Flypaper Effects in Asset Allocation - CiteSeerX

0 downloads 0 Views 154KB Size Report
May 18, 2007 - Madrian and Shea (2001) show that the employer specification of .... Madrian, Brigitte C. and Dennis F. Shea (2001). “The Power of Suggestion: ...
PRELIMINARY AND INCOMPLETE

Flypaper Effects in Asset Allocation

James J. Choi Yale University and NBER David Laibson Harvard University and NBER Brigitte C. Madrian Harvard University and NBER

May 18, 2007

We thank Hewitt Associates for providing the data analyzed in this paper. We are particularly grateful to Lori Lucas, Pam Hess, Yan Xu and Greg Tabackman, some of our many contacts at Hewitt. We appreciate the research assistance of David Borden, Ananya Chakravarti, Chris Nosko, and Neel Rao. Choi acknowledges financial support from the Mustard Seed Foundation. Choi, Laibson, and Madrian acknowledge individual and collective financial support from the National Institute on Aging (grants R01-AG-021650, P30-AG012810, and T32-AG00186).

Flypaper Effects in Asset Allocation

Abstract: We document a flypaper effect in individual investors’ asset allocation. We study a firm that twice changed the rules governing the asset allocation of its 401(k) employer matching contribution flows. These changes were economically neutral because after the matching contributions were made, employees could freely reallocate their match account assets. But we find that money sticks where it hits. Employees rarely reallocated their match account, and they did not adjust the asset allocation in their own-contribution account within the 401(k) to compensate for their match account allocation. Thus, these rule changes caused dramatic overnight shifts in employees’ asset allocations. We argue that mental accounting and investor inertia generate the flypaper effect. The flypaper effect may explain some of the anomalously high voluntary holdings of employer stock in 401(k) plans.

James J. Choi Yale School of Management 135 Prospect Street P.O. Box 208200 New Haven, CT 06520-8200 [email protected]

David Laibson Department of Economics Harvard University Littauer Center Cambridge, MA 02138 [email protected]

Brigitte C. Madrian Kennedy School of Government Harvard University 79 JFK Street Cambridge, MA 02138 [email protected]

2

In this paper, we document a flypaper effect in individual investors’ asset allocation. Like the traditional flypaper effect of public finance (Hines and Thaler, 1995), we find that money sticks where it hits: investors given securities in kind hold onto those securities for a long time with minimal corresponding offsets in other parts of their portfolio. We identify the asset allocation flypaper effect by exploiting two natural experiments at one large firm. Throughout the sample period, for each dollar an employee contributed to the 401(k) up to a limit, the company made an additional matching contribution to the employee’s 401(k). The company maintained the own-contribution and matching moneys in separate accounts, as is customary. Before March 2003, all matching contributions were made in the form of the employer’s stock. Once the matching contribution was made, the employee was free at any time to trade out of the employer’s stock into the other seven investment options in the 401(k). Starting in March 2003, the company no longer required employees to accept matching contributions in employer stock. Instead, employees who enrolled in the 401(k) were required to explicitly specify the asset allocation for their matching contribution flows—the mix of securities within the 401(k)’s investment menu they would receive as their matching contributions. This change was completely neutral from an economic perspective. Even though the prior regime made matching contribution in employer stock, employees could immediately reallocate their match account portfolios after the contribution was made. Therefore, any asset allocation feasible in the new regime could be easily replicated under the old regime. However, the change was far from neutral in practice. Employees who enrolled in the plan during February 2003 (one month before the change) held 24% of their own-contribution balances and 95% of their matching contribution balances in employer stock at year-end 2003. Employees who enrolled in the plan during March 2003 held 20% of their own-contribution balances and only 28% of their matching-contribution balances in employer stock at year-end 2003. Integrating the two 401(k) accounts, the fraction of total balances held in employer stock at year-end 2003 drops from 56% to 23% across the two enrollment cohorts. Looking at contribution flow allocations yields a nearly identical story. In April 2005, the company made a second change. The company automatically set the match flow allocation equal to the own-contribution flow allocation for all employees who had enrolled before March 2003 and who had not actively specified a match flow allocation. Nearly

