Data source

6We analyze a cross-sectional dataset, collected in 2005 from a French CAC 40 listed bank whose employees are eligible to a company savings plan. Our sample represents 44,649 employees who invest (or not) in the company-based savings plan. Information is captured at the employee level. We only observe employees’ investment in the company-based plan. Our data provides information dealing with the composition of employees’ company savings: stocks, bonds, and monetary assets, as well as diversified funds comprising a variable proportion of these three latter categories. In our sample, the number of investment options varies from 2 to 27 across the 84 firm’s subsidiaries. Each subsidiary belongs to one of the bank’s five business lines: retail banking; specialized financial services; asset management/insurance/private banking; corporate banking and investment banking. Despite differences across business lines, the company’s policy is homogenous and only differs regarding to the plan menu offered. One subsidiary is in charge of payments, buybacks, transfers and information associated with the plan.
Our data source is noteworthy for several reasons. Most publications rely on questionnaires (Bergstresser and Poterba, 2004; Bertaut and Starr-McCluer, 2002; Heaton and Lucas, 2000; Poterba and Samwick, 2003; Tracy et al., 1999). Yet, data collection by questionnaire can be accompanied by selection bias, with respondents liable to form a specific sub-category of the population whose characteristics one wishes to measure (Kennickell, 1998). In France, there is little data on the distribution and the total amount of company-based savings. This lack of availability of empirical data is a major obstacle to studying company-based savings (Balligand and Foucault, 2000). This explains why most empirical studies took place in the United States where administrative data are available to researchers through 401(k) investments. Since administrative data are not systematically collected by foreign public authorities, empirical studies rely more on one-company case studies. Degeorge et al. (2004) investigate the determinants of investments in France Telecom’s stocks by its employees, and focus on a very specific context: the partial privatization of the company in 1997. Our study takes place in a more general context, and is not restricted to company stocks investments. We also explore employees’ investment in several other asset categories.

Measures

7We have four dependent variables of interest: (i) the participation in each asset category; (ii) the amount invested in each asset category; (iii) the number of asset categories selected; and (iv) the number of funds selected. Participation is measured using six dichotomous variables describing investment in the following asset categories: stocks overall, non employer’s stocks, company stocks, bonds assets, monetary assets, and diversified assets. Continuous variables measure conditional amounts invested in each asset category. The number of asset categories and funds selected are count variables.

8We control for several socio-demographic variables (age, gender, Paris region residency) and human capital characteristics (salary, time with company, financial expertise, and education). Age and time with the company are measured in years. Salary is the gross annual fixed wage in Euros. Education is a dichotomous variable measuring whether the employee holds a Master’s Degree or higher [2]. In order to control for disparities in saving behaviors according to geographic location, a dichotomous variable measures if the employee lives in the Paris region. Since our sample comes from a bank, we are able to estimate financial expertise. Financial expertise is assessed using a dichotomous variable measuring if employees hold a job requiring knowledge of portfolio management (asset management, finance and investment banking, private equity, financial services and risk support function). To measure how the number of funds offered by a given subsidiary affects employees’ behaviors, we create three groups: 1-5 funds, 6-10 funds, and >10 funds.

Statistical analysis

9Objective 1 is to describe employees’ saving behaviors for each asset category offered within the plan. We first compare investors’ characteristics to non-investors’ characteristics, using Chi-squared tests (Table 1). We then compute the distribution of employee savers by asset category, to identify the number of employees (and associated percentage of the population) investing in a given asset category. We use two different distributions: the percentage represented by each asset category in the company savings plan (see Table 2A), and the value (in €) of investments in each asset category (see Table 2B). We also study these distributions across employees’ demographic characteristics: gender, salary, age, time with company and residence (see Tables 3A and 3B). We run means comparison tests to control whether investment patterns significantly differ between groups.

