A review of international experience

Yüklə 376.41 Kb.
PDF просмотр
ölçüsü376.41 Kb.
1   ...   20   21   22   23   24   25   26   27   28

Behavioral and Design Issues

Matching Contributions and 
Savings Outcomes: 
A Behavioral Economics Perspective
Brigitte C. Madrian
Including a matching contribution increases savings plan participation and contributions, 
although the impact is less significant than that of nonfinancial approaches. Conditional on 
participation, a higher match rate has only a small effect on savings plan contributions. In 
contrast, the match threshold has a substantial impact, probably because it serves as a natu-
ral reference point when individuals are deciding how much to save and may be viewed as 
advice from the savings program sponsor on how much to save. Other behavioral approaches 
to changing savings plan outcomes—including automatic enrollment, simplification, plan-
ning aids, reminders, and commitment features—potentially have a much greater impact 
on savings outcomes than do financial incentives, often at a much lower cost.

common feature of schemes designed to increase individual savings is providing a 
matching contribution to create an incentive for participation in the program and 
induce higher levels of savings. The vast majority of employer-sponsored savings plans 
include an employer match, as do many employer-sponsored health savings accounts. 
The saver’s credit, a feature of the U.S. tax code designed to encourage savings by lower-
income households, also provides a government match to individual savings. Many field 
experiments aimed at encouraging savings have also included a match in their experimen-
tal design. This rich set of experience informs the understanding of behavioral responses 
to various matching contribution arrangements.
Traditional economic models point to financial incentives, such as a matching con-
tribution, as the logical mechanism to increase savings plan participation. The first part of 
the chapter summarizes the literature on the impact of providing a match on savings plan 
outcomes, including participation, contributions, and net worth. The evidence comes 
from a variety of sources, including observational data from surveys, natural experiments
and large-scale field experiments. Although the empirical evidence largely supports the 
predictions of traditional economic models, these models fail to incorporate the many 
psychological frictions that impede savings, including present bias, complexity, inatten-
tion, and temptation, which in many cases exert a much stronger impact on savings out-
comes than do financial incentives. Traditional economic models also fail to characterize 
some significant behavioral aspects of savings outcomes, including inertia and the impor-
tant role of focal points. The second part of the chapter evaluates the literature on other, 
nonfinancial approaches to increasing individual savings.
The evidence suggests that matching contributions increase savings plan participa-
tion and contributions, although the impact is less significant than that of nonfinancial 

approaches. Conditional on participation, a higher match rate has only a small effect 
on savings plan contributions. In contrast, the match threshold has a substantial impact, 
probably because it serves as a natural reference point when individuals are deciding how 
much to save and may be viewed as advice from the savings program sponsor on how 
much to save. Automatic enrollment, simplification, planning aids, reminders, and vari-
ous commitment devices potentially have a much greater impact on savings plan partici-
pation and contributions, often at a much lower cost.
Impact of Matching Contributions on Savings Outcomes: Theory 
In traditional models, the impact of a match on savings outcomes depends in part on the 
structure of the match. The simplest form is a flat match rate on all incremental savings 
(for example, all new contributions are matched 100 percent). In practice, offering an 
unlimited match is expensive for the party providing the match; as a consequence, savings 
schemes typically limit the contributions that are matched (for example, all contribu-
tions up to $1,000 are matched 100 percent, and contributions above that level are not 
Savings schemes with more complicated match structures are common. For exam-
ple, the match might be tiered, with contributions up to $500 matched 100 percent, 
contributions of $501–$1,000 matched 50 percent, and contributions above $1,000 not 
matched. Alternatively, contributions might be matched only after a certain level of con-
tributions is reached (for example, contributions below $500 are not matched, contribu-
tions of $501–$1,000 are matched 100 percent, and contributions above $1,000 are not 
In standard economic models of intertemporal decision making, adding a matching 
contribution or increasing the generosity of a match, whatever its form, should increase 
participation in a savings scheme through a substitution effect. The match makes con-
suming income more expensive than saving it, motivating individuals to substitute saving 
for consumption in response to the match.
The theoretical impact on individuals already contributing to the savings plan, how-
ever, is ambiguous. Consider, for example, introducing a scheme in which contributions 
are matched only up to a certain threshold. Such a scheme would increase contributions 
for individuals who were not previously participating, as some of these nonparticipants 
may be induced to start saving by the match. In contrast, individuals who were already 
contributing in excess of the match threshold are predicted to respond to the new match 
by reducing their contributions, through an income effect. The match on their existing 
contributions acts like an additional source of income, some of which individuals use to 
increase their consumption and correspondingly reduce their savings. Their combined 
own plus matching contributions, however, should still be higher than before the match.
The impact on individuals previously contributing at or below the match threshold 
is ambiguous; they are affected by both the income and substitution effects described 
above. Because they are saving below the match threshold, the match creates an incentive 
to substitute additional savings, up to the match threshold, for consumption. But the 
match on contributions already made acts like additional income, some of which will be 
used to increase consumption and reduce contributions.

