The Flypaper Effect in Individual Investor Asset Allocation

Transcription

1 The Flypaper Effect n Indvdual Investor Asset Allocaton James J. Cho Yale Unversty and NBER Davd Labson Harvard Unversty and NBER Brgtte C. Madran Harvard Unversty and NBER November 21, 2007 We thank Hewtt Assocates for provdng the data analyzed n ths paper. We are partcularly grateful to Lor Lucas, Pam Hess, Yan Xu, and Greg Tabckman, some of our many current and former contacts at Hewtt. We thank semnar audences at Brgham Young Unversty, Brown, HKUST, Natonal Unversty of Sngapore, Northwestern, Sngapore Management Unversty, UCLA, and Yale for helpful comments. We apprecate the research assstance of Davd Borden, Ananya Chakravart, Chrs Nosko, and Neel Rao. Cho acknowledges fnancal support from the Mustard Seed Foundaton. Cho, Labson, and Madran acknowledge ndvdual and collectve fnancal support from the Natonal Insttute on Agng (grants R01-AG , P30-AG012810, and T32-AG00186).

2 The Flypaper Effect n Indvdual Investor Asset Allocaton Abstract: We document a flypaper effect n asset allocaton: securtes receved n knd stck where they ht. We study a frm that twce changed the rules governng the securtes n whch ts 401(k) matchng contrbutons were ntally nvested. Both of these rule changes were economcally neutral: employees were always free to mmedately reallocate ther match account balances. However, we fnd that most employees nether reallocate ther match balances, nor offset employer-ntated changes n the match allocaton by adjustng the allocaton of ther own contrbutons. Consequently, these rule changes caused dramatc shfts n partcpants 401(k) portfolo rsk. After examnng several alternatve explanatons for ths flypaper effect, we conclude that t s largely due to a combnaton of passvty and mental accountng. James J. Cho Davd Labson Yale School of Management Department of Economcs 135 Prospect Street Harvard Unversty P.O. Box Lttauer Center New Haven, CT Cambrdge, MA james.cho@yale.edu dlabson@harvard.edu Brgtte C. Madran Kennedy School of Government Harvard Unversty 79 JFK Street Cambrdge, MA brgtte_madran@harvard.edu 2

3 In ths paper, we document a flypaper effect n the asset allocaton of ndvdual nvestors. Lke the flypaper effects of publc fnance, corporate fnance, and ntra-household consumpton (see Hnes and Thaler, 1995, for a lterature revew, and Duflo and Udry, 2004, for evdence on the flypaper effect n ntra-household consumpton), we fnd that money stcks where t hts (ths phrase s attrbuted to Arthur Okun). Indvdual nvestors gven securtes n knd hold onto those securtes for a long tme wth mnmal offsets n other parts of ther portfolo. As a result, n-knd securty transfers can cause dramatc shfts n the rsk profle of an nvestor s portfolo. Ths s nconsstent wth nvestors havng a clear target asset allocaton for ther entre portfolo whch they mantan through perodc rebalancng. We dentfy the asset allocaton flypaper effect by explotng two natural experments at a large frm. Throughout the perod that we consder, for each dollar an employee contrbuted to the company s 401(k) savngs plan up to a lmt, the company made an addtonal matchng contrbuton. The employee s own contrbutons to the 401(k) and the company s matchng contrbutons were mantaned n separate accounts. Before March 2003, all matchng contrbutons were made n the stock of the employer. After the match was receved, however, all 401(k) partcpants were free to mmedately trade out of the employer stock and nto any of the other avalable nvestment optons. In March 2003, the company mplemented the frst polcy change that we study n ths paper: t ceased requrng that 401(k) partcpants accept matchng contrbutons n the form of employer stock. Instead, gong forward, new partcpants were requred upon enrollment to explctly choose an asset allocaton for ther matchng contrbuton flows, just as they dd for ther own-contrbuton flows. 1 Employees already enrolled n the 401(k) plan could also freely choose the asset allocaton for ther future matchng contrbuton flows, but f they dd not, ther matchng contrbutons contnued to be drected entrely nto employer stock. Note that ths plan change was economcally neutral. Any asset allocaton of balances the allocaton that determnes nvestment returns that was feasble n the new regme could be replcated under the old regme. In practce, however, the change was far from neutral. Employees who enrolled n the plan durng February 2003 (one month before the rule change) chose to drect 23% of ther 2003 own-contrbuton flows to employer stock; n contrast, 95% of 1 Throughout ths paper, we draw a dstncton between the asset allocaton for contrbuton flows and the asset allocaton for balances. The former s the mx of assets n whch ncremental 401(k) contrbutons are ntally nvested. The latter s the mx of all accumulated assets currently held n the portfolo. 3

