SW. .25....

 

s. t.
in:

uﬁ. 3.

:235331 4
.53. .L.—At:.!vxlv
A3035».
:3 .. aha... .
. 1:1. . a}:
mo. .JblﬂW-ﬁ. : trauma”
.3 hr;
I - Eu‘uitin

.353.
lizlasnxbri l.
r. 5» it»: t
.9... (a.

‘la :IL 3
It
2

in

a. 1
F... J: 1.1:.“
Lastllii113

2....

r \ ..
.1 niﬁhxﬁletl [$.42
.- .. . . .: .
1.. n: )pl :Ian!

Fixiafi :-
«a!

.
w
6:55;;
53 5.046.:
. .

 

 

{STE ..
=3). . :2

 

 

 

. .2 .
. .l ‘12:?!
.21. 1...:

: n .11“...

 

 

 

‘ C h 5 Michigan State

 

LIBRARY IHIHJUHHlllHllWWIUIIWNIIllellNHlUlJHHIHII

302048 8742

University

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

JUN 2 araaoz

 

OCT 2 3121502

 

a
co

 

 

 

 

 

 

 

 

 

 

 

 

11/00 chlRCIDatoDmpBS-p.“

ESSAYS ON RETIREMENT EDUCATION AND PRIVATE PENSIONS
By

Leslie Ann Muller

A DISSERTATION
Submitted to
Michigan State University

in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

2000

ABSTRACT
ESSAYS ON RETIREMENT EDUCATION AND PRIVATE PENSIONS
By

Leslie Ann Muller

This dissertation examines the effects of taking a retirement class on two choices
employees must make within their deﬁned contribution (DC) pension plans: whether to
save their DC balance upon separation from their employer and how to allocate their
assets within their pension plan. As DC plans are increasingly becoming the primary
pension plan offered by employers, the choices employees make will affect their
retirement security. With the future of Social Security uncertain, employer-sponsored
pension plans will be playing an increasing role in providing retirement income for the
future elderly.

The ﬁrst chapter, Retirement Education and Pension Preservation: Does
Retirement Education Teach Individuals to Save Pension Distributions? examines
whether taking a retirement class affects an individual's decision to save or spend his
pension distribution. Most DC plans allow the participant to take his pension balance
with him upon job separation. Spending these distributions can have negative effects on
retirement saving, and educating workers about the importance of saving their pension
distributions may be one way of retaining some or all of these funds for retirement. This
paper provides econometric evidence of the effect of retirement education on the use of

pension distributions for DC participants. I ﬁnd that retirement education substantially

increases the probability Of saving a distribution for participants age 40 and under, while
decreasing the probability of saving a distribution for college graduates and females.
These important differentials are concealed by estimating the effect of retirement
education on participants generally.

I examine the effect of retirement education on asset allocation within a DC plan
in Chapter 2, Investment Choice in Private Pensions: The Eﬂects of Retirement
Education on Asset Allocation. Increasing numbers of DC plans are being set up as self-
directed plans, which allow participants to choose how their pension assets are invested.
The portfolio one chooses can have substantial effects on the amount of pension income
available during retirement. Recently, both pension plan sponsors and policy makers
alike have become concerned as to whether participants have the ﬁnancial knowledge to
make such important decisions. This paper provides econometric evidence of the effect
of retirement education on asset allocation in self-directed DC plans. I ﬁnd that those
with an extreme degree of risk aversion do not re-allocate their portfolios after taking a
retirement class, whereas those with a Slightly lower degree of risk aversion invest a
larger percentage of their assets in equities after attending a class. Participants generally
do not alter their behavior after attending a class. It is possible that this result is due to
the heterogeneity in ﬁnancial instruction or the inability of the model to detect small

changes in asset allocation.

ACKNOWLEDGMENTS

I would like to thank many individuals for their advice, support, and guidance, for
without them this dissertation could not have been completed.

I would like to ﬁrst thank my advisor, Leslie Papke, for taking an interest in my
work and sharing her knowledge of pension economics with me. She answered my
questions in a thorough and thoughtful manner, and provided me with a rich base of
knowledge from which to further my studies.

I would also like to thank David Neumark for his suggestions and help through
these years. He always made time for me in his busy schedule, and for this I will always
be grateful.

I also appreciate the input of Jeff Wooldridge in the writing of this paper. His
knowledge of econometrics has given me a better understanding of econometric analysis
and its various applications to policy, which will prove to be most valuable to me in my
future work.

A thanks goes to those professors and fellow graduate students who took the time
to add valuable suggestions to the writing of this paper. Scott Adams, Christine Amsler,
Heather Bednarek, Hailong Qian and Daiji Kawaguchi have been much help to me. A
special thanks goes to Kathleen Beegle, who ﬁelded questions long-distance when I was
in a bind.

Finally, I would not be where I am today without the support of my husband,
Chris, and my parents. Both have provided me with emotional and ﬁnancial support that

I will never be able to repay.

iv

TABLE OF CONTENTS

LIST OF TABLES .............................................................................................................. vi

INTRODUCTION ............................................................................................................... 1
Introduction ............................................................................................................. 1
Economic and Financial Knowledge ....................................................................... 4
Myopia in Retirement Planning ............................................................................... 6
Individual Rules for Saving Behavior ..................................................................... 8
Conclusion ............................................................................................................. 10
References .............................................................................................................. 12

CHAPTER I

RETIREMENT EDUCATION AND PENSION PRESERVATION:
DOES RETIREMENT EDUCATION TEACH INDIVIDUALS TO

SAVE PENSION DISTRIBUTIONS? .............................................................................. 14
Introduction ........................................................................................................... 14
Background ........................................................................................................... 17
Previous Research ................................................................................................. 26
Does Retirement Education Promote Saving of
Distributions? Evidence from the 1992 HRS ........................................................ 29
Econometric Results on the Effects of Retirement Education
on Pension Preservation ........................................................................................ 37
Conclusion ............................................................................................................. 54
References ............................................................................................................. 56

CHAPTER 2

INVESTMENT CHOICE IN PRIVATE PENSIONS: THE EFFECTS

OF RETIREMENT EDUCATION ON ASSET ALLOCATION ..................................... 74
Introduction ............................................................................................................ 74
Background ............................................................................................................ 77
Previous Research .................................................................................................. 88

Retirement Education and Asset Allocation in the 1992 HRS ............................... 92
Econometric Results on the Effects of Taking a Retirement Class
on DC Asset Allocation ....................................................................................... 100
Conclusion ........................................................................................................... 1 14
References ............................................................................................................. 1 17

LIST OF TABLES

Table 1.1 - Summary Statistics .......................................................................................... 59
Table 1.2 - Uses of Pension Distributions Upon Job Separation ....................................... 60

Table 1.3 - Percentage of Individuals who Attended a Retirement
Meeting and Saved their Pension Distributions, by Group ................................................ 61

Table 1.4 - Disposition of Pension Distributions by Retirement
Meeting Attendance ........................................................................................................... 62

Table 1.5 - Effects of Retirement Education on Pension Preservation.
Saving deﬁned as All Types of Saving .............................................................................. 63

Table 1.6 - Effects of Retirement Education on Pension Preservation.
Saving deﬁned as Retirement Saving ................................................................................ 65

Table 1.7 - Effects Of Retirement Education on Pension Preservation.
Full Sample Versus CNW. Saving deﬁned as All Types of Saving ................................. 67

Table 1.8 - Effects of Retirement Education on Pension Preservation.
Full Sample Versus CNW. Saving deﬁned as Retirement Saving ................................... 69

Table 1.9 - OLS Estimates of the Effect of Retirement Education on
Pension Preservation by Age, Education, and Gender ....................................................... 71

Table 1.10 - OLS Estimates of the Effect of Retirement Education on
Pension Preservation for those with Combinations of Age, Education,

and Gender Characteristics ................................................................................................ 72

Table 1.11 - IV Estimates of the Effects of Retirement Education on
Pension Preservation by Age, Education, and Gender ...................................................... 73

Table 2.1 - Summary Statistics ........................................................................................ 121

Table 2.2 - Proportion of Participants with Each Type Of Asset
Allocation, by Individual Characteristics ........................................................................ 123

Table 2.3 - Proportion of Participants who Attended a Retirement
Meeting and Respective Asset Allocation, by Individual Characteristics ....................... 124