The effects would be similar for increasing the match rate while maintaining the 
same match threshold. The effects of increasing the match threshold while keeping the 
match rate constant are more complicated. Such a change should have no effect on people 
contributing below the old threshold. It should increase contributions by people at the 
old threshold (a substitution effect), have an ambiguous effect on people above the old 
threshold but at or below the new threshold (opposing income and substitution effects), 
and decrease contribution rates by people above the new threshold (an income effect).
Impact of Matching Contributions on Savings Outcomes: Evidence
What is the evidence on how people actually respond? Estimating the impact of a match-
ing contribution on savings outcomes requires introducing some variation in the extent or 
structure of the match. The research has used three sources of match variation: naturally 
occurring cross-sectional variation (for example, differences in the match rate or match 
threshold in employer-sponsored savings plans); natural experiments, or changes in the 
structure of the match, within a savings scheme; and experimental variation generated by 
researchers, in which some individuals are offered a match, or a more generous match, and 
others are not.
The advantage of naturally occurring cross-sectional variation is that there can be con-
siderable heterogeneity in the types of matching incentives different individuals face. For 
example, the match rates in employer-sponsored 401(k) savings plans in the United States 
range from no match to match rates as high as 200 percent, and the match thresholds range 
from 1 percent of salary to $17,000 a year.
 This type of variation can be useful if, for exam-
ple, one wants to simulate what would happen under a match structure that is very different 
from what is currently used. A severe limitation of using this type of variation, however, is 
that it may be difficult to disentangle the impact of differences in the match structure on 
individual behavior from other factors that might also affect outcomes. For example, indi-
viduals who have a strong savings motive may seek employment in firms that offer a savings 
plan with a generous match, whereas individuals with a weak savings motive may select 
into firms with a less generous or no match (or no savings plan at all). If this type of sorting 
occurs, the estimated relationship between the match and savings outcomes will be biased.
The advantage of natural and field experiments is that there are generally fewer con-
cerns about the endogeneity between the generosity of the match and individual savings 
preferences. In field experiments, individuals are usually randomly assigned to receive dif-
ferent match structures. With natural experiments, concerns about endogeneity can be 
minimized by focusing on the same group of individuals before and after a policy change, 
essentially holding savings motives fixed. The limitation of field and natural experiments 
is that they typically examine a much smaller range of variation in matching schemes, with 
only two, or perhaps three, different types of match. The generalizability of the results 
from these studies is limited by the extent of the variation that is actually analyzed. These 
studies also typically focus on a specific group of individuals (for example, employees at a 
single firm, customers of a particular financial services provider, or low-income workers), 
limiting the extent to which the results can be generalized.
Most of the empirical studies on matching and savings outcomes have exploited 
the naturally occurring variation in the match rates of employer-sponsored savings plans 

