Understanding the role that
Understanding the role that
In this analysis, I use the 1992 Health and Retirement Study (HRS) to examine the individual and job characteristics that are associated with asset choice in defined contribution plans. The HRS is one of the few surveys that records whether participants have choice over their defined contribution assets. I find that men and women are equally likely to have investment choice in their pension plan and that choice does not vary by marital status. Employees' ability to choose their own investments increases with education and family net worth. Workers in managerial positions are more likely to have choice than are workers in other occupations, and employees in larger establishments and larger firms are also more likely to have investment choice.
I also examine whether having choice over pension investments increases contributions to defined contribution plans. As more workers rely on defined contribution plans, a key issue in retirement income security is to encourage sufficient contributions. (Often employee contributions are required before the employer will make a matching contribution.) A key econometric issue in estimating the effect of choice is the potential endogeneity of choice; that is, do participants with some financial sophistication work at firms that offer plans with investment choice? If so, then it will be difficult to infer the effect of choice on the general population. I address this issue econometrically in two different ways. My preferred estimates indicate that a participant with choice contributes over 8.5 percentage points more annually to their defined contribution plan than a comparable, randomly selected participant without choice. This effect is estimated fairly precisely, and it is the largest effect on contributions. Single and married women are estimated to contribute more (0.83 and 1.03 percentage points, respectively) than married men. Older participants contribute more, but the effect is quite small economically. I also find that the benefits of tax-deferred saving are distributed fairly evenly across income levels. Finally, I submit these results to a series of robustness checks.
From a policy perspective, it is important to understand which plan features encourage employee participation in defined contribution plans. This article adds to a growing literature that suggests that plan attributes other than the employer match rate can play a role in increasing participant contributions. Loan provisions and asset choice may encourage contributions even as employers reduce or eliminate matching provisions in their
The striking growth of
Recent work points to the importance of plan features in encouraging plan participation and contribution rates. These characteristics include the presence and size of an employer match rate, the fraction of salary matched, participant choice over asset allocation, and loan provisions. Plan features may have unintended consequences as well. For example, the structure of many of the
Participant-friendly features, such as asset choice and loan provisions, may encourage or maintain participation levels during periods in which companies discontinue their contributions to
The behavior of participants in self-directed individual retirement accounts is also relevant in the discussions of adding personal accounts to Social Security. Further, if future Social Security benefits are reduced, there will be an increased role for personal saving in financing retirement with tax-deferred plans that are not related to employment, such as individual retirement accounts (IRAs).
This article
Prior to the availability of
Companies are encouraged to provide a diverse offering of assets by section
Perhaps due in part to this regulation, participant direction grew during the 1990s. Wiatrowski (2000) summarizes the trends in investment choice for full-time employees using several years of the Bureau of Labor Statistics' National Compensation Survey. He reports that, in 1985, 90 percent of full-time employees had investment choice over their own contribution, and 48 percent had control over their employers' contribution. By 1997, there was a slight drop in the percentage who could control their own contributions (87 percent), but more than 65 percent had choice over their employers' contribution. Using these data, Wiatrowski also finds that a smaller percentage of participants may choose company stock as an investment option. In 1985, for example, 70 percent of employees could choose employer stock for their contribution, and 61 percent could choose employer stock for their employer's contribution. In 1997, those figures were 42 and 25, respectively.
Recent work relates plan features to participation. Using the 1998 Survey of Consumer Finances, Munnell, Sundén, and Taylor (2001) find that the ability to borrow from the plan increases the
Other work indicates that many participants do not change the default choices firms make for them. Choi and others (2002), using administrative data from three companies with automatic enrollment in their
This section presents estimates of the determinants of pension plan investment choice and the effect of asset choice on contributions. I use data on pension participants of preretirement age from Wave 1 (1992) of the Health and Retirement Study (HRS). The respondents are aged
The 1992 HRS is a detailed survey that includes pension questions on up to three defined benefit and three defined contribution plans. I restrict the sample to those defined contribution participants who answer the following question about each of their defined contribution plans: Were you able to choose how the money in your account is invested? In the typical regression, there are 1,690 individual defined contribution plan participants and 180 multiple plans, for a total of 1,870 observations.
Note that the HRS question does not explicitly distinguish employer from employee contributions, so it is possible that the employee would report having choice over his or her contributions, even if the employer directs the firm's contributions. Unfortunately, there is no firm-reported pension information in the publicly available HRS data.
