# Retirement Savings Inequality: Different Effects of Earnings Shocks, Portfolio Selections, and Employer Contributions by Worker Earnings Level

by
Social Security Bulletin, Vol. 78 No. 3, 2018

After the Great Recession of 2007–2009, 64 percent of higher-earning workers and 56 percent of lower earners experienced increases in their accumulated retirement savings. For our 2009–2011 study period, we match Survey of Income and Program Participation data to Social Security Administration earnings records to examine retirement savings outcomes by earnings level and to identify factors that may underlie differences. The number of years with an earnings loss of 10 percent or more, the number of nonemployment spells, a decrease in employer contributions to a worker's defined contribution retirement plan, and less diversified investment portfolios barely affect the accumulated savings of higher earners, but are associated with decreased savings for lower earners. These differences may contribute to a growing retirement wealth gap.

Joelle Saad-Lessler is the Associate Industry Professor in the School of Business at Stevens Institute of Technology. Teresa Ghilarducci is the Bernard L. and Irene Schwartz Professor of Economics and Director, Schwartz Center for Economic Policy Analysis at the New School for Social Research in New York. Gayle Reznik is with the Office of Retirement Policy, Office of Retirement and Disability Policy, Social Security Administration.

Acknowledgments: A previous version of this article was published as Schwartz Center for Economic Policy Analysis at The New School for Social Research Working Paper No. 2017-9 (http://www.economicpolicyresearch.org/images/docs/research/retirement_security/DB_Wealth_Inequality_WP.pdf). The authors thank the National Endowment for Financial Education for research funding; and seminar participants at Rutgers University–New Brunswick and Baruch College of the City University of New York, and reviewers at the Social Security Administration's Office of Retirement and Disability Policy, for their valuable comments.

The findings and conclusions presented in the Bulletin are those of the authors and do not necessarily represent the views of the Social Security Administration.

## Introduction

 DC defined contribution IHS inverse hyperbolic sine IRA individual retirement account IV instrumental variable OLS ordinary least squares SIPP Survey of Income and Program Participation SSA Social Security Administration

This study examines how the accumulated discretionary retirement savings of workers differed by earnings level in the first years after the Great Recession. We look specifically at workers' combined holdings in individual retirement accounts (IRAs), Keogh plans, and, in particular, employer-sponsored defined contribution (DC) plans.1 Analyzing the economic experiences of workers during 2009–2011 reveals that higher earners were more likely to accumulate greater retirement savings than lower earners were. Sixty-four percent of workers at the top of the earnings distribution experienced an increase in retirement savings compared with 56 percent of those at the bottom. Higher earners may have fared better because of more favorable economic and life events and because higher and lower earners exhibit different voluntary contribution behaviors (Gist and Hatch 2014).

This study uses panel data to investigate changes in retirement savings from 2009 through 2011, and the determinants of those changes, by workers' earnings levels.2 Understanding how workers' earnings levels predict their ability to increase their retirement savings could inform changes to DC plan features that might help lower earners save in volatile economic conditions and slow or reverse the growth in the retirement wealth gap.

We report four key findings. First, each instance of annual earnings loss of 10 percent or more through 2009 was associated with a loss of retirement savings of $450 during 2009–2011 for lower earners, while the effect was negligible for higher earners. Second, for every week a worker was not employed during 2009–2011, lower earners lost$55 in retirement savings, but nonemployment spells did not affect higher-earning workers' savings. Third, diversification in retirement-asset allocation increased the savings of lower earners but had no significant effect on higher earners' savings.3 And fourth, employer contributions increased lower earners' DC plan wealth but had no significant effect on higher earners' DC plan wealth.

This article consists of six sections, beginning with this introduction. The second section describes the importance of examining changes in retirement savings for workers of different earnings levels; the third and fourth sections respectively describe the data and the estimation strategies. The fifth section presents the results of decomposition and regression analyses and discusses the robustness checks; the final section examines policy implications and concludes.

## Factors Affecting Retirement Savings Differ by Earnings

Previous studies have examined how earnings affect retirement wealth accumulation. Dushi, Iams, and Tamborini (2011) reported that earners at the lower end of the earnings distribution are much less likely to participate in DC pensions, and that those who do participate contribute a lower share of their earnings than higher earners do. Smith, Johnson, and Muller (2004) found that retirement-plan participation rises with increases in own earnings, family income, and age; and with being a homeowner, the birth of a child, and having a spouse with health problems. Conversely, changing jobs, having unemployment spells, and having greater numbers of children reduced participation. That study also found that retirement-plan participation responds to plan design features, such as whether loans are allowed and whether the employer matches contributions. Dushi, Iams, and Tamborini (2013) were the first to evaluate the effect of a significant earnings loss—defined as a drop of 10 percent or more—on retirement savings. They found that DC plan participants experiencing such a significant earnings loss during the Great Recession of 2007–2009 were more likely to have stopped contributing to their plan by 2009 than were those with stable earnings, and that overall, their contributions decreased substantially. Dushi and Iams (2015) similarly found that significant earnings losses and job changes depressed contributions during the Great Recession and in the preceding 2-year period. In Ghilarducci, Saad-Lessler, and Reznik (2017), we found that inertia in contribution behavior depended on whether workers' earnings decreased or increased during 2009–2011.

This study expands the understanding of the role of earnings in retirement wealth accumulation beyond DC plan participation and contribution rates. Using panel data from the Survey of Income and Program Participation (SIPP), we identify how workers' earnings levels affected the resilience of their retirement savings accumulation during 2009–2011.

## Data

We construct a study sample from SIPP data matched to Social Security Administration (SSA) longitudinal earnings records. We merge data from waves 1 through 11 of the 2008 SIPP panel and identify respondents who remained continuously in the sample, were aged 25–61 in 2009, had nonzero retirement savings in 2009, and worked in both 2009 and 2011.

Data on retirement savings are from responses to SIPP wave 4 and wave 10 questions about the market value of IRA, Keogh plan, and DC plan (such as 401(k), 403(b), and thrift plan) account balances held by respondents in 2009 and in 2011. Data on DC plan design features, including plan type, employer contribution provisions, preretirement loan provisions, and choice of investment allocation are from the SIPP Retirement Expectations and Pension Plan Coverage modules, which were fielded in waves 3 and 11 (in 2009 and in 2011–2012, respectively). The combined data on pension plan design features from wave 3 and on retirement savings from wave 4 are compared with combined data on savings from wave 10 and design features from wave 11.4 This process yields panel data on changes in retirement savings and pension plan design features over a 3-year period. Details on variable construction are available in Appendix A.

A 3-year panel is short—which limits opportunities to make definitive conclusions—but it contains high-quality comprehensive data on the determinants of short-term changes in retirement savings. One particular advantage of using the 2008 SIPP panel is that it fielded the Retirement Expectations and Pension Plan Coverage module in two waves, providing longitudinal data on every aspect covered therein. Previous SIPP panels fielded that rich module only once.