Table 2.4 - Percentage Point Increase in Stock if Attended a Retirement Class,
by Plan. OLS Estimates ................................................................................................... 125

vi

Table 2.5 - Percentage Point Increase in Stock if Attended a Retirement Class,
by Plan. Ordered Logit Estimates .................................................................................... 128

Table 2.6 - Effects of Attending an Employer-Sponsored Class or
an Outside Retirement Class on Percentage Point Increase in Stock, by Plan ................ 131

Table 2.7 - Percentage Point Increase in Stock if Attended a Retirement Class,
by Years until Retirement and Degree of Risk Aversion, separately .............................. 132

Table 2.8 - Percentage Point Increase in Stock if Attended a Retirement Class,
by Years until Retirement and Degree of Risk Aversion, together ................................ 133

vii

INTRODUCTION

The private pension environment in America is rapidly changing. Whereas
twenty years ago most employees could plan on retirement income from a deﬁned beneﬁt
(DB) pension plan from their employer, this is often no longer the case. Deﬁned
contribution (DC) plans, the most common being the 401(k) salary reduction plan, are
quickly becoming the primary retirement plan offered by companies. Unlike the DB
plan, where the employer periodically sets aside money for the employee to be paid out in
an annuity at retirement, DC plans allow the responsibility of retirement security to be
shared by both the employer and the employee. Employees must decide whether to
participate in the DC plan, and if they participate, how much money they would like to
contribute. In addition, employees can decide what to do with their money if they
separate from their employer, and in many cases how to invest their pension
contributions.

In the early to late 19803, when companies increasingly began to offer 401(k)
plans to their workers, most employees did not have to make the choices they do today.l
Information about the plan was limited, and often restricted to a lengthy plan description
that most employees were unable to understand. Valuations (calculations of the current
value of the DC account) were available once or twice a year. Employees often had to
ﬁle with the plan sponsor at least a month in advance to change their contribution rate or
re-allocate their assets, often using information over 2 months old to make these
decisions. For those who could choose their investments (these plans are often called

'self-directed' plans), choices were Often limited to one or two types of assets, either

company stock, Guaranteed Investment Contracts (GICS), or some other type of ﬁxed-
income asset.

Today employees have more freedom when making decisions about DC plans.
Participants enjoy more ﬂexibility in changing contribution rates and many are able to
access the value of their account on a daily basis. The number of investments the
employee has to choose from is also greater. A 1995 RogersCasy study found that 74
percent of surveyed companies offer six or more investment Options (Shultz, 1996). In
addition, many plans are now offering loans (56 percent in 1998) (V anDerhei et al.,
2000), and DC plans allow the employee to take his balance with him upon job
separation.

But with greater freedom comes greater responsibility. In the new age of DC
plans, not all employees are making decisions that will offer them retirement security.
For instance, in 1993 only 65 percent of employees whose employer sponsored a DC plan
participated in the plan (Employee Beneﬁt Research Institute (EBRI), 1995). In addition,
sponsors of self-directed plans are concerned about the investment choices participants
are making. A 1994 survey of the largest DC plans in the country by Phoenix-Hecht
found that only 8 percent of plan sponsors feel that participants are well qualiﬁed to make
sound investment decisions (Phoenix-Hecht, 1994a). Furthermore, in 1996, 21 percent of
DC plan dollars were not reinvested in a tax-deferred account when the participant
separated from his job (Dickemper and Yakoboski, 1997).

It has been suggested that individuals are not making sophisticated choices
because they lack the ﬁnancial knowledge to do so. Traditional life cycle models of

saving require that an individual understand many ﬁnancial concepts, including the real

 

' The information in the following two paragraphs follows Hurt (1998).

2

rate of interest, inﬂation, and the beneﬁts of compounding. More speciﬁc models of
portfolio choice also require an understanding of risk/retum trade-offs and saving time
horizons.

Currently, both employers and policy makers are emphasizing retirement
education as one way to increase ﬁnancial knowledge and awareness about retirement
saving. In the early 19905 many ﬁrms initiated retirement education programs, and by
1994 over 88 percent of large employers reported offering some form of retirement

education (Pensions and Investments, 1995). In addition, the government initiated a

 

public saving and pension campaign through the Department of Labor (DOL) in 1995
aimed at increasing public awareness on the need to save for retirement.

Retirement education programs promote sophisticated saving choices in several
ways. Foremost, they provide ﬁnancial knowledge the participant may be lacking. They
also may enhance individual investment skills through providing ﬁnancial planning
training and guidance. In addition, the mere fact that the individual is exposed to
information on retirement saving may cause him to realize the importance of securing his
retirement and initiate a savings plan.

In this paper I will examine evidence that shows that many Americans may
beneﬁt from some form of retirement education. In the next section I will outline survey
results that show the state of ﬁnancial knowledge in the United States. In Section 3, I
will show that many individuals do not realize they are not preparing adequately for
retirement (or they are not making use of all available information). In Section 4, I
examine the manner in which individuals appear to be making saving choices, and

evaluate some of these methods. I offer concluding remarks in Section 5.

2. ECONOMIC AND FINANCIAL KNOWLEDGE

The economic and ﬁnancial illiteracy of Americans ranges from the inability to
handle the most elementary calculations to lack of knowledge in solving more complex
retirement saving problems. For example, existing studies show that only 20 percent of
adults can determine the correct change using prices from a menu (Bemheim, 1998).
Given that many people lack simple skills needed to perform everyday tasks, it is
doubtful that they possess the ability and knowledge needed to make optimal saving
decisions.

Bemheim (1998) uses a 1993 survey conducted by Merrill Lynch to test the
ﬁnancial and macroeconomic knowledge of individuals ages 29 to 47. The survey
contains eleven questions that cover such topics as the current rate of inﬂation, average
returns between different types of investments, and the effects of compounding interest.
Overall, individuals performed quite poorly, with more than 80 percent of the sample
answering at least ﬁve of the eleven questions incorrectly.

As I noted in the previous section, most DC plan sponsors do not feel their
employees have the knowledge to make sound investment decisions. Looking at the
Merrill Lynch results, their hesitation is not surprising. Nearly two-thirds of the sample
would not even guess as to the current level of the Dow Jones average, and 42 percent of
respondents could not choose the correct reason as to why returns might differ between
mutual funds and federally insured Certiﬁcates of Deposits. Additionally, nearly one-
third of individuals answered that $1,000 left in the bank for 30 years at 8 percent interest

would earn less than $5,000. The inability to appreciate the beneﬁts of compound

interest not only affects asset allocation decisions, but may also increase the chance that
DC participants who change jobs may spend their balances prematurely.

The results of a study done by EBRI (1996) using the 1995 Retirement
Conﬁdence Survey (RC S)2 reinforces the inadequacy of knowledge demonstrated in the
Merrill Lynch sample. The RCS included ﬁve questions (four true and false, one
multiple choice) covering subjects related to investment returns, goals for retirement
security, and average length of retirement. Although most of the individuals (80 percent)
knew what percentage of current income one would need to retire and maintain the same
standard of living, a Signiﬁcant amount of individuals could not correctly answer
questions on asset returns. Only 55 percent knew that over the past 20 years government
bonds provided a lower rate of return than the domestic stock market. Additionally, only
38 percent knew that company stock is not typically a less volatile option than investment
in a diversiﬁed portfolio of stocks.

In addition to general retirement knowledge, a number of individuals admit they
do not understand many concepts related to retirement (tax-deferred) accounts. In the
1997 RCS, 35 percent of those participating in a DC plan did not know the maximum
they are allowed to contribute (Dickemper and Yakoboski, 1997). Of those currently
eligible to make a contribution to an IRA, only 50 percent report they have a very clear or
somewhat clear understanding of the eligibility rules for making an IRA contribution
(Dickemper and Yakoboski, 1997). Relaying this type of straightforward information via
retirement education materials may increase the number of individuals who take

advantage of the tax-deferred nature of these accounts.

3. MY OPIA IN RETIREMENT PLANNING

In the 1998 RCS, 36 percent of respondents say they have not started to save for
retirement. Of these individuals, 21 percent report their reason for not saving as "just
have not thought about retirement", while 20 percent report they believe retirement will
"work itself out." If workers cannot realize their ﬁnancial vulnerabilities, ﬁnancial
knowledge alone will have little success in changing their behavior.