In this section, I relate individual characteristics, and the limited characteristics of firms available in the HRS, to the participant-reported ability to choose pension investments.4 Table 1 presents summary statistics for this HRS sample. About 59 percent of the 1,983 respondents report having choice over investments in their pension plan, and they contribute, on average, about 5.05 percent of salary. Single women account for about 13 percent of the sample; single men, about 7 percent; married women, 32 percent; and married men, about 48 percent. The average age in the sample is 54 years. About 58 percent of this sample reports having an individual retirement account, and 42 percent have a defined benefit pension plan in addition to their defined contribution plan—either with the current or a former employer.
Dependent variable | Mean | Standard deviation |
Observations |
---|---|---|---|
Choice | .589 | .492 | 1,983 |
Contribution percentage | 5.050 | 4.831 | 1,981 |
Single female | .128 | .334 | 1,981 |
Single male | .071 | .257 | 1,981 |
Married female | .318 | .466 | 1,969 |
Age | 54.456 | 4.850 | 1,981 |
Education | 13.488 | 2.542 | 1,981 |
Income (dollars) | |||
25,000–50,000 | .302 | .459 | 1,981 |
50,000–100,000 | .441 | .497 | 1,981 |
More than 100,000 | .168 | .374 | 1,981 |
Net worth (dollars) | |||
50,000–100,000 | .189 | .391 | 1,981 |
100,00–250,000 | .351 | .478 | 1,981 |
250,000–500,000 | .182 | .386 | 1,981 |
More than 500,000 | .094 | .292 | 1,981 |
Employee has— | |||
Defined benefit plan | .421 | .494 | 1,981 |
Individual retirement account | .581 | .494 | 1,981 |
Industry dummies | |||
Agriculture and mining | .046 | .210 | 1,908 |
Manufacturing | .275 | .446 | 1,908 |
Transportation | .096 | .295 | 1,908 |
Wholesale | .048 | .213 | 1,908 |
Retail | .074 | .262 | 1,908 |
Fire and services | .403 | .491 | 1,908 |
Public administration | .059 | .235 | 1,908 |
Occupation dummies | |||
Management | .238 | .426 | 1,922 |
Professional, specialist, and technical | .200 | .400 | 1,922 |
Sales | .079 | .269 | 1,922 |
Clerical | .209 | .406 | 1,922 |
Services | .049 | .217 | 1,922 |
Farming and construction | .027 | .162 | 1,922 |
Machine operators | .198 | .399 | 1,922 |
Number of employees in participant's work location | |||
Less than 100 | .450 | .498 | 1,913 |
100–499 | .276 | .447 | 1,913 |
500 or more | .273 | .446 | 1,913 |
Number of employees in participant's firm a | |||
Less than 100 | .166 | .372 | 1,904 |
100–499 | .162 | .369 | 1,904 |
500 or more | .672 | .470 | 1,904 |
SOURCE: Author's calculations based on data from the 1992 Health and Retirement Study. | |||
a. Participant's estimate of firm's employment in all locations. |
What individual and job characteristics are associated with defined contribution plans that offer asset choice? Column 1 of Table 2 presents estimates of a linear probability model of choice as a function of individual and firm characteristics. The ability to choose pension investments does not appear to vary by sex or marital status (married men are the omitted category). Asset choice increases with years of education: a participant with 4 more years of education is 6.4 percentage points more likely to have investment choice than is an otherwise comparable participant. Family net worth between $250,000 and $500,000 is also associated with a higher probability of having choice (9.2 percentage points higher).