3

all the previously passive employees remained passive in the face of this change, and matching contribution flows to employer stock plummeted overnight. The asset allocation flypaper effect is primarily driven by two forces: inertia and narrow framing within mental accounts. Much other research has documented high levels of individual investor inertia (Samuelson and Zeckhauser, 1989; Madrian and Shea, 2001; Choi, Laibson, Madrian, and Metrick, 2002, 2004a; Choi, Laibson, and Madrian, 2005a,b). In the flypaper context, inertia manifests itself in the failure to trade out of employer stock in the match account. Given that employees choose to hold only 28% of their match account and 23% of their total 401(k) portfolio in employer stock when required to choose matching-contribution flow allocations, it is remarkable that 95% of pre-March 2003 enrollees’ match account balances is held in employer stock. Mental accounting is the psychological segregation of subsets of the wealth portfolio (Thaler, 1985, 1990, 1999). Because money is fungible, a rational agent making a financial decision should consider its impact on her entire wealth portfolio, not just its effect on a subset of her portfolio. Mental accounters instead engage in “narrow framing” (Kahneman and Lovallo, 1993; Barberis, Huang, and Thaler, 2006), making decisions for money within each mental account without considering the money they have in other mental accounts. An enrollee in the first regime who was not a narrow framer but who understood that he was going to be passive in the future would have decreased his own-contribution allocation to employer stock to compensate for the fact that he would never change the 100% employer stock allocation in his match account. In fact, February 2003 enrollees have a higher own-contribution allocation to employer stock than March 2003 enrollees (24% versus 20%). The flypaper effect can help explain the high levels of employer stock ownership in 401(k) plans, which are puzzling because they run contrary to the logic of diversification. Fidelity Investments (2002) reports that 44% of employer matching contributions are made in employer stock, like in the first enrollment regime of our study company. Even though 66% of these companies do not impose restrictions on selling the employer stock, the flypaper effect causes diversification to be rare in practice even when allowed.1

1

See also Choi, Laibson, and Madrian (2005b) for evidence on passivity in the face of relaxed diversification restrictions.

4

The flypaper effect cannot explain why employees choose to voluntarily hold positive amounts of employer stock in their own-contribution accounts. However, it can explain why employer stock holding in own-contribution accounts is not lower in companies that require employees to hold employer stock in their match account than companies that have no such requirement.2 Narrow framing causes employer stock held in the match account to be accretive rather than a substitute for employer stock held in the own-contribution account, since employees do not adjust their own-contribution allocation decisions to account for the high employer stock holdings in their match account. Our paper is related to other work that has shown that narrow framing affects asset allocations. The reluctance to realize paper losses (the “disposition effect”) has been documented in many settings (e.g. Odean, 1998; Grinblatt and Keloharju, 2001; Wermers, 2003; Coval and Shumway, 2005; Frazzini, 2006) and is hard to justify for an investor who cares only about the total wealth portfolio. Risk aversion over small gambles is another common behavior that somebody who only cares about total consumption would not exhibit (Rabin, 2000). The disposition effect and risk aversion over small gambles can be optimal for agents who experience utility from gains and losses per se. The narrow framing result in this paper differs in that it is difficult to construct a plausible hedonic function that justifies such dramatic portfolio changes across the two plan regimes studied. It seems much more likely that employees at our firms engaged in narrow framing due to cognitive shortcomings. Our paper is also related to research that seeks to explain the determinants of employer stock holding in defined contribution savings plans. Benartzi (2001) and Huberman and Sengmueller (2002) show that past employer stock returns positively predict current employer stock holdings, suggesting that employees naively extrapolate past employer stock returns into the future. Choi, Laibson, and Madrian (2005b) show that even when holding requirements for employer stock are lifted, the average portfolio response is small. Huberman (2001) argues that employer stock holdings are motivated by the comfort of investing in an asset with which one is familiar, while Cohen (forthcoming) argues that employees’ loyalty to their company also plays a role.

2

Holden and Vanderhei (2001) report that employees with employer stock holding requirements invest 33% of their own-contribution 401(k) balances—which are not subject to holding requirements—in employer stock. In contrast, employees without such requirements invest only 22% of their 401(k) in employer stock.