It is possible that the mere presence of retirement education - via employer-
sponsored classes or pamphlets, or just through public advertising - may incite some of
these individuals to start thinking about retirement. It is through this recognition effect
that retirement education may affect saving choices.3 I will outline two studies - one by
Bemheim (1998) and one done by EBRI using the 1999 RCS - that exemplify this overly-
optimistic attitude of many Americans regarding retirement and saving preparedness.

Bemheim (1998) uses a 1993 Merrill Lynch survey to examine the relationship
between self-rated ﬁnancial preparedness for retirement and actual ﬁnancial status. He
measures self-rated preparedness in three ways: whether the respondent feels his ﬁnances
are secure, whether he feels prepared for retirement, and whether he feels he will have a
greater standard of living in retirement. Bemheim measures actual ﬁnancial status by
segmenting the respondents into wealth-to-eamings quartiles and adjusting each group
for ﬁnancial vulnerability (including factors such as age, marital status and education).

He then compares the individual's own rating of preparedness with the actual ratios.

 

2 The RC8 is an annual survey covering 1,000 randomly-selected individuals ages 25 and over.

3 Cagan (1965) and Katona (1965) were the ﬁrst to put forth the idea of a recognition effect. They
hypothesized that those with pensions save more because the mere existence of the pension led them to
think about saving for retirement.

There is some evidence that individuals correctly assess their vulnerabilities, as
the fraction of individuals who describe themselves as very or somewhat well prepared
for retirement rises with the adj usted-wealth quartiles. However, patterns of overly
optimistic behavior also exist. Among those with the lowest actual preparedness, 54
percent regard their personal ﬁnances as secure or fairly secure. Forty-ﬁve percent of
this group believes they are very or somewhat well prepared for retirement, while only 37
percent expect to achieve a lower standard of living in retirement.

The 1999 RCS also measures self-assessed and actual retirement preparedness.
EBRI (1999) uses responses on actual ﬁnances and readiness from the survey and
constructs a retirement readiness scale4 with which to compare self-assessed retirement
conﬁdence. Of those reporting being very conﬁdent in their preparedness, only 25
percent are rated as being very well prepared for retirement, and 26 percent are rated as
being very poor to adequately prepared. Additionally, of those reporting being somewhat
conﬁdent, 50 percent are rated as actually being very poor to adequately prepared.

It is encouraging to note, however, that retirement education may increase actual
preparedness. The 1998 RCS reports that of individuals who say they have set aside
retirement savings, 33 percent list having a retirement plan or educational material at
work as a reason they started saving. Eighteen percent list a ﬁnancial advisor, and 17
percent credit print media articles or television as providing incentive to save.
Additionally, 33 percent of workers in an EBRI/Greenwald study said using employer-
provided educational material led them to increase the amount of their contributions to

their DC plan, and 44 percent said it led them to change their pension portfolio allocation

(EBRI, 1995). Although examination of correlations in the raw data does not imply
causation, the use of retirement education appears to be a promising way of affecting

saving behavior.

4. INDIVIDUAL RULES FOR SAVING BEHAVIOR

If individuals are not acting in ﬁnancially sophisticated ways, then how are they
making their saving decisions? Not all advice is good advice - be it from a ﬁnancial
planner, a retirement class, or the media. Determining an amount to save and how to
invest this money is a complex matter, one that may differ depending on the
characteristics and life circumstances of the individual. If an individual lacks the
economic and ﬁnancial knowledge to make sound decisions, he is likely to follow "rules"
of behavior that reﬂect his ﬁnancial illiteracy.

Several patterns of unsophisticated saving behavior can be observed in the

literature. Bemheim (1994a) uses data from a 1993 survey for The Equitable to examine

 

whether individuals act in a sophisticated manner by considering both current and future
earnings when making ﬁnancial decisions. The data provides information on current
income, whether the individual expects future income growth, and saving targets. The
standard life cycle theory predicts that when future income potential is higher, current
saving should be lower, as there is not as much need to save today for the future.
However, Bemheim ﬁnds just the opposite in the data: a higher percentage of those with

more potential future income growth are attempting to save (85 and 78 percent,

 

‘ The scale is computed from responses on actual saving amounts for retirement, if the individual has
estimated his needs for retirement, if he has established an investing or saving program for retirement, and
responses to various aspects of retirement preparedness.

respectively). In addition, mean and median saving rate targets are uncorrelated with
expected income growth.

Feenberg and Skinner (1989) present evidence that shows how ﬁnancial
information can still result in unsophisticated behavior if individuals do not interpret the
information correctly. They present data on individual IRA limits and actual IRA
contributions from the early 19805, after the liberalization of IRAs. Banks and other
ﬁnancial institutions put out massive advertising campaigns to get people to take
advantage of the tax-deferred status of these accounts, often using concrete examples to
promote the tax savings. Most of these ads used the $2,000 contribution limit in their
examples. The result was as ﬁnancial institutions had hoped: increased IRA
contributions. However, out of the 14 million families that contributed in 1984, 5.5
million contributed exactly $2,000 when they actually faced either a $2,500 or $4,000
contribution limit. It is not known for certain that these families did not intentionally
contribute below their limit, but it is highly unlikely that they would all choose to
contribute exactly $2,000.

Sometimes it is not so easy to determine whether a given behavior is
unsophisticated, because economic and ﬁnancial professionals cannot agree themselves
on a model that best describes optimal behavior. Most ﬁnancial professionals advise
individuals to follow a “rule of thumb” to determine the optimal amount of equity in their
portfolios, based on their age. They propose that an individual invest a proportion of
stocks equal to 100 minus his current age. So if a worker is 55 years old, he should

invest 45 percent of his portfolio in equities.

Bodie (1995) is one ﬁnancial economist who disagrees with the conventional
advice. He argues that the riskiness of holding equities actually increases the longer the
holding period when compared to a risk-less asset (such as default-free zero coupon
bonds). Theories that suggest that stocks are less risky in the long run base their claim
only on the probability of a shortfall (the difference between the value of the stock and
the bond at maturity), but do not factor in the magnitude of the shortfall. Using an
Option-pricing model to evaluate the cost of insuring against a shortfall, Bodie ﬁnds that
holding stocks is more risky compared to the risk-less asset as the holding period
lengthens.

Examination of age/equity patterns in the literature does not Show one speciﬁc
pattern relating equity holdings with age. While Papke (1998) and Goodfellow and
Schieber (1997) ﬁnd evidence of decreasing equity holdings with age, Muller

(Dissertation, Chapter 2) and Poterba and Wise (1998) ﬁnd no discemable relationship.

5. CONCLUSION

Evidence suggests that many individuals not only do not possess adequate
ﬁnancial knowledge to make sophisticated decisions, but also do not realize their
ﬁnancial vulnerabilities. Advice on retirement saving may still result in unsophisticated
behavior if individuals are unable to use the information and apply it to their own
situations. Furthermore, the situation is complicated if ﬁnancial professionals offer rule
of thumb advice or do not agree on what choices are optimal.

Much has been done in the area of retirement education in the past few years to

address some of these issues. Along with the savings campaign initiated by the federal

10

government, in 1992 the DOL ﬁnalized section 404(c) of ERISA. Section 404(c) limits
plan Sponsor liability with regard to advice given in retirement education seminars, given
sponsors follow certain guidelines. These guidelines list speciﬁc requirements pertaining
to asset allocation and require plan sponsors to provide certain types of retirement
education. Compliance is not mandatory, but as a 1998 KPMG Peat Marwick survey
Shows, most sponsors (those with 94 percent of plans) intend to comply with the
guidelines. With the fear of legal liability greatly diminished, more companies may offer
retirement education. Additionally, the requirements for education put forth in the
guidelines should streamline information given in employer-sponsored retirement

education across ﬁrms.

ll

REFERENCES

Bemheim, B. Douglas, "Financial Illiteracy, Education, and Retirement Saving," in
Living with Deﬁned Contribution Pensions: Remaking Responsibility for Retirement,
edited by Olivia S. Mitchell and Sylvester J. Schieber, 38-68. Philadelphia,
Pennsylvania: Pension Research Council, the Wharton School of the University of
Pennsylvania, and the University of Pennsylvania Press, 1998.

Bemheim, B. Douglas, "Personal Saving, Information, and Economic Literacy: New
Directions for Public Policy," in Tax Policy for Economic Growth in the 1990s,
Washington, DC: American Council for Capital Formation, 1994a, pp. 53-78.

 

Bodie, Zvi, "On the Risk of Stocks in the Long Run," Financial Analysts Journal, 51:
May/June 1995, pp. 18-22.