Dependent variable | Choice (1) |
Percentage of salary contributed |
||||
---|---|---|---|---|---|---|
Ordinary least squares (2) |
Two-stage ordinary least squares (3) |
|||||
Choice | 2.936 | *** | 8.551 | *** | ||
(.214) | (2.027) | |||||
Single female | -.050 | .579 | * | .825 | * | |
(.041) | (.355) | (.429) | ||||
Single male | .020 | .547 | *** | .441 | ||
(.049) | (.395) | (.450) | ||||
Married female | -.032 | .832 | *** | 1.032 | *** | |
(.030) | (.284) | (.339) | ||||
Age | -.00011 | .076 | *** | .080 | *** | |
(.002) | (.021) | (.026) | ||||
Education | .016 | *** | -.036 | -.132* | ||
(.006) | (.053) | (.072) | ||||
Income (dollars) | ||||||
25,000–50,000 | -.016 | .932 | *** | .995 | ** | |
(.045) | (.350) | (.440) | ||||
50,000–100,000 | .011 | 1.175 | *** | 1.059 | ** | |
(.048) | (.392) | (.479) | ||||
More than 100,000 | .015 | 1.005 | ** | .845 | ** | |
(.056) | (.497) | (.598) | ||||
Net worth (dollars) | ||||||
50,000–100,000 | .045 | .657 | ** | .352 | ||
(.038) | (.302) | (.402) | ||||
100,00–250,000 | .043 | 1.045 | *** | .732 | ** | |
(.036) | (.302) | (.399) | ||||
250,000–500,000 | .092 | ** | 1.806 | *** | 1.234 | ** |
(.042) | (.387) | (.504) | ||||
More than 500,000 | .057 | .900 | * | .589 | ||
(.053) | (.488) | (.566) | ||||
Employee has— | ||||||
Defined benefit plan | .134 | *** | 1.451 | *** | .540 | |
(.025) | (.228) | (.423) | ||||
Individual retirement account | .025 | .141 | .049 | |||
(.026) | (.227) | (.273) | ||||
Number of employees in participant's work location | ||||||
100–499 | .072 | ** | ||||
-.030 | ||||||
500 or more | .071 | ** | ||||
-.031 | ||||||
Number of employees in participant's firm a | ||||||
100–499 | .074 | * | ||||
-.043 | ||||||
500 or more | .107 | *** | ||||
-.038 | ||||||
Industry and occupation dummies | Yes | Yes | Yes | |||
Constant | .151 | -4.320 | -5.513 | |||
(.176) | (1.503) | (1.886) | ||||
Observations | 1,865 | 1,888 | 1,865 | |||
R2 | .1084 | .1980 | ||||
SOURCE: Author's calculations based on data from the 1992 Health and Retirement Study. | ||||||
NOTE: Standard errors that are robust to heteroskedasticity and to correlation across mulitple plans for an individual are in parentheses. | ||||||
* = statistically significant at the 10 percent level. ** = statistically significant at the 5 percent level. *** = statistically significant at the 1 percent level. |
||||||
a. Participant's estimate of firm's employment in all locations. |
The individual characteristic with the largest economic effect on the choice probability is having a defined benefit plan, either with the current employer or with a previous job. Having a defined benefit plan raises the probability of having asset choice in one's defined contribution plan by 13.4 percentage points. This suggests that individuals with a taste for saving, as evidenced by their pension participation, may prefer defined contribution plans with self-direction features. Such features allow them to choose pension assets to achieve their preferred asset allocation across tax-deferred and non-tax-deferred accounts.
The probability of having choice also varies for a few industry and occupation categories (these coefficients are not reported in Table 2). Workers in public administration are estimated to be 17 percentage points more likely than workers in agriculture and mining to have choice in their pension plan (none of the other industry coefficients differed from the omitted category). Workers in service, farming and construction, and machine operator occupations are estimated to be significantly less likely to have choice than workers in managerial positions (from 10 to 24 percentage points).
I also include two sets of firm-size dummies in the linear probability model of choice. The first set is based on the number of employees in the participant's work location, and the second set is the participant's estimate of the firm's employment in all locations. The four dummy variables that are included are statistically significant in the choice equation, suggesting that participants in larger firms (more than 99 employees) are between 7 and 10 percentage points more likely to have investment choice in their pension plan.
Since choice is a binary variable, I also estimate a probit model containing the same explanatory variables as in Table 2. The estimates, reported in column 1 of Table 3, are qualitatively similar to those from the linear probability model. In particular, the direction and statistical significance of the coefficients are the same. As in the linear probability model, firm size has an important effect on choice.