Table 1

Description of employees’ demographic characteristics

Description of employees’ demographic characteristics

Table 2A

Distribution of employee savers by asset category – Percentage of the company savings plan

Distribution of employee savers by asset category – Percentage of the company savings plan

Table 2B

Distribution of employee savers by asset category – Value in Euros

Distribution of employee savers by asset category – Value in Euros

Table 3A

Statistics by group - Percentage of the company savings plan

Statistics by group - Percentage of the company savings plan

Table 3B

Statistics by group – Value in Euros

Statistics by group – Value in Euros

10Objective 2 is to examine how human capital variables influence amounts invested and participation in each asset category. We use Heckman’s two-step regressions to study which factors are related to the amount invested in each asset category, conditional on the decision to invest in this category (see Table 4A). The Heckman’s two-step statistical approach offers a means of correcting for non-randomly selected samples and selection bias. Indeed, company based plans investors may be less risk averse than non-investors, and therefore choose to invest larger amounts. The first stage models the probability of investing in each asset category, e.g. stocks overall, non employer’s stocks, company stocks, bonds, monetary and diversified assets. The second stage models how much is invested in each asset category conditional on the probability of investing. We test for the presence of selection bias in that process controlling for the inverse Mills’ ratio statistic in the second equation. This parameter is defined as the product of the cross-equation correlation and the standard error of the residuals from the outcome equation. Both steps (investment, conditional amount invested) control for age, time with company, salary, financial expertise, Paris region residency, education, and gender.

Table 4A

Determinants of the amounts invested in the asset categories – Results of the Heckman’s two-step regressions

Determinants of the amounts invested in the asset categories – Results of the Heckman’s two-step regressions

Table 4B

Determinants of the amounts invested in the asset categories – Marginal effects after the Heckman’s two-step regressions

Determinants of the amounts invested in the asset categories – Marginal effects after the Heckman’s two-step regressions

11Objective 3 is to estimate to what extent investment options offered by the employer are associated with employees’ portfolio breadth. Portfolio breadth is measured by the number of asset categories selected (Table 5A, first column), and the number of funds selected (Table 5A, second column). We use count dependent variables and Poisson regressions. To account for the employer’s influence on employees’ investment choices, we control for three dummy variables measuring the number of funds offered. We control for the same independent variables as before: age, time with company, salary, financial expertise, Paris region residency, education, and gender. In sensitivity analysis, we use the ratio of the number of funds selected over the number of funds offered as an additional dependent variable consistently with Huberman and Jiang (2006).
To demonstrate the magnitude of obtained effects for objectives 2 and 3, we compute marginal effects (see Tables 4B, 5B).

Table 5A

Determinants of portfolio breadth – Results of Poisson regressions

Determinants of portfolio breadth – Results of Poisson regressions

Table 5B

Determinants of portfolio breadth – Marginal effects after the Poisson regressions

Determinants of portfolio breadth – Marginal effects after the Poisson regressions

Objective 1

13Table 2A shows the distribution of employees according to the proportion of company savings invested in each asset category. Large proportions (between 28% and 91%) of employees do not hold each of the asset categories (x = 0). Monetary assets and company stocks appeal respectively 5% and 4% of employees to invest 100% of their savings. A large proportion of savers do not invest in bonds funds (91%). The most popular asset category is the diversified funds category (18% for80 ? x ? 100).

14Table 2B displays the distribution of employees according to the value (in Euros) of company savings invested in the asset category. Diversified assets, monetary assets and company stocks are the most widespread assets within lower ranges (€0 – €5,000 and €5,000 – €10,000). Few employees (4%) are holding more than €50,000 in company stocks.

15Table 3A shows proportions invested in each asset category for different groups of employees. All means comparison tests are significant, showing that the means of the groups are different. The low proportion of company savings hold in bonds assets is a characteristic of the entire sample, since this proportion never exceeds 4.48%. Diversified funds are popular among employees with lower income (37.21%) and senior employees (34.47%). The increase in the proportion of employee savings across age is curvilinear, and high annual salary levels are associated with larger rates of stockholding. Exposure of lower paid employees to investment in company stocks is high (30.31%) compared to other stocks holdings (0.5%). This trend is less pronounced for employees with higher income. Overall, the proportion of stockholding is uniformly distributed across age.
Table 3B displays value (in €) invested in each asset category across groups of employees. Male employees and those living in the Paris region hold significantly higher amounts of monetary assets. Differences in amount invested between men and women are much more pronounced for the most risky asset categories. Compared to women, men invest twice the amount in stocks. Overall, higher annual salary is associated with higher assets holding. Age and time with the company similarly affect the distribution of amounts invested in each asset category.

