Paul Farber ’13
I find evidence of a significant race effect in employer stock levels in pension accounts and find that reported risk preference is not an indicator of company stock holdings. While there have been widespread decreases in employer stock holdings in the past decade, there are still significant issues to be addressed, such as the correlation between employer stock holdings and race and the dissociation between risk preferences and the amount of risky employer stock held in retirement accounts.
The collapse of Enron and WorldCom at the turn of the century was a dramatic illustration of the risk to employees of investing in the stock of their employer. Holding large amounts employer stock in 401k plans is a puzzle that has been studied over the past decade. The investigation into employer stock holdings in 401k plans began with Benartzi (2001), who showed that allocations to plans were positively correlated with the returns of the employer’s stock, with the strongest results for 10-year returns, and weaker results for shorter time horizons, by using SEC fillings to look at flows into retirement accounts of firms in the S&P 500 who allocate employer stock directly from their treasury into retirement accounts. Liang and Weisbenner (2002) confirm these results for a larger number of companies over a longer period of time. Choi, Laibson, Madrian, and Metrick (2003) also show that there is an over extrapolation of past returns of employer stock on initial flows into retirement plans using individual-level data for almost 100,000 individuals at three large firm’s from 1992 to 2000. They further refine this phenomenon by showing that individuals act as contrarian investors when reallocating their portfolio, moving out of company stock after high returns. This suggests the strong relationship found by Benartzi (2001) may be weaker than expected; however, absent frequent rebalancing, confirmed by Agnew, Balduzzi and Sundén (2003), the results using cross-sectional snapshots of holdings and contribution fractions gives good estimations of the magnitude of the problem.
The majority of this research looks at data from 2000 and before, when new participants in 401k plans with company stock as an option were contributing around 23 percent of their assets to employer stock. This has since declined dramatically to 7.9 percent in 2007 as compiled by Alonso, Holden, and VanDerhei (2009) using EBRI data. With the total assets in employer stock at 10.6 percent in 2007 and almost half of participants holding no employer stock, there has been a significant shift in employer stock as a fraction of total assets.
This reduction comes starting in 2003, coinciding with research by Meulbroek (2005) using a Sharpe-ratio approach that employees who invest one-quarter of their assets in company stock sacrifice 42 percent of the stock’s market value relative to holding a well-diversified equity portfolio. Ramaswamy (2002) also finds that the cost of holding employer stock is cost prohibitive using option-pricing techniques to price the risk of insuring the stock. These costs are illustrated by the dramatic bankruptcy of Enron and a collapse of the retirement accounts of employees.
Figure 1 from the Bureau of Labor Statistics shows the percentage of employer sponsored retirement accounts that allow employee contributions to employer stock and those that allow employer contributions to employer stock. There has been a significant reduction in plans that allow employee contributions to employer stock. While this is a significant improvement in terms of diversification, there have been several suggestions for steps to take to limit employer stock holdings even further. While the Pension Protection Act of 2006 mandates that all employer stock purchased by employees be immediately diversifiable and employer stock in matches be diversifiable within three years, Beshears, Choi, Laibson, and Madrian (2008) show that default options play a significant role in determining asset allocations in 401k retirement accounts.
The purpose of this paper is to empirically investigate the characteristics of people who hold employer stock and those who hold higher amounts than others. Are people aware of the riskiness of their choices, seen in the data as a correlation between risk preferences and amount of company stock held? Are there important demographic characteristics that suggest a worry trend of at-risk groups making poor investment decisions?
First I look at reported risk preferences for a variety of different groups. This measure has often been seen as a puzzling finding in household surveys, in particular the Survey of Consumer Finances. Lusardi (2006) proposes that the presence of a large majority stating they are unwilling to take any financial risk may be capturing not only high risk aversion but also the fact that many respondents do not know or understand risk diversification. However, much of the literature shows a correlation of reported risk preferences with measures of observed risk preferences. Sunden and Surette (1998) include the investor’s stated risk preferences as explanatory variables in an equation explaining allocation of defined contribution pensions into mostly stocks or mostly bonds and find a significant correlation. Schooley and Worden (1996) also conclude that stated risk preferences are consistent with observed portfolio allocations by using one-way analysis of variance methods. Research by Jianakoplos (2002) illustrates that, although contradictions can be found in the stated versus observed financial risk taking of households (a measure defined as risky financial assets/total financial assets), the measures are consistent at the ordinal level. Because of changes in the SCF questions, the 2007 data does not include an easy way to determine risky versus risk-free retirement account assets, making it difficult to reconstruct a very accurate observed risk metric. Therefore, I will be using the reported risk as a suggestive metric instead of as a robust risk metric.