in the United States to examine the impact of matching on savings outcomes. Most of 
these studies find, consistent with theoretical predictions, that matching increases savings 
plan participation rates (Andrews 1992; Bassett, Fleming, and Rogrigues 1998; Clark 
and Schieber 1998; Clark and others 2000; Dworak-Fisher 2008; Even and Macpher-
son 1997 and 2005; GAO 1997; Huberman, Iyengar, and Jiang 2007; Mitchell, Utkus, 
and Yang 2007; Papke and Poterba 1995). Some studies, however, find no relationship 
between matching and savings plan participation (Kusko, Poterba, and Wilcox 1998; 
Papke 1995). 
In evaluating how matching affects savings plan contributions, the empirical evi-
dence is less decisive (as noted above, the theoretical predictions are also not unam-
biguous). A few studies find a positive relationship between matching and savings plan 
contributions (Andrews 1992; Even and Macpherson 1997; Kusko, Poterba, and Wilcox 
1998; Papke and Poterba 1995). One, Basset, Fleming, and Rodrigues (1998), finds no 
relationship between matching and savings plan contributions. Several studies estimate 
that a higher match is associated with lower contributions (Clark and others 2000; Mitch-
ell, Utkus, and Yang 2007; Munnell, Sundén, and Taylor 2001; VanDerhei and Holden 
2001). Some studies find heterogeneous effects. Huberman, Iyengar, and Jiang (2007) 
find that a higher match increases contributions for low-income individuals but decreases 
contributions for middle- and high-income individuals. Papke (1995) and GAO (1997) 
find a positive effect of the match rate on contributions when the match rate is low but a 
negative effect on contributions when the match rate is high.
The most careful and convincing study using naturally occurring variation in match 
rates is Engelhardt and Kumar (2007). This study has several attractive features:
• It is the only study that appropriately accounts for the nonlinear savings incen-
tives generated by the employer match.
• It uses administrative data on savings plan contributions and earnings (from tax 
authority records on earnings and savings plan contributions) and on the struc-
ture of the employer match (from employer plan documents) to accurately model 
the incentives that individuals face and to get more accurate measures of their 
choices than is the case in self-reported survey data.
• It accounts for factors other than the employer match that might also influence 
savings outcomes, including taxes and alternative savings opportunities that may 
be equally or more attractive—specifically, individual retirement accounts (IRAs).
The biggest limitation of this study is that the data come from the Health and Retire-
ment Study and thus focus on older individuals (the average age is 55), whose behavior 
may differ from that of younger people. 
Engelhardt and Kumar estimate that increasing the match rate by 25 percentage 
points (for example, from $0.25 per $1 to $0.50 per $1 contributed) raises savings plan 
participation by 5 percentage points and increases contributions by plan participants by 
$365 (in 1991 dollars). They estimate that responsiveness to the employer match increases 
with the reported education level of respondents. Their overall conclusion is that neither 
participation nor contributions are very responsive to changes in the employer match and 
that “matching is a rather poor policy instrument with which to raise retirement saving” 
(p. 1921).

Duflo and others (2006) report the results of a field experiment on matching and 
savings outcomes. This study offered clients of the U.S. tax preparation firm H&R Block 
the opportunity to use their federal tax refund to open an IRA. Some individuals were 
offered the opportunity to open such an account with no match; others were offered a 
match of either 20 percent or 50 percent on contributions up to $1,000. Figure 15.1 
shows the fraction contributing to an IRA and the amount contributed by those who 
chose to open an account. Only 3 percent of the study participants in the no-match group 
elected to open an IRA. With a 20 percent match, 8 percent opened an IRA, and with a 
50 percent match, 14 percent opened an IRA.
FIGURE 15.1  Evidence on the effect of matching and saving from the H&R Block experiment
a. Percentage contributing to IRA 
b. Amount contributed to IRA
no match 
20% match  50% match 

no match 
20% match  50% match 
SOURCE: Dufl o and others 2006.
The magnitude of the effects estimated by Duflo and others (2006) is strikingly 
similar to that estimated by Engelhardt and Kumar (2007), even though the two studies 
examined different mechanisms (saving out of a tax refund versus enrolling in an employer-
sponsored savings plan) and different types of individuals (middle-income H&R Block 
clients versus older Health and Retirement Study survey respondents). Engelhardt and 
Kumar estimate that increasing the match rate by 25 percent of contributions increases 
savings plan participation by about 5 percentage points; Duflo and others estimate that 
increasing the match rate from 0 to 20 percent of contributions increases savings plan 
participation by 5 percentage points, and increasing the match rate from 20 percent to 
50 percent of contributions increases participation by 6 percentage points.
Mills and others (2008) report the results from a different multiyear field experi-
ment on saving in individual development accounts (IDAs) in the United States. Lower-
income families (income of less than 150 percent of the poverty level) were randomly 
assigned to either a treatment or a control group. Members of the treatment group were 
allowed to open an IDA to which contributions of up to $750 per year were potentially 
matched. Members of the control group were not allowed to open an IDA. One difference 
between this program and other savings schemes is that contributions were matched upon 

