Defined Contribution Pension Plans: Determinants
of Participation and Contributions Rates
Gur Huberman
&
Sheena S. Iyengar
&
Wei Jiang
Received: 21 October 2005 / Revised: 22 August 2006 / Accepted: 05 January 2007
# Springer Science + Business Media, LLC 2007
Abstract Records of 793,794 employees eligible to participate in 647 defined contribution
pension plans are studied. About 71% of them choose to participate in the plans, and of the
participants, 12% choose to contribute the maximum allowed, $10,500. The main findings
are (other things equal) (1) participation rates, contributions and (most remarkably) savings
rates increase with compensation; on average, a $10,000 increase in compensation is
associated with a 3.7% higher participation probability and $900 higher contribution; (2)
women
’s participation probability is 6.5% higher than men’s and they contribute almost $500
more than men; (3) participation probabilities are similar for employees covered and not
covered by DB plans, but those covered by DB plans contribute more to the DC plans; (4) the
availability of a match by the employer increases employees
’ participation and contributions;
the effect is strongest for low-income employees; (v) participation rates, especially among
low-income employees, are higher when company stock is an investable fund.
Keywords 401(k) plans . defined contribution
1 Introduction
Defined contribution pension plans in general, and 401(k) plans in particular are important
vehicles for retirement savings. Although a handful of studies have considered individual
and plan-level attributes that affect participation in such plans, or, for participants, levels of
contribution, these studies either used only survey data (Papke
2004a
; Munnell et al.
2001
;
J Finan Serv Res
DOI 10.1007/s10693-007-0003-6
G. Huberman
:
W. Jiang (
*)
Finance and Economics Division, Columbia Business School, 3022 Broadway,
New York, NY 10027, USA
e-mail: wj2006@columbia.edu
G. Huberman
e-mail: gh16@columbia.edu
S. S. Iyengar
Management Division, Columbia Business School, New York, NY, USA
e-mail: ss957@columbia.edu
Even and Macpherson
2003
), or employee records for very few firms (Kusko et al.
1998
;
Clark and Schieber
1998
; Agnew et al.
2003
) or used plan-level data (Papke
2004b
; Papke
and Poterba
1995
). With information on almost 800,000 employees eligible to participate in
647 such plans (include those who chose not to participate), this study provides a
comprehensive picture of the variables associated with individual participation in and
contribution to 401(k) plans.
Individual-level data are important because in general, it is inappropriate to estimate a
relation on an aggregate level and then infer that an analogous relation holds at the
individual level. In some cases, even the sign of certain sensitivity estimates could be
reversed (See a discussion in Freedman
2001
). Records of non-participants afford
particularly powerful analysis of the participation and contribution decisions.
Some
—but not all—of the qualitative relations reported here are straightforward in light
of the incentives faced by employees. For instance, the presence of an employer match
should increase employees
’ participation in a 401(k) plan. The data afford going beyond
qualitative observations. Specifically, they allow precise estimation of the sensitivities of
employee behavior to explanatory variables which are important in their own right, and
very useful to designers of retirement savings plans and policy makers at the firm and
national levels.
This study goes beyond estimating overall relations between choice variables
—
participation and level of contribution
—and individual and plan attributes. It explores
why and how sensitivities of the choice variables to the attributes differ between the
participation and contribution decision, and it also considers how the sensitivities of the
choice variables to the attributes vary with compensation. The main results can thus be
summarized while enumerating the main attributes, both of the plans and individuals.
Although individual characteristics such as gender and age are clearly exogenous, one
cannot rule out the possibility that plan design could be catering for the aggregate
characteristics and preferences of plan employees (see Mitchell et al.
2006
, for an analysis
of plan design), or that individual employees self-select into employers who offer plans that
suit their retirement savings needs. Without reasonably long panel data it is difficult to tease
out such effects using identification based on changes (such as the fixed effects method).
All individual-level analyses in the paper control for plan aggregate characteristics (such as
plan-average compensation, etc.) to mitigate the endogeneity concern at the individual
level. It is possible, however, that some of the effects of plan policies documented here
might capture cross-sectional relation rather than a definite causal relation due to
unobserved plan-level heterogeneity. In interpreting these results, we also discuss the
plausibility of alternative hypothesis.
Plan designs have strong effects on savings outcome (see a recent review by Choi et al.