 

Cagan, Phillip, "The Effect of Pension Plans on Aggregate Saving: Evidence from a
Sample Survey," Occasional Paper 95, New York: Columbia University Press, 1965.

Dickemper, Jennifer, and Paul Yakoboski, “Increased Saving but Little Planning: Results
of the 1997 Retirement Conﬁdence Survey,” Employee Beneﬁt Issue Brief no. 191,
November 1997.

Employee Beneﬁt Research Institute, The 1999 Retirement Conﬁdence Survey (RCS)
Summary of Findings [Online] Available
http://www.ebri.org/rcs/ l999/1999_results.html, June 20, 2000.

 

 

Employee Beneﬁt Research Institute, The 1998 Retirement Conﬁdence Survey (RC S)
Summary of Findings [Online] Available http://www.ebri.org/rcs/ l998__results.html,
June 20, 2000.

 

 

Employee Beneﬁt Research Institute, The 1997 Retirement Conﬁdence Survey (RCS)
Summary of Findings [Online] Available
http://www.ebri.org/rcs/1997/1997_results.html, June 20, 2000.

 

 

Employee Beneﬁt Research Institute, “Can We Save Enough to Retire? Participant
Education in Deﬁned Contribution Plans,” Employee Beneﬁt Research Institute Issue
Brief no. 160, April 1995.

Employee Beneﬁt Research Institute, “Participant Education: Actions and Outcomes,”
Employee Beneﬁt Research Institute Brief no. 169, January 1996.

Feenberg, Daniel, and Jonathan Skinner, "Sources of IRA Saving," Tax Policy and the
Economy, 1989, pp. 25-46.

 

l2

REFERENCES (continued)

Goodfellow, GP, and SJ. Schieber, "Investments of Assets in Self-Directed Retirement
Plans," in Positioning Pensions for the Twenty-ﬁrst Century, edited by MS. Gordon,
08., Mitchell, and M.M. Twinney, 67-90. University of Pennsylvania Press, 1997.

 

Hurt, Ronald D., "The Changing Paradigm of 401(k) Plan Servicing," in Living with
Deﬁned Contribution Pensions: Remaking Responsibility for Retirement, edited by
Olivia S. Mitchell and Sylvester J. Schieber, 193-207. Philadelphia, Pennsylvania:
Pension Research Council, the Wharton School of the University of Pennsylvania, and
the University of Pennsylvania Press, 1998.

 

Katona, George, Private Pensions and Individual Saving, Ann Arbor: University of
Michigan Institute for Social Research, 1965.

 

KPMG Peat Marwick, Retirement Beneﬁts in the 19903: 1998 Survey Data. Washington,
DC, KPMG Marwick, 1998.

 

Papke, Leslie B, "How Are Participants Investing Their Accounts in Participant-Directed
Individual Account Pension Plans?" Michigan State University Working Paper, 1998.

Pensions and Investments. "Employees Getting More: Investment Education, Planning
Help on the Increase," January 23, 1995, p. 74.

 

Phoenix-Hecht, Plan Sponsor Survey: Large Deﬁned Contribution Plans Respondent
Report. Research Triangle Park, NC: Phoenix-Hecht, 1994a.

 

Poterba, James M., and David A. Wise, "Individual Financial Decisions in Retirement
Saving Plans and the Provision of Resources for Retirement, " in Privatizing Social
Security, edited by Martin F eldstein, 85-105. Chicago: University of Chicago Press,
1998.

 

Schultz, Ellen E., "Executives See Trouble in Employee's Nest Eggs," Wall Street
Journal, March 27, 1996, C1.

VanDerhei, Jack, Holden, Sarah and Carol Quick, "401(k) Plan Asset Allocation,

Account Balances, and Loan Activity in 1998," Employee Beneﬁt Research Institute
Issue Brief no. 218, February 2000.

13

Chapter 1: RETIREMENT EDUCATION AND PENSION PRESERVATION:
DOES RETIREMENT EDUCATION TEACH
INDIVIDUALS TO SAVE PENSION DISTRIBUTIONS?

INTRODUCTION

AS deﬁned contribution (DC) plans have become increasingly common over the
past two decades, lump sum distributions from these pension plans have also become
more common. DC plans give employees the ﬂexibility to take the balance of their
account with them upon job separation. For those who elect this option, they can choose
to maintain these funds in accounts designated for retirement, invest them in other
savings vehicles, or spend them. If spent pension savings are not replaced by other
saving, retirement wealth for the future elderly is more likely to be insufﬁcient to
maintain a desired standard of living in old age.5 As people live longer, pension
retirement savings will play an increasingly important role in ﬁnancing retirement.

Most previous research on the preservation of pension funds has focused on the
demographic and ﬁnancial characteristics of those individuals who spend their
distributions upon job separation (Andrews, 1991; Atkins, 1986; Bassett, Fleming and
Rodrigues, 1998, henceforth BFR; Poterba, Venti and Wise, 1998, henceforth PVW).
While identifying the characteristics of these individuals is important, it is also necessary
to look at what can be done to halt this leakage out Of the private pension system. Two
pieces of tax legislation in 1986 and 1992 include penalties for pension withdrawals

before the age of 55. Although tax penalties are somewhat successful in lowering the

 

5Engelhardt (1999) estimates that spent distributions result in a decrease in retirement consumption by
approximately 6-8 percent, given that Social Security and pensions account for three-quarters of retirement
income.

14

number of employees spending their pensions before retirement, statistics show many
individuals still do not save their balances. Using data from the 1996 Retirement
Conﬁdence Survey (RCS), Yakoboski (1997) reports that 10 percent of workers said they
had the Opportunity to take their pension balances with them upon job separation in the
last three years. Of those who elected to take the distribution, 50 percent reported
spending some or all of the money on consumption. This results in approximately 5
percent of employee pension distributions being used for consumption.

Traditional economic models of saving assume that individuals have the
knowledge to make utility maximizing saving decisions, or at least they act as though
they do. These models predict that the consumption/saving choice individuals make —
although it may be to consume a pension distribution rather than save it — is the utility-
maximizing choice for that individual. However, if choices are based upon limited
information and ﬁnancial knowledge, these decisions may not actually be the best ones
for individuals and their retirement security.

Educating workers about the necessity to save for retirement is one way of
increasing pension preservation upon job separation. This education can come in the
form of private ﬁnancial planners, employer-sponsored materials, or retirement classes
offered at institutions such as community colleges. Bemheim and Garrett (1996) ﬁnd
that employer-sponsored education increases general saving rates by approximately 2
percent and increases retirement wealth by $2,176. Thus, educating individuals about
their future income security may be a valuable way to reach employees who currently
spend their pensions before retirement.

In this paper I examine the relationship between retirement education -

15

speciﬁcally retirement meetings — and the saving of pensions upon separation from a job.
I include both employer-sponsored meetings and those offered by public and private
institutions in my study. I analyze this relationship using two deﬁnitions of saving - a
deﬁnition that includes only re-investment in tax-deferred saving vehicles (Individual
Retirement Accounts (IRAS), new employer pension plans, etc.) and a broader deﬁnition
that includes tax-deferred vehicles, general saving vehicles (stocks, bonds, saving
accounts, etc.) and paying off debt. Because different types of individuals may not react
in the same way to attending a retirement class, I also evaluate the effects of retirement
education for speciﬁc types of individuals especially important to policy makers.
Previous research on pension distributions identiﬁes younger individuals, females, and
those without college degrees as less ﬁnancially secure in retirement, making the effects
of retirement education on these groups particularly relevant.

I use a cross-section of 938 individuals from the 1992 Health and Retirement
Study (HRS) who participated in a DC plan in a previous job. This is the ﬁrst study to
test the relationship between retirement education and pension preservation, while
controlling for characteristics at the time of the pension distribution such as reason for job
separation and income and marital status at the time of separation from employment. In
addition, this data set is unique in that it provides a rich source from which to construct
proxy variables to control for unobservable saving tastes.

The next section provides background information on pension distributions, their
tax treatment, and an overview of retirement education and why it may be effective in
slowing the ﬂow of money out of the pension system. In Section 3, I summarize the

literature on pension distributions and retirement education. I provide an overview of the

16

data in Section 4. Section 5 provides econometric evidence of the effect of retirement
education on pension preservation. I ﬁrst estimate the effect of retirement education on
the use of pension distributions with and without additional control variables. I then add
interaction terms to account for differences in the effects of retirement education on
pension preservation for sub groups who may be at risk of ﬁnancial insecurity at
retirement. I also discuss possible selection bias in the model and present two ways to

address this bias. I offer concluding remarks in Section 6.