I then perform several regressions to determine the marginal effects of a variety of variables on having a pension. I find that whites, those with higher income, as well as those who report having a high risk preference and a good understanding on the question are positively correlated with having employer sponsored retirement accounts.
I then look at determinants of holding company stock and find little consistent evidence of significant variation among pension holders. However, when looking at the amount of company stock held I find significant predictors. Those who are white tend to hold significantly less company stock as are those who are young, likely because there were fewer opportunities to invest in employer stock when they began holding employer sponsored retirement accounts. I also find that those who report they have more adequate retirement income hold more company stock, either a concerning trend that they view holding company stock as a risk-reducing measure or that they hold other retirement assets that allow them to hold riskier assets in their employer sponsored plan. I also find that those who report shopping around for investments are less likely to hold employer stock.
The Survey of Consumer Finances (SCF) is a triennial survey conducted by the Board of Governors of the Federal Reserve System in cooperation with the Statistics of Income Division (SOI) of the Internal Revenue Service. The SCF is intended to provide reliable information on financial characteristics of U.S. households. Detailed information is collected on all types of assets and liabilities, income, employment history, pensions, demographic characteristics, the use of financial services, and a variety of opinion variables on risk preferences and perceived financial literacy.
Each survey wave has combined an area probability sample with a high-income oversample, which allows the SCF to provide accurate information on broad population characteristics while also offering in-depth information on wealthy households. There has been several thousand man-hours searching for errors in the data and resolving those errors, which results in a robust data set of survey data.
The SCF also uses a multiple imputation method to approximate the distribution of any missing data in the survey. While the actual number of respondents is 4,418, there are 22,090 observations in the data. To avoid inflated statistical significance results arising from the implicates inflating the number of true independent observations, I compute correct standard errors as designated at the beginning of the codebook. 
Since I am estimating structural models of people within the survey and not estimating descriptions of the population, I will follow standard practice among SCF users and perform analysis on an unweighted basis. Also, all summary statistics will be reported on a weighted basis, which will prevent broad aggregates, such as income, from being overestimated, but may cause there to be outliers in variables that only pertain to a small sample, which have been evaluated on an individual basis and, if substantial, will have the unweighted estimates presented.
I first present summary statistics describing the population in Table 1a. As expected, the results follow population statistics reported in the census such as three-quarters of the people being white and the age distribution roughly following that described in the census. With the weighted sample exhibiting similar characteristics to the population as we know it, I began to look at differences between the entire sample and those who hold employer sponsored pensions.
Because of the controversy surrounding pension valuation, instead of attempting to value employer stock holdings from all pensions held, I look at the first reported pension accounts of both the head of the household and their spouse and if only one has reported holding company stock, they are included and if they both hold company stock, the one with the greater percentage of company stock held is used. All other pension funds (listed as the second, third and all other remaining in the data) are more highly skewed towards funds types that are non-traditional retirement accounts (rows 4 through 6 in Table 1b).
There is a roughly consistent racial distribution with 76 percent of pension holders being white. There is a significant change in the age distribution with 50 percent versus 38 percent being within the ages 45 and 64 and also increases in the 35 to 44 range and significant decreases in the under 35 category (62 to 10 percent) and the over 65 category. This is consistent with the idea that those above 65 are retired and from 35-64 people have settled into a job that has an employer sponsored pension plan and have thought about planning for retirement. For those under the age of 35, they may not be making a high enough income to put a significant portion away into a pension or are making investments in housing or children instead of saving. There is also a change in income distribution for those who hold pensions. Nearly 35 percent of pension holders are in the top income quintile with only 3 percent of pension holders being from the bottom quintile. This is consistent with the story put forth by Scholz, Seshadri, and Khitatrakun (2006) that lower income households can depend on social welfare programs in retirement so they do not need to hold retirement accounts.