Objective 2

16Results of Heckman’s two-step regressions are reported in Tables 4A and 4B. Inverse Mills’ ratios are significant for all models, revealing the presence of selection biases based upon the significant correlation between the error terms in the two stages. This suggests that the two equations measuring the probability of investing in each asset category and the conditional amounts invested cannot be modeled separately. We trace investment behaviors in the following asset categories: stocks, bonds, monetary assets, diversified funds, non employer’s stocks and company stocks. Stocks, non-employer’s stocks, and company stocks are the most risky investment options whereas bonds, monetary, and diversified assets are the safest investment options.

17The first stage of Heckman regressions models the probability of participating in each asset category. Age, salary and financial expertise are associated with a higher probability of risky asset investment. Men and employees with high education level (? Masters’ Degree) have a lower probability of risky asset investment. Time with the company is not associated with a higher probability of investment in company stocks, but is associated with an increase in the probability of investment in stocks overall and non-employer’s stocks. Living in the Paris region is associated with an increase in the probability of investment in non employer’s stocks, but it is also associated with a decrease in the probability of company stocks investment. Though, Paris region residency is negatively associated with the probability of investment in risky assets. Other factors associated with the probability of investment in bonds and monetary assets are: increased age (+), increased time with company (–), increased salary (+), presence of financial expertise (+), and being a man (+). Determinants of investment behavior in diversified assets are specific. Salary and financial expertise are associated with a smaller probability of investment in diversified funds. On the contrary, education is positively related to diversified assets investment. For all other assets (risky and non risky), best educated employees have lower probability of investment, and financial expertise is associated with a higher probability of investment.

18The second stage models amounts invested in each asset category, conditional on participation. Among employees who decide to invest in risky assets, increased age is positively associated with invested amounts. An increase in time with the company is associated with an increase in amounts invested in non employer’s stocks, and a decrease in amounts invested in company stocks. Employees with high salary invest lower amounts in riskier assets. Conditional amounts invested in non-risky assets increase with age, time with company, salary, and education. Employees with financial expertise invest lower amounts in non-employer’s stocks than employees who work in areas not related to finance. Education is also positively linked to amounts invested in non employer’s stocks. Financial expertise is not associated with amounts invested in bonds and diversified assets, but has a negative association with monetary assets investment. Again, education and financial expertise have asymmetric associations with investment behaviors. Men invest higher amounts in company stocks, bonds and monetary assets, when women are more likely to invest in diversified assets. Living in the Paris region is associated with larger amounts invested in stocks overall and company stocks, and lower amounts invested in non-employer’s stocks.
Table 4B reports marginal effects computed after two-step models estimations. Financial expertise was associated with an incremental increase in 6.4% in the probability of investment in stock overall, 13.7% in the probability of investment in non employer’s stocks, 2.7% in the probability of investment in company stocks, 7.1% in the probability of bonds investment, and 14.8% in the probability of monetary assets investment. Compared to other employees, those with financial expertise have 15.3% lower probability of investing in diversified assets. Employees have a higher probability of increasing their investment in riskier assets, e.g. stocks overall and company stocks.

Objective 3

19Results of Poisson regressions on the portfolio breadth measures are displayed in Table 5A and 5B. In Table 5A (column 1), the association between the number of funds offered and the number of assets selected is not significant. In column 2, where the dependent variable is the number of funds selected, these coefficients are positive and significant at the 1% level. This indicates that the number of funds offered is associated with higher number of funds selected, but not with a greater number of assets selected. Living in the Paris region is associated with a larger number of funds selected, and a smaller number of assets selected. Portfolio breadth is more pronounced among older, better paid employees with greater financial expertise. Being a man and holding at least a Master’s Degree is associated with lower portfolio breadth within the company based savings plan. Results displayed in Table 5B confirm that the number of funds offered has a large influence on the number of funds categories chosen. Compared to employees who have less than 6 funds offered, employees who have between 6 and 10 investment choices have 6.43% more chance of investing into a larger number of funds.
In sensitivity analyses (not reported), we measure the utilization rate of funds offered by using the ratio of the number of selected funds over offered funds. This additional dependent variable is also used in Huberman and Jiang (2006) to investigate the number of funds chosen. Obtained coefficients have same sign and significance as those already reported in Table 5A. Age, salary and financial expertise are associated with increased utilization rates. Being a male, living in the Paris region, and holding a Masters’ Degree are associated with a decrease in the ratio. The only variable that differs is time with the company, which has a positive association with the ratio, while the coefficient is negative in Poisson regressions.

