoning research field. Findings in this area can help lenders provide better services and improve their risk management models and may be also helpful for policy-makers to design better policies in the future.

CRediT authorship contribution statement

Edina Berlinger: Conceptualization, Validation, Writing – review & editing. Hubert János Kiss: Conceptualization, Methodology, Validation, Writing – original draft. Sára Khayouti: Methodology, Software, Validation, Data curation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table B1

Forbearance, time discounting, and locus of control (logit regressions controlling for self-reported income).

	Dependent variable: takeup of loan forbearance (=1 if yes)
	(1)	(2)	(3)	(4)	(5)		(6)
Delta (discount factor)	−2.289***	−2.218***	−1.791**	−1.958**	−1.684*		−1.507
	(0.694)	(0.726)	(0.891)	(0.917)	(0.999)		(1.022)
Present bias	0.385	0.339	0.271	0.169	0.081		0.025
	(0.298)	(0.307)	(0.369)	(0.382)	(0.411)		(0.431)
Risk (bet in HUF)	−0.00001	−0.015	−0.006	−0.015	−0.024		−0.022
	(0.00004)	(0.037)	(0.043)	(0.044)	(0.017)		(0.048)
Internal locus of control	−0.023	−0.017	−0.023	−0.021	−0.036		−0.038
	(0.022)	(0.023)	(0.028)	(0.029)	(0.032)		(0.033)
Demographic controls		✓	✓	✓	✓		✓
Financial status			✓	✓	✓		✓
Financial difficulty				✓	✓		✓
Savings					✓		✓
Effects of COVID-19							✓
Constant	1.481*	1.228	0.805	0.993	2.244		2.343
	(0.848)	(1.146)	(1.392)	(1.425)	(1.583)		(1.877)
Observations	324	324	215	215	206		201
Log Likelihood	−206.098	−201.844	−143.591	−137.421		−124.683	−121.587
Akaike Inf. Crit.	424.195	429.687	323.182	314.843	295.365		295.174

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.

Weighted logit regressions on the takeup of loan forbearance using the stepwise regression method.

The delta discount factor represents time discounting from the (β, δ)-model, while present bias is a dummy variable (=1, if β < 1).

Locus of control is the aggregated value from seven 5-point Likert scale questions, where higher scores correspond to a stronger internal locus of control.

Higher values of risk preference represent more risk-tolerant respondents.

Demographic controls: gender, age, settlement type (the capital, other towns, or village), and education level (primary education, no high-school graduation, high-school graduation, tertiary education).

Financial status: employment (employed, unemployed, inactive), perceived financial situation (good, okay, bad), income.

Financial difficulty: whether the respondent has problems paying their loans (if they had any) or utility bills.

Savings: measured by the number of months the respondent could live off of their savings).

Effects of COVID-19: see definitions in Section 2 (Data), lower values indicate that the pandemic had a worse effect on the respondent.