5

The remainder of this paper proceeds as follows. Section I describes the details of the two plan changes at the company we study. Section II describes our data. Section III analyzes the plan changes’ impact on employees’ asset allocation. Section IV discusses the mechanisms underlying the flypaper effect. Section V concludes.

I. 401(k) Savings Plan Features at a Large U.S. Corporation We identify the flypaper effect in 401(k) asset allocation at a large publicly traded company in the retail sector. In December 2005, this firm had locations in all fifty U.S. states, as well as the District of Columbia and Puerto Rico. Table 1 lists the 401(k) plan features at this company. The company offers a generous employer match of 150% on the first 1% of pay contributed and 50% on the next 4% of pay (for a total matching contribution of 3.5% of pay for participants contributing at least 5% of pay). At year-end 2005, 59% of eligible employees participated in the 401(k). Besides the two plan changes that are the focus of this paper’s analysis, there were three other plan changes that merit mention. First, before 2003, employees were eligible to participate in the 401(k) savings plan starting 12 months after hire, provided they had worked at least 1,000 hours at the company. In April 2003, the eligibility requirement was reduced to 90 days of employment, although eligibility for the employer match was still restricted to those with 12 months of tenure and 1,000 hours of service.3 We will show that the measured flypaper effect is identical when we restrict our sample to an enrollment window where eligibility requirements remained constant. The second noteworthy plan change was an increase in the maximum employee contribution rate from 15% to 50% of pay in January 2002. This change affects only the 4% of sample participants who contributed between 16% and 50% of pay when the higher maximum is in effect. The third plan change also occurred in January 2002: an expansion in the number of 401(k) investment options. Prior to January 2002, participants had six investment options from which to choose, including employer stock. In January 2002, the company added two investment options, allowing participants to choose among a stable value fund, a balanced fund, five equity funds, and employer stock.

3

For a small number of employees, the eligibility change did not take effect until later in the year.

6

Because we identify the flypaper effects from discontinuities in employer stock allocations chosen around March 2003 and April 2005, these January 2002 plan changes should not contaminate our analysis. We also do not observe significant changes in employer stock allocations around the January 2002 enrollment window.

II. Data Description Our data come from Hewitt Associates, a large benefits administration and consulting firm. The data are a series of year-end cross-sections from 2002 to 2005 of all employees eligible for the 401(k) plan at our study company. These cross-sections contain demographic information such as birth date, hire date, gender, and compensation. They also contain point-in-time information on 401(k) savings outcomes including participation status in the plan, date of first participation, contribution rates (on a monthly basis), asset allocation, and total balances. We impose three sample restrictions. First, we consider only employees who enrolled in the plan between November 1998 and December 2005. Prior to November 1998, the plan’s matching contribution structure was different. Second, we drop 401(k) participants who are ineligible to receive matching contributions—those who have not completed 12 months at the company and 1,000 hours of service. We infer match eligibility from whether participants actually received a matching contribution. In a given year, about 16% of participants do not receive a matching contribution and are presumed ineligible for the match. Third, we attempt to eliminate employees who joined the company as a result of an acquisition because these individuals may not be comparable to employees who joined the company organically. Unfortunately, we cannot identify precisely which individuals became company employees through an acquisition. We drop individuals whose initial appearance in our data does not correspond to when they would have become eligible given their coded hire date. For example, employees hired by the company in 2002 should first appear in our data in 2003 because the service requirement for plan eligibility was 12 months prior to April 2003. Employees who first appear in our data in 2003 but who were hired prior to 2002 are presumed to have been acquired sometime in 2003 and thus excluded from our analysis. In 2003, the eligibility requirement was lowered from 12 months to 90 days of service. Thus, starting in 2003, we exclude individuals hired prior to the last 90 days of the calendar year who do not appear in that year’s data (e.g., employees hired in June 2004 who do not appear in the 2004 data). We also

