BORROWING FROM THE FUTURE? 401(K) PLAN LOANS AND LOAN DEFAULTS

Abstract

Most employers permit 401(k) plan participants to borrow from their retirement plan assets. Using an administrative dataset tracking over 800 plans for five years, we show that 20 percent of workers borrow at any given time, and almost 40 percent borrow at some point over five years. Also, workers borrow more when a plan permits multiple loans. Ninety percent of loans are repaid, but 86 percent of workers changing jobs with a loan default on the outstanding balance. We estimate that $5 billion per year in defaulted plan loans generate federal revenues of $1 billion annually, more than previously thought.

I. INTRODUCTION

Defined contribution (DC) retirement plans in the U.S. generally are accorded tax-deferred status as a means to encourage saving for old age.¹ Yet the majority of such plans also give workers access to their money prior to retirement, including the availability of a borrowing feature and other pre-retirement withdrawals.² The U.S. tax code typically discourages such preretirement access by imposing a tax liability and, if under age 55 and separating from service from a DC plan, an additional 10 percent penalty tax on amounts withdrawn (including unpaid loan balances). Nevertheless, estimates of aggregate premature withdrawals from all tax-deferred accounts amount to 30–45 percent of annual total retirement plan contributions.³ Such sizeable outflows relative to inflows raise the important question of how such plan liquidity features influence retirement security and government revenues from tax-deferred plans.

A few recent papers have examined the demographic and financial aspects of 401(k) borrowers (Li and Smith, 2010; Beshears et al., 2012), but no previous study has explored how employer loan policy affects whether and how workers default on plan loans. This is an important topic since most DC participants in the U.S. have the option of borrowing from their retirement accounts,⁴ and loan defaults upon job change can erode retirement nest eggs. Accordingly, here we address several questions regarding the factors shaping borrowing from retirement accounts. First, we ask whether and how participants’ borrowing patterns respond to employer plan loan policies. Second, we investigate who defaults on plan loans, and how these patterns are related to employer loan policy. Third, we consider the effect of borrowing on employee plan contributions. Last, we assess the policy consequences of having people borrow from their retirement plans and then default on these 401(k) loans.

Drawing on a rich administrative dataset of over 800 retirement plans for which we have information on plan borrowing and loan defaults, we demonstrate that retirement account loans are quite common. Specifically, one-fifth of DC participants have an outstanding loan at any given date, and nearly 40 percent borrow over a five-year period. One reason employers allow plan loans is that they believe it encourages higher retirement plan contributions by making tax-deferred retirement accounts more liquid (Mitchell, Utkus, and Yang, 2007). Moreover, participants must repay their 401(k) loans on a set schedule by law, usually through payroll deduction. We estimate that fully 90 percent of loans are repaid in a timely way. Nevertheless, the remaining one in ten loans is not repaid, and such loan defaults can erode retirement savings.⁵

Consistent with a standard lifecycle model of saving and borrowing, we find that liquidity-constrained households are more likely to borrow from a 401(k) plan: those who are younger, with lower-income and lower wealth outside of their retirement accounts. Higher unemployment rates in a state do lead to more borrowing, but financial market volatility reduces loans. The effect of higher loan interest rates is negative but economically not significant, in part due to the fact that 401(k) loan interest is paid to the participant’s own account. We also show that employer loan policy influences 401(k) borrowing. When a plan sponsor permits multiple loans rather than only one at a time, the initial fraction of account wealth borrowed is smaller, consistent with a buffer-stock model where participants reserve the option to borrow more in case of future consumption shocks (Carroll, 1992).⁶ Yet the total fraction of account wealth borrowed is higher when a plan allows multiple loans, which is suggestive of a plan-related encouragement to borrow (i.e., an “endorsement effect”). Since it is possible that firm loan policy might reflect endogenous differences in credit demand across groups of workers, we undertake various tests to allay these concerns.

Turning to loan defaults, we find that while most plan loans are repaid, defaults are strongly associated with leaving a job and having a plan loan outstanding. Thus 86 percent of employees who leave their employers with a plan loan outstanding fail to repay their loan balances, reducing retirement savings and exposing them to income tax and possibly an additional penalty tax. It is this group in particular where loan defaults appear to undermine long-term retirement savings. We also evaluate whether the economic turmoil of 2008–09 dramatically changed 401(k) plan borrowing and default patterns, and we find little evidence of any substantial behavioral change. In particular, borrowing rates fell slightly during the global financial crisis, while defaults rose by an economically negligible amount.

Although most plan loans are repaid, it is possible that borrowing could affect retirement adequacy in other ways, for example, by leading employees to reduce plan contribution rates. Accordingly we analyze plan contributions six months prior to and after employees took a plan loan, and we show that employee contributions drop by very little, by only 6 percent of dollars contributed ($20 per month at the mean). Yet this saving response is quite heterogeneous, with about one-third of borrowers reducing contributions meaningfully, while two-thirds keep contributions steady or even raise them after borrowing.⁷

Finally, we use our results to offer a new assessment of how 401(k) loan defaults influence government tax revenue. Our estimate of loan defaults amounts to around $5 billion per year, a value far larger than prior authors’ estimates (which relied on incomplete data).⁸ Nevertheless, this is still an order of magnitude lower than annual employee cash-outs of retirement plan assets, which the GAO (2009) reported amounted to $74 billion (in 2006). The small relative size of loan defaults is relevant to the question of whether plan borrowing should be further restricted (Leonard 2011).

In what follows, Section II provides an overview of 401(k) loan rules, and Section III reviews related studies. Section IV describes the data and develops our hypotheses. In Section V we present empirical results on borrowing, and in Section VI we provide results on loan defaults. Section VII reports our findings on savings and borrowing. Section VIII provides our estimate of the aggregate tax revenue impact of loan defaults, and Section IX concludes.

II. 401(K) LOAN RULES

A 401(k) loan is not a credit instrument in the conventional sense, but rather an arrangement allowing plan participants to gain access to their retirement account accumulations under certain conditions. As outlined in U.S. Treasury regulations (GAO, 2009), a participant may only borrow up to half of his account balance to a maximum loan of $50,000 (in nominal terms). The participant must also agree at the time of the loan to replenish the withdrawn funds plus interest in accordance with a standard flat-dollar amortizing loan schedule, typically through payroll deduction. Employers can also impose their own requirements on plan loans, including whether 401(k) loans are permissible at all; as a practical matter, 90 percent of active contributors have access to plan loans in the U.S. Sponsors may also determine the number of individual loans allowed, whether loans must be for some minimum amount (e.g., $1,000), and what the participant must pay in terms of an interest rate. In general, plan and regulatory rules interact as follows: if a 401(k) plan offers a loan feature with a minimum required loan amount of L_min, the participant with an account balance W_401k seeking to borrow loan amount L must satisfy two conditions:

$$L\le \frac{1}{2}{W}_{401k}\mathrm{ and }{L}_{\min }\le L\le \$50,000.$$

For a typical L_min = $1,000, a participant will not be eligible to borrow until such time as his account reaches or surpasses $2,000; at that point, he may borrow up to half of his account balance. The 50 percent limit binds until the participant’s account balance exceeds $100,000; above that, the maximum amount withdrawn cannot exceed $50,000. If a plan only allows one loan at a time, the borrower must fully repay his current loan before he can borrow again. Yet some plan sponsors allow participants to take out multiple loans (two, three, or even more) in increments L₁, L₂, and so on, with L = ∑L_i. In such cases, borrowers can hold as many plan loans as permitted at any given time, as long as the total amount outstanding does not exceed the cap described above.