Findings

20Our results show the need for modeling separately each step of employees’ saving behavior, as investment steps (participate in offers, conditional amount invested) are correlated [4]. We emphasize three main results: (i) presence of extreme strategies, (ii) positive influence of human capital, and (iii) influence of the plan menu on the number of funds selected by employees. In this section, we discuss to what extent these effects are expected and make sense, and if these results are surprising given predictions of saving theories.

21In the context of an experimental setting, Benartzi and Thaler’s (2001) find that employee savers seem to consider company stocks as a separate asset category. Our results are consistent with this finding. The distribution of employees by different stockholding proportions is homogeneous, but biased towards company stocks investments, which are distributed uniformly across employees. Our results show large investment in company stocks, and emphasize the existence of other extreme behaviors matching findings from Agnew et al. (2003). In our sample, significant proportions of employees do not invest in the company-based savings plan at all and do not hold any bonds at all.

22This finding can be explained by several factors. First, it can be argued that matching contributions and discounts on the stocks’ price may influence investment choices. However, we believe matching contributions have little influence on our results, since they are not only restricted to company stocks and apply to all other asset categories. Second, marketing efforts to promote company stocks investment may have introduced a bias towards this investment option. Marketing effort, which cannot be measured in our study, relied on several supports such as the intranet, stock price simulators, movies, posters, CD-ROMs and information meetings. However, the influence of marketing on our results may be reduced by the fact that all employees shared similar information. No sub-population was targeted by marketing campaigns. Third, high company stocks investments can be explained by the fact that investors face fixed costs due to the research of information about investment alternatives. Haliassos and Bertaut (1995) and Vissing-Jorgensen (2003) show that even moderate costs can discourage participation. Employees would choose to invest in their company stocks because they have privileged information about that investment category.

23Our results confirm the presence of gender differences in investment strategies. Barber and Odean (2001) show that overconfidence can lead men to adopt different strategies than women, particularly by exposing themselves more to risk. Our results provide additional information, suggesting that women are more likely to invest in company based plans, while men are more likely to invest larger amounts.

24Our analyses show that investment is associated with several measures of human capital: age, salary, education, time with company, and financial expertise. We find the presence of a positive effect of salary that can be seen as an implicit emolument of risk-free assets (Cocco et al., 2005). Following that theory, employees with high salaries would be more likely to invest in risky assets. To the extent that human and financial capital are correlated in the long-run (Benzoni et al., 2005), salary would have a negative effect on employees’ risk exposure.

25Our results confirm the importance of controlling for several proxies of human capital. We find that the effect of age, time with the company, education, and financial expertise, differ according to the asset category modeled. The negative effect of education falls within the discussion on rationality and savers’ limited information. Degeorge et al. (2004) show that the current value of human capital is a function of the current salary amount, its rate of growth, and the temporal horizon during which it will be paid. According to Heaton and Lucas (2000) and Viceira (2001), human capital is associated with an increase in real risk aversion, and leads to rational savings behavior. Employees with higher education may prefer to invest outside the company-based plan, as they have more information than other employees. Conversely, less educated employees prefer contractual savings, for which information research costs are low. The freezing of sums invested and restriction on payments certainly influence employees’ investment strategy. Company based savings can, from this standpoint, attract employees who encounter difficulties in keeping to the saving objectives they have set themselves (Thaler and Shefrin, 1981). Thaler and Benartzi (2004) show that restrictive measures on employees’ choices may correct behaviors that the neo-classical theory would describe as irrational. Such behaviors often result in insufficiently diversified portfolio and inadequate saving rate. The reason for investing in a constraining company savings plan would be a search for rationality, and a desire to properly protect oneself against the risk of future impoverishment (Thaler and Sunstein, 2003; Madrian and Shea, 2001).
Results from Poisson regressions suggest that the number of funds offered by the plan is significantly associated with the number of funds used by plan participants. Although consistent with experiments of Benartzi and Thaler (2001), our findings differ from results obtained by Huberman and Jiang (2006). Huberman and Jiang (2006) find no significant relationship between the number of funds chosen and the number of funds offered. They explain that 401(k) plans participants first consider fund categories (stocks, bonds, diversified, monetary), and then choose individual funds within each category. We do not have the same pattern in the plan we study where funds appear directly.