Table 1a also presents, conditional on having a pension, those who hold company stock. We look at those who hold company stock from their current employer because, while company stock holdings from previous employers are likely to represent large undiversified portions of their holdings, it is no longer correlated with their labor income, which makes it less costly than that from a current employer. Figure 2 shows a graph of company stock holding conditional on having company stock in the pension. The general trend is as would be expected: downwards sloping with a mean of around 20 percent, the mean of which is remarkably similar to the statistic put together by the EBRI on company stock holdings. The racial distribution again remains consistent and the age distribution becomes more skewed towards younger age buckets. This is consistent with the principle that, as a young person, you can hold equities and other risky assets to gain higher returns. Also becoming more skewed is the distribution of income. The percentage of people in both the top and second highest quintile increase by 5 percent, both increases coming from the 2nd and 3rd quintiles and leaving the lowest quintile unchanged. This increase perhaps reflects the idea that those with company stock options are at the largest of companies making the highest salaries.
Perhaps the most interesting results in Table 1a are the average percent of company stock held in their pension. Here we see that males typically hold 7 percent more in employer stock than females (36 to 28 percent), perhaps because of a male tendency to participate in riskier activities. Another interesting result is that whites hold nearly 15 percent less company stock than other races. The exact cause of this is hard to pin down, but further regression analysis will be performed later in the paper.
There is also a relationship between age and percent of company stock held. Conditional on having company stock, the older groups have higher percentages of stock. This is consistent with the historical trend of company stock, having remained exceptionally high for many years when these older pension holders entered into pensions and the extensive documentation by Choi, Madrian, and Laibson (2005, 2009) of a strong flypaper effect for asset allocations in savings accounts.
Of note is the multimodal distribution of the percentage of company stock held in pensions with peaks occurring at 1, 50, and 100 percent. The first mode is likely due to the tendency for people to be generally diversified, holding many different uncorrelated assets to reduce risk. The second mode at 50 percent of assets is consistent with the literature on employer matching policy (Brown, Liang, Weisbenner 2004, Profit Sharing/401(k) Council of America 2002) that suggests 30-40 percent of employers with company stock options required their match to be made in employer stock. This assumes that the matching threshold exhibits an “anchoring effect” that serves as a salient starting point in the decision of which contribution rate to select. (Kahneman and Tversky 1974) While this paints a pretty picture, what has been called judgment for factual questions in the psychology literature may be at play where responses recalling specific numbers may tend to be biased towards approximations that center around very rounded numbers, such as 50 (Tourangeau, Rips, Rasinski 2000). The final mode at 100 percent was especially intriguing. The bucket at 100 percent includes over 20 percent of those who hold employer stock. At first look it seems that they could be profit-sharing accounts that commonly have high percentages of company stock, but Table 1b shows that very few plans are profit sharing or ESOP plans and upon further examination only a single observation is both invested 100 percent in stock and either a profit sharing or ESOP plan.
Table 1a also presents summary statistics of the characteristics of those who hold 100 percent of pension assets in company stock. As far as race is concerned, there are 15 percent fewer whites than in the sample of those who hold company stock. There is some change in the age distribution. For ages under 35 there is a decrease of 10 percent and for those over 65 there is an increase of 6 percent. There is also a significant shift in the income distribution, with there being 20 percent more respondents from the top quintile and 15 percent less from the second highest quintile.
Table 2 reports the reported risk preference of respondents. Consistent with other studies we see that the percent of respondents who reported that they are not willing to take any financial risks is very high at 42 percent. However, when looking at those with employer sponsored retirement accounts, there is a significance of 20 percent fewer reporting they are not willing to take any financial risk. This reduction results in an equal increase of 10 percent in both the average and above-average reported risk levels. It is also clear that the percentage of those who reported they took “substantial financial risks expecting to earn substantial returns” remains constant through all groups, including the top decile of those who hold company stock. This seems to fit well with the extensive literature by Lusardi that many respondents do not know or understand risk diversification. However, the more interesting result is that risk preferences remain consistent over groups that should exhibit significantly different risk levels.