Tax rules require a series of loan repayments PMT according to this schedule: $\sum _{t=1}^t=n\frac{PMT}{{(1+i)}^t}$, where the loan interest rate is i and n refers to the number of periods over which the loan must be repaid.⁹ The loan repayment is taken from the participant’s after-tax salary. One portion of the payment stream represents principal repayment, while another represents interest. Loan interest payments are unlike the traditional cost of credit, in that the participant is effectively repaying himself: the loan re-payments are deposited to the participant’s account pre-tax. Hence, a higher interest rate will lead to larger account accumulations, ceteris paribus. In exchange for agreeing to these repayment terms, the participant can spend his pre-tax L on consumption with no immediate income tax consequences. In other words, when the plan loan is exercised, the participant avoids paying current taxes as well as an early withdrawal penalty on the amount withdrawn from his pre-tax retirement account. In most circumstances, the opportunity cost of a 401(k) loan will be meaningful less than the cost of paying market rates on credit cards or secured lines of credit like a car loan (Li and Smith 2010)¹⁰, both because of a lower quote rate and the fact that in a 401(k) loan the interest rate is paid to the borrower’s account.

When the borrower leaves his job, any remaining balance due on his 401(k) loan, L_bal, typically converts to a balloon payment. If he leaves his job and does not repay L_bal within 60–90 days, the loan is considered to be in default. It must then be reported to the IRS as a taxable distribution from the plan at that time, producing a tax liability of L_bal(τ+.1). Specifically, the 10 percent penalty is imposed on the amount borrowed if it is not returned to the account and if the participant is under age 59½. Workers age 60+ are not subject to the additional penalty. It is worth noting that τ≈0 for many low- and middle-income households today due to a variety of tax credits, so the expected cost of default may be simply 0.1*L_bal for those younger than age 59½.¹¹

Because 401(k) loans are not conventional borrowing arrangements but rather represent withdrawals from one’s own savings, these are not subject to credit underwriting and not reflected on credit reports. Accordingly, a 401(k) loan can be a convenient way of accessing credit, particularly for the credit-constrained. Moreover, loan defaults have no credit reporting consequences, so defaulting on a 401(k) loan has no effect on a borrower’s ability to take a new loan from a new plan. For this reason, taking plan loans—even with the possibility of defaulting—may well be optimal when workers are liquidity-constrained and have few other options for low-cost credit.

III. PRIOR STUDIES

Saving and borrowing from a 401(k) plan is usefully examined against the broader literature on the impact of tax-advantaged retirement saving on total saving.¹² Focusing just on 401(k) plans, several prior studies have examined aspects of borrowing behavior. The GAO (1997) concluded that allowing plan loans raises participation rates. Several others have reported that making loans available also boosts saving on the intensive margin, in the form of higher DC plan contribution rates (Munnell, Sunden, and Taylor, 2001/02; Holden and Vanderhei, 2001; Mitchell, Utkus, and Yang, 2007).

In terms of the characteristics of borrowers, Sunden and Surette (2000) and Li and Smith (2010) used Survey of Consumer Finances data and found that people who borrowed from their 401(k) accounts had higher DC account balances, but lower total financial assets, higher debt, and were more credit-constrained. Turning to what people use the loans for, Utkus and Young (2011) reported that about 40 percent of borrowers surveyed used plan loans for bill or debt consolidation, while some 30 percent used them for home improvement and repair. They also found that the least financially literate borrowers used 401(k) loans for consumption rather than investment purposes. Weller and Wenger (2014) identified a link between 401(k) borrowing and health status, with households in poorer health being more likely to borrow. Beshears et al. (2012) used participant-level information to show that plan borrowing followed a hump-shape age profile. Their analysis did not address loan defaults, the role of employer loan policy, or the interplay between savings and loan-taking.¹³

In a different context, Gross and Souleles (2002a, 2002b) examined people who borrowed using credit cards, and they found that individuals exhibited “buffer stock” behavior. Specifically, people tended to not borrow up to the maximum they were allowed to take, so as to leave a margin in case of emergency. That study reported credit card interest rates averaging 16 percent, which is far higher than the average 401(k) loan interest rate of just over seven percent (in our dataset, about which we say more below). Such a large difference in borrowing rates suggests that employees with access to plan loans could benefit from substituting lower-cost 401(k) loans for much higher-cost credit card debt.¹⁴ Nevertheless, Li and Smith (2010) reported that many people had substantial credit card debt, even when a plan loan would have been less costly. They suggested that this seemingly illogical behavior could have been influenced by financial advisers’ negative views of 401(k) loans,¹⁵ along with a mental accounting perspective where 401(k) accounts could be thought of as ‘restricted’ for retirement purposes rather than used for current consumption.¹⁶

IV. METHODOLOGY

A. Preliminary Evidence

Our empirical analysis uses a rich administrative dataset for DC plan participants covering the five-year period from July 2004 to June 2009. The dataset includes 882 defined contribution (mostly 401(k)) plans. To assess peoples’ propensity to borrow from their plans, we use a time-varying sample of over 900,000 participants observed monthly, with over 55 million observations.¹⁷ We observe an average of over 13,000 new plan borrowers each month (for a total of nearly 800,000 borrower observations). Variables available include plan characteristics and participant demographic/financial characteristics. We also observe information on loan default behavior for workers terminating employment.

In any given month, an average of 1.38 percent of eligible participants took out a new plan loan (see Table 1). The average amount borrowed per loan was just over $8,300 (in $2015); the mean total amount borrowed from the plan was around $12,400. Loan interest rates varied by plan, though most pegged the interest rate to the Prime Rate plus one percent. Loan interest rates were only modestly higher for borrowers than for the entire participant sample. The average age of borrowers was 42, slightly younger than the average participant age of 45. We also see that borrowers had about eight years of tenure and somewhat lower income, lower non-retirement financial wealth,¹⁸ and half the plan account balance, as compared to all loan-eligible participants. Borrowers were also more likely to be in plans where multiple loans were allowed. During the period of the global financial crisis, defined here as September 2008-June 2009, fewer participants borrowed from their retirement accounts.

Table 1.

Characteristics of 401(k) Loan-Eligible Participants and Plan Borrowers

	Loan-Eligible Participants		Plan Borrowers
	Mean	Std dev.	Mean	Std dev.
Outcomes of interest
% participants taking a new loan	1.38	11.65	NA	NA
Loan amount ($)	NA	NA	9,614	10,668
Total amount borrowed ($)	NA	NA	12,378	12,725
Plan design factors
Multiple loans allowed (%)	0.48	0.50	0.58	0.49
Loan interest rate (%)	7.16	1.78	7.34	1.88
Participant characteristics
Age	44.7	10.7	41.6	9.7
Age < 25 (%)	0.20	0.40	0.26	0.44
Age 35–45 (%)	0.28	0.45	0.34	0.47
Age 45–55 (%)	0.32	0.47	0.30	0.46
Age 55–60 (%)	0.12	0.32	0.08	0.26
Age > 60 (%)	0.08	0.27	0.03	0.16
Male (%)	0.52	0.50	0.44	0.50
Tenure (years)	8.5	7.4	8.2	6.9
Tenure**2	127.4	211.3	114.8	187.0
Income < $35,000 (%)	0.08	0.27	0.14	0.34
Income $35,000-$87,500 (%)	0.49	0.50	0.61	0.49
Income > $87,500 (%)	0.43	0.50	0.26	0.44
Low wealth < $25,000 (%)	0.43	0.50	0.65	0.48
Medium wealth $25,000 -$100,000 (%)	0.31	0.46	0.25	0.43
High wealth > $100,000 (%)	0.26	0.44	0.10	0.30
401(k) account balance ($)	95,157	169,930	53,793	94,116
Ln 401(k) account balance	10.47	1.50	9.98	1.31
Macroeconomic variables
Financial turmoil period (%)	0.19	0.39	0.16	0.37
Lagged state unemployment rate (%)	5.3	1.3	5.3	1.3
No. observations	55,175,718		797,720

Source: Authors’ tabulations.