withdrawal, with the rate of the match dependent on the purpose of the withdrawal. Con-
tributions withdrawn to purchase a home were matched 200 percent, whereas contribu-
tions withdrawn for other qualified purposes, such as education, starting a business, home 
improvement, or retirement saving, were matched 100 percent. Contributions withdrawn 
for nonqualified purposes were not matched.
Overall, the results indicate that there is no significant relationship between IDA 
participation and net worth (figure 15.2). For most of the distribution, the effect is small 
but negative; in the upper and lower quantiles, the point estimates are positive, and some-
times large, but never statistically significant. These results challenge the effectiveness of 
match-based savings schemes for increasing the net worth of very low-income families. 
FIGURE 15.2  Impact on Net Worth of Opening and Contributing to an Individual Development 
Account after Three Years

5  10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 
treatment effects at wave 3 ($) 
quantile treatment effect 
average treatment effect 
95% confidence interval for quantile treatment effect 
SOURCE: Mills and others 2008.
Choi and others (2002, 2004b, 2006) adopt the natural experiment approach to 
analyze the impact of matching on savings outcomes. They examine two companies with 
employer-sponsored savings plans that changed their employer match: one added a match 
to a plan that did not previously have one, and one increased its match threshold while 
keeping its match rate constant. This approach uses individual behavior before the changes 
as a control for employee behavior after the changes in the matching formulas as a way to 
address concerns about the endogeneity of individual savings preferences with respect to 
the generosity of the employer match.
The first company (Firm A) introduced a 25 percent match on employee contribu-
tions up to 4 percent of income in October 2000; before that date, the plan offered no 
match. Using data on employees hired up to 26 months before the plan change and up 
to 14 months after the plan change, Choi and others estimate a hazard model of the time 

from hire to the date of initial savings plan participation. They find that the introduc-
tion of the employer match increased the rate at which employees enrolled in the savings 
plan by about 25 percent. However, because participation rates at this company were low 
before the introduction of the match, the absolute magnitude of the estimated participa-
tion increase was not large. For example, their model predicts that the 25 percent match 
adopted by this firm leads to a 4.7 percentage point increase in savings plan participation 
for 40-year-old men with three years of tenure. This effect is roughly in line with the effect 
estimated by Engelhardt and Kumar (2007) and Duflo and others (2006).
The second company (Firm B) increased the match threshold in its savings plan 
in January 1997 while keeping its match rate constant. Before January 1997, unionized 
employees received a 50 percent match on the first 5 percent of income contributed to the 
savings plan, and nonunion employees received a 50 percent match on the first 6 percent 
of income contributed. In January 1997, the match threshold for both groups of employ-
ees was increased by 2 percent—from 5 percent to 7 percent of pay for union employees 
and from 6 percent to 8 percent of pay for nonunion employees. Contributions up to the 
new threshold were still matched at 50 percent.
Using data on employees hired up to one year before and one year after the plan 
change, Choi and others estimate a hazard model of the time from hire to the date of 
initial savings plan participation. They find no significant impact of the increase in the 
match threshold on savings plan participation. This result is consistent with the theoreti-
cal arguments outlined earlier, which posit that an increase in the match threshold does 
not affect the marginal incentives to participate in the savings plan. As expected, Choi and 
others find no effect on participation of such a plan change.
The more interesting results in Choi and others (2002, 2004b, 2006) address the 
impact of the match threshold on savings plan contributions. Figure 15.3 shows the distri-
bution of contribution rates in the savings plan at Firm A for participants who joined the 
plan when it had no match and for participants who joined the plan after it introduced a 
25 percent match on employee contributions up to 4 percent of income. With no match, 
the most frequently chosen contribution rates were 5 percent, 10 percent, and 15 percent 
of income—numbers that are multiples of 5. After the employer match, many partici-
pants also chose contribution rates that were multiples of 5. In addition, there was a large 
increase in the fraction of participants who made a 4 percent contribution, the new match 
threshold. In the absence of an employer match, very few employees chose to participate 
in the savings plan at a 4 percent contribution rate; with the employer match, the 4 per-
cent match threshold became the modal contribution rate.
The distribution of contribution rates at Firm B, which increased its match thresh-
old, exhibits a similar pattern. Figure 15.4 shows the distribution of contribution rates 
to the savings plan for two groups of participants: those who joined the plan in the nine 
months before the increase in the match threshold, and those who joined the plan over a 
similar period of time after the increase in the match threshold. As in figure 15.3, there 
are clear spikes in the distribution of contribution rates both before and after the change 
in the match threshold at multiples of 5 (5 percent, 10 percent, 15 percent, 20 percent, 
and 25 percent of pay). And, as in figure 15.3, the modal contribution rate under both 
distributions is at the match threshold: 5 percent or 6 percent of pay before the change in 
the match threshold and 7 percent or 8 percent of pay after the match threshold. 