4 ther 2003 matchng contrbuton flows were drected nto employer stock. Ths resulted n ther holdng 24% of ther own-contrbuton balances and 94% of ther match balances n employer stock at year-end 2003 (ten months after the rule change). In contrast, employees who enrolled n the plan durng March 2003 (the frst month after the rule change) chose to drect 20% of ther 2003 own-contrbuton flows and 27% of ther 2003 matchng contrbuton flows to employer stock, resultng n 20% of ther own-contrbuton balances and 27% of ther match balances beng held n employer stock at year-end Integratng the two 401(k) accounts, the fracton of total 2003 contrbutons gong nto employer stock an undversfed, hghly rsky asset dropped by more than half from 56% to 23% across the two enrollment cohorts, and the fracton of total year-end 2003 balances held n employer stock fell smlarly, from 56% to 22%. In Aprl 2005, the company made a second change. For partcpants who had not yet actvely elected an asset allocaton for ther matchng contrbuton flows, the company automatcally set the matchng contrbuton flow asset allocaton equal to what partcpants had prevously selected for ther own-contrbuton flows. The vast majorty of partcpants who enrolled before the frst plan change had not chosen an asset allocaton for ther matchng contrbuton flows by ths tme, and most also remaned passve n the face of ths second change. As a result, matchng contrbuton flows to employer stock plummeted overnght for these ndvduals to the much lower level that they had selected for ther own contrbutons. The asset allocaton flypaper effect s prmarly drven by two forces: passvty and narrow framng wthn mental accounts. Passvty among ndvdual nvestors has been extensvely documented (Samuelson and Zeckhauser, 1988; Madran and Shea, 2001; Cho, Labson, Madran, and Metrck, 2002, 2004a; Cho, Labson, and Madran, 2005a,b; Agnew, Balduzz and Sundén, 2003; Mtchell, Mottola, Utkus, and Yamaguch, 2006). In the flypaper context, passvty manfests tself as the falure of nvestors to reallocate whatever securtes are deposted nto ther match account. When the match s made n employer stock, 94% of match account balances reman n employer stock. Smlarly, when the company changes match flow allocatons to equal own-contrbuton flow allocatons, almost nobody exercses ther freedom to reverse ths change. Passvty, however, cannot explan why decsons made at enrollment a pont when people have overcome ther passvty and are takng an acton about the asset allocaton of total contrbuton flows dffer between the frst two match allocaton regmes. 4

5 We nterpret ths dfference across the frst two regmes as arsng from mental accountng, whch s the psychologcal segregaton of subsets of the wealth portfolo (Thaler, 1985, 1990, 1999). Indvduals then engage n narrow framng (Kahneman and Lovallo, 1993; Barbers, Huang, and Thaler, 2006) wthn these mental accounts, makng decsons for each account wthout consderng ther other accounts. Because money s fungble, a ratonal agent should nstead consder a fnancal decson s mpact on her entre wealth portfolo, not just ts effect on a subset of her portfolo. Before March 2003, enrollees only had to choose a flow allocaton for ther own-contrbuton account, makng t easy for them to gnore the match flow allocaton when makng that decson. Startng n March 2003, enrollees were forced to smultaneously choose flow allocatons for both accounts, makng the two salent and encouragng them to be psychologcally ntegrated. Consstent wth ths story, February 2003 enrollees drected 23% of ther own-contrbuton flows to employer stock, whle March 2003 enrollees drected 23% of ther combned own-contrbuton and match flows to employer stock. Partcpants appear to have allocated roughly one quarter of ther contrbutons to employer stock n whatever set of portfolos was salent, whether that set was narrow (own contrbutons) or broad (matchng and own contrbutons). A pre-march 2003 enrollee who was not a narrow framer would have chosen a lower employer stock allocaton for her own-contrbuton flows to compensate for the fact that she was constraned to a 100% employer stock allocaton n her match flows. The flypaper effect can help explan the hgh levels of employer stock ownershp n some 401(k) plans, an outcome that runs contrary to the logc of dversfcaton. Fdelty Investments (2002) reports that 44% of employer matchng contrbutons are made n employer stock, lke n the frst regme of our study company. Even though 66% of companes that match n employer stock do not restrct the subsequent sale of employer stock, the flypaper effect causes dversfcaton to be rare even when allowed. 2 Our results complement those of Card and Ransom (2007), who study contrbutons to unversty defned contrbuton penson plans. Usng across-plan comparsons, they fnd that lump-sum contrbutons that are labeled employer contrbutons crowd out dscretonary employee contrbutons to the plan less than contrbutons labeled mandatory employee 2 See also Cho, Labson, and Madran (2005b) for evdence on passvty n the face of relaxed dversfcaton restrctons. 5

6 contrbutons. They nterpret ths dfference as arsng because employees consder mandatory employee contrbutons a closer substtute for dscretonary employee contrbutons than employer contrbutons, due to mental accountng. Our paper s also related to other work that has shown that narrow framng affects asset allocaton. The reluctance to realze paper losses (the dsposton effect ) has been documented n many settngs (e.g. Odean, 1998; Grnblatt and Keloharju, 2001; Wermers, 2003; Coval and Shumway, 2005; Frazzn, 2006) and seems nconsstent wth nvestors carng only about ther total wealth portfolo. Lkewse, rsk averson over small gambles s a common behavor that s hard to ratonalze for ndvduals who are concerned only wth total wealth (Rabn, 2000). Fnally, our paper s adds to the lterature on the determnants of employer stock holdng n defned contrbuton savngs plans. Benartz (2001) and Huberman and Sengmueller (2004) show that past employer stock returns are postvely correlated wth current employer stock contrbuton allocatons, suggestng that employees navely extrapolate past employer stock returns nto the future. Cho, Labson, Madran, and Metrck (2004b) fnd that contrbuton flow allocatons to employer stock are ncreasng n recent returns, but reallocaton of balances nto or out of employer stock s on average contraran wth respect to recent returns. Benartz (2001) suggests that when companes requre matchng contrbutons to be made n employer stock, employees nterpret ths as an endorsement of employer stock that then drves employees to nvest more of ther own contrbutons n employer stock. (We wll present evdence n Secton IV that the endorsement effect s not a prmary drver of the flypaper effect). Cho, Labson, and Madran (2005b) show that even when employees are allowed to dversfy out of employer stock, the average portfolo response s small. Huberman (2001) argues that employer stock holdngs are motvated by the comfort of nvestng n a famlar asset. Lastly, Cohen (forthcomng) shows that employees loyalty to ther company also plays a role n ther wllngness to hold employer stock. The remander of ths paper proceeds as follows. Secton I descrbes the features of the 401(k) savngs plan at our study company, and Secton II descrbes our data. Secton III analyzes the mpact of the two match polcy changes on asset allocaton outcomes. Secton IV dscusses the mechansms underlyng the flypaper effect. Secton V concludes. 6