To maximize the accuracy of the survey results, SIPP respondents are asked to check their records before they begin answering questions regarding their income.5 Respondents can report the value of their retirement accounts as a number or a range, and when they provide a range, we impute the precise value. In our sample, about 50 percent (or more) of IRA, Keogh, and DC plan account balances are imputed (Table 1). Notwithstanding the need for some imputations, the SIPP offers the best panel data available for workers aged 25–61 because, along with demographic information, it includes characteristics of each person's DC plan, including asset allocation and contribution rates and levels.

Table 1. Selected characteristics of SIPP respondents' retirement savings, by plan or account type, 2009 and 2011
Characteristic DC plan IRA Keogh plan
Percentage reporting plan balances within a range of values (data requiring imputation)
2009 50 46 72
2011 54 51 68
Percentage with a nonzero account balance
2009 84 45 3
2011 75 43 1
Average account balance (current $) 2009 40,818 19,459 1,207 2011 47,812 24,742 741 Change 2009–2011 6,994 5,283 -466 SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records. Table 1 also shows that the majority of SIPP respondents in the 2008 panel had nonzero balances in DC plans (84 percent in 2009, 75 percent in 2011). Close to half held IRAs (45 percent in 2009, 43 percent in 2011) and negligible shares held Keogh accounts (3 percent in 2009 and 1 percent in 2011). We restricted the study sample to respondents with a nonzero balance in 2009 in any of these types of retirement savings. Finally, Table 1 shows that DC plan wealth dominated the average balances of the three savings vehicles, accounting for about two-thirds of retirement savings, followed by IRAs (about one-third), then Keogh plans (less than 2 percent). Not surprisingly, the average nominal dollar amounts of account-balance changes during 2009–2011 were ranked in the same order. The linked SSA records (which are based on Internal Revenue Service Form W-2 records) contain data on current and lifetime earnings and annual employee contributions to DC retirement plans.6 The SSA records allow us to calculate the number of years in which a person's annual earnings fell by 10 percent or more and the standard deviation (or volatility) of annual earnings over the person's working life. We also use the SSA data to calculate the earnings brackets, changes in earnings, and employee DC plan contribution rates and amounts for 2009, 2010, and 2011. Employer contribution rates are self-reported by SIPP respondents and employer contribution amounts are derived from those SIPP data.7 Withdrawals, rollovers, and DC plan balances in 2009 and 2011 are also from the SIPP data. The 2008 SIPP panel contains 36,578 persons who were surveyed continuously from 2008 through 2012 and who had matched records in the administrative data. Of these, 19,017 persons had jobs in 2009 and 2011 and 10,554 had nonzero retirement savings in 2009. Restricting the sample to those who were aged 25–61 in 2009 leaves us with a sample of 9,508 respondents. We divide earners into three groups based on the 2009 bend points for the Social Security retirement benefit formula.8 Lower earners, comprising the bottom 55 percent of the earnings distribution, earned less than$53,796; higher earners, making up the top 12 percent of the earnings distribution, earned more than $106,800 per year. ## Estimation Strategies This section describes the decomposition and regression analyses. It also discusses the inverse hyperbolic sine (IHS) transformation of DC plan wealth, the possible endogeneity of employer contributions, proxies for portfolio allocation, and the descriptive data. ### Decomposition Analysis Factors affecting DC plan wealth include employee and employer contribution rates, account withdrawals, and rollovers related to job changes. Additional factors affecting DC plan wealth include management fees and the market performance of plan holdings; we refer to these factors as portfolio allocation effects. We use a decomposition analysis to evaluate the effect of each factor on DC plan wealth changes. Specifically, we divide the dollar amount of each of the following factors by the change in DC plan wealth from 2009 to 2011: employee contributions; employer contributions; plan withdrawals; rollovers; and residual effects, combining DC plan balance changes attributable to all other factors. We measure the effect of each determinant of change in DC plan wealth during 2009–2011 for two groups: workers whose plan balance increased and those whose balance decreased. ### Regression Analysis An ordinary least squares (OLS) regression identifies the factors affecting retirement savings of higher and lower earners using a model described in the following equation: $sinh −1 (Y)=a+bX+e$ , where Y is the change in retirement savings during 2009–2011, sinh−1(Y) is the IHS transformation of Y, and e is an error term. X comprises determinants of changes in DC plan wealth, including the employee and employer contribution rates in 2009, changes in the employee and employer contribution rates during 2009–2011, and measures of portfolio allocation in 2009. As noted above, the latter reflect market gains or losses and management fees. Very few people withdrew funds or rolled their accounts over in the study period, but when positive withdrawals and rollovers were reported, the values were extreme; therefore, we do not include indicators of withdrawals and rollovers in the regressions.9 The model is run separately for workers in the bottom 55 percent of the earning distribution and those in the top 12 percent to identify how the structure of DC plans and life events interact differently for higher and lower earners. ### The “Difference in Initial Wealth” Problem Pence (2006) noted the possible problem in evaluating changes in levels of wealth when initial levels of wealth are not the same across groups and when asset values change over time.10 We address the difference in initial wealth by transforming changes in DC plan holdings using the IHS transformation (Burbidge, Magee, and Robb 1988; Pence 2006). The IHS transformation is similar to the log transformation used by Poterba, Venti, and Wise (2015), but IHS allows for negative values of the transformed variable. Our dependent variable is negative when DC plan wealth decreases over the study period (which occurs for almost half of the sample), making a log transformation impractical. The IHS transformation scales the change in DC plan wealth to the initial balances, and thereby reduces the influence of outlying values of the dependent variable. This approach also reduces the effect of measurement error in self-reported DC plan balances and changes to that wealth. The coefficients estimated from the regression equation are converted into marginal effects using median values of DC plan wealth, which further reduces the effect of outlying values of the dependent variable. Standard errors for the marginal effects are computed using a bootstrap method with 50 replications. ### Employer Contributions Errors in employer contributions arise because those data are self-reported by workers rather than the employers themselves. If the reporting errors are random, they are absorbed in the error term and do not induce bias in the estimated coefficients. However, if self-reported errors are correlated with earnings, the estimated coefficients are biased. Such reporting errors may be endogenous if, for example, higher earners pay closer attention to their retirement accounts than lower earners do and therefore report their employers' contributions more accurately. We test for the endogeneity of self-reported employer contribution rates using average employer contribution rates and the shares of workers with nonzero employer contributions for each state, industry, and education level as instruments for employer contribution rates. If we cannot reject the null hypothesis, that means the employer contribution rate is exogenous. For the sample of lower earners, we cannot reject the null hypothesis that employer contributions are exogenous because the results of an instrumental variable (IV) corrected approach do not differ significantly from those of an OLS approach. We use OLS regression for the lower-earner sample because it yields more efficient estimates than IV regression does. For the sample of higher earners, we reject—at the 1 percent level—the null hypothesis that employer contributions are exogenous. Therefore, for this sample, we use instruments for employer contributions in an IV regression. ### Portfolio Allocation Proxies The SIPP asks respondents how their IRA, Keogh account, and DC plan balances are invested. For each account type, a respondent may report up to four investment choices, but not the amounts invested, which limits our knowledge of workers' portfolio allocation. Given this limitation, we create a variable measuring the degree of risk, with which we classify portfolio investment choices as either “safe” or “involving some risk.”11 The number of investment types reported indicates the portfolio's diversification. We supplement the portfolio risk and diversification measures with indicators of a worker's risk and return preferences. These include the number of years in which a worker lost more than 10 percent of earnings through 2009 and of weeks not worked during 2009–2011, as well as educational attainment, retirement savings as of 2009, lifetime earnings, and home equity. We regard these factors as indicators of risk tolerance because they reflect either an appetite for risk or an ability to withstand negative financial outcomes. We also consider the change in the number of children living in the family and the worker's responsibility for the household's financial well-being—the latter measured as a household income ratio, or the worker's personal income as a percentage of household income. These factors may indicate the presence of or changes in liquidity constraints, which can affect risk-taking and lead to more or less conservative investment strategies. Having high or low earnings affects wealth accumulation in textured and complicated ways. Access to trusted and accurate financial information and networks depends in part on socioeconomic status and community (Chong, Dow, and Phillips 2010). Compared with nonwhite workers, those who are white tend to have more access to and engagement with financial institutions and social networks and better information about investing; these factors may in turn affect the level of fees and the composition of the portfolio. The effect of having a business degree12 also varies by earnings class, though it is treated uniformly as a proxy for financial literacy regardless of socioeconomic class. Financial literacy is linked to choosing appropriate savings rates and asset allocations, discerning fees, and assessing risk (Lusardi and Mitchell 2014). ### Descriptive Data Higher earners were more likely than lower earners to have their retirement savings increase from 2009 to 2011 (64 percent versus 56 percent; Table 2). This was because the average DC plan contribution rate for higher earners in 2009 (6.11 percent of earnings) was more than twice that for lower earners (2.80 percent). In addition, the employers of higher earners also contributed at a higher average rate (2.86 percent) than did employers of lower earners (2.15 percent). Table 2. Summary statistics for lower- and higher-earner study samples Variable a Lower earners Higher earners Difference in means Mean Standard error Mean Standard error Retirement savings characteristics Percentage of workers whose retirement savings increased during 2009–2011 56 64 . . . Accumulated savings ($) 34,137 55,668 147,787 137,650 113,650***
DC pension plan contribution rate (%) of—
Employee 2.80 4.68 6.11 5.47 3.31***
Employer 2.15 5.93 2.86 6.41 0.71***
Percentage-point change during 2009–2011 in DC pension plan contribution rate of—
Employee 0.13 2.94 -0.06 3.78 -0.19*
Employer 0.46 7.92 0.22 6.8 -0.24
Percentage of workers whose retirement investments involve some risk 82 38 86 35 4***
Number of retirement asset types 1.49 0.90 2.02 1.19 0.53***
Sociodemographic characteristics
Percentage—
With associate's degree or higher 59 49 90 30 31***
With business degree 11 31 24 43 13***
Female 57 50 24 42 -33***
Married 66 47 80 40 14***
White with U.S. citizenship 78 42 80 40 2**
Self-reporting fair or poor health; or having a mental or work-limiting or -preventing disability 8 27 3 16 -5***
Who have ever received transfer payments 8 28 2 14 -6***
Home equity ($) 44,247 73,606 107,350 114,139 63,103*** Number of times divorced 0.20 0.48 0.14 0.41 -0.06*** Number of own children living with the family 0.95 1.09 1.19 1.17 0.24*** Household income ratio (personal income as a percentage of household income) b 58.32 b 32.71 80.63 22.09 22.31*** Job/career characteristics Percentage— Working at a large firm (100+ employees) c 66 47 77 42 11*** Usually working at least 35 hours per week 81 39 87 34 6*** Unionized 15 36 7 26 -8*** Job tenure (years) 8.90 7.95 11.61 8.94 2.71*** Weeks not worked during 2009–2011 3.37 12.51 1.66 7.29 -1.71*** Years with an earnings loss ≥ 10 percent 5.25 3.40 4.60 3.05 -0.65*** Change in earnings during 2009–2011 ($) 3,100 19,251 6,281 177,759 3,181
Lifetime earnings ($) 820,108 605,518 3,243,030 2,455,141 2,422,922*** Volatility of annual earnings ($) d 16,138 d 27,367 76,934 76,958 60,796***
SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records.
NOTES: Numbers of observations are 5,139 (lower earners) and 1,083 (higher earners), unless otherwise noted.
. . . = not applicable.
* = statistically significant at the 10% level; ** = statistically significant at the 5% level; *** = statistically significant at the 1% level.
a. As of 2009, unless otherwise noted.
b. Based on 5,125 observations.
c. Based on 5,059 lower-earner observations and 1,071 higher-earner observations.
d. Based on 5,138 observations.