7

drop employees hired during the last 90 days of a calendar year who do not appear in the subsequent year’s data (e.g., employees hired in November 2003 who do not appear in the 2004 data). Because the empirical analysis of the March 2003 plan change consists of comparing asset allocations of employees who enrolled in the 401(k) before and after March 2003, we want to establish the comparability of these two groups. Table 2 presents demographic statistics as of year-end 2003 for employees who enrolled from March 2002 to February 2003 (column 1) and statistics as of year-end 2004 for employees who enrolled from March 2003 to February 2004 (column 2). The two groups are similar in average age (39 years), gender composition (about 38% female), and proportion married (33% and 30%). Average and median incomes differ by no more than 2% between groups. These numbers suggest that there were no significant changes in the company’s workforce composition around March 2003. The only demographic characteristic that changes significantly across the two cohorts is the tenure distribution. The lower average tenure among employees enrolling between March 2003 and February 2004 is consistent with the change in the plan eligibility rules discussed above. Columns 3 and 4 give the demographic characteristics of all employees at the company as of year-end 2004 and all U.S. retail sector employees in the March 2005 Current Population Survey. Relative to the entire firm’s workforce, employees who became 401(k) participants around the time of the 2003 plan change are a little older, somewhat higher paid, and more likely to be married. These differences are consistent with our excluding employees who are either ineligible to participate in the savings plan or who choose not to. Such employees will younger and lower paid. Relative to the aggregate retail sector, the company we study employs fewer women, probably due to the type of goods Company A sells. Our company’s employees are also paid somewhat more than the average retail worker. The firm we study is very large, so this pay gap is consistent with Brown and Medoff (1989), who find that large firms tend to pay more than small firms. .

III. Empirical Results A. March 2003 Plan Change Starting in March 2003, new participants were required to specify the asset allocation for their matching contribution flows at the same time they chose their own-contribution flow

8

allocations. The company’s enrollment systems would not allow the employee to complete the enrollment process unless the matching flow allocation was explicitly chosen. Prior to March 2003, employer matching contributions were made in the form of employer stock. However, employees could immediately trade out of the employer stock in their match account and into the plan’s other investment options. The solid lines in Figure 1 show, by month of plan enrollment, the average fraction of contribution flows during 2003 that was allocated to employer stock within employees’ own contributions (grey line) and employer matching contributions (black line).4 The dashed lines are analogous series for contribution flow allocations during 2004. Across all enrollment cohorts, the average fraction of own-contributions allocated to employer stock was 37% in 2003 and 34% in 2004. Few employees changed their owncontribution flow allocation between years; 87% of participants observed at both year-ends 2003 and 2004 have the same flow allocation to employer stock at both points in time. Therefore, most of the variation in flow allocation across enrollment cohorts reflects variation in allocation decisions made at the time of enrollment. Consistent with the findings of Benartzi (2001) and Choi, Laibson, Madrian, and Metrick (2004), the own-contribution flow allocation to employer stock tends to be higher among employees who enrolled when the company’s stock had performed well in the recent past (not shown in the figure). The most important thing to notice is that there is no discontinuity in own-contribution flow allocations to employer stock between the cohort enrolling immediately before the March 2003 plan change and the cohort enrolling immediately afterwards. The picture for match flow allocations is strikingly different. The average fraction of contribution flow allocated to employer stock for employees who enrolled prior to March 2003 was 98% in 2003 and 94% in 2004. There is remarkably little variation in this average by enrollment month. Only 9% of employees who enrolled while the match had to be taken entirely in employer stock had changed their match flow allocation 22 months after the company lifted this restriction. In contrast, employees who enrolled in the savings plan in March 2003 or later allocated much less of their matching contribution flow to employer stock: 34% on average in

4

All figures weight each employee equally. Dollar-weighting the averages produces visually indistinguishable results.