7 I. 401(k) Savngs Plan Features at a Large U.S. Corporaton The company we study s a large publcly traded company n the retal sector. In December 2005, ths frm had locatons n all ffty U.S. states, as well as the Dstrct of Columba and Puerto Rco. Table 1 lsts the major 401(k) plan features at ths company. Employees must actvely opt nto savngs plan partcpaton there s no automatc enrollment. The company offers a generous employer match of 150% on the frst 1% of pay contrbuted and 50% on the next 4% of pay, resultng n a total matchng contrbuton of 3.5% of pay for partcpants contrbutng at least 5% of ther pay. At year-end 2005, 59% of elgble employees partcpated n the 401(k). Besdes the two 401(k) plan changes dscussed n the ntroducton that are the focus of ths paper s analyss, there are three other plan changes that mert menton. Frst, before Aprl 2003, employees were elgble to partcpate n the 401(k) startng 12 months after hre, provded they had worked at least 1,000 hours at the company. In Aprl 2003, ths elgblty requrement was reduced to 90 days of employment, although elgblty for the employer match was stll restrcted to those wth 12 months of tenure and 1,000 hours of servce. 3 The elgblty change affects savngs plan partcpaton for employees wth lower levels of tenure, and one mght worry that t confounds our estmates of the effect of the March 2003 plan change. We wll show n Secton III, however, that ths elgblty change does not n fact alter our conclusons; the measured flypaper effect s vrtually dentcal when we restrct our sample to an enrollment wndow over whch elgblty requrements were constant. The second noteworthy plan change was an ncrease n the maxmum employee contrbuton rate from 15% to 50% of pay n January Ths change affects only the 4% of sample partcpants who contrbuted between 16% and 50% of pay after the hgher maxmum took effect. The thrd plan change also occurred n January 2002: an expanson n the number of 401(k) nvestment optons. Pror to January 2002, partcpants had sx nvestment optons from whch to choose, ncludng employer stock. In January 2002, the company added two nvestment optons, allowng partcpants to choose among a stable value fund, a balanced fund, fve equty funds, and employer stock. Because we dentfy the flypaper effect from dscontnutes n employer stock allocatons chosen around March 2003 (the frst plan change) and Aprl 2005 (the second plan change), these two January 2002 plan changes should not contamnate our 3 For a few small sub-groups of employees, ths elgblty change dd not take effect untl later n

8 analyss. Moreover, we do not observe sgnfcant changes n employer stock allocatons between employees enrollng just before and just after January II. Data Descrpton Our data come from Hewtt Assocates, a large benefts admnstraton and consultng frm. The man data are a seres of year-end cross-sectons from 2002 to 2005 of all employees elgble to partcpate n the 401(k) plan at the study company. These cross-sectons contan demographc nformaton such as brth date, hre date, gender, and compensaton. 4 They also contan nformaton on each ndvdual s 401(k), ncludng partcpaton status n the plan at year-end, date of frst partcpaton, monthly contrbuton rates, total dollars contrbuted to each nvestment opton durng the calendar year, and balances held n each nvestment opton at yearend. The contrbuton flow and balance allocaton nformaton s gven separately for the owncontrbuton account and the match account. We also make some lmted use of data from yearend We mpose three restrctons to obtan the plan partcpant sample used n our analyss. Frst, we consder only employees who enrolled n the plan between November 1998 and December We exclude partcpants who enrolled pror to November 1998 because the matchng contrbuton n the plan was sgnfcantly changed at that tme, and the focus of our analyss s on the asset allocaton of matchng contrbutons. Second, we drop 401(k) partcpants who are nelgble to receve matchng contrbutons those who have not completed 12 months at the company and 1,000 hours of servce. We nfer match elgblty from whether partcpants actually receved a matchng contrbuton. In a gven year, about 16% of partcpants do not receve a matchng contrbuton and are presumed nelgble for the match. 6 Fnally, we purge the data of partcpants who are lkely to have joned the company as a result of acqustons (the frm made several acqustons between 2002 and 2005) because these ndvduals may not be 4 We only have data on compensaton n the 2004 and 2005 year-end extracts. 5 The 1998 data are only avalable for a random sub-sample of employees who were at the frm at year-end Of the plan partcpants n our year-end 2004 data extract who dd not receve matchng contrbutons, 72% had less than 1 year of tenure, 23% had 1 to 2 years of tenure, and only 5% had more than 2 years of tenure, a pattern that s consstent wth the elgblty requrements for recept of an employer match. 8