Figure 1 illustrates the monthly and cumulative percentage of loan-eligible participants having one or more outstanding loans in our sample. One striking point is that about 20 percent of active participants had a loan outstanding in any given month, so loan originations were approximately offset by loan repayments or defaults. Over the entire five-year period of our study, the cumulative proportion of participants borrowing from their retirement plan amounted to nearly 40 percent. In other words, instead of the same participants taking repeated plan loans, many different participants borrowed from their retirement accounts over a longer time horizon.

Figure 1.

Percentage of Participants with 401(k) Loans Outstanding

Source: Authors’ calculations.

We are also interested in the impact of employer plan design on participant borrowing. Figure 2 presents the mean proportion of new plan borrowers over the five-year period, where we compare plans which offered only a single loan at a time, with those permitting multiple loans. When only a single loan was allowed, an average of 1.10 percent of eligible participants took a new loan per month. With multiple loans, the average rose to 1.69 percent per month.

Figure 2.

Loan Incidence by Number of Plan Loans

Source: Authors’ calculations.

B. Hypotheses

We seek to examine how plan loan policies influence plan borrowing and default patterns. As noted above, the buffer-stock model suggests that cautious borrowers with the option of multiple loans will remain below the maximum borrowing limit to allow for future consumption shocks. In a 401(k) setting, participants will be restricted by employer policy regarding both the number of loans allowed at one time, and the total dollar amounts that can be borrowed. In our dataset, 40 percent of plans covering 52 percent of participants permitted workers to take out two or more loans at once. In such cases, buffer-stock participants would be predicted to be more likely to borrow from their plan but take smaller loans, compared to plans permitting only a single loan. We also hypothesize that the availability of multiple loans could be seen as an employer “endorsement” (Benartzi, 2001) of plan borrowing. If this were true, we would anticipate that aggregate borrowing would be higher when multiple loans are permitted.

Permitting multiple loans may or may not affect default behavior at job termination. On the one hand, defaulting depends only on whether the borrower can pay off his outstanding balance, regardless of how many loans he has taken. In that case, there may be no relationship between defaults and multiple loans. On the other hand, taking multiple loans could indicate lack of self-control or inability to manage one’s finances. If so, those who took several loans might be more likely to default. These are two distinct sources of variation in default outcomes, so results pertinent to participant heterogeneity provide no direct insights into the effect of plan policies.

Employers also have control over another plan feature that may affect borrowing behavior, namely the interest rate charged to plan borrowers. A lower rate may make borrowing more attractive because it increases the perceived spread between a 401(k) loan and other forms of borrowing, and it reduces the impact on take-home pay. Yet because workers are repaying themselves through a 401(k) loan, they may prefer a higher rate.

Setting the interest rate higher reduces the worker’s spendable salary more, but it also repays the worker’s retirement account faster over time. The two are economically off-setting effects (a higher interest rate leads to a reduction in salary and a correspondingly higher repayment to one’s account), although whether one is more salient to the participant versus the other is an open question.

Another issue we explore is whether plan borrowing and loan default rates changed materially during the financial crisis of 2008–09. With respect to borrowing, the predicted impact is ambiguous: on the one hand, employees may have become more cautious and borrowed less, but on the other hand, they might have sought additional loans due to financial insecurity or household financial shocks. Regarding defaults, there are again two potentially competing effects: voluntary job changes would be expected to decline during a recession, reducing the incidence of default. Yet involuntary job losses rise, raising the risk. Again, empirical analysis is required to discern the net effect.

In addition to our focus on salient employer plan design features, we also control on demographic and financial factors that could affect participant borrowing and default behavior. Naturally age is important, as borrowing would be expected to be higher among the credit-constrained young and then decline with age. Yet in 401(k) accounts, borrowing is conditioned on the employee’s account balance which rises with both age and salary. Therefore we would anticipate a hump-shared age profile for borrowing (as in Beshears et al., 2012) since the ability to borrow rises with age and salary and job tenure, but the demand for plan borrowing falls with age. Li and Smith (2010) also noted that liquidity-constrained households are more likely to rely on 401(k) borrowing. Using our much more extensive dataset, we examine the robustness of this finding. Moreover, we hypothesize that liquidity constraints are also likely to drive loan defaults on job termination, since the outstanding balance accelerates as a balloon payment.

V. MULTIVARIATE ANALYSIS OF PLAN BORROWING AND LOAN SIZE

To investigate borrowing patterns from 401(k) accounts we estimate multivariate regression models of the following form:

$$BORRO{W}_{i,j,t}=\delta +\alpha ’POLIC{Y}_{j,t}+\beta ’PARTICIPAN{T}_{i,j,t}+\gamma ’MACR{O}_{i,t}+\mu ’PLA{N}_j+\pi ’PLA{N}_{-}MEAN{S}_j+{\epsilon }_{i,j,t}$$

where BORROWi,j,t refers to a vector of several outcomes including the probability of borrowing from one’s plan, the size of new plan loans, and the total amount borrowed by the ith participant in the jth plan in month t. We examine all loan-eligible participants, defined as those having assets at least twice the minimum loan amount set by the plan and not otherwise subject to any other IRS or plan limit (whether in terms of dollars or number of loans allowed). The POLICY _j.t vector includes a flag for whether the plan permitted multiple loans, the loan interest rate, and plan size (number of participants) in each period. The PARTICIPANT_i,j.t vector refers to participant characteristics including age, sex, job tenure, income, account balance, and nonretirement household wealth.¹⁹ The MACRO_j.t vector controls for the lagged three-month average state-specific unemployment rate,²⁰ and a flag indicates the financial crisis period (September 2008-June 2009). Finally, we control in all models for firm-level effects (the PLAN_,j vector, which includes industry and plan size).

One challenge we face is the possibility that a firm’s loan policy might be endogenenous. A conventional empirical approach to handle this issue would include plan-level dummies to control for unobserved firm-level fixed characteristics. Nevertheless in our setting, a key variable of interest, loan policy, is itself a plan-level characteristic and so correlated with plan-level dummies. To address the problem of possible omitted fixed effects, and following Chamberlain 1985, we also incorporate mean participant and firm-level characteristics as explanatory variables (PLAN_MEANS). These controls include average age, proportion male, tenure, and balances, as well as plan-level income and non-retirement wealth indicators. As noted in Huberman, Iyengar, and Jiang (2007), an element of plan-level policy that is orthogonal to these mean participant and plan characteristics can be viewed as itself exogenous, and it is this element that our specification captures.

A. Determinants of Plan Borrowing

Our first empirical regression permits us to evaluate the determinants of the probability of a participant taking a new loan in month t. Coefficient estimates and marginal values from our multivariate Probit model appear in Table 2, where the mean value of the dependent variable is 1.38 percent per month.

Table 2.

Factors Associated with the Probability of a Participant Borrowing from Plan

		Probability of taking a new loan (Probit)
	Mean	Estimate		(SE)	Marginal effect (%)
Plan design factors
Multiple loans allowed	0.48	0.137	***	0.001	0.32
Loan interest rate (%)	7.16	−0.008	***	0.000	−0.02
Participant characteristics
Age 35–45	0.28	0.038	***	0.001	0.09
Age 45–55	0.32	−0.041	***	0.001	−0.09
Age 55–60	0.12	−0.160	***	0.002	−0.32
Age > 60	0.08	−0.412	***	0.003	−0.64
Male	0.52	−0.041	***	0.001	−0.09
Tenure	8.51	0.045	***	0.000	0.11
Tenure**2	127.42	−0.001	***	0.000	0.00
Income < $35,000	0.08	0.103	***	0.002	0.26
Income > $87,500	0.43	−0.099	***	0.001	−0.23
Low wealth	0.43	0.171	***	0.001	0.41
High wealth	0.26	−0.143	***	0.002	−0.30
Ln(401(k) account balance)	10.47	−0.123	***	0.000	−0.07
Macroeconomic variables
Financial turmoil period (%)	0.19	−0.014	***	0.002	−0.03
Lagged state unemployment rate (%)	5.30	0.013	***	0.000	0.03
Industry controls		Yes
Plan mean variables		Yes
Intercept		−2.08	***	0.019
No. observations		55,175,718
Mean of dependent variable		0.0138
Pseudo R-squared		0.0141

Note:

^**Significant at 0.05 level,

^***significant at 0.01 level.