2. BACKGROUND
Pension Preservation

Pensions can be categorized into two broad types: deﬁned beneﬁt (DB) and DC
plans. The amount of beneﬁt in a DB plan is usually determined by a formula that
includes years of service and average annual salary. Beneﬁts are paid in a stream of
annuities beginning at a speciﬁed age. Portfolio managers decide how the assets in the
pension fund are invested. DB plans were by far the most common type of plan until the
19805.

The typical DB plan does not allow for pre-retirement distributions upon job
separation. In 1988, only 10 percent of employers allowed their pre-retirement
employees the Option to take their pension balance with them upon separation
(Fernandez, 1992).6 Pension law does allow employers the option, however, to

unilaterally cash out employees with balances Of less than $5,000, regardless Of the

 

6 Technically, employees with traditional DB plans do not can-y balances, since each employee does not
have an individual account. The balance referred to here is calculated using a formula that reﬂects the
amount of service completed upon separation.

17

employee’s preference to remain in the plan.7

The design of DC plans is quite different. The 401(k) Plan, Employee Stock
Ownership Plans (ESOPS), and Simpliﬁed Employee Pension Plans (SEPS) are a few of
the common types of DC plans. DC plans establish an individual account for every
employee into which the employee, the employer, or both may make contributions.
Contributions and interest income are tax-deferred until withdrawal. In addition, many
DC plans allow the employee to make his own investment choices. The implementation
of 401(k) plans in 1981 and their subsequent growth has accounted for most of the
increased use of DC plans in recent years.

Most DC plan sponsors allow employees to take the balance of the account with
them upon separation, regardless of the reason for separation. Employees may choose to
keep the ﬁmds in a former employer's pension plan, transfer the funds to the new
employer's plan, roll the money directly into a tax-deferred account such as an IRA, or
take the distribution as a lump sum.8 Those who take the lump sum distribution can then
spend the money, invest it in a general saving vehicle, or put it into an IRA. Leaving
pension funds with the previous employer, transferring it to a new employer's account, or
transferring it directly into another tax-deferred vehicle is called a direct rollover or
ﬁduciary-to-ﬁduciary transfer. Taking a lump sum distribution and then putting it into an
IRA is called an indirect rollover. In the remainder of this paper, I will use the term
pension distribution to refer to all pension balances that are available upon job separation,

whether the money is spent or saved. I will use the term lump sum distribution or

 

7 The Employee Retirement Income Security Act (ERISA), passed in 1974, originally set the maximum
unilateral cashout at $1,750. This limit was raised to $3,500 by the Retirement Act of 1984. Effective
January 1, 1998, the limit was raised to $5,000.

8 Chang (1996) notes that, as of 1988, most pension plans did not allow rollovers from other pension plans.

18

cashout to refer speciﬁcally to those distributions for which the employee actually
withdraws the pension funds, and then decides whether to spend them or put them into a
general or tax-deferred savings vehicle.9

In the current economic environment where workers are increasingly mobile, the
popularity of the DC plan is likely to continue. In 1975, 13 percent of workers were
covered by a primary DC plan; by 1987 that number rose to 30 percent (Beller, 1992).
With this increase in DC plans the availability of distributions has also increased. The
number of workers reporting the availability of a distribution from any retirement plan
was 47.8 percent in 1983. By 1993, 71.5 percent reported having this option (Scott and
Shoven,]996)

As the Option to cash out pension balances upon job separation has become more
widespread, the percentage of all workers electing to maintain their balances in a tax-
deferred plan has also grown in recent years, albeit modestly. In 1983, 45 percent of
individuals reported saving their balances in tax-deferred accounts. In 1993, that number
rose to 50 percent (Scott and Shoven, 1996). Paul Yakoboski (1997) reports similar
trends using Hewitt Associates data from 1993 and 1996. In 1993, 35 percent of
individuals rolled over their pension funds into a tax-deferred account. By 1996, that
number had risen to 40 percent.

Speciﬁc groups of individuals are more likely to spend their distributions. Lower-
income or less-educated individuals, as well as those with smaller distributions, are less

likely to save their pension distributions in either a retirement or general saving vehicle

 

9 If a worker separates from his job and has the option to do whatever he wishes with his money, he is
choosing what to do with his pension distribution. If that worker chooses to actually take a check from the
employer for his pension and then decides what to do with the money, he takes a lump sum distribution. A
lump sum distribution is a particular form of a pension distribution.

19

(Atkins, 1986; BFR, 1998; PVW, 1998). Females are less likely than males to save their
distributions in a tax-deferred vehicle (Hurd, Lillard and Panis, 1998, henceforth HLP).
Chang (1996) concludes that although a 10 percent penalty on consumption of
distributions increased the probability of rollovers for high-income groups (those with
annual incomes exceeding $39,999), there was no effect for low-income groups.

Groups that are more likely to spend their pension distributions are among those
most vulnerable to inadequate retirement security. Low wage earners tend to also have
less formal education, which means decreased potential for ﬁrture wage growth and
subsequent lower retirement wealth. The probability of spending a distribution decreases
with age (Atkins, 1986; PVW, 1998) and younger recipients tend to have lower account
balances at the time of j Ob separation (Piacentini, 1990). A younger worker who receives
a distribution will most likely change jobs more often in his lifetime than an older
worker, allowing him more opportunities to spend his distribution. Spending a small
pension distribution from one job may not appear to affect retirement security by much,
but repeated distributions that are not saved result in signiﬁcantly smaller retirement
wealth.10 Females are also at greater risk for inadequate retirement security. Divorced
and widowed households headed by women in their 505 and 605 have approximately two-
thirds the median net worth” of Similar households headed by men (Smith, 1995).]2
Increasing numbers of individuals with inadequate personal retirement wealth increases
pressure on Social Security and may translate into higher future outlays for government

programs such as Medicaid and Supplemental Security Income.

 

'0 See Piacentini (1990) for calculations of retirement income loss due to spending of pension distributions.
” Net worth includes IRA and Keogh accounts, but does not include Social Security or private pension
wealth.

”This difference in net worth is found only for white males and females.

20

I must emphasize that in some cases spending the pension distribution may be the
individually optimal thing to do. However, from a public policy perspective, inadequate

personal retirement security for future generations is an issue that must be addressed.

Tax Legislation

Government policies aimed at preserving pension distributions have been
predominately in the form of tax penalties. The Tax Reform Act of 1986 (TRA86)
contains the ﬁrst changes in the tax code to address pension distributions. TRA86
imposes a 10 percent penalty for those under age 55 '3 on the distribution amount if the
distribution is not re-invested in a tax-deferred retirement vehicle within 60 days of
receipt. For example, a 45 year-old with a pension balance of $50,000 would have to pay
$5,000 in penalty tax if he spent or invested his distribution in a non-tax-deferred savings
instrument, or put it into a tax-deferred instrument later than 60 days after receipt.

Congress added a withholding tax to pension distributions not directly rolled over
into another tax-deferred account in 1992. The Unemployment Compensation
Amendments Act imposes a 20 percent withholding tax for those under age 55 on the
amount of the pension not transferred directly into another tax-deferred retirement
account. This law allows the federal government to withhold 20 percent of a distribution

whenever a qualiﬁed recipient takes a lump sum distribution.”

 

'3 Withdrawals made after job separation are subject to the penalty if the individual is under age 55 at time
of withdrawal. Exemptions to the penalty exist. See: US Congress, Joint Committee on Taxation, 1987.

'4 The 20 percent withholding tax is structured differently than the 10 percent penalty tax. The 10 percent
tax is an excise tax. The 20 percent tax is a federal income withholding tax. Twenty percent is taken out of
the distribution and if the individual then puts the pension money back into a tax-deferred account, he is
credited for this amount in that year’s federal income tax. However, in order to not incur the 10 percent
penalty on the 20 percent withheld, the individual must come up with the additional 20 percent and invest
this in the retirement account as well. If he does not, 10 percent of the 20 percent withheld is taken in
penalty tax.