9 comparable to employees who joned the company organcally. 7 Our results are qualtatvely smlar, however, even f we nclude partcpants who were potentally acqured. Because the emprcal analyss of the March 2003 plan change conssts of comparng the asset allocaton of employees who enrolled n the 401(k) before ths change wth those who enrolled after, we want to establsh the comparablty of these two groups. Table 2 presents demographc statstcs as of year-end 2003 for employees who enrolled n the 401(k) durng the month before and the month after the 2003 plan change (columns 1 and 2), as well as for employees who enrolled durng the sx months before and the sx months after the plan change (columns 3 and 4). The broader group of enrollees who joned n the sx months before and the sx months after the plan change are smlar n average age (38 years) and gender composton (61% male), and average and medan ncomes dffer by no more than 4% between groups. The more recent enrollees are slghtly less lkely to be marred (27% versus 33% for the earler cohort) and, as expected, have a lttle less tenure at the company (2.3 years versus 3.0 years). The subset of employees who enrolled n the month before and the month after the plan also look smlar, although the later group s slghtly older than the earler group. These numbers ndcate that there were no sgnfcant changes n the company s 401(k) partcpant composton around March III. Emprcal Results: The Impact of Match Polcy on Asset Allocaton A. March 2003 Plan Change Pror to March 2003, all employer matchng contrbutons to the 401(k) plan were made n the form of employer stock, although employees had the ablty to mmedately trade out of 7 Unfortunately, our data do not explctly dentfy how employees joned the frm. To screen out acqured ndvduals, we drop employees whose ntal appearance n our data does not correspond to when they would have become elgble to partcpate f they were full-tme employees, gven ther coded hre date. For example, full-tme employees hred by the company n 2002 should frst appear n our data n 2003 because the servce requrement for plan elgblty was 12 months pror to Aprl Employees who frst appear n our data n 2003 but who were hred pror to 2002 are presumed to have been acqured sometme n 2003 and are excluded from our analyss. In 2003, the elgblty requrement was lowered from 12 months to 90 days of servce. Thus, startng n 2003, we exclude ndvduals hred pror to the last 90 days of the calendar year who do not appear n that year s data (e.g., employees hred n June 2004 who do not make ther frst appearance untl the 2005 data). We also drop employees hred durng the last 90 days of a calendar year who do not appear n the subsequent year s data (e.g., employees hred n November 2003 who do not make ther frst appearance untl the 2005 data). 8 The larger number of post-march 2003 enrollees relatve to pre-march 2003 enrollees s due to seasonal enrollment patterns coupled wth the company s growng sze over the sample perod. Enrollees average asset allocatons do not follow a seasonal pattern when there are no plan changes, ndcatng that seasonalty does not confound our plan change effect estmates. 9

10 the employer stock n ther match account and nto any of the plan s other nvestment optons. Startng n March 2003, all partcpants were gven the ablty to choose an asset allocaton other than 100% employer stock for ther future matchng contrbuton flows. Employees enrollng n the 401(k) after ths change were requred to specfy an asset allocaton for ther matchng contrbuton flows durng enrollment, when they were also choosng an asset allocaton for ther own-contrbuton flows. The company s enrollment system would not allow the employee to complete the enrollment process unless an asset allocaton was explctly chosen for both ownand matchng contrbuton flows. In contrast, employees who enrolled pror to the plan change were not requred after the plan change to explctly specfy an asset allocaton for future matchng contrbuton flows, although they had the opton to do so. If they dd not make an actve electon, the company contnued to drect these partcpants matchng contrbutons entrely nto employer stock. Fgure 1 shows the mpact of the 2003 plan change on asset allocaton. The sold lnes show, by month of plan enrollment (the x-axs), the average fracton of 2003 contrbuton flows that was allocated to employer stock for employees own contrbutons (the grey lne) and for ther employer matchng contrbutons (the black lne). 9 Smlarly, the dashed lnes show ownand matchng contrbuton flow allocatons to employer stock n We frst consder the allocatons to employer stock for the employees own contrbutons n 2003 and 2004 (the grey sold and dashed lnes). Across all enrollment cohorts, the average fracton of own-contrbuton flows allocated to employer stock was 37% n 2003 and 34% n Few employees changed ther own-contrbuton flow allocaton between these two years; 87% of partcpants observed at both year-end 2003 and year-end 2004 have the same annual contrbuton allocaton to employer stock at both ponts n tme. Thus, most of the varaton n contrbuton flow allocatons across enrollment cohorts reflects varaton n allocaton decsons made at the tme of enrollment. Indeed, among November to December 1998 enrollees who were partcpants n the 401(k) plan at both year-end 2003 and fve years earler at year-end 1998, 76% had the same own-contrbuton allocaton to employer stock n both years, and the average own-contrbuton flow allocaton to employer stock was vrtually unchanged over ths perod, 59% n 1998 and 61% n All fgures weght each employee equally. Analogous fgures wth dollar-weghted averages are qualtatvely smlar and are avalable from the authors by request. 10

11 Consstent wth the fndngs of Benartz (2001) and Cho, Labson, Madran, and Metrck (2004b), the own-contrbuton flow allocaton to employer stock tends to be hgher among employees who enrolled when the company s stock had performed well n the recent past (to mantan the anonymty of the study company, stock market performance s not shown n the fgure). The most mportant thng to notce, however, s that there s no dscontnuty n owncontrbuton flow allocatons to employer stock between the cohort enrollng mmedately before the March 2003 plan change and the cohort enrollng mmedately afterwards. The pcture for match flow allocatons (the black lnes) s strkngly dfferent. The average fracton of annual matchng contrbutons allocated to employer stock for employees who enrolled between November 1998 and March 2003 was 98% n 2003 and 94% n There s remarkably lttle varaton n ths average by enrollment month. At year-end 2004, 22 months after the company gave employees the opton to drect ther matchng contrbuton flows to somethng other than employer stock, only 9% of partcpants who enrolled pror to March 2003 had done so. In contrast, employees who enrolled n the savngs plan n March 2003 or later allocated a much lower fracton of ther matchng contrbuton flow to employer stock: 34% on average n 2003 and 35% n The sharp dscontnuty n the matchng contrbuton allocatons to employer stock between employees enrollng before the March 2003 plan change and those enrollng afterwards s readly apparent n Fgure 1. The top panel of Table 3 shows the magntude of the effect that the 2003 plan change had on the contrbuton flows allocated to employer stock. If we measure the effect of the 2003 plan change by comparng February 2003 enrollees to March 2003 enrollees, we obtan a 67.9 percentage pont declne n the fracton of matchng contrbuton flows allocated to employer stock. The dfference between these two enrollment cohorts offers the cleanest comparson avalable, snce nether group was affected by the Aprl 2003 reducton n the watng perod for plan elgblty. 10 However, broadenng the before and after groups to nclude employees who enrolled n the two months before and the two months after the plan change, or the sx months before and the sx months after the plan change, yelds very smlar estmates of 67.6 and 66.5 percentage ponts respectvely. Interestngly, pre-march 2003 enrollees average owncontrbuton allocaton to employer stock and post-march 2003 enrollees average total 10 Although the watng perod for plan elgblty was reduced n Aprl 2003, there was no change n the watng perod for elgblty to receve matchng contrbutons. Because our sample s restrcted to partcpants who receve a match, ths Aprl 2003 plan elgblty change should have lttle effect on our estmates. 11