Plan means included to control for unobserved fixed effects.

See text.

Turning first to the impact of employer loan policy, the data show a sizeable effect on borrowing when a plan offers multiple loans. The availability of multiple loans in a plan raises the monthly borrowing rate by 0.32 percentage points relative to a mean of 1.38 percentage points—a relatively large change of 23 percent. Meanwhile, plans charging higher loan interest rates did not have materially different borrowing patterns: a 1 percent increase in interest rates reduced the mean 1.38 percent loan-taking rate by 0.02 percentage points. Two competing effects are at work here: a higher interest rate results in a larger reduction in a participant’s take-home pay, while it also leads to higher account accumulations over time. On net, 401(k) loan demand proved quite insensitive to the price of plan loans.

The next portion of Table 2 indicates that loan demand was highest among those age 35–45, implying that plan borrowing is a function of both lifecycle demands for credit and the ability to accumulate a sufficient balance from which to borrow. Demand for 401(k) borrowing is second highest among those under age 35 (the reference category), and it decreases among the older age groups. Workers age 60+ do pay income tax but no penalty on plan loans when they default; in our dataset, the lower penalty did not appear to encourage borrowing among the older age group. Loan demand was higher among those with low incomes and lower non-retirement wealth, factors associated with being liquidity-constrained. Job tenure was also quite positively associated with taking a loan, probably because long-time employees have time to learn more about the feasibility of plan loans.

One concern voiced by policymakers is that plan borrowing might have ramped up during the financial crisis, but our evidence suggests otherwise.²¹ Specifically, during that period, participants were 0.03 percentage points less likely to take a new loan per month. Local unemployment did have a statistically significant impact, in that a one percentage point increase in the unemployment rate was associated with a 0.03 percentage point increase in monthly plan borrowing. Nevertheless, the magnitudes of both factors are quantitatively small, perhaps because of heterogeneity in the responses. That is, some households may have boosted borrowing in response to a negative income shock, while others curtailed borrowing in the face of greater economic uncertainty.

B. Determinants of Amounts Borrowed

Next we turn to the intensive margin of borrowing, examining both the size of new loans and aggregate amounts borrowed from the plans. Table 3 provides descriptive statistics. The median new loan came to $5,600 (in $2015), while the total amounts borrowed (including those having multiple loans) ranged from $1,555 at the 10^th percentile to $32,124 at the 90^th (again in $2015). Also shown are loan values as a fraction of 401(k) account wealth, with the median total loan amounting to nearly 40 percent of worker plan balances.

Table 3.

Characteristics of Amounts Borrowed

Percentile	New loan amount ($)	Total amount borrowed ($)	New loan as % of account balance	Total loan as % of account balance
10th	1,257	1,555	0.06	0.09
20th	1,955	2,520	0.12	0.16
30th	2,873	3,760	0.18	0.24
40th	4,027	5,466	0.23	0.31
50th	5,627	7,391	0.29	0.39
60th	7,506	10,406	0.35	0.46
70th	10,877	14,227	0.42	0.50
80th	15,418	20,809	0.48	0.54
90th	23,912	32,124	0.51	0.65
Mean	9,614	12,378	0.42	0.53
Std. Dev.	10,668	12,725	5.68	6.03

Source: Authors’ tabulations.

Table 4 reports linear regression results with the same controls as before for new loan amounts and total borrowing, expressed in natural logs. The next new loan represents the marginal amount of any new borrowing (a flow measure); the total amount borrowed is a sum of the new loan taken plus any prior amounts borrowed (a sum of the new flow plus the existing stock of loans outstanding). All variables are measured in the month when the new loan is taken, and our sample includes only borrowers. Because we control on participant 401(k) account balances, these results can be interpreted as the relative proportion borrowed from participant accounts. The results reflect the average monthly effect over our sample period.

Table 4.

Factors Associated with the Size of Plan Loans and Total Amount Borrowed

		Ln ($ new loan amt.) (OLS)			Ln ($ total amt. borrowed) (OLS)
	Mean	Estimate		(SE)	Estimate		(SE)
Plan design factors
Multiple loans allowed (dummy)	0.58	−0.172	***	0.002	0.202	***	0.002
Loan interest rate (%)	7.34	−0.030	***	0.001	−0.021	***	0.001
Participant characteristics
Age 35–45	0.34	0.070	***	0.002	0.086	***	0.002
Age 45–55	0.30	0.047	***	0.002	0.061	***	0.002
Age 55–60	0.08	0.010	**	0.004	0.015	***	0.004
Age > 60	0.03	−0.085	***	0.006	−0.129	***	0.006
Male	0.44	0.047	***	0.002	0.048	***	0.002
Tenure	8.23	0.019	***	0.000	0.039	***	0.000
Tenure**2	114.84	−0.001	***	0.000	−0.001	***	0.000
Income < $35,000	0.14	−0.038	***	0.002	−0.038	***	0.002
Income > $87,500	0.26	0.036	***	0.002	0.029	***	0.002
Low wealth	0.65	−0.051	***	0.002	−0.044	***	0.002
High wealth	0.10	0.044	***	0.003	0.038	***	0.003
Ln(401(k) account balance)	9.98	0.596	***	0.001	0.576	***	0.001
Macroeconomic variables
Financial turmoil period (%)	0.16	−0.078	***	0.003	−0.053	***	0.003
Lagged state unemployment rate (%)	5.27	−0.007	***	0.001	−0.004	***	0.001
Industry controls		Yes			Yes
Plan mean variables		Yes			Yes
Intercept		5.077	***	0.033	4.584	***	0.033
No. of observations		797,720			797,720
Mean of dependent variable		8.502			8.764
R-squared		0.551			0.577

Note: 2015$.

^**Significant at 0.05 level,

^***significant at 0.01 level.

Plan means included to control for fixed effects.

Source: Authors’ calculations.

In terms of loan policy, results show that the new loan amounts were smaller in plans allowing multiple loans. This is evidence of buffer-stock behavior in the multiple-loan plans. Yet the total amount borrowed was also higher in multiple loan plans, illustrating that workers see the multiple-loan policy as an endorsement encouraging 401(k) borrowing at the margin. At a given point in time in our sample, having a multiple-loan policy leads to participants borrowing a higher fraction of retirement wealth than otherwise. Plan interest rates charged are negatively and significantly associated with borrowing amounts, but the estimates are quantitatively small.

Participant characteristics also matter. Workers age 35–45 borrowed the largest fraction of their wealth, both for new loans and in total, while employees age 45–55 were the second highest group. Interestingly, participants age 55–60, who are exempt from any 10 percent penalty tax if they separate from service, were slightly more likely to borrow than the reference group, participants under age 25. However, workers 60+ borrowed at the lowest rates of all age groups. This may be because such employees are often able to make penalty-free withdrawals of 401(k) savings while working, unrelated to the loan feature. We also see that higher-paid and wealthier participants borrowed larger fractions of their plan wealth, compared to middle/ lower-income and wealth groups. Hence, while our earlier results showed that lower income and lower wealth households were more likely to borrow at the extensive margin, the higher income/wealth participants borrowed proportionally more, conditional on borrowing.