Eighty-six percent of higher earners held investments involving some risk, compared with 82 percent of lower earners. Higher earners can presumably take on more risk because they have fewer liquidity constraints and more financial information. Higher earners also have more diversified portfolios (averaging 2.02 asset types) than lower earners (1.49 asset types). To the extent that risk exposure yields higher returns and that diversification protects against market losses, the differences in risk and diversification choices contribute to a greater likelihood of an increase in retirement balances for higher earners than for their lower-earning peers.

Factors associated with lower retirement savings include having educational attainment of less than an associate's degree, being female, being unmarried, having been divorced, being in poor health, having received means-tested transfer payments, working at a small firm, working part-time, and having a short job tenure (Johnson, Mermin, and Uccello 2006; Smith, Johnson, and Muller 2004; Tamborini, Purcell, and Iams 2013). Lower earners are more likely to have these characteristics than are higher earners.

Table 2 shows that lower earners had more time (3.37 weeks) not employed during 2009–2011 than higher earners had (1.66 weeks). The majority of nonemployment spells reflected time spent out of the labor force as opposed to time unemployed.13 Workers in poor health, who were relatively older, and who increased their educational attainment were more likely to have had nonemployment spells whereas higher earners, workers with longer job tenure, those caring for more children, and those who earned a larger share of the household income were less likely to have had any nonemployment spells from April 2009 to March 2012 (Table 3).