12 contrbuton allocaton to employer stock (matchng plus own-contrbuton) are nearly dentcal; the two numbers are no more than a percentage pont apart, regardless of the comparson groups used. The hstogram n Fgure 2 shows that ths smlarty holds not only for the means, but also for the dstrbutons of pre-march 2003 enrollees own-contrbuton allocatons and post-march 2003 enrollees total contrbuton allocatons to employer stock. There s no reason to beleve that ndvduals enrollng mmedately before the 2003 plan change had systematcally dfferent nvestment preferences than those enrollng mmedately afterwards. As Table 2 shows, the observable characterstcs of these employees are very smlar, and the 401(k) plan admnstraton team member wth whom we corresponded was not aware of any shft n enrollee characterstcs around March If the presumpton of comparable nvestment preferences s correct, then both groups should have the same target level of total employer stock exposure, and the lower matchng contrbuton flow nto employer stock for the post-change enrollees should be offset by an ncrease n ther own-contrbuton flow to employer stock. Ths s not what we observe. Table 3 and Fgure 1 show that own-contrbuton flows to employer stock n fact fall by a small amount after the plan change (3.3 percentage ponts usng the one-month comparson groups). The result s that the combned own- and employer match contrbuton flow to employer stock falls by 33.4, 33.2, and 31.3 percentage ponts usng the one, two, and sx-month comparson groups n Table 3. Even though contrbuton flow allocatons are dramatcally dfferent for those enrollng after the 2003 plan change than for those enrollng before, the allocaton of balances, not flows, s what ultmately determnes portfolo returns. It s possble that employees were reallocatng ther 401(k) assets after contrbutons were made n order to undo the dscrepances n flow allocatons. Recall that even pror to March 2003, employees had the ablty to transfer ther accumulated match balances out of employer stock. Fgure 3 shows that such ex post rebalancng was not an mportant factor. The fracton of total balances held n employer stock at year-ends 2003 and 2004 looks remarkably smlar to the contrbuton flow allocatons to employer stock n Fgure 1. For employees who enrolled pror to March 2003, the vast majorty of employer match balances are nvested n employer stock even at year-end 2004, 22 months after the plan change. Ths fndng s consstent wth the results of Cho, Labson, and Madran (2005b), who document that when employees are gven the ablty to dversfy out of employer stock, almost none do. The fact that balance allocatons closely track 12

13 flow allocatons even for those who enrolled early n the sample perod (e.g. 1998) demonstrates that flow allocaton decsons (whch are largely made upon enrollment and rarely revsted) are not much more lkely to be reversed as 401(k) balances grow larger and the absolute dollar consequences of the 401(k) asset allocaton ncrease. The bottom panel of Table 3 shows the estmated mpact of the 2003 plan change on the balances held n employer stock. We compare holdngs of groups who enrolled n the month before and the month after the plan change, the two months before and the two months after the plan change, and the sx months before and the sx months after the plan change. As suggested by Fgure 3, the balance results are nearly dentcal to the contrbuton flow results. The fracton of match balances held n employer stock falls by between 65.7 and 67.6 percentage ponts an effect only slghtly smaller than that measured for contrbuton flow allocatons whereas the fracton of own-contrbuton balances held n employer stock falls by at most 3.8 percentage ponts. Integratng the own-contrbuton and employer match accounts, the mpact on total 401(k) balances s a 32.0 to 34.2 percentage pont reducton n employer stock holdngs. We have also estmated the regresson-adjusted mpact of the 2003 plan change for each of the comparson groups and accounts lsted n Table 3, controllng for demographc characterstcs. 11 These unreported results dffer lttle from the raw effects the flypaper effect s not arsng due to dfferences n the demographc composton of employees before and after the 2003 plan change. Moreover, we also fnd that the magntude of the flypaper effect s smlar across dfferent demographc subgroups. Of partcular note, even hgher-ncome partcpants who are probably the most fnancally lterate exhbt a flypaper effect that s comparable n sze to the company-wde average flypaper effect. Fnally, the flypaper effect s not arsng from the fact that some partcpants may not value the employer match (and thus pay lttle attenton to t) because they have not yet met the company s vestng requrements we fnd that the magntude of the flypaper s smlar even when we restrct the sample to employees whose matchng contrbutons are fully vested upon enrollment. 12 One mght expect that startng n March 2003, when new enrollees were requred to smultaneously choose an asset allocaton for ther own-contrbuton and matchng contrbuton 11 The regresson-adjusted results control for gender, martal status, age, tenure, ncome, and plan balances. These results are avalable from the authors upon request. 12 Plan partcpants at our study company are 100% vested n ther employer matchng contrbutons once they reach 3 years of servce. If they leave the frm before reachng 3 years of servce, they forfet all of ther accumulated employer match balances. 13