(
2004c
)). Among plan-level attributes, the first important one is coverage in a defined
benefit (DB) plan. Comparing two similar individuals, the one with a DB plan is already
saving for retirement, and his propensity to forego current consumption and liquidity in
favor of consumption during retirement should be lower. On the other hand, an extreme
form of mental accounting will render rights within a DB plan completely irrelevant to
choices in a DC plan. Employees covered by a DB plan who have this form of mental
accounting will participate in and contribute to their DC plans as if they were not covered
by a DB plan. (Shefrin and Thaler
1992
, and Thaler
1999
, describe and analyze instances of
mental accounting.) Moreover, the need to save for retirement may be more salient among
those covered by a DB plan, or the savings-prone individuals are more likely attracted to
employers that offer both DB and DC plans. The combined effect will lead to the surprising
result that those covered by DB plans make stronger usage of 401(k) plans. The data are
J Finan Serv Res
consistent with this last, and very counterintuitive result: other things equal, participation
rates of those covered and not covered by DB plans are similar; contributions of those
covered by DB plans are higher.
Many employers
–539 plans in the study’s sample of 647 plans–offer to match
employees
’ contributions to DC plans. These matches are powerful incentives to
participate, and indeed, participation rates are higher in the presence of a match. The
incentive effect of the match is strongest for the lowest-income employees, and it decreases
with compensation. In fact, at low-income levels (annual compensation between $10,000
and $20,000), a 100% employer match could increase participation probability by nearly
20%; at higher incomes (above $90,000), the incentive effect drops to about 5%.
The data indicate that the presence of a match increases contributions, primarily by
increasing participation. In fact, among participants, the presence of a match seems to have
no effect on contributions of low-income employees and, surprisingly, negative effect on
contributions of those earning between $40,000 and $130,000. Participants
’ tendency to
contribute at the upper limit on employer
’s match may be responsible for this counterintuitive
finding. (Employers typically limit their matches to 5
–6% of a participant’s salary.)
The strong effect of matching programs on participation, especially of low-income
employees, offers an immediate suggestion for a policy that encourages retirement savings
in self-directed savings plans such as IRAs: the government could match the savers
’
contributions. Such a matching program can be more intense for low-income individuals if
wealth redistribution is a secondary goal.
The inclusion of company stock in the plan
’s menu of investable funds guarantees the
presence of a familiar option in the menu. Huberman (
2001
) argues that familiarity breeds
investment. In fact, one of his examples is company stock in 401(k) plans. Other studies
that consider the impact of company stock on asset allocation in 401(k) plans include Benartzi
(
2001
), Choi et al. (
2004a
,
b
,
c
), Huberman and Sengmuller (
2004
), Liang and Weisbenner
(
2002
), Mitchell and Utkus (
2003
), Meulbroek (
2002
), Ramaswamy (
2002
), Holden and
VanDerhei (
2003
), and Poterba (
2003
). One theme common to these studies is that being
associated with bad portfolio selection, the presence of company stock in the investable
funds is bad for participants.
Overlooked thus far has been the potential salutary effect of including company stock in
the investable funds: participation probability may be higher, presumably because eligible
employees feel more comfortable participating when a familiar option is available.
Empirically, this is the case. Participation probabilities are higher, especially for low-
income employees. For employees who earn less than $35,000, the presence of company
stocks as an investment option increase participation by more than 5%. The effect
diminishes for employees who earn more than $40,000.
Compensation and gender are the more interesting individual attributes. The progres-
sivity of the income tax code entails stronger incentives to participate and contribute to
those who earn more. Moreover, low-income employees are more likely to have, or
anticipate having liquidity constraints which will deter them from participating or
contributing large sums to a 401(k) plan, where the money is locked up until retirement.
Additionally, low-income employees expect higher salary replacement rates from social
security upon retirement than high-income employees. This anticipation lowers the desire to
save for retirement.
The data indeed show that controlling for all other variables, participation probability
typically increases by almost 4% and contributions increase by about $900 for an increase
of $10,000 in compensation. Moreover, savings rates
—the ratios of contributions to
compensation
—increase with compensation.
J Finan Serv Res
Gender matters in saving decisions, adding to prior findings of gender differences in
financial decisions (see, e.g., Barber and Odean
2001
, and Bajtelsmit and Bernasek
1999
).
Holding other variables the same (especially compensation!) women
’s participation
probabilities are 6.5% higher and their contributions are close to $500 higher.
This gender difference has at least two explanations, which are not mutually exclusive.
One, that women have a stronger preference for saving, perhaps because they typically live
longer than men. Two, the unit of decision is the household, and in many cases women are
secondary wage earners whose incomes supplement those of their husbands. In these cases
the women
’s recorded incomes are substantially lower than their households’ incomes and
their behavior is likely to reflect their households
’ incomes. (Nationally, according to
Business Week, in 70% of the married households the husbands earn more than the wives.)
The next section describes the data and the econometric model. Section
3
reports the
overall evidence and Section
4
reports how estimates vary with compensation. Section
5
discusses the findings.