9

2003 and 35% in 2004. There is a sharp discontinuity between employees enrolling in the month before the March 2003 plan change and those enrolling immediately afterwards. The top panel of Table 3 shows that if we estimate the plan change effect by comparing February 2003 enrollees with March 2003 enrollees, we obtain a 67.3 percentage point fall in the fraction of matching contribution flow allocated to employer stock. This comparison is the cleanest, since neither group of enrollees was affected by the April 2003 eligibility change at the time of their enrollment. However, broadening the before and after groups to include employees who enrolled in the two months before and the two months after the plan change, or the six months before and the six months after the plan change, yields very similar estimates of 67.0 and 65.8 percentage points respectively. There is no reason to believe that individuals enrolling immediately before the plan change had systematically different investment preferences than those enrolling immediately afterwards. Therefore, both groups should have the same target level of total employer stock holding, and the post-change group’s reduction in matching contribution flow to employer stock should be offset by an increase in own-contribution flow to employer stock. This is not, however, what we observe. Table 3 and Figure 1 show that own-contribution flows to employer stock fall by a small amount after the plan change (3.1 percentage points using the one-month comparison groups). The result is that total 401(k) flows to employer stock falls by 33.0, 32.7, and 30.9 percentage points using the one, two, and six-month comparisons, respectively. Even though contribution flow allocations changed dramatically for those enrolling after the plan change, the allocation of balances, not flows, is what ultimately determines portfolio returns. It is possible that employees were reallocating their 401(k) assets after contributions were made in order to undo the discrepancies in flow allocations. Recall that even prior to March 2003, employees had the ability to transfer their accumulated matching balances out of employer stock. Figure 2 shows that ex post rebalancing was not an important factor. The fraction of total balances held in employer stock at year-ends 2003 and 2004 looks remarkably similar to the contribution flow allocations to employer stock. For employees who enrolled prior to March 2003, the vast majority of employer match balances are invested in employer stock even at yearend 2004. This finding is consistent with the results of Choi, Laibson and Madrian (2005), who

10

document that even when employees are given the ability to diversify out of employer stock, the vast majority do not do so. The bottom panel of Table 3 shows the estimated impact of the 2003 plan change on the balances held in employer stock. We compare holdings of groups who enrolled in the month before and the month after the plan change, the two months before and two months after the plan change, and the six months before and the six months after the plan change. As suggested by Figure 2, the balance results are very similar to the contribution allocation results. The fraction of match balances held in employer stock falls by between 61.4 and 63.8 percentage points—an effect only slightly smaller than that measured for contribution flow allocations—whereas the fraction of own-contribution balances held in employer stock falls by at most 1.7 percentage points. Integrating the own-contribution and employer match accounts, the impact on total 401(k) balances is a 31.4 to 33.6 percentage point reduction in employer stock holdings. One might expect that starting in March 2003, when new enrollees were required to simultaneously choose own-contribution and match allocations, it would be natural to choose the same allocation for both accounts. However, we see in Table 3 that on average, post-change enrollees allocate 7 to 8 percentage points more to employer stock in their match account than in their own-contribution account. Most people do not behave this way; Table 4 shows that about three-quarters of post-change enrollees choose the same asset allocation for both accounts. For the quarter of post-change enrollees with different allocations at year-end 2004, 36% allocate everything to employer stock in the match account; 38% allocate less than 100% in the match account but, nonetheless, more in the match account than in their own-contribution account; and the remaining 26% allocate more to employer stock in their own-contribution account than in their match account. Very few (17%) of the post-change enrollees with a similar asset allocation in both accounts have 100% allocated to employer stock in both accounts. It is unclear why those who choose different allocations tend to allocate more to employer stock in their match account. The demographics of those who choose similar allocations and different allocations are fairly similar, although those who allocate more to employer stock in their matching account are slightly younger (1.3 years) and slightly higher paid (about $2,400). One potential explanation is that employees are willing to take more risk in their match account, which they perceive as “house money” (Thaler and Johnson, 1990; Clark, 2002). This

11

explanation assumes that employees do perceive employer stock as being riskier than the other investments available in the plan. Several surveys document that individuals on average believe that employer stock is less risky than a well-diversified stock mutual fund (John Hancock, 2002; Choi at el., 2005; Benartzi et al., 2007). The house money explanation may also be inconsistent with the fact that the likelihood of holding more employer stock in the match account does not decrease among participants who were vested in the employer match when they first joined the 401(k). (The presumption is that the match account feels less like house money once the participant is vested.) Another possible explanation is that a gift-exchange motive causes employees to hold more employer stock in their match account. Because the match may feel like a gift from the employer, employees feel the urge to reciprocate the kindness by holding more employer stock in their match account. Cohen (forthcoming) argues that employee loyalty to their employer causes them to hold employer stock.