Dependent variable | Choice (1) |
Percentage of salary contributed (instrumental variables) (2) |
||
---|---|---|---|---|
Choice | 9.096 | *** | ||
(2.001) | ||||
Single female | -.137 | .847 | * | |
(.110) | (.441) | |||
Single male | .067 | .427 | ||
(.132) | (.463) | |||
Married female | -.085 | 1.052 | *** | |
(.083) | (.347) | |||
Age | -0.00062 | .081 | *** | |
(.0068) | (.027) | |||
Education | .045 | *** | -.141 | * |
(.016) | (.073) | |||
Income (dollars) | ||||
25,000–50,000 | -.047 | 1.001 | ** | |
(.120) | (.455) | |||
50,000–100,000 | .019 | 1.046 | ** | |
(.129) | (.494) | |||
More than 100,000 | .037 | .827 | ||
(.156) | (.615) | |||
Net worth (dollars) | ||||
50,000–100,000 | .123 | .325 | ||
(.101) | (.415) | |||
100,00–250,000 | .117 | .703 | * | |
(.099) | (.411) | |||
250,000–500,000 | .263 | ** | 1.182 | ** |
(.120) | (.517) | |||
More than 500,000 | .155 | .560 | ||
(.150) | (.581) | |||
Employee has— | ||||
Defined benefit plan | .371 | *** | .452 | |
(.068) | (.422) | |||
Individual retirement account | .068 | .040 | ||
Number of employees in participant's work location | ||||
100–499 | .202 | ** | ||
(.082) | ||||
500 or more | .203 | ** | ||
(.090) | ||||
Number of employees in participant's firm a | .191 | * | ||
100–499 | (.113) | |||
.283 | *** | |||
500 or more | (.101) | |||
Industry and occupation dummies | Yes | Yes | ||
Constant | -.952 | -5.636 | ||
(.484) | (1.940) | |||
Observations | 1,865 | 1,865 | ||
SOURCE: Author's calculations based on data from the 1992 Health and Retirement Study. | ||||
NOTE: Standard errors that are robust to heteroskedasticity and to correlation across mulitple plans for an individual are in parentheses. | ||||
* = statistically significant at the 10 percent level. ** = statistically significant at the 5 percent level. *** = statistically significant at the 1 percent level. |
||||
a. Participant's estimate of firm's employment in all locations. |
Pension investment choice may affect the level of participation in a pension plan. Some behavioral theories of saving suggest that a change in the economic environment, such as requiring participants to choose assets in a pension plan, may stimulate saving.5 With employer-provided pensions in place, individuals are more likely to learn that others think saving is important. This section presents models explaining the employee's contribution percentage to the defined contribution plan as a function of choice and of individual and employment characteristics. In this HRS sample, the average contribution percentage is 5.05 percent of salary, with a standard deviation of 4.81. The median contribution is 5.00 percent, and the mean of those who contribute is 7.10 percent, with a standard deviation of 4.27. About 29 percent (573) report a zero contribution.
Simple tabulations indicate that employees are more likely to participate when investment choice is present. While 50.37 percent of those without choice report a zero contribution, only 13.76 percent of those with choice report a zero contribution.
Table 2 also presents estimates of linear models of the percentage of salary the participant contributes to his or her defined contribution plan. The ordinary least square estimates are in column 2. Standard errors are corrected for heteroskedasticity. Controlling for individual characteristics, financial characteristics, and industry and occupation, choice over pension assets is estimated to increase the annual contribution by 2.9 percentage points. This increase, a 43 percent increase relative to the unconditional mean contribution of 5.05 percent, is an economically large effect and is precisely measured.
Having a defined benefit plan or an individual retirement account also increases the percentage contributed. These dummy variable coefficients indicate a taste for saving: a participant with a defined benefit plan is predicted to contribute about 1.5 percentage points more to his or her defined contribution plan. The sign of the coefficient on the IRA dummy is also positive but is imprecisely measured. The percentage of salary contributed is greater for each included category of income and net worth relative to the omitted categories (income less than $25,000 and net worth less than $50,000), but the difference is generally close to 1 percentage point. (The largest effect is 1.8 percentage points for the highest net worth category—$250,000 to $500,000.) This suggests that the benefits of tax-deferred saving, in percentage terms at least, are fairly evenly spread among the medium- to high-income participants.
These OLS estimates indicate that participant-direction of pension assets has a statistically significant and economically large effect on pension contributions. However, one might argue that choice is an endogenous variable in these regressions. That is, participants with some financial sophistication and taste for saving join firms that offer plans with investment choice. Unobserved saving heterogeneity may remain in the error term despite my attempt to control for saving propensity by including the ownership of an individual retirement account and participation in a defined benefit plan. Pension plans with participant-direction features may be more common in certain industries and occupations. I include those dummy variables as well to allow individuals to sort on that basis. Unfortunately, there are a limited number of pension plan features in the publicly available HRS data.
Ideally, we could find one or more instrumental variables (IV) for the choice variable in the contribution equation. Such a variable must be exogenous in the contribution equation, that is, it must be properly omitted from the equation and uncorrelated with unobservables, such as taste variables, in that equation. In addition, the IV candidates must be partially correlated with choice. Instrumental variables are difficult to come by without some kind of natural experiment that would exogenously cause some firms to offer choice when they might not have otherwise. In the publicly available HRS, the possibilities are rather limited. Nevertheless, the results for the linear probability choice models are suggestive. In particular, firm size has a significant effect on choice. It may also be reasonable to assume that although individuals with a taste for saving may select different industries and occupations, there is no systematic sorting of those with a taste for saving into certain firm sizes, either by their work location or number of employees in the entire firm. As with most applications of instrumental variables, this assumption can be questioned. However, I can partly test exogeneity of the firm-size variables using a test of the overidentifying restrictions; the results are reported below.