Figure 15.5 examines the impact of the increase in the match threshold of the Firm B 
savings plan for individuals participating in the plan before the match threshold changed. 
It shows how the contribution rates of these participants evolved over time after the plan 
change. The sample in figure 15.5 is restricted to employees contributing to the Firm B 
savings plan nine months before the increase in the match threshold. As in figure 15.4, a 
large proportion of participants (more than 45 percent) start with a contribution rate of 
5 percent or 6 percent of pay. The switch from the old threshold to the new threshold is 
clearly apparent: there is an immediate shift from the old threshold (5 percent or 6 per-
cent of pay) to the new threshold (7 percent or 8 percent of pay) when the match thresh-
old change occurred, in January 1997, and a slower adjustment over the next three years, 
as more and more participants shifted from the old to the new threshold. In contrast, 
the fraction of participants at the other contribution rates remained fairly stable over the 
entire time period.
The patterns in figures 15.3, 15.4, and 15.5 reveal the behavioral nature of savings 
plan participation. The fact that the contribution rates spike at multiples of 5 suggests 
an important role for focal points in savings choices. When individuals face compli-
cated decisions, such as deciding how much to save, they adopt heuristics to simplify the 
decision-making process. This pattern of contribution rate outcomes suggests that one 
such heuristic is to winnow the set of potential contribution rates to a subset of the pos-
sible options—in this case, those that are multiples of 5. The predominance of the match 
threshold in the distribution of contribution rates suggests that it also serves as a focal 
point in participants’ considerations about how much to save. The kink in the budget set 
generated by the match threshold would be expected to result in bunching at the match 
threshold, absent any behavioral considerations. But it is likely that the match threshold 
gets additional consideration as participants evaluate how much to save because it serves 
as a natural focal point (precisely because it is where the financial incentives to save 
change); individuals may also view the match threshold as carrying an implicit recom-
mendation about how much they should save; this endorsement effect would further 
reinforce the focal nature of the match threshold. Finally, the slow movement of existing 
participants away from the old match threshold and toward the new match threshold in 
figure 15.5 suggests inertia on the part of savings plan participants. Such inertia in sav-
ings plan outcomes has been well documented (see Beshears and others 2008 for a review 
of this literature). It is also consistent with participants’ anchoring on the original match 
Perhaps the most surprising finding in the literature on matching and savings plan 
outcomes is that even with a match, participation rates are often surprisingly low (Choi, 
Laibson, and Madrian 2011). Collectively, the research on matching and savings out-
comes suggests that at best, increasing the match rate on savings leads to small increases in 
participation and contributions conditional on participation. The more important match-
related tool is the match threshold, which serves as a strong focal point as individuals 
decide how much to save. A lower match rate with a higher match threshold may be a 
more effective way to increase individual contributions than a higher match rate with a 
lower match threshold—that is, providing a match of 25 percent on contributions up to 
10 percent of pay will induce individuals to save more than a match of 50 percent up to 
5 percent of pay at a similar (or lower) cost to the organization providing the match.