B. April 2005 Plan Change In April 2005, plan participants who had not actively selected their match contribution flow allocations had their future match contribution flows automatically changed from 100% in employer stock to the flow allocation that participants had elected for their own contributions. At year-end 2004 (roughly four months before this plan change and 22 months after the March 2003 plan change), 91% of the pre-March 2003 enrollees still had 100% of their matching contributions flows directed to employer stock. As a consequence, this plan change affected a large fraction of pre-March 2003 enrollees. The impact of the 2005 plan change is seen in Figure 3 by comparing the contribution flow allocation of plan participants before the plan change and after the plan change. Our data separates 2005 matching contributions by whether they occurred from January 1 through March 31, 2005—before this plan change—or from April 1 through December 31, 2005. For own contributions, we only have an aggregate figure for the entire 2005 calendar year. We therefore assume that the calendar-year 2004 contribution allocation to employer stock was still in effect for the first three months of 2005. We then calculate what the own-contribution allocation to employer stock must have been for the last nine months of the year (after the plan change).

12

Note that the fraction of own-contributions allocated to employer stock (the solid and dashed gray lines in Figure 3) is very similar both before and after the 2005 plan change, regardless of the initial enrollment date. Similarly, the fraction of matching contributions allocated to employer stock (the solid and dashed black lines) is similar before and after the 2005 plan change for those employees who enrolled after the March 2003 plan change. Recall that these employees were required to make an active flow allocation decision for their matching contributions. The picture is very different for those employees who enrolled before the March 2003 plan change. Contribution allocations for the first three months of 2005 (before the plan change) are still almost entirely allocated to employer stock, even though 22 months have elapsed since these employees were allowed to choose their own matching contribution flow allocations. As a result, when the 2005 plan change occurred, the fraction of matching contribution flow going to employer stock dropped from 94% to 38%. This 55.9 percentage point effect is about 10 percentage points smaller than the decline in the share of matching contribution flow allocated to employer stock following the March 2003 plan change. There are two factors that account for the smaller magnitude of this plan change. The first is that pre-March 2003 participants had a much higher fraction of own-contributions allocated to employer stock than did post-March 2003 plan participants (39% versus 26%, according to Table 5). This caused a higher fraction of post-change contributions to go to employer stock (and thus a smaller decline relative to before the plan change), since the plan change mechanically mapped this higher own-contribution allocation to employer stock to the matching contribution as well. The second factor is that the 2003 plan change affected all new participants going forward, whereas this plan change affected only the 91% of pre-March 2003 participants who had not already actively chosen their employer matching contribution flow allocation.5 The March 2003 plan change had an equally large impact on both the fraction of annual matching contributions allocated to employer stock and on the accumulated match balances allocated to employer stock for post-March 2003 enrollees. In contrast, the decline in the fraction of matching balances allocated to employer stock as a result of the 2005 plan change is much smaller than the decline in the fraction of matching contributions allocated to employer stock. 5

Only 0.07% of these individuals who had 100% of their January to March 2005 matching contributions allocated to employer stock electively chose this allocation as evidenced by having less than 100% of either their 2003 or 2004 contributions allocated to employer stock.

13

The fraction of matching balances allocated to employer stock falls from an initial level of 93% at year-end 2004 before the plan change to 81% at year-end 2005 after the plan change, a difference of only 12 percentage points (relative to the 56 percentage point difference observed for matching contributions allocated to employer stock). This divergence from the 2003 plan change results should not be surprising. The tremendous inertia in both contribution and balance allocations before the 2005 plan change implies that the asset allocation for balances at any point in time will largely mirror contribution allocations at the same point in time (which, due to inertia, also tend to mirror contribution allocations at earlier points). This is particularly true for employees with low levels of tenure where, in the extreme, the asset allocation for the balances of new participants is determined (almost entirely) by their most recent contribution allocation (with only slight discrepancies for short-term differences in market returns across different assets). The participants whose asset allocation outcomes underlie the measurement of the 2003 plan change effects in Table 3 all have low levels of tenure (