14 flows, t would be natural to choose the same allocaton for both flows. However, we see n Table 3 that on average, post-change enrollees allocate 7 percentage ponts more to employer stock n ther match account than n ther own-contrbuton account. In fact, about three-quarters of post-change enrollees do have dentcal allocatons n both accounts. It s unclear why the remanng quarter choose dfferent allocatons, and why they tend to allocate more to employer stock n ther match account. The demographc characterstcs of those who choose the same versus dfferent allocatons are smlar, although those who allocate more to employer stock n ther match account are slghtly younger (1.3 years) and slghtly hgher pad (about $2,400). One potental explanaton s that employees are wllng to take more rsk n ther match account, whch they perceve as house money that can be lost wthout much regret (Thaler and Johnson, 1990; Clark, 2002). Ths explanaton assumes that employees perceve employer stock to be a rsker nvestment than the other optons avalable n the plan. However, several surveys document that ndvduals on average beleve that employer stock s less rsky than a welldversfed stock mutual fund (John Hancock Fnancal Servces, 2002; Benartz et al., 2007). We also fnd that the lkelhood of holdng more employer stock n the match account than n the own-contrbuton account does not depend on the vestng status of partcpants. If the match feels less lke house money once the partcpant s vested, then the lack of such a relatonshp s further evdence aganst the house money hypothess n ths context. Another possble explanaton s a gft-exchange motve. If the match s perceved to be a gft from the employer, then employees may recprocate the kndness by holdng more employer stock n ther match account. Cohen (forthcomng) argues that employee loyalty ncreases the holdng of employer stock. B. Aprl 2005 Plan Change We turn now to the second plan change that affected the asset allocaton of matchng contrbutons. In Aprl 2005, the company dentfed partcpants who had not yet actvely elected an asset allocaton for ther matchng contrbuton flows (and as a result had 100% of ther matchng contrbuton flows drected nto employer stock). The company then set the asset allocaton for these partcpants matchng contrbuton flows equal to the asset allocaton that they had selected for ther own-contrbuton flows. Partcpants who had already actvely chosen an asset allocaton for ther matchng contrbuton flows were unaffected. 14

15 Ths plan change affected almost all of the partcpants who had enrolled before March By the end of March 2005 (two years after the 2003 plan change), 91% of these partcpants had receved all of ther 2003, 2004, and January to March 2005 matchng contrbutons n employer stock. In contrast, post-march 2003 enrollees were requred to select an asset allocaton for ther matchng contrbutons when enrollng n the plan and were thus untouched by the change. We measure the mpact of the Aprl 2005 plan change by comparng partcpants contrbuton allocatons before the plan change and after the plan change. Our data separate 2005 matchng contrbuton flows nto those made before ths plan change (January 1 to March 31, 2005) and those made after ths plan change (Aprl 1 through December 31, 2005). For owncontrbuton flows, however, we have only an aggregate fgure for the entre year. We therefore assume that the 2004 own-contrbuton flow allocaton to employer stock was stll n effect for the frst three months of Gven the hgh levels of nerta n employer stock allocatons we have already documented, ths assumpton seems reasonable. We then calculate what the owncontrbuton allocaton to employer stock must have been for the last nne months of 2005 (after the plan change). Fgure 4 shows that the fracton of 2005 matchng contrbuton flows allocated to employer stock s smlar before Aprl 2005 (the sold black lne) and after Aprl 2005 (the dashed black lne) for partcpants who enrolled after the frst plan change n Recall that these employees were requred to make an actve allocaton decson for ther matchng contrbutons. The pcture s very dfferent for employees who enrolled before the March 2003 plan change. Ther matchng contrbutons n the frst three months of 2005 (the sold black lne) are stll almost entrely allocated to employer stock, even though 22 months have elapsed snce they were allowed to choose ther own matchng contrbuton flow allocaton. In contrast, ther matchng contrbutons n the last nne months of 2005 (the dashed black lne) are dramatcally lower and, as expected, mrror ther own-contrbuton allocatons. Table 4 shows that among those who enrolled between November 1998 and February 2003, the fracton of matchng contrbuton flows allocated to employer stock plummeted 55.7 percentage ponts from 94.1% to 38.4% startng n Aprl Ths 56 percentage pont declne s about 10 percentage ponts smaller than the declne n the share of matchng contrbuton flows allocated to employer stock followng the 2003 plan change. There are two factors that account 15