Table 3. OLS regression estimates of determinants of work interruptions and earnings losses
Category and determinant a Nonemployment spells (weeks not employed April 2009–March 2012) Earnings shocks (years with earnings loss ≥ 10 percent)
Coefficient Standard error Coefficient Standard error
Incremental variables
Health
Years since start of long-term work-limiting disability -0.13* 0.07 0.02 0.02
Demographics
Number of own children living with the family -0.23** 0.11 0.12*** 0.03
Number of divorces -0.03 0.24 0.32*** 0.06
Years of age 0.02* 0.01 0.21*** 0.00
Change in the number of children living with the family during 2009–2011 0.13 0.28 0.13** 0.07
Job
Job tenure (years) -0.10*** 0.01 -0.08*** 0.00
Finances
Household income ratio (worker income as a percentage of household income) b -0.01** 0.00 0.00*** 0.00
Home equity ($10,000) 0.00 0.00 0.00*** 0.00 Categorical variables Health Experienced change in health condition 0.68 0.44 0.47*** 0.11 Poor health (self-reported) 1.43*** 0.55 0.46*** 0.13 Education Has associate's degree or higher 0.24 0.25 0.32*** 0.06 Increased attainment level during 2009–2011 1.45*** 0.58 0.10 0.14 Has business degree -0.39 0.33 -0.18** 0.08 Demographics Is female 0.00 0.24 -0.26*** 0.06 Is married -0.19 0.28 -0.28*** 0.07 Is white -0.03 0.29 0.26*** 0.07 Is a U.S. citizen -0.81 0.70 1.84*** 0.17 Job Works at a large firm (100+ employees) 0.04 0.25 -0.93*** 0.06 Usually works 35 or more hours per week 0.26 0.29 -0.35*** 0.07 Finances Is a higher earner -0.80*** 0.18 -0.60*** 0.04 Has ever received transfer payments 0.08 0.46 0.84*** 0.11 Has ever received lump-sum pension payment 0.03 0.33 -0.23*** 0.08 Intercept 4.90*** 1.03 -3.82*** 0.25 R2 0.01 0.41 SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records. NOTES: Number of observations = 9,231. * = statistically significant at the 10% level; ** = statistically significant at the 5% level; *** = statistically significant at the 1% level. a. As of 2009, unless otherwise noted. b. Percentage-point increments. Lower earners experienced more years with a significant earnings loss over their working lives (5.25 on average) than higher earners did (4.60; Table 2).14 Workers in relatively poorer health, with more children living with them, and who had more divorces had experienced more episodes of significant earnings loss over their careers (Table 3). Being a higher earner, having a business degree, working full time in a large firm, having long job tenure, and being married and female decreased the likelihood of suffering episodes of earnings loss over one's career. Age was associated with more years with significant earnings loss because older workers have had more time to experience such episodes. Earnings losses were also more common among workers with an associate's degree or higher—probably because they had left employment to attend school. Being a recipient of transfer payments was associated with more episodes of earnings loss, but any causation is likely in the opposite direction because workers whose earnings decrease are more likely to be eligible for aid. White workers and U.S. citizens reported more episodes of earnings loss than nonwhites and noncitizens. This may reflect a greater tendency of nonwhite and noncitizen workers to engage in informal labor markets with unreported earnings; the volatility of their actual earnings would therefore not be indicated in the administrative data. ## Results This section reports the results for the decomposition analysis and regressions, as well as the robustness checks of the results. ### Decomposition Results The decomposition reveals that among workers whose DC plan balances increased, employee and employer contributions were more important than portfolio allocation effects in explaining the change (Table 4). However, among workers whose balances declined, portfolio allocation effects dominated the change: They accounted for 373 percent of the total loss for lower earners and for 221 percent of the loss for higher earners. At the same time, employer contributions were more instrumental in holding back losses for lower earners than for their higher-earning peers; they represented a −160 percent counterweight to the overall loss for lower earners and −86 percent for higher earners.15 Withdrawals explained 6 percent of the loss for lower earners but had no significant effect on DC plan balances among higher earners. (Rollovers had no effect and are not reported.) Among workers who experienced balance gains, the decomposition found that withdrawals were more pronounced for lower earners than for higher earners but no other factors differed significantly. Table 4. Decomposition estimates for causes of change in retirement savings for lower and higher earners with an increase or a decrease in DC plan balance during 2009–2011 Variable Lower earners Higher earners Difference in means Mean effect (%) Standard error Mean effect (%) Standard error Workers with increase in DC plan balance Portfolio allocation effects a -30 24.55 -37 6.18 7 Plan withdrawals -1 0.19 0 0.08 1* Employee contributions 64 15.97 104 5.60 40 Employer contributions 67 14.88 33 1.30 34 Observations 2,759 649 . . . Workers with decrease in DC plan balance b Portfolio allocation effects a 373 47.51 221 4.48 152* Plan withdrawals 6 1.12 0 0.01 6*** Employee contributions -160 37.35 -86 3.16 74 Employer contributions -118 29.15 -35 1.51 83* Observations 2,257 389 . . . SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records. NOTES: Rounded mean-effects percentages do not necessarily sum to 100. . . . = not applicable. * = statistically significant at the 10% level; ** = statistically significant at the 5% level; *** = statistically significant at the 1% level. a. Portfolio management fees and market performance. b. Mean-effect percentages indicate factor contributions to the loss in DC plan value; thus, positive percentages indicate DC plan losses and negative percentages indicate plan gains. ### Regression Results Table 5 shows estimates of the statistically significant determinants of change in retirement savings separately for lower and higher earners.16 The regression results reveal that for each year a lower-earning worker had experienced an earnings drop of 10 percent or more through 2009, her or his retirement savings declined by$450 during 2009–2011. Recall that lower earners averaged about 5 years with significant earnings losses over their career (Table 2) so their average retirement savings were reduced by a total of about $2,250 because of these earnings shocks.17 Note that the earnings-loss effect on retirement savings is limited to lower earners because the effect is not statistically significant for higher earners. Table 5. Regression-estimated changes in retirement savings during 2009–2011 attributable to selected factors: IHS results, lower and higher earners (2011 dollars) Determinant a Lower earners b Higher earners c Coefficient Standard error Coefficient Standard error Incremental variables Years with earnings loss ≥ 10 percent (lifetime) -450*** 139 -75 851 Weeks not employed 2009–2011 -55** 27 -339 310 Number of retirement asset types held in portfolio 1,194*** 487 1,277 1,286 DC pension plan contribution rate d of— Employee 729*** 91 1,077*** 359 Employer 182** 88 -54 459 2009–2011 increase in DC pension plan contribution rate d of— Employee 360*** 127 705 471 Employer 141* 75 200 349 Lifetime earnings ($10,000) 66*** 18 39 25
Change in number of children living with the family during 2009–2011 1,568** 714 -579 4,201
Household income ratio (worker income as a percentage of household income) d -28*** 10 -79 88
Retirement savings ($10,000) -2,099*** 85 -2,399*** 156 Home equity ($10,000) 199*** 56 367** 155
Categorical variables
Respondent—
Is white and a U.S. citizen 4,054*** 792 11,118** 5,048
Has associate's degree or higher 1,669** 755 14,439*** 5,359
Observations 5,045 1,071
R2 0.23 0.20
SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records.
NOTES: Table omits variables for which results were not statistically significant.
* = statistically significant at the 10% level; ** = statistically significant at the 5% level; *** = statistically significant at the 1% level.
a. As of 2009, unless otherwise noted.
b. Data are OLS regression estimates for workers in the lowest 55 percent of the earnings distribution. Testing the hypothesis that variables are exogenous, Durbin χ2 = 1.69, p = 0.43 and Wu-Hausman F = 0.84, p = 0.43.
c. Data are IV regression estimates for workers in the highest 12 percent of the earnings distribution. Instruments for employer contribution include average employer contribution rates for a participant's state, industry, and education level; and the fraction of workers with nonzero employer contributions at the state and industry-education levels. Testing the hypothesis that variables are exogenous, Durbin χ2 = 12.87, p = 0.0016 and Wu-Hausman F = 6.29, p = 0.0019. Testing for overidentifying restrictions, Sargan χ2 = 4.06, p = 0.40 and Basman χ2 = 3.93, p = 0.41.
d. Percentage-point increments.