Column 3 of Table 2 contains two-stage least squares (2SLS) estimates of the contribution equation, where the four firm-size dummies included in column 1 are used as instrumental variables for choice. The 2SLS estimate of the effect of choice is substantially larger than the OLS estimate, suggesting that a participant with choice contributes over 8.5 percentage points more annually to his or her defined contribution plan than a comparable participant without choice. This effect is estimated fairly precisely, and it is the largest effect. Single and married women are estimated to contribute more (0.83 and 1.03 percentage points, respectively) than married men. Older participants contribute more, but the effect is quite small economically. Participants with higher income and higher net worth are estimated to contribute about 1 percentage point more than those in the lowest income and net worth categories, as in the OLS estimates. The 2SLS coefficient on the defined benefit indicator is positive but not statistically significant. Apparently, the real causal effect of having a defined benefit plan is zero.
The difference between the OLS and 2SLS estimates of the choice coefficient are practically large. Nevertheless, because the 2SLS standard error is about 10 times larger than the OLS standard error, the difference between OLS and 2SLS could be due to sampling error. That is not the case here. I use a regression-based Hausman test, made robust to heteroskedasticity, to determine whether the difference between OLS and 2SLS is statistically significant. The statistic is computed by obtaining the reduced form residuals from the linear probability model for choice and then including them as a regressor in the contribution equation. The expanded equation is estimated by OLS, and the heteroskedasticity-robust t statistic on the reduced form residuals is a valid test statistic. Assuming that the firm-size dummies are exogenous, the null hypothesis is that choice is exogenous. A significant t statistic on the reduced form residuals rejects exogeneity of choice (Wooldridge 2000). When I carry out this test, the coefficient on the reduced form residual is −5.513 and its
The story about unobserved taste for saving being positively correlated with choice means we would expect OLS to have an upward bias. On average, then, we would expect the 2SLS estimate to be smaller than the OLS estimate. There are several reasons the opposite might occur. First, of course, firm size might not be exogenous, in which case the 2SLS estimates could have an upward bias. (I offer a test of this below.) If firm size is positively correlated with unobserved taste for saving, we expect an upward bias. Because the correlation between choice and firm size is not perfect, a modest amount of correlation between firm size and taste for saving can lead to a large asymptotic bias in 2SLS.6 A second possibility is that choice is measured with error, in which case OLS could have a downward bias. Unfortunately, the direction of bias for OLS is unclear in this application, as choice is a binary variable that cannot satisfy the classical errors-in-variables model. Still, when IV estimates are unexpectedly higher than OLS estimates, measurement error is often cited as a possibility.
A third, more subtle possibility comes from the literature on treatment effect. In a simple bivariate setting, Imbens and Angrist (1994) characterize the probability limit of the IV estimator of the treatment effect of a binary endogenous explanatory variable. In my application, the Imbens and Angrist results imply that the IV estimator consistently estimates the average effect for participants whose choice status is induced by a change in firm size. It could be that this effect is larger than the effect for the population as a whole.7
Since I have one endogenous variable—choice—and four instruments (the firm-size dummies in column 1 of Table 2), I am able to test the three overidentifying restrictions. The heteroskedasticity-robust regression-based statistic is 5.036. Under the null hypothesis that all instrumental variables are uncorrelated with the structural error, this is the outcome of a
Finally, rather than using a standard 2SLS procedure, where the reduced form of choice is linear, I use the fitted choice probabilities from the probit model in Table 3 as a single instrumental variable for choice in the contribution equation. This is the most efficient instrumental variable, since the fitted probabilities are the best predictor of choice. These instrumental variable estimates are reported in column 2 of Table 3 (they are generally similar to the 2SLS estimates in Table 2).
In this paper, I use an instrumental variable approach to the problem of potential endogeneity of investment choice in a contribution equation. My preferred estimates indicate that a participant with choice contributes over 8.5 percentage points more annually to his or her defined contribution plan than does a comparable participant without choice. This is an economically large effect: the unconditional mean of contributions is about 5 percentage points of salary. I also find that the benefits of tax-deferred saving are distributed fairly evenly across income levels.
From a policy perspective, it is important to understand which plan features encourage employee participation in defined contribution plans. This article adds to a growing literature that suggests that plan attributes other than the employer match rate can play a role in increasing participant contributions. Loan provisions and asset choice may encourage contributions even as employers reduce or eliminate matching provisions in their