21

Has tax legislation affected the consumption of funds meant for retirement?
Because the portability of pensions is a relatively recent phenomenon, very few studies
have been done on the effects of tax legislation. Chang’s (1996) study on the effects of
the 10 percent tax penalty on the saving of pensions is the only econometric evidence to
date. Her estimates suggest that the penalty increases the probability of rolling over a
pension distribution by 4 percent for high-income groups. The penalty has no statistically
signiﬁcant effect on low-income groups, particularly those who are under 55 years of
age. Scott and Shoven’s (1996) tabulations also document differences in rollovers across
income groups. For individuals who earn more than $30,000 annually, the propensity to
cash out declined from 1983 to 1988, and continued on a downward trend from 1988 to
1993. But for those with incomes below $30,000, the cashout propensity declined from
1983 to 1988 and then rose from 1988 to 1993. Although the tax penalty may affect the
behavior of certain types of individuals, other types of workers still fail to roll over their
distributions.

To date, a similar study has not been done for the 1992 legislation. However,
Fronstin et a1. (1997) tabulate the use of lump sum distributions ﬁom the 1993 EBS and
compare them with similar tabulations from the 1996 RCS. The number of lump sum
distributions invested in retirement accounts before and after the 1992 legislation is
remarkably similar: 45.5 percent for distributions received from 1987 to 1993 (EBS) and
46 percent for distributions received from 1993 to 1996 (RCS). The numbers suggest
that the withholding tax has not had much effect; however, it is premature to conclude

this without further econometric evidence.

22

Retirement Education

The availability of retirement education, like that of pension distributions, has
grown remarkably in the past 20 years. Although private ﬁnancial planners have been
available for many years, the availability of retirement education at community colleges
and in the workplace has just grown substantially throughout the 19805 and 19905. In
1994, 88 percent of large employers offered ﬁnancial education, with two-thirds of these

programs added after 1990 (Pensions and Investments, 1995).

 

Retirement education takes many different forms. Some ﬁrms use only written
materials such as brochures and pamphlets, while others offer workshops and seminars,
or one-on-one ﬁnancial advising. Most recently some sponsors have added Internet sites
and interactive software speciﬁcally designed to inform individuals about retirement
saving. In 1993, approximately 65 percent of pension plan sponsors provided
newsletters, and 44 percent offered seminars to all workers (Bayer, Bemheim, and
Scholz, 1996, henceforth BBS).

Retirement education includes a variety of topics, from the basics of investing and
deﬁnitions of ﬁnancial terms, to subjects such as asset allocation, the effects of
compounding and inﬂation, and retirement income projections. In 1995 nearly 100
percent of employer-sponsored programs covered asset allocation and 95 percent
explained risk and risk tolerance. Eighty-eight percent discussed basic investment
terminology and explained the characteristics of the pension plan, while 73 percent
showed employees how to calculate income for retirement. But only 39 percent
speciﬁcally addressed the impact of spending pension distributions before retirement

(Employee Beneﬁt Research Institute (EBRI), 1995).

23

The growth of retirement education has lagged behind the growth of DC plans. In
the past, employers have expressed concern for the legal liability that may result from
providing ﬁnancial education. Because most DC plans offer participants the ability to
choose how to invest their pension balances (commonly referred to as self-directed
plans), employees often look to employers to advise them on how best to allocate their
assets.15 Advising employees is risky - employers who offer anything that resembles
advice may be held liable for losses incurred by the employee, even if the employee
ultimately is the one who makes the investment decision.

In addition, not until recently has the government provided clear guidelines for
employers to follow when organizing a retirement education program. In 1992 the
Department of Labor ﬁnalized section 404(c) regulations of ERISA that outlines ways for
employers to minimize their legal risk. Although following these guidelines does not
completely protect a ﬁrm from legal liability, it does provide a defense against allegations
of misdirected advice. These guidelines have allowed ﬁrms that otherwise would not

have offered retirement education to implement programs at the workplace.

Why Should Retirement Education Affect Pension Preservation?

Americans demonstrate a clear need for ﬁnancial guidance. Only 33 percent of
surveyed adults understand the basics of compounding interest, while 42 percent can not
identify why federally-insured certiﬁcates of deposit have a lower rate of return than

privately held mutual funds (Bemheim, 1998). Individuals are also uninformed about

 

'5 A Bureau of Labor Statistics survey published in 1990 reported that 90 percent of participants in
particular types of DC plans had self-directed plans. For further information see: United States Department
of Labor, Bureau of Labor Statistics, "Employee Beneﬁts in Medium and Large Firms, 1989," Bulletin
2363, June 1990.

24

retirement planning. Results from the 1997 RCS Show that only 36 percent of workers in
the United States have any idea of how much they will need to save for retirement, and of
these individuals, 24 percent cannot provide an actual ﬁgure. Furthermore, of those who
participate in a 401 (k) plan, 35 percent do not know their contribution limit (Dickemper
and Yakoboski, 1997).

With little knowledge about ﬁnancial matters and retirement, workers may be
making different spending and saving decisions than they would had they more
information. An informed decision about the use of pension distributions involves
knowledge about the employee's pension plan, how much income is needed for
retirement, and the advantages of compound interest, for example. But does educating
individuals change behavior?

Evidence suggests that exposure to ﬁnancial information and advice does affect
saving behavior. It is likely that formal education (i.e., high school, college) increases
general economic literacy and helps to develop the skills needed to process ﬁnancial
information. Bemheim and Scholz (1992) ﬁnd that those with more formal education are
more likely to Show evidence of sophisticated ﬁnancial planning. When comparing the
average person with and without a college degree, those with college degrees save more
adequately for retirement — that is, they make decisions that are more consistent with the
life cycle model.

Evidence also shows that retirement education - employer-sponsored classes in
particular - affects savings behavior. Bemheim and Garrett (1996) ﬁnd that individuals
who make use of retirement education offered by their employer have a general saving

rate 2.2 percent higher than those who do not use educational material. A 1996 EBRI

25

analysis reports that 39 percent of workers in pension plans said that using employer-
provided materials or attending seminars led them to increase their contribution rate to
the plan. In addition, 53 percent of workers reported that additional advice on ﬁnancial
planning by their employers would lead them to save more for retirement. Thus it is
reasonable to suggest that retirement education may have an effect on general saving

behavior, and speciﬁcally, the choice to save a pension distribution.

3. PREVIOUS RESEARCH
Who Saves and Who Spends their Pension Distributions?

While there are no studies that address the impact of retirement education on
pension preservation, previous work on pension distributions has examined the
characteristics of those individuals who spend their pensions and those who save them.
BFR (1998) and PVW (1998) both use the EBS in cross-section analyses. BFR include
distributions from DB and DC plans and ﬁnd that those with a college degree are 11
percent more likely than those without a degree to roll over their distributions into a tax-
deferred account. They also ﬁnd that as family income increases, so does the probability
of rollover. For each additional $1,000 in family income, the rollover probability
increases by 3 percentage points.

PVW (1998) also examine the factors affecting the decision to save a pension.
They use the 1993 EBS as their primary data set, and support these ﬁndings with data
from the 1992 HRS. They ﬁnd that older individuals, workers with a college degree and
those with higher current labor income are more likely to save their pensions. Workers

with higher pension balances at time Of receipt are also more likely to save their

26

distributions: the probability of rolling over a distribution of more than $25,000 is 25
percentage points higher than that of a distribution worth less than $1,000.

Chang (1996) uses the 1988 EBS to study the effect of the TRA86 10 percent
penalty on increasing pension preservation. She uses both traditional probit estimation
and a difference-indifference approach in her analysis. Results from the probit

speciﬁcations Show that a one percent increase in the tax price of spending the

distribution raises the probability of rollover among high-income groups by 0.4 percent.I

A one percent increase in the tax price for low-income groups raises the probability by
only 0.2 percent. Using the difference-in-difference approach, a one percent increase
raises the probability of rollover by 1.3 percent for high-income individuals, but has no
statistically signiﬁcant effect on the use of the distribution for low-income individuals.
The ﬁnding that low-income individuals do not respond to tax penalties for
spending pension distributions is important, considering low-income workers are more
likely to spend their distributions in the ﬁrst place. The tax penalty does not affect the
behavior of those who most need to preserve their pensions, resulting in increased

economic vulnerability in old age and perhaps more impoverished elderly in the future.