Table 3a explores the respondents’ level of suspicion of the interview, as reported by the survey taker. This measure of suspicion captures the pre-interview suspicion, and can be viewed as a measure that reflects the respondents’ level of financial distrust. This variable and other related ones were not asked prior to the 1989 SCF, but analysis since then has shown that it may be a fairly accurate measure with respect to expected demographics such as higher levels of suspicion among non-white households, the old, and households that report being risk-averse (Bergstresser and Beshears 2010). Most households (54 percent) were not recorded by the interviewer as appearing suspicious. Approximately one-third of households appeared somewhat suspicious, while close to ten percent appeared very suspicious of the interview process. Table 3b shows that, controlling for other characteristics, pre-interview suspicion was relatively unchanged across varying income. Reported suspicion appears higher among non-white households than among white households and among the old than among the young. Households with higher education appear to have lower levels of suspicion.
Table 3a also shows the respondents’ apparent comprehension of the interview question. This measure can be viewed as a measure that reflects the financial literacy of the respondent. Since there are no questions included to directly measure financial literacy such as those designed by Lusardi and Mitchell (2006) for the 2004 Health and Retirement Study, I include Table 3b, which shows that, consistent with the large volume of literature on financial literacy, race, education, and gender are predictive of understanding the questions as reported by the interviewer. I find that nonwhites are more likely to not understand the questions. I also find that low income and low levels of education suggest worse understanding of the questions.
The main results are presented in Table 4a and Table 4b. Regression (1) and (2) explore the determinants of being a participant in an employer sponsored retirement account. Both the OLS and probit results are shown and have consistent marginal effects; however, significance is different in some cases, so the probit regression model (2) will be used. Since the marginal effects are reported for the probit model, the coefficients are easily interpretable. The results suggest that a 10 percent increase in income is associated with a 3.5 percent increase in having an employer sponsored retirement account. Being white and having a higher education is also associated with having an employer sponsored retirement account. Also, it appears that having low risk preferences is associated with not being in a retirement account. The results also suggest that those who have a poor understanding of the questions are less likely to have an employer sponsored retirement account. There is certainly potential reverse causality bias in this instance because having a pension account could require one to make financial decisions and make one more aware of the terminology used in the survey. Additional independent variables include age, education, the household’s taste for shopping around for investments, and suspicion of the interview.
In regression (3) and (4) from Table 4a I explore the existence of company stock within a person’s pension fund. The sample is somewhat smaller than the sample in regressions (1) and (2); only retirement account holders are included in the regression. I add independent variables for contribution percent and a variable of the response to the question, “Using any number from one to five, where one equals totally inadequate and five equals very satisfactory, how would you rate the retirement income you (receive or expect to receive) from Social Security and job pensions?”
The marginal effects from the probit model in regression (4) suggest that the four years from a college education would lead to a 2.4 percent decrease in the likelihood of owning company stock in one’s retirement account. Also, with 10 percent significance, a change in the risk preference from high to low would result in a 5 percent decrease in the probability of holding company stock. While age and the log income are significant in the OLS regression (3), they become very insignificant in the probit model. The varying significance between the OLS and probit estimations leads to a hesitant acceptance of the suggested coefficients of the better fit (double the r-squared value) probit regression (4).
Regressions (5) through (8), in Table 4b, limit the sample to households that reported holding company stock in their retirement assets. The sample size is smaller in these regressions. Regression (5) looks at just risk preference as an independent variable, but, as seen in regression (3) in Table 3, risk preference for the entire sample is highly correlated with demographics, a result that holds for all samples in these results. This correlation means the risk coefficient is picking up the many demographic effects. Regression (6) is another simplified regression including only demographic characteristics and log income as independent variables. Regression (7) looks at these demographic characteristics, log income, and risk preferences as independent variables. Risk preference now has the expected sign indicating lower risk tolerance would be associated with a decrease in company stock holdings; however, this result is not significant and remains only suggestive.