C. Endogeneity Tests

Thus far, we have assumed that employers design their plan loan policies independently of employee demand. But a possible alternative explanation for the observed effect of plan loan features on participant borrowing could be that plan sponsors structure their firms’ loan policies in response to employee liquidity needs. That is, some firms might attract employees with an inherently higher demand for credit (due to lifecycle reasons or behavioral biases), in which case they might offer multiple loan features to attract such individuals. If so, the positive correlation between participant borrowing and the number of plan loans allowed would reflect plan sponsor anticipation of worker borrowing needs, rather than worker reactions to employer plan design preferences.

While we cannot completely rule out this possibility, we have undertaken two tests for endogeneity, on which we now report. For instance, if plan sponsors did set loan policy in anticipation of participant needs, we might expect that plans which allowed multiple loans would differ systematically from single-loan plans, along observable characteristics. By contrast, if differences in borrowing behavior were due to participants reacting independently to plan loan policies, there should be no systematic differences across plans. To test this hypothesis, we run the following regression:

$$MULT{I}_{-}LOA{N}_{,j}=\delta +\alpha ’PLA{N}_{-}CHA{R}_j+{\epsilon }_{,j}.$$

The dependent variable takes the value of one if the plan allowed its participants to hold multiple loans, and 0 if it allowed only a single loan. The PLAN_CHAR_j vector consists of characteristics including the mean and standard deviation of participant age, tenure, household income, financial wealth, the plan loan interest rate, the number of participants in the plan, and controls for the firm’s industrial sector. Our hypothesis is that there should be no statistically significant difference in observable characteristics between plans allowing multiple loans and plans allowing only one loan. What we find is that only factor, the standard deviation of tenure, is occasionally significant (full results appear in Online Appendix Table 3), while no other factors are statistically significant. Accordingly, we conclude that sponsors do not structure plan loan policies to meet participant characteristics associated with borrowing needs.²²

As another robustness test, we hypothesize that if plans allowed multiple loans due to participant demand, the proportion of participants holding multiple loans in those plans should be relatively large. Instead, only 13.8 percent took additional loans when allowed to do so; in other words, 86.2 percent of eligible borrowers took only a single loan, even when their plans allowed multiple loans. We do not observe the counterfactual statistic for participants in single-loan plans; that is, we cannot measure the relative demand for additional loans among participants in single-loan plans. Yet we can examine the demand for subsequent loans by sequential borrowers in single-loan plans. Sequential borrowers are those who, in single-loan plans, take one loan, repay it, and then take a new loan immediately. We find that only 13.9 percent of participants were sequential borrowers in single-loan plans, a rate virtually identical to the 13.8 percent of multiple-loan takers in plans allowing more than one loan. This suggests that participant loan demand for multiple loans at the extensive margin is not substantially different across plans, irrespective of whether they permit only a single or multiple loans. It is on the intensive margin of multiple-loan loan demand, as reflected in the fraction of account wealth borrowed, that the two types of plans differ.

VI. DETERMINANTS OF DEFAULTS ON PLAN LOANS

Next we explore the determinants of 401(k) loan defaults, beginning with the summary statistics appearing in Table 5. In a single year, about one-fifth of loan-eligible employees in our sample had one or more loans outstanding. But among participants terminating employment with outstanding loans, 86 percent failed to repay their outstanding loans (the remainder paid their account loans and hence avoided default). Since participants defaulting on their plan loans totaled around 10 percent of all participants with outstanding loans, we estimate that about 90 percent of participants repaid their loans over the five-year period observed.²³

Table 5.

Default Patterns from 401(k) Loans: Full Period and by Year

	Full Period	7/04–6/05	7/05–6/06	7/06–6/07	7/07–6/08	7/08–6/09
N active participants	6,590,397	1,102,216	1,205,973	1,331,734	1,490,100	1,460,374
% of active participants w/ loan outstanding	20.6	21.5	21.0	20.7	19.6	20.3
Terminations
% of those w/ loan terminating	11.2	10.3	11.0	10.6	10.6	13.1
Default rates
% of those terminating w/ loan who default	86.0	84.8	84.2	85.4	87.3	87.6
Defaults as a % of loans outstanding	9.6	8.7	9.3	9.1	9.3	11.5

Source: Authors’ tabulations.

Statistics comparing participants who defaulted versus repaid their loans appear in Table 6, along with data on all borrowers and all loan-eligible plan participants. The sample of defaulters consisted of 130,295 participants in 401(k) plans who terminated employment with at least one loan outstanding.²⁴ Compared to other plan borrowers, they were somewhat younger, had shorter job tenure, and held lower balances. Those who defaulted on their loans also had lower income, lower balances, and had less non-retirement wealth, compared to those who repaid their loans on leaving their jobs.

Table 6.

Characteristics of Participants Defaulting on, or Repaying, their Loans upon Job Termination

	All Participants	Participants w/ Loan	Participants Terminating with Outstanding Loans
			All	Defaulting	Repaying
Plan design factors
Multiple loans allowed (%)	43.3	49.7	47.6	47.1	50.6
# Loans taken	N/A	1.3	1.3	1.3	1.2
Loan interest rate (%)	6.89	6.88	7.08	7.08	7.10
Participant characteristics
Mean age	43.6	44.1	42.9	42.4	46.1
Male (%)	49.0	50.6	48.5	47.7	53.5
Mean tenure	8.0	10.5	8.6	8.1	11.3
Mean income ($)	84,371	73,252	67,369	64,958	81,503
Low wealth (%)	50	64	65	67	52
Medium wealth (%)	29	25	24	24	29
High wealth (%)	21	11	11	10	19
Mean account balance ($)	80,573	70,158	48,034	38,957	104,188
Loan balance ($)	N/A	8,946	6,949	6,709	8,429
No. observations	6,590,397	1,354,900	151,458	130,295	21,163

Source: Authors’ tabulations.

To compare employees who terminated employment and defaulted on their 401(k) loans with those leaving employment who repaid their loans in full more rigorously, we next estimate a multivariate Probit model where the dependent variable, D_i,j,t, refers to the probability of individuals terminating their jobs and defaulting. As indicated in Table 7, the mean of the dependent variable is 86 percent.²⁵ Regressors are identical to those in our previous examinations of loan probabilities and amounts borrowed. In addition, we also control for the borrowers’ remaining outstanding loan balances.

Table 7.

Determinants of the Probability of Participants Defaulting on 401(k) Plan Loans Outstanding upon Job Termination

		Probability of defaulting on 401(k) loans outstanding (Probit)
	Mean	Estimate		(SE)	Marginal effect (%)
Plan design factors
Multiple loans allowed	0.48	−0.063	***	0.010	−1.15
Loan interest rate (%)	7.08	−0.004		0.003	−0.07
Participant characteristics
Age 35–45	0.31	−0.029		0.013	−0.53
Age 45–55	0.26	−0.040	***	0.014	−0.73
Age 55–60	0.10	0.021		0.019	0.37
Age > 60	0.07	0.070	***	0.022	1.23
Male	0.48	0.010		0.011	0.19
Tenure	8.47	−0.003		0.002	−0.05
Tenure**2	118.21	0.000		0.000	0.00
Income < $35,000	0.15	0.044	***	0.015	0.78
Income > $87,500	0.27	−0.097	***	0.011	−1.82
Low wealth	0.65	0.115	***	0.011	2.15
High wealth	0.11	−0.140	***	0.015	−2.73
Ln(401(k) account balance)	9.88	−0.400	***	0.007	−3.34
Ln Loan balance	8.11	0.282	***	0.004	82.37
Macroeconomic variables
Financial turmoil period (%)	0.22	0.033	**	0.017	0.60
Lagged state unemployment rate (%)	5.36	−0.003		0.004	−0.06
Industry controls		Yes
Plan mean variables		Yes
Intercept		2.324	***	0.134
No. observations		151,458
Mean of dependent variable		0.86
Pseudo R-squared		0.121

Note:

^**Significant at 0.05 level,

^***significant at 0.01 level.

Plan means included to control for fixed effects.