Does Retirement Education Affect Saving? Evidence in the Literature
There are no known econometric studies on retirement education outside of the
workplace. However, there are two studies that address the relationship between

employer-sponsored retirement education and savings behavior. The ﬁrst, by Bemheim

and Garrett (1996), uses a sample of 2,055 individuals, ages 30 through 48 in 1994. The

 

6

'6 Workers who do not roll over their pensions must pay income tax on the amount of the distribution in the

year received. Because TRA86 also lowered income tax rates for most levels of income, the tax price of a
non-rollover must also incorporate these lower marginal income tax rates.

27

survey is sponsored by Merrill Lynch and includes information on both the availability
and use of employer-sponsored education. The type of retirement education offered
(written materials or seminars) is not speciﬁed, so the analysis estimates the effects of the
typical retirement education program in the workplace.

Bemheim and Garrett use median values of the dependent variable to estimate the
effects of employers offering and employees participating in retirement education on the
household saving rate, retirement saving rate, total wealth, and retirement wealth. The
results suggest a relatively small, positive relationship between the presence of retirement
education and the household saving rate and retirement education and retirement wealth.
Workers whose employers oﬂer retirement education save 1.7 percent more than those
not offered education, and have $2,488 more in retirement wealth. Employees who use
retirement education materials save 2.2 percent more than those who do not, while
accumulating $2,176 more in retirement wealth. Coefﬁcient estimates on both the
offering and usage of retirement education in speciﬁcations using total wealth and the
retirement saving rate are statistically insigniﬁcant. The fact that the general savings rate
increases more than the retirement saving rate suggests that perhaps retirement education
not only affects saving for retirement, but has a spillover affect on saving in general.
Alternatively, individuals could be using general saving vehicles to save for retirement.

Bemheim and Garrett also test the hypothesis that attending a retirement meeting
is correlated with unobserved saving tastes. They conclude that those with lower
propensities to save are more likely to be offered retirement education and to use it in the
workplace.

BBS (1996) use a 1993-1994 longitudinal survey of 1,100 employers selected by

28

KPMG Peat Marwick to study the effects of employers offering retirement education on
participation and contribution rates in 401(k) plans. Their data allows them to examine
the effects of both seminars and written material, as well as the frequency in which
education is offered. When estimating these effects for both highly compensated (HC)
and non-highly compensated (NHC) employees”, they ﬁnd that participation rates are
11.5 percent higher for NHC employees and 10.5 percent higher for HC employees when
retirement education is offered, given that the seminars are offered frequently. ‘8
4. DOES RETIREMENT EDUCATION PROMOTE SAVING OF
DISTRIBUTIONS? EVIDENCE FROM THE 1992 HRS

The 1992 HRS is the ﬁrst wave of an ongoing survey of 12,057 men and women,
most between the ages of 51-61 in 1992.19 Besides including standard demographic and
ﬁnancial information, the survey also includes questions speciﬁcally relating to
retirement. Detailed pension information and expectations about the retirement years
make for a rich source of data for analyzing retirement issues. As I am interested in how
retirement education affects DC participants who received a pension distribution, my
sample consists of the 938 individuals who participated in a DC plan in their previous
job.

Although the HRS is a longitudinal study, I only use the 1992 wave. The

questions addressing retirement education are only asked during the 1992 survey, and

 

'7 Highly compensated employees are deﬁned as having annual earnings of $100,000 or more, owning
more than 5 percent of the company, or having annual earnings Of $65,000 or more if this amount is in the
top quintile of the ﬁrm’s salary distribution.

'8 The offering of retirement education has no effect if the education is in the form of written materials, or
if seminars are Offered occasionally.

'9 Because spouses of respondents of this age group are also asked the same questions, only 9,673
individuals ﬁt into this age category. Since my analysis looks at a past event, a pension distribution, the age
range at which these distributions occurred is larger than the age range of the respondents at the time of the
Intervrew.

29

although using information from subsequent waves may add new lump sum distribution
recipients to the sample, information on attendance in retirement classes after 1992 is not
available. Given the acceleration of retirement education in the 19905, there is a high
probability of undocumented participation for those who attended a meeting between
1992 and 1996.

Table 1.1 reports summary statistics for the sample of individuals for whom I am
able to classify their response in the two savings categories and who also answered the
question on attending a meeting. Out of 640 individuals”, 19 percent report having
attended a meeting. Of those who attended a meeting, 72 percent (14 percent of the
sample) took this meeting from either their employer or their spouse's employer. Forty-
nine percent of the sample is female and 25 percent of the respondents earned at least a
college degree. The mean age when the distribution was received is 49 years of age. The
average labor income of both spouses at the time of the distribution is $50,247, while the

mean pension balance at the time of the distribution is $33,797.21

Pension Preservation in the HRS

All respondents were asked if they had a previous job for which they worked at
least ﬁve years. If they did, they were asked if they had an employer-sponsored pension
plan at that job and if the plan was characteristic of a DB plan, a DC plan, or had
characteristics of both. Because only a small percentage of DB plan participants

historically have had the choice on whether to take a distribution, only those with DC

 

2° Tables 1-4 include only individuals in the sample who answered the retirement meeting question and
whose pension distribution can be classiﬁed as either general savings, retirement savings, or paying off
debt. Because many individuals did not answer one or both of these questions, the sample size for the
regressions drops from 938 to 640.

30

plans are included in the analysis.22

The HRS then asks each respondent what he did with his pension upon job
separation. Options include rolling it over into a new employer's pension plan, leaving it
with the former employer to accumulate, rolling it directly into an IRA, converting it to
an annuity, or receiving a distribution.23 For those who elected to receive a distribution,
the use of the distribution is recorded.24 Choices include spending it, saving/investing it
in general saving vehicles or paying Off debt, or investing it in an IRA. Only one use for
each respondent is recorded.

I analyze the effect of retirement education on saving a distribution, using two
deﬁnitions of saving. The broad deﬁnition includes saving the distribution in either a
tax-deferred (or retirement) or a non-tax-deferred (or general) savings vehicle. Included
in this deﬁnition are direct rollovers into new employer accounts, IRAs, or annuities,
leaving the distribution with a former employer, taking a lump sum distribution and
investing it in an IRA or general saving vehicle, or paying off debt.25 A general saving
vehicle is any investment vehicle that is not tax-deferred, including but not limited to,
stocks, bonds, certiﬁcates of deposit, and bank saving accounts. Henceforth, this
deﬁnition of saving will be referred to as all types of saving. The narrower deﬁnition
includes types of saving included in the broad deﬁnition except for saving in general

vehicles and paying off debt. Henceforth, this deﬁnition of saving will be referred to as

 

2' All dollar variables are in 1993 dollars.

22 Individuals whose pension plans had characteristics of both DB and DC plans are omitted from the
analysis, given that it is not possible to sort out those plans that typically allowed distributions (DC types)
from those that did not (DB types).

23 Other options include losing the beneﬁts, having beneﬁts in litigation and waiving pensions beneﬁts in
exchange for other beneﬁts. Respondents who chose these options are omitted from the sample.

2" Whether the individual was eligible to receive a distribution is not given. Because DC plans have
historically had a high rate of portability, the omission of data on eligibility is not an issue here.

31

retirement saving.

Employing two different deﬁnitions of saving - one including not only tax-
deferred accounts but also general savings accounts - is important to this analysis. One
characteristic of tax-deferred accounts is their lack of liquidity. Withdrawal of funds
from a tax-deferred account such as an IRA before age 59-1/2 brings with it a substantial
tax penalty. For this reason, individuals, especially younger ones, may keep funds they
have mentally designated for retirement in non tax-deferred instruments. Individuals who
invest their pension in general saving vehicles should not be classiﬁed with those who
simply spend their distributions on consumption.

Table 1.2 reports the uses for pension balances for the sample. The top two rows
report the percentage of respondents who saved their distribution in either all types of
saving vehicles or retirement vehicles only. Sixty-nine percent of respondents saved their
pensions in all types of savings vehicles, while 60 percent saved their pensions in a tax-
deferred account. Subtracting these two ﬁgures shows that 9 percent of DC participants
saved their distributions in a general saving vehicle or paid off debt. The remaining rows
report the speciﬁc use of the distribution. Thirty-one percent spent their distributions on
consumption (using the broad deﬁnition of saving). Of those who did not spend their
distributions, the most frequent uses include leaving the money in a previous employer's
account (30 percent) and rolling over the funds into an IRA (22 percent).

At least a third of DC participants report spending their distributions - 31 percent
if saving is deﬁned broadly, 40 percent if saving is deﬁned as only retirement saving.