Regression (8) includes contribution percent and a variety of opinion variables included the previous regression. With a continuous dependent variable of the percent of company stock conditional on having stock, coefficients are interpreted as a coefficient percent increase in company stock held. The coefficient estimate of -16.07 on the white variable indicates that being white resulted in holding 16.07 percent less company stock on average. With respect to the variable on level of shopping around for investment, a change in the level of shopping from almost no shopping to a great deal of shopping was associated with a 20.5 percent decrease in company stock held. Also, contradictory to what would be expected, a change in the response to the question of adequate retirement from inadequate to very satisfactory would result in an 18.17 percent increase in the company stock held. While risk preference comes in with the expected sign, more risk adverse would result in a decrease in company stock holdings; however, it is not significant.
As discussed earlier, the top decile of company stock holders looked to be an interesting group. Regression (10) uses a dependent binary variable that equals 1 for those who are in the top decile of company stock holders. White comes in extremely significant with the probability of being in the top decile decreasing by 15.8 percent among white respondents. Adequate retirement income comes in with 10 percent significance, suggesting that a change from inadequate to very satisfactory was associated with a 20 percent increase in the probability that they are in the top decile. The similarity of these results to regression (8), the OLS regression with the same independent variables, suggest that the top decile may be a significant driver of the results in regression (8).
Regression (9) removes the top decile of company stock holders and reruns the same regression as (8). These results show no variables coming in as significant other than age at the 10 percent significance level. This confirms the belief that those in the top declie drive much of the significance in the previous regressions.
The evidence in Table 4 suggests that there is a significant correlation between race and being in the top decile of company stock holders. As regression (9) and (10) illustrate, the race effect goes away when the top decile of people, with respect to percentage of employer stock held, are not included and is extremely significant in (10) which shows, with high significance, that whites are 15.8 percent less likely to be in the top decile of those who employer stock holdings. This suggests that the race effect is concentrated almost entirely in those who own 100 percent employer stock. I present one explanation of this result.
Madrian and Shea (2001) document a strong race effect in default behavior resulting from participant inertia and from employee perception of the default as investment advice. They show that whites are less likely to remain at the default option condition on participation. The sample that I am looking at is those who have enrolled in an employee sponsored retirement plan, so conditional on being enrolled, there is a greater number of minorities, especially those who are black, who tend to stay at the default rate. Along with evidence already present on the flypaper effect in asset allocation, the presence of the race effect in company stock holdings is plausibly caused by some default and flypaper effect.
Also of interest is the insignificant risk preference as an indicator of either holding company stock or, conditional on holding stock, the percentage of company stock held. It enters with the correct sign in all regressions with other controls, which suggests there is a potential tie between reported risk preferences and company stock holdings. As Jianakoplos (2002) documented, reported risk preferences are significant at the ordinal level but not the quantitative level. This reduced predictive capability can be controlled for using more complex econometric methods, which is necessary in future research to pinpoint the cause of these interesting findings on risk preference.
Among households with a non-white head of house, who are older, who report having adequate retirement income, or who have reported doing little shopping for investments were more likely to hold higher percentages of company stock conditional on having company stock, controlling for a wide range of other factors. Furthermore, these relationships seem to be strongly driven by those in the top decile of company stock holdings. These results can possibly be explained by an interaction between structural characteristics, such as default options and employer matchings, and employer stock holdings. This is reinforced by the results from reported risk preferences, which should be a strong predictor of holding a risky asset such as company stock, coming in as only suggestive, not significant.
Alonso, Luis, Sarah Holden, and Jack VanDerhei, 2009 “401(k) Plan Asset Allocation,
Account Balances, and Loan Activity in 2008,” EBRI Issue Brief, no. 335, and ICI Perspective, Vol. 15, no. 2, October 2009.
Agnew, Julie, Pierluigi Balduzzi, and Annika Sundén. 2003. “Portfolio Choice and Trading in a Large 401(k) Plan.” American Economic Review, 93(1): 193–215.
Benartzi, Shlomo, 2001, “Excessive extrapolation and the allocation of company stock to retirement accounts”, Journal of Finance 56(5), 1747-1764.