FIGURE 15.3  Distribution of contribution rates at a firm that added an employer match: Firm A

1 2 3 4 5 6 7 8 9 10 
year-end contribution rate (%) 
no match    
match (25% up to 4% of pay)   
no match: spikes at 5%, 10%, 15% 
% of employees
match: spike
at 4% match
SOURCE: Choi and others 2006.
FIGURE 15.4  Distribution of initial contribution rates at a firm that changed its match threshold: 
Firm B
% of participants

1  2  3  4 5–6 
7–8 9  10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 
initial contribution rate 
initial participants
April–December 1996
(threshold 5% or 6%)  
initial participants 
April–December 1997
(threshold 7% or 8%) 
SOURCE: Choi and others 2004b.
FIGURE 15.5  Evolution of contribution rates over time: Firm B
% of participants

SOURCES: Choi and others 2002, 2006.

Complementary and Alternative Approaches to
Increasing Saving
The literature on behavioral economics and savings plan outcomes suggests several alter-
native, and potentially more cost-effective, strategies to increase individual saving. This 
section reviews some of these approaches.
By far the most effective method to increase participation in defined contribution savings 
schemes is automatic enrollment. The research on participation in employer-sponsored 
savings plans in the United States shows that participation rates are substantially higher 
when the default is enrollment in the savings plan (that is, individuals must opt out if they 
prefer not to save) than it is when individuals must take action to participate in the sav-
ings plan. The impact of automatic enrollment on participation rates can be sizable. In the 
first study of the impact of automatic enrollment on savings outcomes, Madrian and Shea 
(2001) document a 50 percentage point increase in savings plan participation for newly 
hired employees (less than 15 months of tenure) at a large employer that switched from an 
opt-in to an opt-out automatic enrollment regime. Other studies also document signifi-
cant increases in participation as a result of automatic enrollment (see Beshears and others 
2008; Choi and others 2002, 2004a, 2004b; Nessmith and others 2007). The impact of 
automatic enrollment is greatest for groups with the lowest saving rates initially: younger, 
lower-income workers.
Matching is not completely irrelevant in plans that have automatic enrollment. A 
more generous match is associated with higher participation rates, with effects that are 
roughly in line with those discussed earlier in the context of savings schemes without 
automatic enrollment.
Beshears and others (2010) take two different approaches to evaluating the impor-
tance of the match in employer-sponsored savings plans that have automatic enrollment. 
First, they examine a firm that replaced its employer match of 25 percent on the first 
4 percent of pay contributed to the plan with a noncontingent employer contribution 
(that is, the firm made a savings plan contribution on behalf of all employees, regardless 
of whether employees made any contributions of their own to the savings plan). They 
estimate that eliminating the employer match reduced participation by at most 5–6 per-
centage points, an effect very similar to that estimated by Engelhardt and Kumar (2007), 
Duflo and others (2006), and Choi and others (2002, 2004b, 2006) for similar changes in 
the match rate in savings plans without automatic enrollment.
The second approach taken by Beshears and others (2010) in evaluating the impact 
of matching in savings plans with automatic enrollment is to exploit variation in the 
match structure both within (for firms that changed their matching policy) and across a 
sample of nine firms with employer-sponsored savings plans with automatic enrollment. 
This analysis is potentially confounded by endogeneity between the generosity of the 
match and employee saving preferences; in addition, the sample of firms included in the 
analysis is small. With these caveats in mind, Beshears and others find that a 1 percentage 
point increase in the maximum potential match as a fraction of salary is associated with a 
2–4 percentage point increase in savings plan participation (figure 15.6). Based on these 

estimates, decreasing the match rate from the modal match in employer-sponsored sav-
ings plans in the United States of 50 percent on the first 6 percent of pay to 25 percent on 
the first 6 percent of pay (a reduction in the match rate of 25 percentage points) is pre-
dicted to reduce savings plan participation under automatic enrollment by 3–6 percentage 
points. This estimate aligns with that from the single firm case study discussed in Beshears 
and others (2010); it is also consistent with the studies of similar match changes in savings 
plans without automatic enrollment discussed earlier.
FIGURE 15.6  Matching contributions and savings plan participation in firms with automatic 
0 1 2 3 4 5 6 7 8 9 
1   ...   20   21   22   23   24   25   26   27   28

Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©azkurs.org 2016
rəhbərliyinə müraciət

    Ana səhifə