For each week a lower-earning worker was not employed during 2009–2011, retirement savings decreased by $55. Because lower earners were not employed for an average of about 3 weeks in 2009–2011, the total average reduction in their retirement savings was$165. Note, again, that statistically significant negative effects of nonemployment spells on retirement savings were limited to lower earners.

For each type of asset that workers held in their portfolios, lower earners gained, on average, $1,194 in account value. Because the lower earners invested in an average of 1.5 asset types, their total average increase in retirement savings was$1,791. This diversification effect was only significant for lower earners.

White citizens had average retirement-savings increases from 2009 to 2011 of $4,054 for lower earners and$11,118 for higher earners. We attribute these increases in large part to portfolio allocation because white workers are associated with having greater access to a network of potential advisors on portfolio choices (Chong, Dow, and Phillips 2010).18

Each percentage point in employee contribution rates to DC pension plans in 2009 was associated with increases in retirement savings during 2009–2011 of $729 and$1,077 for lower and higher earners, respectively. Because lower earners contributed an average of 2.8 percent and higher earners contributed 6.1 percent, the total average contribution-rate effects were $2,041 for lower earners and$6,570 for higher earners. The change in low-earner employee contribution rates during 2009–2011 was associated with a $360 increase in retirement savings. Given the 0.13 percentage-point average increase in the employee contribution rate among low earners, their savings increased by an average of$47 over the period.

Each percentage point in 2009 employer contribution rates was associated with significantly higher retirement savings for lower earners only, in the amount of $182. Because the average employer contribution rate for low earners was 2.15 percent, the total average increase for lower earners was$391. Per percentage point, the change in employer contribution rates during 2009–2011 led to an additional $141 for low earners, which added$65 to their mean retirement savings because their average employer contribution rate increased by 0.46 percentage points. The significant effects of employer contribution rates—both their initial levels and their increases—highlight their importance in helping lower earners save for retirement.

Higher education, measured as having at least an associate's degree, led to a $14,439 increase in retirement savings for higher earners, but the effect was a much smaller$1,669 for lower earners. Interestingly, having a business degree did not boost retirement savings (not shown).

Lifetime earnings affected the retirement savings of lower earners only. For each $10,000 in lifetime earnings, lower earners' savings increased by$66. The average lifetime earnings of lower earners ($820,108) resulted in a total average change of$5,412.

Adding children to the family during 2009–2011 was associated with a $1,568 average increase in the retirement savings of lower earners for each child, but was not associated with an increase for higher earners. Previous studies (Smith, Johnson, and Muller 2004; Butrica and Smith 2014) found that the birth of a child positively affects retirement-plan participation and contributions; those studies did not differentiate workers by earnings level. The household income ratio—the respondent's personal income as a percentage of total household income—fluctuates when the partner's income changes. Higher ratios may reflect liquidity constraints on workers' ability to contribute to their retirement accounts, resulting in smaller retirement wealth accumulation. For each percentage point in the household income ratio, balances for low earners declined by$28. Given the mean ratio of 58.32 percent among low earners, the total average decline in retirement savings attributable to their household income ratio was $1,633. This result was not duplicated for higher earners, who are less likely to face liquidity constraints. Higher initial retirement savings were associated with larger losses, a result also found in Gustman, Steinmeier, and Tabatabai (2012). We find that lower earners suffered a loss of$2,099 for every $10,000 of retirement wealth held in 2009, while higher earners experienced a loss of$2,399 for every $10,000 of retirement balances held in 2009. Because lower earners in 2009 held average retirement savings of$34,137 and higher earners held $147,787, the total average loss was$7,165 for lower earners and $35,454 for their higher-earning counterparts. These results probably reflect the greater exposure of higher initial balances to market swings and portfolio losses in a downturn. Finally, every$10,000 of home equity was associated with increases in retirement savings of $199 and$367 for lower and higher earners, respectively. With average home equities of $44,247 for the former and$107,350 for the latter, the resulting total average increases were $880 for lower earners and$3,940 for higher earners. Because respondents can borrow against their home equity or sell the house and convert equity to cash, home equity can indicate the presence of liquidity constraints. This effect is more potent for higher earners, who typically have higher average home equity values and can lever that equity to contribute more toward their retirement accounts. Lower earners may have more pressing priorities that limit their opportunities to save for retirement.