Offering multiple loans is associated with a default rate that is one percentage point below the average (of 86 percent). This is an economically negligible effect. There are also many significant demographic factors, but most of are of negligible economic significance. For example, workers age 45–55 were less likely to default by 0.73 percentage points relative to the mean (or less than one percent). Perhaps due to the elimination of additional penalty after age 60+, older workers were more likely to default, but the effect is small, only 1.23 percentage points. The most substantial factor associated with defaulting is the size of workers’ loans.

Though permitting multiple loans has only a small effect on default rates overall, it is possible that default behavior could differ between people who took only a single loan versus those who took multiple loans. To better understand the role that plan design plays in influencing default behavior on employment termination, we next categorize borrowers into three groups: (1) those allowed only a single loan; (2) those permitted to take multiple loans but who took only one single loan; and (3) those with multiple loans. Table 8 evaluates the extended model controlling for aggregate loan balances. We hypothesize that, if default rates varied across groups solely due to loan balances, these variables should be statistically insignificant; alternatively, if we found a significant effect of these regressors, it would suggest a relationship between the number of loans allowed/taken and default behavior. Again we note that results driven by participant heterogeneity are not informative the effects of plan policies, but they are of interest in their own right.

Table 8.

Factors Associated with the Probability of Loan Defaults upon Job Termination: Extended Model

		Probability of defaulting on 401(k) loans outstanding (Probit)
	Mean	Estimate		(SE)	Marginal effect (%)
Plan design factors
# loans allowed > 1, #loans taken > 1 (dummy)	0.25	0.145	***	0.013	2.51
# loans allowed > 1, #loans taken = 1 (dummy)	0.24	−0.178	***	0.011	−3.41
Loan interest rate (%)	7.08	−0.004		0.003	−0.07
Participant characteristics
Age 35–45	0.31	−0.029		0.013	−0.53
Age 45–55	0.26	−0.041	***	0.014	−0.75
Age 55–60	0.10	0.021		0.019	0.38
Age > 60	0.07	0.082	***	0.022	1.42
Male	0.48	0.015		0.011	0.27
Tenure	8.47	−0.006	***	0.002	−0.11
Tenure**2	118.21	0.000	**	0.000	0.00
Income < $35,000	0.15	0.041	***	0.015	0.72
Income > $87,500	0.27	−0.091	***	0.011	−1.69
Low wealth	0.65	0.109	***	0.011	2.01
High wealth	0.11	−0.134	***	0.015	−2.58
Ln(401(k) account balance)	9.88	−0.385	***	0.007	−3.18
Ln Loan balance	8.11	0.261	***	0.004	77.57
Macroeconomic variables
Financial turmoil period (%)	0.22	0.037	**	0.017	0.66
Lagged state unemployment rate (%)	5.36	−0.003		0.004	−0.06
Industry controls		Yes
Plan mean variables		Yes
Intercept		2.411	***	0.135
No. of observations		151,458
Mean of dependent variable		0.860
Pseudo R-squared		0.124

Note:

^**Significant at 0.05 level,

^***significant at 0.01 level.

Plan means included to control for fixed effects.

Our findings show that employees permitted to take multiple loans but who held just one loan were less likely to default by a statistically significant 3.41 percentage points, whereas participants taking multiple loans were 2.51 percentage points (or 23 percent) more likely to default. In other words, given two participants with the same 401(k) total debt, the employee who took one loan was less likely to default, compared to his counterpart holding multiple loans. The amount borrowed remains the most economically significant variable.

In sum, defaults are widespread among those leaving jobs with a loan, but few control variables have economically meaningful effects on the mean default rate of 86 percent, other than the total loan balance. Accordingly, other unobserved factors could be driving pension loan defaults, such as financial illiteracy, high employee discount rates, or lack of self-control.²⁶ This could imply that many people borrowing from their retirement plans were simply unaware of the consequences of job termination for their 401(k) loans, so for them, loan defaulting was accidental and unexpected.

VII. 401(K) SAVING AND BORROWING

In our sample, 90 percent of loans were repaid, so taking a loan depleted retirement savings only for the 10 percent of participants changing jobs and failing to repay their outstanding loan balances. Yet plan borrowing could still influence retirement security, if it prompted participants to reduce their ongoing plan contributions. To assess this possibility, we turn next to an evaluation of how loan-taking influenced plan contribution rates.

We begin with descriptive statistics comparing borrowers’ contribution patterns six months prior to taking a new loan and six months afterward. In Panel A of Table 9, we see that the mean contribution amount (for all loans) was $362 per month before taking a loan, and $340 after taking a loan. On average, plan borrowing lead participants to cut contributions by $22 per month or 6 percent. But in Panel B, the savings effect prove to be rather heterogeneous. In the all-loans category, around one-third of participants cut their contributions by 10 percent or more when taking a plan loan, while two-thirds have a smaller reduction, no reduction at all, or an increase.

Table 9.

Employee Contributions Pre/Post Loan (2015$)

A. Overall contributions by borrowers		Mean ($)	Std Dev. ($)
1 st loan	Prior to Loan (6 mo.)	383	438
	After Loan (6 mo.)	359	411
	Difference	−25	295
2nd+ loan	Prior to Loan (6 mo.)	329	343
	After Loan (6 mo.)	311	324
	Difference	−17	214
All loans	Prior to Loan (6 mo.)	362	404
	After Loan (6 mo.)	340	380
	Difference	−22	266
B. Distribution of contributions by borrowers
		10th ($)	20th ($)	30th ($)	40th ($)	50th ($)	60th ($)	70th ($)	80th ($)	90th ($)
1 st loan	Prior to Loan (6 mo.)	78	118	156	198	246	309	398	544	858
	After Loan (6 mo.)	73	113	149	189	236	295	377	507	782
	Difference	−164	−64	−28	−10	−1	5	17	39	93
2nd+ loan	Prior to Loan (6 mo.)	72	112	150	190	234	284	353	462	686
	After Loan (6 mo.)	66	105	142	180	223	271	338	440	647
	Difference	−129	−53	−23	−9	−1	4	14	34	80
All loans	Prior to Loan (6 mo.)	76	116	154	195	241	298	378	509	786
	After Loan (6 mo.)	70	110	146	186	231	284	360	479	726
	Difference	−149	−59	−26	−10	−1	4	16	37	87

Source: Authors’ tabulations.

It is worth noting that changes in contributions in our data set could arise for two different reasons: people might have actively changed their plan contribution rates, or their earnings could have changed, in turn lowering contributions. Both cases would make plan borrowing appear to be associated with reduced plan contributions, although in only the former case would the participant have intentionally changed plan contributions. Since our income statistics classify people into annual income brackets, we cannot control tightly for the impact of short-term earnings changes on plan contributions.

To explore pre/post loan contribution patterns, we use a difference-in-difference approach examining changes in contributions upon taking a loan. Table 10 summarizes results from our model, and in particular indicates how firms’ loan policies influenced contribution patterns when people did borrow from their plans.

Table 10.