2 Data Description and Model Set-up
2.1 Data
The Vanguard Group provided 926,104 participation and contribution employee records
(including employees who were eligible but chose to not participate) in defined contribution
(DC) pension, mostly 401(k) plans for the year 2001. The data contain 647 plans in 69
industries (by SIC two-digit codes). All plans required eligible employees to opt into the plan.
Other concurrent studies using the same dataset including Iyengar et al. (
2004
, on the effect
of offered choices on 401(k) participation), Mitchell et al. (
2005
and
2006
, on the effect of
plan design on plan-level savings behavior; 2006, on the determinants of 401(k) plan
design), Huberman and Jiang (
2006
, on the relation between offerings and choices for
individual 401(k) participants), and Iyengar and Kamenica (2006, on choice overload and
401(k) asset allocation).
For the purpose of this research, excluded from the data were observations in at least one
of the following categories: (1) The employee was hired after January 1, 2001 (9.6% of the
observations). This exclusion criterion ensures that the person is employed for the whole
year of 2001; (2) The person is less than 18 years old (0.02% of the observations). (3) The
annual compensation is less than $10,000 or above $1 million (7.51% of the observations)
to avoid the influence of extreme outliers. 793,794 observations survive. The
Appendix
offers more details on the construction of variables.
The all-sample participation rate is 71%, and about 76% of the eligible employees have
positive balances (comparable to the national average participation rate of 76% reported by
the Profit Sharing/401(k) Council of America
2001
,
2002
). The average individual pre-tax
contribution rate for the whole sample and that for the highly compensated employees
(defined as those who earned $85,000 or more in 2001) were 4.7 and 6.3%, respectively,
compared to the national averages of 5.2 and 6.3% (Council of America
2001
,
2002
). In
summary, the savings behavior of employees in the Vanguard sample seems representative
of the overall population of eligible employees.
In the sample, 63% are male, and the mean age is 43. Figures
1
and
2
plot the sample
’s
age and compensation histograms, respectively. Compensation mean and median are
$61,150 and $47,430, respectively. In comparison, the same figures from the Survey of
Consumer Finance (SCF) are $70,700 and $43,200. The average compensation is $65,900
J Finan Serv Res
for the 1998 SCF 401(k) eligible employee sample. In 2001, the maximum compensation
for defined contribution plan purpose was $170,000, and therefore the compensation
variable used in the regressions is winsorized at $170,000 (about 3% of the sample). Other
information about individual characteristics includes tenure and financial wealth of the
nine-digit zip neighborhood the employee lives in. A company called IXI collects retail and
IRA asset data from most of the large financial services companies. IXI receives the data
from all the companies at the nine-digit zip level, and then divides the total financial assets
by the number of households in the relevant nine-digit zip area to determine the average
0
20
40
60
80
100
120
140
160
180
18-22
23-27
28-32
33-37
38-42
43-47
48-52
53-57
58-62
63-67
68-72
73-
Number (in 1,000s)
Employees
Participants
Figure 1 Age distribution of eligible employees and participants
0
20
40
60
80
100
120
140
10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-
100
100-
120
120-
140
140-
160
160-
180
180-
200
over
200
Dollar (in 1,000)
N
u
mb
er
(i
n
1,
000)
Employees
Participants
Figure 2 Compensation distribution of eligible employees and of participants
J Finan Serv Res
assets for each neighborhood. There are 10
–12 households in a nine-digit zip area on
average. Subsequently, IXI assigns a wealth rank (from 1 to 24) to the area.
The records break down contributions to DC plans into three parts: employee pre-tax
contribution, employee after-tax contribution, and employer contribution (including
employer match). All the work reported here uses employee before-tax contributions to
be comparable to most other research in 401(k) savings. In this study an employee is
considered as a participant in a DC plan in 2001 if she contributes a positive amount before
tax. By this criterion, participation rate is 71%, while 75% of the accounts have positive
balances in tax-deferred accounts. (The employees who made no contribution in 2001 but
had positive balances are probably those who had made contributions in earlier years but
not in 2001 or those working for employers who make contributions but they choose not to
contribute.)
Most plans ask employees to specify their deferral rates at the beginning of the year. The
maximum contribution allowed in 2001 was the lower of $10,500, or 25% of
compensation. Some plans impose additional limits on contributions made by highly
compensated employees (HCEs, defined as those who earned $85,000 or higher in 2001) to
ensure that the DC plans do not overly disproportionately benefit the high-income people
(see, e.g., Holden and VanDerhei
2001
). The mean deferral rate is 5.2%, and 12% of the
participants contributed the maximum amount. Figure
3
plots the relation between
participation/maximum contribution
1
and compensation. Both participation probabilities
and the probability of contributing the maximum increase with compensation. The majority
of those earning $30,000 or above participate. The majority of employees who earn about
$130,000 or above contribute the maximum. Nonetheless, about 9% of the high-income
employees do not participate at all.