16 Dd pre-march 2003 enrollees offset these dramatc match flow allocaton changes by ncreasng the employer stock allocatons of ther own-contrbuton flows? Fgure 4 shows that they dd not. The fracton of own-contrbuton flows allocated to employer stock n 2005 before the second plan change (the sold gray lne) s very smlar to the fracton allocated to employer stock after the plan change (the dashed gray lne). As a result, Table 4 shows that the total flow allocaton (matchng plus own-contrbuton) to employer stock fell from 62.4% to 37.3% among pre-march 2003 enrollees after the 2005 plan change. Note that these effects hold regardless of partcpants enrollment dates, suggestng the flypaper effect s not drven by partcpants havng nsuffcent plan balances to care about the 401(k) or by a lack of experence wthn the plan. Moreover, as wth the 2003 plan change, the effects of the Aprl 2005 plan change are smlar n magntude f we restrct our sample to partcpants whose match balances were fully vested at the tme of the change. Despte the large drop n matchng contrbuton flow allocatons to employer stock among pre-march 2003 enrollees, Fgure 5 and Table 4 show that the 2005 plan change caused a much smaller declne n the share of match balances held n employer stock, from 93.4% before the plan change (at year-end 2004) to 81.4% after the plan change (at year-end 2005), a dfference of only 12.0 percentage ponts. Ths dscrepancy between the mpact on contrbuton flow allocatons and the mpact on balance allocatons s due to the fact that the plan change only affected the allocaton of contrbuton flows gong forward, not the allocaton of balances already accumulated before Aprl If partcpants made no actve 401(k) allocaton changes n response to the plan change, the fracton of balances allocated to employer stock would dmnsh only gradually over tme, wth a more rapd declne for partcpants whose contrbuton flow s large relatve to ther prevously accumulated balances. Consstent wth near-unversal passvty, for the smaller magntude of the 2005 plan change effect. The frst s that the 2003 plan change affected all new enrollees gong forward, whereas the 2005 plan change affected only the 91% of pre-march 2003 enrollees who had not already actvely chosen ther matchng contrbuton flow allocaton. The second s that we measure the effect of the 2003 plan change and that of the 2005 plan change on dfferent samples of enrollees. The effect of the 2003 plan change s computed by comparng the 2003 allocatons of employees who enrolled n the one, two, or sx months before March 2003 to those who enrolled n the one, two, or sx months after March We measure the effect of the 2005 plan change by comparng the January to March 2005 allocatons of employees who enrolled n the 401(k) from November 1998 to February 2003 to those same enrollees allocaton over the Aprl to December 2005 perod. Note that employees who enrolled pror to September 2002 had a much hgher fracton of ther own contrbutons allocated to employer stock than those who enrolled later, probably due to the stock s strong performance n earler years. As a result, when the company set matchng contrbuton allocatons equal to own-contrbuton allocatons, pre-september 2002 enrollees matchng contrbuton allocatons to employer stock declned by less. If we had restrcted our sample to September 2002 to February 2003 enrollees when computng the 2005 plan change effect, the two plan changes effects would have a smlar magntude. 16

17 f we stratfy pre-march 2003 enrollees nto thrds based on the rato of ther 2005 contrbutons to ther year-end 2004 balances, the declne n the fracton of balances held n employer stock from year-end 2004 to year-end 2005 s 17 percentage ponts for partcpants wth the hghest rato of contrbutons to balances, 13 percentage ponts for those n the mddle, and 7 percentage ponts for those wth the lowest rato of contrbutons to balances. In contrast to the dsparate mpact of the 2005 plan change on contrbuton flow allocatons to employer stock and balance allocatons to employer stock, we found lttle dscrepancy between the effects of the 2003 plan change on these two outcomes. The reason for the smlarty n the two effects for the 2003 plan change s that we dentfed the 2003 plan change effect by comparng the cohort who enrolled mmedately before the 2003 plan change to the cohort who enrolled mmedately afterwards. The later cohort never contrbuted to the plan under the old regme, and hence had no accumulated balances under the old regme to attenuate the effect of the 2003 plan on balances. When dentfyng the 2005 plan change effect, we are comparng pre-march 2003 enrollees balance allocatons at year-end 2004 (when they had at least 22 months of accumulated contrbutons) to ther balance allocatons at year-end Another dfference between the 2003 and 2005 plan changes s the magntude of the dscrepancy between the fracton of own- and matchng contrbutons allocated to employer stock among affected employees. Recall that after the 2003 plan change, post-march 2003 enrollees allocated 7 percentage ponts more of ther matchng contrbutons to employer stock than they dd of ther own contrbutons. Table 4 shows that ths dfference remans about 6 to 7 percentage ponts just pror to and after the 2005 plan change. For the pre-march 2003 enrollees, however, ths gap s a much smaller 1.5 percentage ponts after the 2005 plan change. Ths dscrepancy n outcomes between the 2003 and 2005 plan changes s largely mechancal. As noted above, 91% of the pre-march 2003 enrollees were affected by the Aprl 2005 plan change. These affected employees had ther match flow allocaton gong forward set equal to ther own-contrbuton flow allocaton. We would thus expect almost no dfference between the match flow allocaton and the own-contrbuton flow allocaton to employer stock for ths group. The small dscrepancy between the fracton of contrbutons allocated to employer stock from Aprl to December 2005 n the own- and matchng contrbuton accounts s due to the 9% of pre-march 2003 enrollees who were not affected by the 2005 plan change. Consstent wth 17

18 the pattern observed for post-march 2003 enrollees, the employer stock flow allocaton s hgher n the match account (36.4%) than n the own-contrbuton account (18.9%) for ths group. IV. Dscusson: Passvty, Mental Accountng, Ignorance, and Endorsement Effects What drves the asset allocaton flypaper effect? The most obvous force s passvty. Before the 2003 plan change, savngs plan partcpants had the ablty to transfer ther match account balances out of employer stock, yet very few chose to do so. At year-end 2002, two months before the frst plan change, 97% of partcpants n our sample had 100% of ther match balances allocated to employer stock. Surprsngly, the fracton of partcpants at year-end 2002 wth all ther match balances nvested n employer stock s not senstve to tenure wthn the savngs plan; newer and older partcpants alke were almost completely passve wthn ther match account. We also see after the 2005 plan change that almost all employees who had not prevously made an actve match flow allocaton decson passvely accepted the company s change to ther match flow allocaton. The March 2003 rule change generated large asset allocaton dfferences between enrollment cohorts that had a materal mpact on ther portfolo rsk. We have already noted that employees enrollng on ether sde of ths frst rule change look very smlar n ther observable characterstcs, so t s unlkely that enrollment cohorts mmedately before and after the rule change had extremely dfferent nvestment preferences. The Aprl 2005 rule change affected the asset allocaton of pre-march 2003 enrollees. We doubt that the pre-march 2003 enrollees optmal asset allocaton changed n Aprl 2005 n a way that happened to concde wth the result of passvely acceptng the company s match allocaton choce. The passvty we document s consstent wth other evdence n the savngs lterature (Samuelson and Zeckhauser, 1988; Madran and Shea, 2001; Cho, Labson, Madran, and Metrck, 2002, 2004a; Cho, Labson, and Madran, 2005a,b; Agnew, Balduzz, and Sundén, 2003; Mtchell, Mottola, Utkus, and Yamaguch, 2006). However, passvty cannot be the entre explanaton for the dramatc shft n portfolo allocatons caused by the frst plan change n March An employee n the process of enrollng n her 401(k) plan has momentarly overcome passvty. The cost of explctly statng an asset allocaton preference s already sunk, so there s no reason not to choose an allocaton that s closest to what the employee thnks s optmal. An enrollee before March 2003 thus 18