### Robustness Checks

We test for the robustness of the findings by dividing the sample into two age groups and examining whether the results remain statistically significant. Table 6 presents the results, omitting all determinants with no significant associations for either lower or higher earners. Among lower-earning workers aged 25–49, each year of substantial earnings loss (10 percent or more) reduced retirement savings by $550, while each week not employed over the study period reduced savings by$92. For each type of asset held in one's portfolio, savings increased by $785. These results are statistically significant for lower earners aged 25–49 (and for the entire sample of lower earners), but not for higher earners. Table 6. Robustness check: Regression-estimated changes in retirement savings during 2009–2011 attributable to selected factors: IHS results, lower and higher earners, by age group (2011 dollars) Determinant a Lower earners b Higher earners c Coefficient Standard error Coefficient Standard error Age 25–49 Incremental variables Years with earnings loss ≥ 10 percent (lifetime) -550*** 198 411 1,243 Weeks not employed 2009–2011 -92*** 28 -363 461 Number of retirement asset types held in portfolio 785* 416 -579 2,489 DC pension plan contribution rate d of— Employee 750*** 134 1,788*** 575 Employer 172** 78 18 924 2009–2011 increase in employee DC pension plan contribution rate d 422*** 169 1,499** 767 Lifetime earnings ($10,000) 75** 32 -9 63
Household income ratio (worker income as a percentage of household income) d -25** 11 138 124
Retirement savings ($10,000) -2,475 138 -2,729*** 243 Home equity ($10,000) 247 96 241 260
Categorical variables
Respondent—
Is white and a U.S. citizen 3,263 707 10,249 6,650
Participates in a defined benefit pension plan -2,497*** 855 4,591 6,140
Observations 3,218 637
R2 0.27 0.20
Age 50–61
Incremental variables
Years with earnings loss ≥ 10 percent (lifetime) -398** 194 -842 861
Number of retirement asset types held in portfolio 1,946** 895 2,382 1,831
Employee's DC pension plan contribution rate d 686*** 163 583 419
2009–2011 increase in employee DC pension plan contribution rate d 324* 180 602 521
Lifetime earnings ($10,000) 44* 24 35 26 Household income ratio (worker income as a percentage of household income) d -37 23 -232** 96 Job tenure (years) 136* 72 -65 226 Retirement savings ($10,000) -1,730*** 149 -1,830*** 133
Home equity ($10,000) 142* 76 373** 191 Categorical variables Respondent— Is white and a U.S. citizen 6,971*** 1,585 11,976 7,831 Has associate's degree or higher 2,464** 1,275 9,800 7,769 Is female -388 1,744 -10,723** 5,474 Has union job -2,506 1,599 -12,218* 6,994 Self-reports fair or poor health; or having a mental or work limiting/preventing disability -2,996* 1,609 -13,438 11,919 Observations 1,827 434 R2 0.20 0.11 SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records. NOTE: * = statistically significant at the 10% level; ** = statistically significant at the 5% level; *** = statistically significant at the 1% level. a. As of 2009, unless otherwise noted. b. Data are OLS regression estimates for workers in the lowest 55 percent of the earnings distribution. c. Data are IV regression estimates for workers in the highest 12 percent of the earnings distribution. Instruments for employer contribution include average employer contribution rates for a participant's state, industry, and education level; and the fraction of workers with nonzero employer contributions at the state and industry-education levels. d. Percentage-point increments. Among lower-earning workers aged 50–61, each year with substantial earnings loss reduced retirement savings by$398. The number of weeks not employed in 2009–2011 had no significant effect on savings for this group and is omitted from the age 50–61 panel. For each type of asset held, savings increased by $1,946, a much larger increase than we see for younger workers, who had less money invested. Again, these results are statistically significant for lower earners, but not for higher earners. ## Discussion and Policy Implications With DC pension plans, workers and employers contribute voluntarily to the worker's account. The worker then constructs an account portfolio, choosing among the investment vehicles the plan makes available. These plans work best for higher-earning workers with stable employment, health, and financial situations and are not as conducive to the saving needs of lower-earning workers, who may experience greater earnings volatility and job insecurity, lack financial literacy, have limited financial networks, and face liquidity constraints. The design of DC pension plans in the United States inadvertently makes lower earners more likely than higher earners to lose DC plan wealth. Earnings volatility is associated with a decrease in DC plan wealth for lower earners but not for higher earners. Dushi, Iams, and Tamborini (2013) found that earnings volatility and nonemployment spells led workers to decrease their contributions and lowered their participation rates in the aftermath of the recession of 2007–2009. Our results suggest that nonemployment spells and years with major earnings losses have longer-lasting effects on workers' DC plan wealth and therefore on retirement savings overall. It may be that workers who suffer a substantial earnings loss cannot maximize their portfolio performance by buying stock when the market is low because those are the times when they are more likely to lose earnings (Weller and Wenger 2009). Workers who experience episodes of earnings decline may reasonably prefer liquidity to investing in what is perceived to be a less liquid retirement account. Declines in earnings may also instill fear in workers that they will be strapped for cash in the future, which further inhibits their willingness to tie up earnings in retirement savings vehicles (Ghilarducci, Saad-Lessler, and Reznik 2017). Regardless of the pathway, the permanent effect of past incidents of substantial earnings loss on DC plan wealth informs policymakers that career risks are faced by all workers, but most acutely by lower earners; and these risks work against lower earners' ability to save for retirement. Our results indicate that a more diversified portfolio is associated with higher DC plan wealth for lower earners. This implies that lower-earning workers may benefit from more vigorous enforcement of regulations requiring employers to provide better investment choices or from prepackaged portfolios that are better-managed and cheaper than target-date funds (Skarbeck 2009; Grant 2014). The finding that higher educational attainment is correlated with increases in retirement savings for all earners may indicate that formal education is a proxy for financial literacy, as it is assumed that financial literacy helps promote higher balances. However, that assumption is challenged by the finding that having a business degree had no effect on retirement savings. Higher employer contributions to DC pension plans helped lower earners increase their DC plan wealth. This highlights the role of employers in helping lower earners save for retirement. However, U.S. employers are not required to contribute to their workers' DC accounts (or even to offer DC plans at all), putting lower earners at a particular disadvantage. Although choosing the worker's DC plan contribution rate may be a family decision, the spouse's income seems to affect the decision for low earners only, not those at the top. A household's reliance on an individual worker's earnings is associated with retirement savings declines among lower earners but not higher earners. In Ghilarducci, Saad-Lessler, and Reznik (2017), we found some evidence that spouses may influence each other's DC plan contributions; but the effect was mostly complementary, not substitutable. In sum, there seem to be three reasons why higher earners' retirement savings were more likely to have increased in our study period. First, lower earners had experienced more weeks of nonemployment during 2009–2011 and more years with earnings losses of 10 percent or more in their lifetime. Second, lower earners had less diversified retirement account portfolios: Higher earners had 2.0 asset types on average, compared with 1.5 for lower earners. Third, higher earners had higher DC plan employee and employer contribution rates in 2009 (6.11 percent and 2.86 percent, respectively) than lower earners had (2.80 percent and 2.15 percent, respectively).19 These findings show that the primary design features of DC pension plans—voluntary employee and employer contributions and individually directed investments—affect people differently based on their economic experiences. The effectiveness of DC pension plans depends on a worker's earnings level. We find stark differences in the resilience of retirement wealth accumulations between high earners and low earners. The results imply that the current design of DC plans disadvantages lower earners in their efforts to save for retirement. ## Appendix A: Additional Notes on the Study Variables Most of this study's variables are derived from data provided by the SIPP 2008 panel; the rest are derived from Social Security administrative records. The latter group consists of the earnings-related variables; specifically, employee's DC pension plan contribution rate, lifetime earnings, change in earnings during 2009–2011, volatility of annual earnings, and years with earnings loss of 10 percent or more. Study variables that are not self-explanatory or are not fully described in the body of this article are listed below with relevant details noted. DC pension plan contribution rate of employee. The amount contributed by an employee in a given year divided by the employee's earnings in that year; the result is then multiplied by 100 to express the value in percentage points. DC pension plan contribution rate of employer. The employer's contribution is reported by the employee in the SIPP as either a rate or an amount. If the employee self-reported the employer's contribution amount, we divide that amount by the employee's self-reported annual earnings and multiply the result by 100 to express the value in percentage points. Participates in a defined benefit (DB) pension plan. Respondent self-reports DB pension coverage at main job, expectation of receiving a DB pension benefits from a previous job, or receipt of DB pension benefits. Ever received transfer payments. Wave 2 respondent self-reports ever receiving Aid to Families with Dependent Children, Supplemental Security Income, or Temporary Assistance for Needy Families (food stamps); followed up in wave 3. Retirement savings. Balances reported by wave 4 and wave 10 respondents. Shown in current-year dollars. Has a union job. Self-reported membership in a union or employee association or coverage by an agreement similar to a union contract at first reported job for current workers or at previous job for retirees. Usually works 35 or more hours per week. Reported in wave 1. Refers to the period beginning with the first 6 consecutive months worked and ending with the current or last previous job. Volatility of annual earnings. The standard deviation of lifetime annual earnings reported in administrative records. Weeks not employed 2009–2011. The number of weeks between SIPP 2008 wave 3 and wave 11 in which the respondent did not work, including any periods out of the labor force. Works at a large firm. A large firm has 100 or more employees at all of its locations combined, or at its single location, as applicable. “Firm” may refer to either the current or the previous employer, depending on whether the respondent is currently working. ## Appendix B Table B-1. Regression-estimated changes in retirement savings during 2009–2011 attributable to selected factors: Results of using untransformed dependent variable, lower and higher earners (2011 dollars) Determinant a Lower earners b Higher earners c Coefficient Standard error Coefficient Standard error Retirement savings characteristics Incremental variables Retirement savings ($10,000) -6,524*** 176 -7,114*** 380
DC pension plan contribution rate d of—
Employee 1,683*** 201 2,037* 1,171
Employer 322 211 13,379** 5,859
2009–2011 increase in DC pension plan contribution rate d of—
Employee 562* 310 1,469 1,333
Employer 247 157 8,108 9,849
Number of retirement asset types held in portfolio -178 473 6,208 3,979
Categorical variables
Respondent—
Participates in a defined benefit pension plan -1,787 2,041 17,632 12,647
Has retirement investments involving some risk 227*** 23 3,638 13,707
Ever received a lump-sum payment from—
Own pension 2,605 2,801 5,480 13,893
Another person's pension 2,354 8,937 -12,098 67,284
Sociodemographic characteristics
Incremental variables
Home equity ($10,000) -1,628 5,867 1,228*** 432 Number of divorces (lifetime) -2,989 1,910 21,917** 11,275 Number of own children living with the family -174 922 -2 4,453 Change in the number of children living with the family during 2009–2011 180 131 -866 13,249 Household income ratio (worker income as a percentage of household income) d -1,817 1,950 -102 262 Categorical variables Respondent— Is married -835** 350 -2,196 13,129 Is female -3,649* 2,218 -4,333 11,411 Is white -127*** 31 25,548* 14,835 Is a U.S. citizen 10,829*** 2,269 -11,008 20,583 Has associate's degree or higher 5,210*** 1,031 30,889* 16,070 Has business degree 4,659** 1,929 18,350* 10,962 Self-reports fair or poor health; or having a mental or work-limiting or -preventing disability -3,859 3,301 -25,085 28,660 Has ever received a transfer payment -3,709 3,309 -15,695 33,119 Is aged 25–49 136 154 -454 967 Job/career characteristics Incremental variables Years with earnings loss ≥ 10 percent (lifetime) 2,455 2,025 373 1,992 Weeks not employed 2009–2011 15 75 -778 954 Lifetime earnings ($10,000) 1,912 2,345 134*** 51
Job tenure (years) 645*** 131 -394 637
Increase in earnings 2009–2011 ($10,000) 3,819* 2,312 -241 287 Volatility of annual earnings ($) 0*** 0 0** 0
Categorical variables
Respondent—
Works at a large firm (100+ employees) 431 3,026 -14,859 12,792
Has union job 1,518 2,539 -38,685** 18,353
Usually works 35 or more hours per week -3,496 2,311 13,221 14,214
Intercept -4,280 8,866 13,621 52,795
SOURCE: Authors' calculations using SIPP 2008 panel data matched to Social Security administrative records.
NOTES: Number of observations = 9,231.
* = statistically significant at the 10% level; ** = statistically significant at the 5% level; *** = statistically significant at the 1% level.
a. As of 2009, unless otherwise noted.
b. Data are OLS regression estimates for workers in the lowest 55 percent of the earnings distribution.
c. Data are IV regression estimates for workers in the highest 12 percent of the earnings distribution. Instruments for employer contribution include average employer contribution rates for a participant's state, industry, and education level; and the fraction of workers with nonzero employer contributions at the state and industry-education levels.
d. Percentage-point increments.