Effect of Offering Multiple Loans on Changes in Employee Contributions

		Change in employee contributions after taking a loan (OLS)
			(1)			(2)
	Mean	Estimate		(SE)	Estimate		(SE)
Plan design
Multiple loans allowed	0.55	4.3	***	1.1		***	1.2
Loan interest rate (%)	7.17	3.2	***	0.3	3.1	***	0.3
Participant characteristics
2nd Loan or more	0.40				4.3	***	0.9
Age 35–45	0.33	4.1	***	1.0	4.1	***	1.0
Age 45–55	0.23	5.0	***	1.2	5.1	***	1.2
Age 55–60	0.06	6.4	***	2.2	6.6	***	2.2
Age > 60	0.02	7.3	*	4.2	7.6		4.2
Male	0.45	−2.9	***	1.1	−2.8		1.1
Tenure	5.14	6.5	***	0.3	6.4	***	0.3
Tenure**2	53.72	−0.2	***	0.0	−0.2	***	0.0
Income < $35,000	0.15	−1.4	*	0.8	−1.5		0.8
Income > $87,500	0.24	−3.2		1.2	−3.1		1.2
Low wealth	0.66	1.9	*	1.0	1.8		1.0
High wealth	0.10	−1.5		2.9	−1.4		2.9
Ln(401(k) account balance)	9.46	−30.6	***	0.8	−30.7	***	0.8
Macroeconomic variables
Financial turmoil period (%)	0.22	6.2	***	1.6	5.6	***	1.6
Lagged state unemployment rate (%)	5.34	0.0		0.4	0.0		0.4
Industry controls		Yes			Yes
Plan mean variables		Yes			Yes
Intercept		−0.5		22.2	2.0		22.2
No. observations		356,672			356,672
Mean of dependent variable		−21.871			−21.871
R-squared		0.015			0.015

Note:

^**Significant at 0.05 level,

^***Significant at 0.01 level.

Plan means included to control for fixed effects.

Source: Authors’ calculations.

Our difference-in-difference model shows that borrowers in plans allowing multiple loans had a somewhat smaller drop in contributions, around $4–5 per month, than their counterparts in single loan plans. Nevertheless, as a percent of total contributions this was a small difference (around one percentage point). Plans charging higher interest rates also experienced slightly less of a decline in contributions. We also learn that participants age 35+ reduced contributions less than did the reference group (participants under age 25). Meanwhile, higher-income participants curtailed their plan contributions by more than did low/moderate-income participants, and workers having more saved in their 401(k) accounts experienced the largest drop in contributions. Thus a one-unit increase in the participant’s log balance (roughly equivalent to an increase from $32,500 to $88,400) produced a $26 per month lower 401(k) contribution. Since this effect controlled on participants’ loan balances, it is not the result of simply having a larger loan to repay.

VIII. ESTIMATED TOTAL LOAN DEFAULTS AND REVENUE CONSEQUENCES

In recent years, several policymakers have proposed legislation to restrict flows of assets out of tax-qualified retirement plans, including plan loans. For example, U.S. Senators Kohl and Enzi proposed the 2011 “Savings Enhancement by Alleviating Leakage in 401(k) Savings Act,” stating in their press release that a “401(k) savings account should not be used as a piggy bank” (Leonard, 2011). In light of this policy concern, we next use our empirical findings to estimate the aggregate annual size of loan defaults from 401(k) plans, along with the tax revenue consequences to the Federal government of plan defaults.

To address this question, prior analysts have relied on the Private Pension Plan Bulletin derived from Form 5500 Annual Reports filed by retirement plans with the Employee Benefits Security Administration of the US Department of Labor (US DOL, 2012). One item reported in that document refers to the “Income Statement of Pension Plans with 100 or More Participants” and it lists the amount of “deemed distribution of participant loans.” Some analysts have incorrectly interpreted this amount as representing the total amount of loan defaults,²⁷ but it actually measures loan defaults only for active plan members due to temporary lay-off, long-term disability, maternity leave, or a leave of absence such as parental leave. Loan defaults due to job termination, which we focus on here, are recorded as offsets to participant account balances at the time of default, reported as “direct benefit payments” in the US DOL’s nomenclature.

To illustrate what a difference this definition makes, we find that only eight percent of the loan defaults observed in our dataset were “deemed” loan distributions. The remaining 92 percent resulted from defaults on job termination, which are the focus of the present analysis. Accordingly, data on “deemed distributions” seriously understate the annual value of retirement plan loan defaults. Applying our sample fractions to the entire private 401(k) system indicates that aggregate system-wide loan defaults amount to roughly $5 billion per year, or over eight times the $600 million in “deemed” loan distributions.²⁸ This is not a small sum, yet it is much lower than the $74 billion of account cash-outs on job termination (in 2006; GAO 2009). Assuming an effective income tax rate of 10 percent and factoring in the 10 percent penalty associated with early distributions, we estimate that the tax revenue flowing to the U.S. Government associated with defaulted DC plan loans to be on the order of $1 billion per year.

IX. CONCLUSIONS

More than two decades ago, Nobel Prize winner Franco Modigliani patented a method for issuing 401(k) credit cards with the aim of making it easier for workers to withdraw from their retirement accounts to cover short-term consumption needs (Vise, 2004). Although the idea of 401(k) credit cards withered under criticism, that proposal highlighted the dual-purpose nature of U.S. defined contribution plans. DC retirement accounts representing a growing fraction of US household wealth are being used by employees to both finance old-age retirement security, and to help cover current consumption needs. The plan loan feature is one of the prominent yet understudied pre-retirement liquidity features of 401(k) plans.

This paper has explored the effects of employer plan loan policy, and we conclude that loan design can and does have an economically meaningful impact on participant borrowing. In our dataset, one-fifth of plan participants had a loan at any given time, while almost 40 percent did so over a five-year period. Participants who borrowed more were also likely to be younger and liquidity-constrained, consistent with a lifecycle model of saving and borrowing. Yet conditional on borrowing, it was higher income/wealth participants who borrowed larger fractions of their 401(k) accounts.

Employer-determined plan loan policy also had a material effect on borrowing behavior. When a plan allowed employees to take out multiple loans, they were more likely to borrow. Individual loans were also smaller, suggestive of a buffer-stock model to managing credit similar to that found in credit cards. That is, given the ability to borrow multiple times, workers were more willing to take the first loan given that they retained slack borrowing capacity against future consumption shocks. Moreover, participants borrowed more as a proportion of retirement savings in multiple-loan plans, despite taking smaller individual loans, suggesting that offering multiple-loans is interpreted by workers as an employer endorsement of plan borrowing. And though we have not explicitly evaluated the idea of a 401(k) credit card, we note that enhancing 401(k) access in that way could strengthen the endorsement effect.

Using our administrative dataset, we show that nine of ten plan loans were repaid but 86 percent of workers with an outstanding loan balance defaulted on their loans when they terminated employment. It is among job-changers with outstanding loans where the impact of loans on retirement savings is strongest. Although liquidity-constrained participants were more likely to default, the size of these effects was small relative to the high overall default rate. This implies that other factors such as low financial literacy, impatience, or inattention, may be at work. In this way, a loan default is similar to the broader problem of cash-outs from DC retirement plans.

When we assessed the interplay between employee plan contributions and borrowing, we found that borrowing was associated with a small drop in monthly contributions, the result of one-third of participants cutting their contributions by 10 percent or more, whereas two-thirds did not. Using our results, we also computed the aggregate effect of loan defaults on retirement savings at around $5 billion per year. We estimate that this produced an annual $1 billion in tax revenue flowing to the U.S. Government due to defaulted DC plan loans each year.

Our research findings should be of interest to policymakers and plan sponsors seeking to evaluate the effectiveness of access features in U.S defined contribution retirement plans. The fact that many workers do borrow from and default on their pension accounts has led some to propose that 401(k) loans should be restricted (Reeves and Villareal, 2008; Weller and Wenger, 2008). Our results imply that such concerns about the effects of plan loans on retirement adequacy seem overstated, particularly when compared to the exit of plan assets due to account cash-outs upon job change. Yet we conclude that offering a single loan in lieu of multiple loans would reduce the incidence of borrowing and the fraction of total wealth borrowed, thereby limiting the impact of future defaults. Additionally, restricting the size and scope of plan loans could reduce the total value of loan defaults.²⁹ Alternatively, firms could permit terminated workers to continue repaying their loans instead of requiring a balloon payment. Of course implementing this could be challenging if employers no longer have an ongoing payroll relationship with terminated employees. Finally, any changes in loan rules must reflect the finding from the existing literature regarding the positive impact of a borrowing feature on contributions, at least in traditional voluntary enrollment 401(k) plans.