19 should have reduced the fracton of her own contrbutons allocated to employer stock upon enrollment to compensate for the fact that all of her matchng contrbutons would be made n employer stock. We nstead see that own-contrbuton allocatons to employer stock among pre- March 2003 enrollees are smlar to own-contrbuton allocatons among post-march 2003 enrollees, even though post-march 2003 enrollees allocate less than a thrd of ther matchng contrbutons to employer stock. Thus, there must be some other factor at play as well. One canddate s the short-sales constrant mposed by the 401(k). Suppose pre-march 2003 enrollees want to reduce ther own-contrbuton flow allocaton to employer stock n order to compensate for ther match flow allocaton. However, they are unable to allocate less than zero to employer stock n ther own-contrbuton flows. Therefore, n the absence of a wllngness to trade n the match account after enrollment, they are constraned to hold a large fracton of ther portfolo n employer stock. Once employees can (and are requred to) specfy a matchng contrbuton flow allocaton, ths short-sales constrant s effectvely relaxed, and they are able to hold less employer stock. Could such a short-sales constrant explan the flypaper effect? Appendx A descrbes the calculaton of how much employer stock allocatons would be expected to drop f only a shortsales constrant were operatve. We fnd that such a constrant cannot come close to quanttatvely generatng a drop equal to what we observe. However, passvty combned wth narrow framng wthn mental accounts (Thaler, 1985, 1990, 1999) can explan all of ths paper s key results (except for the fact that partcpants voluntarly choose to hold employer stock n the frst place, an ssue we do not focus on here). Pror to the 2003 plan change, enrollees made an asset allocaton decson for only ther owncontrbuton flows. Therefore, t was easy to put own-contrbutons nto ther own mental account and to make flow allocaton decsons for these contrbutons whle gnorng the match flow allocaton. After the 2003 plan change, enrollees were forced to smultaneously choose allocatons for both accounts, whch made both salent and encouraged ntegraton of the two nto the same mental account when makng those decsons. Consstent wth ths story, partcpants who enrolled n the 401(k) just before March 2003 allocated about 25% of ther own-contrbuton flows to employer stock, whereas those enrollng just after allocated about 25% of ther total contrbuton flows to employer stock. The average partcpant appears to want to allocate roughly one quarter of ther assets to employer stock n whatever portfolos are 19

20 salent, whether the set of salent portfolos s narrow or broad. Fgure 3 shows that ths scalenvarance apples not only to the mean, but also to the dstrbuton of allocatons to employer stock. That s, the dstrbuton of own-contrbuton flow allocatons to employer stock among pre-march 2003 enrollees s very smlar to the dstrbuton of total contrbuton flow allocatons to employer stock among post-march 2003 enrollees. Note that mental accountng alone cannot explan all of the flypaper effect. Mental accountng explans why choces n the own-contrbuton account do not respond to what happens n the match account when the match account s not salent, whle passvty explans why the match account s allocaton s nfluenced by employer defaults (and changes n those defaults). Of course, wthout drect measurement of partcpant knowledge about the employer match, mental accountng n the pre-march 2003 regme s observatonally equvalent to partcpant gnorance of the match s exstence or ts asset allocaton. Although we fnd some degree of gnorance plausble, gnorance alone cannot account for the magntude of the 2003 flypaper effect. In Appendx B, we calculate a lower bound on the fracton of gnorant partcpants requred to generate the 2003 flypaper effect. Our lower bound estmates range from 92% to 94%, whch we deem an mplausbly large fracton of partcpants that are unaware of a major feature of ther 401(k) plan. There s a fnal canddate explanaton for the 2003 flypaper effect: employees may have perceved the 2003 plan change as removng the company s mplct endorsement of ts stock as an attractve nvestment. The percepton of such an endorsement could explan why so few partcpants dversfed out of employer stock n ther match account before the 2003 plan change. And the removal of the endorsement could have generated the large drop n employer stock allocatons after the change. Past research has documented the exstence of such endorsement effects. For example, Benartz (2001) shows that among companes that offer employer stock n ther 401(k) nvestment menu, savngs plan partcpants allocate 18% of ther own contrbutons to employer stock when the employer match s allocaton s left to the partcpants dscreton, and 29% when the match must be accepted n employer stock, a dfference of 11 percentage ponts. Brown, Lang, and Wesbenner (forthcomng) document a smlar 7 percentage pont effect usng acrossfrm varaton n match allocaton polcy. On the other hand, Brown, Lang, and Wesbenner fnd 20