Limitations

26Our study is faced with several limitations. Focusing on one firm does not allow measurement of variables in relation to potentially significant characteristics of the company such as sector, size and risk. Studying a banking group may raise concerns about the generalizability of results, because employees probably have greater knowledge of financial products and easier access to information than other employees. Furthermore, the association obtained with the variable describing time spent with the company could be biased by the use of a cross-sectional dataset. This would explain the negative association obtained which conflicts with usual results. According to Ameriks and Zeldes (2005), studies using cross sectional data may not fully capture age effects. Our data does not provide information dealing with non-company based financial savings. As we only observe a part of employees’ investment, it would be very difficult to extrapolate behaviors on their total investment allocations. Some employees may have optimal portfolio allocation, which cannot be included in our analysis.

27Finally, our results and conclusions are limited to company-based savings only, and cannot be extended to other financial assets. When comparing our sample to the French population, we find differences in company-based savings levels. According to Rougerie (2006), the median French employee holds €1,800 in 2004 in his company-based savings plan, compared to €9,253 in our sample. Rougerie (2006) also emphasizes that in 2004, the firt decile of the French employees hold less than €300, and the last decile of the same population holds at least €14,200 in their company-based savings plan. In our sample, these amounts respectively are €149 and €57,338. However, Rougerie (2006) concedes that amounts described in her study may be underestimated by the fact that they rely on self-reported information.

Implication for future research

28Our results have several implications for policy and future research. “Libertarian paternalism” would recommend to monitor employees’ investment choices to maximize their well-being without prejudice to their freedom of choice (Thaler and Sunstein, 2003; Thaler and Benartzi, 2004). This position has not, however, remained unchallenged (Mitchell, 2005). According to Campbell (2006), household finance involves wondering what financial instruments households use to achieve their goals. Households hold a large number of assets which are not tradable, such as their human capital, or illiquid, such as their residence, and their capacity to take on debt is limited. This complexity is more conducive to investment mistakes. It can be inferred from Campbell’s comment that several variables determine individual savers’ portfolio choices. In this paper, we attempt to take several of these determinants into account. According to Iyengar et al. (2004), increasing the number of funds can be pernicious. Because of a “choice overload” effect, an abundant offer would result in discouraging saving; thus increasing the number of choices available would have a counter-incentive effect. Benartzi and Thaler (2001) point out that company based plan savers make significantly different portfolio choices when company stocks are offered. Further research into the role of employer stock offer for company based plans beneficiaries will be important.
Our results must be interpreted regarding to the increasing importance of company-based savings plans, which represent a growing and substantial part of French employees’ financial portfolio. Even if company-based savings plans success is recent (Rougerie, 2006), it is all the more noteworthy in that individual possession of other financial assets by French households is decreasing. While the proportion of French households holding securities decreased from 19.3% in 2006 to 18.6% in 2007, the growth of company savings has contributed to the stability of the proportion of French households holding securities (TNS-Sofres, 2007). This latter proportion was 23.8% in 2006 and became 23.6% in 2007 (TNS-Sofres, 2007). In 1998, only 9% of active workers over age 18 had company-based savings. In 2004, 11.5% of French employees (4.7 million people) held such savings, of whom 54% belonged to a company savings plan, and €12.7 billion of employee savings bonuses had been distributed to them. Between December 2004 and December 2007, those participating in company savings plans rose to nearly 65%, increasing in value from €57 to €94 billion. According to the French Financial Management Association, company stocks amounted to 48% of managed company savings assets (AFG, 2009).