Benartzi, Shlomo, Richard Thaler, Stephen Utkus, and Cass Sunstein. 2004. Company stock, market rationality, and legal reform. University of Chicago, Graduate School of Business. Mimeograph.
Beshears, John, James J. Choi, David Laibson and Brigitte C. Madrian (2008). “The Importance of Default Options for Retirement Savings Outcomes: Evidence from the United States.” 2008. In Stephen J. Kay and Tapen Sinha, editors, Lessons from Pension Reform in the Americas, New York: Oxford University Press, pp. 59-87
Boston Research Group, April 2002, “Enron Has Little Effect on 401(k) Participants’ View of Company Stock,” Boston: MA.
Brown J., Liang N., Weisbenner S., 2006,”401(k) Matching Contributions in Company Stock: Costs and Benefits for Firms and Workers,” Journal of Public Economics; 90:1315-46.
Choi, James J., David Laibson, Brigitte Madrian, and Andrew Metrick, 2004 “Employees’ investment decisions about company stock”, Working paper No. 10228, National Bureau of Economic Research.
Choi, James, David Laibson, Brigitte Madrian, and Andrew Metrick. 2004. “Saving for Retirement on the Path of Least Resistance.” In Behavioral Public Finance: Toward a New Agenda, 2006, ed. Edward J. McCaffrey and Joel Slemrod, pp. 304-351. Russell Sage Foundation.
Choi, James J., David Laibson, and Brigitte C. Madrian. 2005. “Are empowerment and education enough: Underdiversiﬁcation in 401(k) plans.” Brookings Papers on Economic Activity, Issue no. 2:151–98. Washington, DC: Brookings Institution.
Choi, James J., David Laibson, and Brigitte C. Madrian. 2009. “Mental accounting in portfolio choice: Evidence from a flypaper effect.” American Economic Review 99, pp. 2085-2099.
Huberman, Gur, and Paul Sengmuller, September 2002, “Company Stock in 401(k) Plans,” Working Paper, Columbia University.
Jianakoplos, Nancy A. 2002. Invest as I say, not as I do? Gender differences in financial risk preferences. International Business & Economics Research Journal. 1:1-8.
Kahneman, Daniel and Amos Tversky (1974), “Judgment under uncertainty: Heuristics and biases,” Science, 185: 1124-1131.
Kruse, L. Douglas, 1993, “Profit Sharing: Does It Make a Difference?” W.E. Upjohn Institute for Employment Research, MI: Kalamazoo
Liang, Nellie and Scott Weisbenner, 2002, “Investor behavior and the purchase of company stock in 401(k) plans – The importance of plan design”, Working paper, University of Illinios.
Mitchell, Olivia S. and Utkus, Stephen P., (2005) “The Role of Company Stock in Defined Contribution Plans.” NBER Working Paper Series, Vol. w9250, 2002.
Meulbroek, L. (2005) Company Stock in Pension Plans: How Costly Is It?, Journal of Law and Economics 48, 443–474.
Ramaswamy, Krishna, 2002, “Company stock and DC plan diversification”, Working paper, The Wharton School of the University of Pennsylvania.
Scholz, John Karl, Ananth Seshadri, and Surachai Khitatrakun, “Are Americans Saving ‘Optimally’ for Retirement,” Journal of Political Economy 114(4), August 2006a: 607-643.
Schooley, Diane K. and Debra Drecnik Worden. “Risk Aversion Measures: Comparing Attitudes and Asset Allocation.” Financial Services Review, 5(2) 1996, pp. 87-99.
Sunden, Annika E. and Brian J. Surette. ” Gender Differences in the Allocation of Assets in Retirement Savings Plans.” American Economic Review, May 1998, pp. 207-211.
Tourangeau, Roger, Lance J. Rips, and Kenneth A. Rasinski. The Psychology of Survey Response. Cambridge: Cambridge UP, 2000. Print.
-  This method reconciles the multiple imputations by computing regressions for each of the implicates. I then take the average of the coefficients of each of the implicates for the actual coefficients and then take the average of the variances of each implicate, add a weight designated by the survey, which includes squared deviations of the coefficient from the average, and take the square root to arrive at the correct standard errors.↩