These findings underscore the reality that DC accounts do provide many workers with pre-retirement liquidity to meet current consumption needs, even though the plans were designed mainly to provide for old-age financial security.

Biography

Timothy (Jun) Lu

Timothy (Jun) Lu is an assistant professor of the Peking University HSBC Business School in China. His research has been supported by the Social Security Administration, the Financial Literacy Center, the Michigan Retirement Research Center, and the Boettner Center for Pensions and Retirement Research/Pension Research Council at the University of Pennsylvania,

Olivia S. Mitchell

Olivia S. Mitchell is a Professor of Insurance & Risk Management, and Business Economics & Policy, of the Wharton School of the University of Pennsylvania, where she also serves as Director of the Pension Research Council, a Wharton School research center and nonprofit entity (pensionresearchcouncil.org). She is a NBER Research Associate and she also serves as an independent Trustee of the Wells Fargo Advantage Funds. Her research has been supported by the Social Security Administration, Netspar, TIAA, the Singapore Management University, the Financial Literacy Center, the Michigan Retirement Research Center, the ARC Centre of Excellence, and the Boettner Center for Pensions and Retirement Research/Pension Research Council at the University of Pennsylvania.

Stephen P. Utkus

Stephen P. Utkus is a full-time employee of Vanguard, a leading investment manager and provider of recordkeeping services to 401(k) plans. He is also a member of the Advisory Council of the Wharton Pension Research Council and a member of the Board of Trustees of the Employee Benefit Research Institute. Vanguard provided the data for this paper under anonymous and restricted access conditions. Vanguard is a Senior Partner of the Wharton Pension Research Council and provides annual funding to the Pension Research Council in support of its research activities.

Jean A. Young

Jean A. Young is a full-time employee of Vanguard, a leading investment manager and provider of recordkeeping services to 401(k) plans. Vanguard provided the data for the current paper under anonymous, restricted access conditions. Vanguard is a Senior Partner of the Wharton Pension Research Council and it provides an annual research grant to the Pension Research Council in support of its research activities

Appendix

Online Appendix Table 1.

Sample Means

	Plans allowing only one loan (N=28,476,193)		Plans allowing multiple loans (N=26,699,525)
	Mean	Std. Dev.	Mean	Std. Dev.
Taking new loan (in a month)	0.01	0.10	0.02	0.13
Plan design
Multiple loans allowed (dummy)	NA	NA	1.00	0.00
Loan interest rate (%)	7.22	1.78	7.10	1.77
Participant demographics
Age < 35	0.20	0.40	0.20	0.40
Age 35–45	0.28	0.45	0.28	0.45
Age 45–55	0.31	0.46	0.33	0.47
Age 55–60	0.12	0.33	0.12	0.32
Age > 60	0.09	0.28	0.07	0.26
Male	0.51	0.50	0.53	0.50
Tenure	7.83	6.98	9.23	7.80
Tenure**2	109.92	195.82	146.10	225.15
Income < $35,000	0.08	0.28	0.08	0.26
Income > $87,500	0.43	0.50	0.43	0.50
Low wealth	0.42	0.49	0.44	0.50
High wealth	0.26	0.44	0.25	0.43
Ln(401(k) account balance)	11.26	1.58	11.59	1.69
Macroeconomic variables
Financial turmoil period	0.20	0.40	0.18	0.38
Lagged state-level unemployment rate (%)	5.37	1.39	5.22	1.29
Plan mean variables
Age	44.85	3.16	44.50	2.98
Male	0.51	0.19	0.53	0.20
Tenure	7.83	2.96	9.23	3.18
Income < $35,000	0.08	0.04	0.08	0.04
Income > $87,500	0.43	0.12	0.43	0.12
Low wealth	0.42	0.15	0.44	0.13
High wealth	0.26	0.12	0.25	0.11
Ln (401(k) acct balance)	11.26	0.49	11.59	0.68
N of participants (/10,000)	1.87	2.14	1.61	1.74
Industry
Agriculture	0.05	0.23	0.15	0.36
Transportation	0.04	0.20	0.05	0.21
Finance	0.20	0.40	0.20	0.40
Retail	0.07	0.26	0.03	0.18
Media	0.11	0.31	0.06	0.23
Business services	0.10	0.30	0.12	0.32
Education	0.07	0.25	0.03	0.17
Public sector	0.00	0.04	0.00	0.03
Other	0.05	0.22	0.05	0.22

Note: Values apply to Table 2. Sample characteristics for other tables are available from the authors on request.

Online Appendix Table 2.

Comparison of Our Sample with EBRI-ICI Participant Data

	2005	2006	2007	2008	2009
A: Loans
% participants w/1+ loan
EBRI/ICI	0.19	0.18	0.18	0.18	0.21
Our sample	0.22	0.21	0.21	0.20	0.20
% of balance borrowed
EBRI/ICI	0.13	0.12	0.12	0.16	0.15
Our sample	0.26	0.26	0.24	0.26	0.29
Mean total amount borrowed ($)
EBRI/ICI	6,821	7,292	7,495	7,191	7,346
Our sample	8,195	8,586	8,723	8,893	8,870
Median total amount borrowed ($)
EBRI/ICI	3,641	4,089	4,167	3,889	3,972
Our sample	4,900	5,170	5,226	5,276	5,250
B: Account Balances
Average account balance ($)
EBRI/ICI	58,328	61,346	65,454	45,519	58,351
Our sample	71,607	77,050	87,012	78,778	65,496
Median account balance ($)
EBRI/ICI	19,398	18,986	18,942	12,655	17,794
Our sample	26,634	27,605	28,652	24,040	22,505
C: Participant Characteristics (2009)
Age	EBRI/ICI	Our sample	Job tenure	EBRI/ICI	Our sample
Median	45	45	Median	6	5.5
% in 20s	0.13	0.12	0–2 years %	0.16	0.24
% in 30s	0.24	0.24	>2–5 %	0.22	0.24
% in 40s	0.29	0.29	>5–10 %	0.23	0.22
% in 50s	0.25	0.26	> 10–20%	0.23	0.21
% in 60s+	0.09	0.09	>20–30 %	0.11	0.07
	1.00	1.00	>30 %	0.05	0.02
				1.00	1.00

Note: EBRI-ICI data as of December, paper sample as of June. During the period, EBR/ICI did not differentiate active versus inactive employees, so both appear in their statistics; our dataset includes only active workers. Source: Authors’ tabulations from Vanguard data and the EBRI-ICI Participant Data Collection Project (Vanderhei et al., 2014).

Online Appendix Table 3.

Determinants of the Probability of a Plan Offering Multiple Loans

	Probability of plan offering multiple loans
	Probit			OLS
	Estimate	(SE)	Marginal Effect (%)	Estimate		(SE)
# Plan participants	−0.034	0.097	−1.3	−0.013		0.037
Plan average age	−0.001	0.012	0.0	0.000		0.004
Plan standard deviation of age	−0.022	0.021	−0.9	−0.009		0.007
Plan average tenure	0.018	0.019	0.7	0.006		0.007
Plan standard deviation of tenure	0.055	0.028	2.2	0.021	**	0.010
% Low income participants (< $35,000)	1.115	0.665	40.5	0.414		0.243
% High income participants (> $87,500)	0.062	0.275	2.0	0.025		0.104
% Low wealth participants	−0.389	0.279	−14.9	−0.140		0.104
% High wealth participants	−0.001	0.367	0.0	−0.001		0.139
Industry Controls	Yes			Yes
Intercept	−0.375	0.541		0.346		0.203
No. observations		882				882
Mean of dependent variable		0.429				0.429
Pseudo R-squared / R-squared		0.04				0.04

Note:

^**Significant at 0.05 level,

^***significant at 0.01 level.

Source: Authors’ calculations.