## Notes

1 Although retirement savings may also include defined benefit pensions, annuities, savings accounts, and other vehicles, we use “retirement savings” to refer exclusively to combined balances in IRAs, Keogh plans, and DC plans.

2 Other studies on retirement savings and preretirement household wealth (Smith, Johnson, and Muller 2004; Johnson, Mermin, and Uccello 2006; Dushi, Iams, and Tamborini 2013; Dushi and Iams 2015) rely on cross-sectional data, which are not as reliable as panel data for tracking financial behavior and outcomes over time.

3 However, lower earners are likely to have less diversified and lower-risk portfolios than higher earners (Kuhnen and Miu 2015).

4 We assume that there are no significant changes in pension plan design features between two contiguous waves because the waves are 4 months apart.

5 Specifically, the SIPP questioner (a Census Bureau field representative) states: “The next part of the interview is about your income since [first reference month] 1st. We want to be as accurate and efficient as we can, so it would be very helpful if you could refer to any records you might have.” In addition, if a respondent says, “hold on while I get my records,” then the SIPP questioner is instructed to let them do so.

6 Data for 1980 and earlier are restricted to earnings in covered employment up to the Social Security taxable maximum.

7 Employer contribution rates are reported in the SIPP data for 2009 and 2011. The employer contribution rate in 2010 is interpolated as the average between the 2009 and 2011 rates. The employer contribution rates are applied to the respondent's self-reported earnings for each month between 2009 and 2011 to yield annual employer contribution amounts for 2009, 2010, and 2011.

8 The Social Security benefit formula uses the average indexed earnings from a worker's 35 highest-earning years. The unadjusted calculation for an eligible worker who claimed retirement benefits in 2009 was equal to 90 percent of the average indexed earnings up to the first bend point of $8,928 per year, plus 32 percent of average indexed earnings between$8,929 and the second bend point of $53,796 per year, and 15 percent of average indexed earnings between$53,797 and the taxable earnings cap of \$106,800. The formula yields a progressive benefit structure. For a low-earning worker, Social Security replaces about 80 percent of final earnings; for a middle-earning worker, it replaces about 40 percent; and for a higher earner, it replaces about 25 percent.

9 Those observations are not dropped from the sample, however, because there is no reason to believe that the rest of the responses are unreliable. We do not include indicators of withdrawals or rollovers in the regression equation only because there are too few observations to develop a reliable estimate of their effect on DC plan wealth.