1
Here we only consider maximum contribution to the IRS limit ($10,500 or 25% of compensation).
Section
3.3
discusses potential plan-specific limits that are lower than the IRS limit.
0
10
20
30
40
50
60
70
80
90
100
10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-
100
100-
120
120-
140
140-
160
160-
180
180-
200
over
200
Compensation ($1,000)
%
Participation
Max-Out
Figure 3 Rates of participation and contributing and maximum at each compensation level
J Finan Serv Res
Figure
4
plots individual annual contributions for the full sample and the sub-sample
with compensation above $85,000 (HCEs). Among all participants, the modal contribution
is between $1,000 and $2,000. Those who earn more than $85,000 (16.7% of the sample)
contribute more than the typical participants, and about 39% of those earning more than
$85,000 contribute the maximum of $10,500. Figure
5
plots the contribution at different
percentiles for employees at different levels of compensation. The figure clearly shows that
high percentiles respond more intensely to increase in compensation, thereby suggesting
that the cross-sectional variance of contributions increases with compensation. Figure
6
0
5
10
15
20
25
30
35
40
45
zero
0-1
1-2
2-3
3-4
4-5
5-6
6-7
7-8
8-9
9-10
10-10.4
Max
Contribution ($1,000)
%
Full Sample
HCEs
Figure 4 Fraction of employees at each contribution level
0
2000
4000
6000
8000
10000
12000
10
20
30
40
50
60
70
80
90
10
0
11
0
12
0
13
0
14
0
15
0
16
0
17
0
18
0
19
0
20
0
21
0
22
0
23
0
24
0
25
0
Com pensation ($1,000)
C
o
ntr
ibuti
on
by
P
e
rc
e
n
ti
le
10%
25%
50%
75%
90%
(For each compensation level, 10th, 25th, 50th, 70th and 90th percentile of contribution in dollars)
Figure 5 Contribution levels at each percentile
J Finan Serv Res
plots the wealth histogram for the general IXI population and for the Vanguard sample.
Evidently, the Vanguard sample is somewhat better off than the general population at lower
to middle wealth ranks.
The records have information about plan policies, including the presence of defined
benefit (DB) plans, the number of investable funds available, employer matching schedule
(match range and match rate), the presence of company stock as an investment option,
whether the employer
’s match is in cash or company stock, and if the latter, the restrictions
on diversification of the employer
’s match. 124 plans (covering 58% of the employees in
the sample) provide own company stocks as an investment option, among which 47
companies match employee contribution with company stocks only. 216 plans (covering
67% of the employees in the sample) offer defined-benefit plans in addition to the defined
contribution plan studied here. The number of funds offered by a plan ranges from 2 to 59
but 90% of the plans offer between 6 and 25 funds. Employers in 539 plans (covering 87%
of the employees in the sample) offer some match to their employees
’ contributions. Most
of them offer to match the employee
’s contribution up to 6% of the employee’s salary, and
the match rates range from 10 to 250%, mostly between 50 and 100%.
Exploratory data analysis is this study
’s main goal. Applying probit, one- and two-sided
Tobit, and censored median regression analyses, the exploration goes beyond simple
tabulation of averages and correlation and linear regression analyses. It affords an
understanding of the decisions made by employees regarding their 401(k) savings.
However, in the absence of a structural model, there is no single preferred specification.
2.2 Model Specification
The dependent variables studied here are: (1) A dummy variable, PART, that equals one if
the individual participates, that is, if he contributes a positive amount to his tax-deferred
account; (2) A dummy variable, MAXOUT, that equals one if the individuals contribute the
maximum amount ($10,500 in 2001) to his tax deferred account; (3) Annual contribution,
CONTRIBUTION, in dollar units or as a percentage of compensation.
The indices i and j represent individuals and plans, respectively. An individual’s benefit
from participating in a DC plan (net of cost) can be expressed as a function of a set of
0
2
4
6
8
10
12
14
16
18
20
zer
o or
ne
gat
iv
e
un
der 1
K
1.
0-
2.
5K
2.
5-
5.
0K
5.
0-
10
K
10
-2
5K
25
-50
K
50
-75
K
75
-10
0K
10
0-
12
5K
12
5-
15
0K
15
0-
175
K
17
5-
20
0K
200
-22
5K
225
-25
0K
25
0-
30
0K
30
0-
35
0K
35
0-
40
0K
40
0-
45
0K
45
0-
50
0K
50
0-
75
0K
75
0K-
1M
ab
ov
e
1M
Dostları ilə paylaş: |