10 Differences in initial levels of wealth across groups may lead to varying saving behaviors, as well as a divergence in investment gains—changes in levels of wealth—across the groups.

11 Investments considered safe are certificates of deposit or other saving certificates, money market funds, U.S. government securities, and U.S. savings bonds. Investments involving some risk are municipal or corporate bonds, stocks or mutual fund shares, and other assets.

12 Indicated by whether the respondent has an associate's degree or a diploma/certificate from a vocational, technical, trade, or business school beyond the high school level in business/office management; or a bachelor's, master's, professional, or doctoral degree in business/management; as of 2009.

13 Ninety-three percent of respondents with nonemployment spells during 2009–2011 did not report being unemployed.

14 To put these numbers in perspective, less than 5 percent of workers in the sample had never experienced a year with a significant earnings loss, while 61 percent of the sample experienced four or more such episodes. Although we do not identify why workers experienced earnings losses, voluntary workforce withdrawal or reductions in work hours to provide unpaid care for family members may be one reason.

15 Decomposition effects could be negative when the change in wealth was negative but contributions were present. This means the factor's effect ran counter to the total effect on wealth. The decomposition effects have to sum to 100 percent of the loss for those who lost DC wealth. Because contributions reduce losses, the ratios of contributions to total loss are negative numbers (−160 percent and −86 percent). This is why the ratios of portfolio effects to total loss exceed 100 percent (373 percent and 221 percent).

16 Appendix Table B-1 presents, for comparative purposes, an alternative version of Table 5 that uses an untransformed dependent variable and includes all determinants regardless of statistical significance.

17 The discussion in the remainder of this section follows the structure of this paragraph in that we cite average values from Table 2, by which we multiply each regression estimate we mention from Table 5. For brevity, we omit repeated references to Table 2 as the source of the average values.

18 Our sample includes many workers who invest in more than one asset type but our data do not specify the amounts invested in each type, preventing full measurement of asset allocation. To adjust, we examined workers who indicated only one asset type and compared their allocation choices. We found that white citizens were much more likely to invest in riskier investment types than were nonwhites or noncitizens. Because we lack specific asset-allocation input data and white-citizen investment choices differed sharply from those of other respondents, the estimates for white respondents in Table 5 are likely a proxy for the unobservable asset-allocation differences by race and citizenship status.

19 Additionally, higher earners receive a higher net-of-tax return on every dollar invested in a DC plan because they receive a higher implicit subsidy via state and federal tax deductions for retirement contributions. A person in the highest tax bracket returns 39.6 cents from the federal deduction for retirement contributions and, on average, 7 cents from a state deduction. If the higher earner also pays lower fees because of scale economies and has better-structured portfolios because of better advice and less risk adversity, then the higher earner's rate of return increases continuously and the accumulated wealth increasingly pulls away from that of a lower earner. This topic merits future research.

## References

Burbidge, John B., Lonnie Magee, and A. Leslie Robb. 1988. “Alternative Transformations to Handle Extreme Values of the Dependent Variable.” Journal of the American Statistical Association 83(401): 123–127.

Butrica, Barbara A., and Karen E. Smith. 2014. “401(k) Participant Behavior in a Volatile Economy.” Journal of Pension Economics and Finance 15(1): 1–29.

Chong, James, James P. Dow, Jr., and G. Michael Phillips. 2010. “Neighbourhood Effects and Asset Allocation.” Journal of Interdisciplinary Economics 23(1): 37–52.

Dushi, Irena, and Howard M. Iams. 2015. “The Impact of Employment and Earnings Shocks on Contribution Behavior in Defined Contribution Plans: 2005–2009.The Journal of Retirement 2(4): 86–104.

Dushi, Irena, Howard M. Iams, and Christopher R. Tamborini. 2011. “Defined Contribution Pension Participation and Contributions by Earnings Levels Using Administrative Data.” Social Security Bulletin 71(2): 67–76.

———. 2013. “Contribution Dynamics in Defined Contribution Pension Plans During the Great Recession of 2007–2009.” Social Security Bulletin 73(2): 85–102.

Ghilarducci, Teresa, Joelle Saad-Lessler, and Gayle Reznik. 2017. “Earnings Volatility and 401(k) Contributions.” Journal of Pension Economics and Finance.

Gist, John R., and Megan E. Hatch. 2014. “Retirement Replacement Rates and Retirement Wealth Inequality among Baby Boomer and Other Birth Cohorts.” The Journal of Retirement 2(1): 55–69.

Grant, Tim. 2014. “The Target Date Paradox: Financial Advisers Warn of Potential ‘Land Mine’ in Bond Portion of Such Funds.” Pittsburgh Post-Gazette (March 7).

Gustman, Alan L., Thomas L. Steinmeier, and Nahid Tabatabai 2012. “How Did the Recession of 2007–2009 Affect the Wealth and Retirement of the Near Retirement Age Population in the Health and Retirement Study?Social Security Bulletin 72(4): 47–66.

Johnson, Richard W., Gordon B. T. Mermin, and Cori E. Uccello. 2006. “When the Nest Egg Cracks: Financial Consequences of Health Problems, Marital Status Changes, and Job Layoffs at Older Ages.” Washington, DC: Urban Institute.

Kuhnen, Camelia M., and Andrei C. Miu. 2015. “Socioeconomic Status and Learning from Financial Information.” Working Paper No. 2015-018. Chicago, IL: University of Chicago, Human Capital and Economic Opportunity Global Working Group.

Lusardi, Annamaria, and Olivia S. Mitchell. 2014. “The Economic Importance of Financial Literacy: Theory and Evidence.” Journal of Economic Literature 52(1): 5–44.

Pence, Karen M. 2006. “The Role of Wealth Transformations: An Application to Estimating the Effect of Tax Incentives on Saving.” Contributions to Economic Analysis & Policy 5(1): 1–26.

Poterba, James, Steven Venti, and David A. Wise. 2015. “What Determines End-of-Life Assets? A Retrospective View.” NBER Working Paper No. 21682. Cambridge, MA: National Bureau of Economic Research.

Skarbeck, Ken. 2009. “More Reforms Sought for Target-Date Funds.” Indianapolis Business Journal 30(17): 26A (June 29).

Smith, Karen E., Richard W. Johnson, and Leslie A. Muller. 2004. “Deferring Income in Employer-Sponsored Retirement Plans: The Dynamics of Participant Contributions.” National Tax Journal 57(3): 639–670.

Tamborini, Christopher R., Patrick Purcell, and Howard M. Iams. 2013. “The Relationship between Job Characteristics and Retirement Savings in Defined Contribution Plans during the 2007–2009 Recession.” Monthly Labor Review 136(5): 3–16.

Weller, Christian E., and Jeffrey B. Wenger. 2009. “What Happens to Defined Contribution Accounts When Labor Markets and Financial Markets Move Together?” Journal of Aging & Social Policy 21(3): 256–276.