These ﬁgures appear low when compared to a study done by Yakoboski (1997), who

 

25 I include paying off debt as saving because, all else equal, it increases net wealth, and therefore, future
income.

32

reports a rate of 54 percent for 50-59 year Olds using 1993 Hewitt data. HLP (1998) also
report a rate of 54 percent when using the 1992, 1994, and 1996 waves of the HRS.
These ﬁgures, however, include only those individuals who take a lump sum distribution
- employees who leave their funds with old employers or transfer them to new employers
are not included in the sample. If I eliminate these people from my sample, the resulting
percentage of individuals who do not save their pensions in a retirement vehicle increases
to 64 percent, slightly higher than the ﬁgures provided by Yakoboski and HLP. The
timing of the receipt of the distribution in my sample may account for the greater
percentage of respondents spending their distributions. While the other studies cover
predominately distributions received in the 19905, the mean year of recipiency in my
sample is 1984. Andrews (1991) ﬁnds that the more recent the distribution, the higher
the probability of it being saved in a retirement vehicle. The 1981 legislation lifting the
restriction on the opening of IRAs to those covered by employer pension plans may have
affected the disposition of cashouts by making it easier to keep pension funds in the

retirement system.

Information on Retirement Education in the HRS

The HRS is the only public data source that contains data on both retirement
education and pension distributions. All respondents are asked two questions about
retirement education: have they ever attended any meetings on retirement or retirement
planning? Of those who answered yes, they are then asked if that meeting was organized
by the respondent's or spouse's employer.

Table 1.3, column (1) reports the percentage of DC participants who took a

33

retirement class, by age, education, and gender. The percentage Of each group that saves
their pensions is then reported in columns (2) and (3). Observing all three columns
shows a general trend among groups - those who are less likely to attend meetings are
also in those groups that have fewer individuals saving their pensions. The most striking
differences appear within the age and educational categories. While half as many non-
college graduates as graduates attended a meeting (15 versus 30 percent), only 66 percent
of non-graduates saved their pensions in all types of saving vehicles, while 80 percent of
graduates saved their distributions. Looking at age, 14 percent of those age 40 and under
attended a meeting, while 20 percent of those over age 40 attended a class. The
percentage of younger workers saving their distributions in all types of saving is 58
percent, compared to 71 percent of older workers.26

Table 1.4 reports the proportion of distributions saved and spent, depending on
whether the respondent attended a meeting or not. (Expressing proportions as
percentages), columns (1a) and (1b) report the percentage of workers who saved their
distribution in all types of saving vehicles, depending on retirement meeting attendance.
Columns (2a) and (2b) do the same for investment in retirement vehicles only. Columns
(1 c) and (2c) report the percentage point difference in saving a distribution between those
who attended a meeting and those who did not. Column (2) is subtracted from column
( 1) in order to determine the percentage of individuals who invested in general saving
vehicles only or paid off debt. I ﬁrst report percentages for all DC participants, and then
proportions by age, gender, and receipt of college degree.

Tabulations show that for the full sample, 73 percent of DC participants who

 

26 The difference in class attendance for college graduates and non-graduates is the only difference that is
statistically signiﬁcant (at the five percent level, not shown on the table).

34

attended a class invested their pensions in any type of saving vehicle versus 68 percent of
those who did not attend a meeting. Sixty-four percent of individuals who attended a
meeting saved in retirement vehicles, versus 59 percent of those who did not participate
in a class.

The differences in pension preservation between those who attended a meeting
and those who did not become more apparent when focusing on sub groups of
individuals. Seventy-eight percent of those age 40 and under saved their distributions in
all types of saving vehicles if they attended a class, versus 54 percent of those who did
not attend a class, with the difference being statistically signiﬁcant at the 10 percent level.
No difference is found when comparing investment in retirement saving vehicles: 36
percent Of those age 40 and under saved in tax-deferred vehicles regardless of class
attendance. The differences between all types of saving and retirement saving show that
much of the investing of this group is done in general saving vehicles: 42 percent of
those age 40 and under who attended classes saved in general vehicles only, versus 18
percent of those who did not attend a class. Small, statistically insigniﬁcant differences
are found for those over age 40, regardless of the deﬁnition of saving used.

Females and males also show different patterns of saving their distributions
depending on whether they attended a class. While the percentages of females saving
their distributions in either type of savings vehicle differs only slightly between those
who attend a meeting and those who do not (2 percentage point decrease for all types of
saving, 3 percentage point decrease for retirement saving), the difference for males is
more striking. Ten percent more males save their distributions if they attended a class,

regardless of the deﬁnition of saving used. Both differences are statistically signiﬁcant at

 

35

the 10 percent level. It is apparent, however, that the difference for all types of saving is
being driven by the difference in retirement saving, with 6 percent of males saving in
general saving vehicles or paying off debt regardless of class attendance.

The effects of retirement education also appear to differ between college and non-
college graduates. A greater percentage of non-college graduates saved their
distributions if they attended a meeting, with 7 percent more saving their distributions in
all types of saving and 4 percent more re-investing their pension balances in retirement
vehicles. On the other hand, a smaller percentage of college graduates saved their
distributions if they attended a meeting, regardless of the deﬁnition of saving used.
Seven percent fewer graduates saved their pensions in all types of saving vehicles, with 4
percent fewer saving in retirement vehicles only. Neither of these differences is
statistically signiﬁcant, however.

The differing effects of retirement education depending on educational level are
reported in two studies by EBRI (1995, 1996). A 1993 EBRI/Matthew Greenwald and
Associates study found that of those workers who read educational material or attended
seminars, 41 percent of those without a college degree report that the seminars led them
to increase the amount of their contributions to their pension plan, compared to 30
percent of college graduates (EBRI, 1995). The 1995 RCS reports similar ﬁndings.
Forty-three percent of those with a high school degree or less who used educational
material increased the amount of contributions to their plan, versus 38 percent of those

with more education (EBRI, 1996).

36

5. ECONOMETRIC RESULTS ON THE EFFECTS OF RETIREMENT
EDUCATION ON PENSION PRESERVATION

In this section I describe the model I use in the econometric analysis and report
the results. I use a linear probability model and estimate it using OLS.27 The model is as

follows (omitting individual subscripts):

Sj = XBJ + (XjR +0)1j(R*y)+ (Dzj (R*C) + (D3j (R*I) + P + 85' , (1)

where SJ- is a dichotomous variable equal to one if the respondent saved his distribution
and equal to zero otherwise, with j=1,2, depending on the deﬁnition of saving used. The
independent variable of interest, R, is a dummy variable equal to one if the respondent
attended a retirement meeting, and zero otherwise. X contains the Observed variables that
are thought to affect the saving decision. The variables (R*y), (R*c), and (R*t) are
interaction terms of the retirement meeting variables with the binary variables age 40 and
under, college degree and female, respectively.28 The variables for age 40 and under,
college degree and female are included in X. P is a vector of proxy variables to control
for unobserved individual saving tastes.

One set of variables included in X merits explanation. I include ﬁve dummy

variables that indicate the reason the individual separated from his job, including quitting,

 

27 I also estimate the model using probit analysis, with similar qualitative results (results not reported in
tables).

28 As is noted previously, individuals with low income and small pension account balances are more likely
to spend their pension distributions, hence workers with these characteristics could also be interesting
candidates with which to interact the retirement meeting variable. However, speciﬁcations using the three
interactions in the model plus an income interaction term show no differing effects of retirement meetings
between low and high income participants. The same result holds when substituting four dummy variables
for the amount in the account in X and including four additional interaction terms. Hence I do not include
these interactions in the model.

37

being ﬁred, or leaving because of disability, retirement, or family issues.29 Individuals
could give more than one reason for separation. To simplify the interpretation of the
coefﬁcient estimates for these variables, I assign only one response per individual.30 The
reason for job separation has not been included in previous models observed in the
pension distribution literature. However, it is relevant because it coincides with the
decision on whether to save the pension and aids in describing the worker's ﬁnancial
situation at the time of the saving choice.

I ﬁrst estimate equation (1) without interaction terms or proxy variables. The
results are presented in Tables 1.5 and 1.6. Column (1) reports the simple relationship
between meetings and pension distributions without any control variables. For all types
of saving, attending a retirement meeting leads to a 5 percent greater likelihood of saving
a distribution, and a 4 percent greater probability when estimating re-investment in
retirement vehicles only. Both estimates are not statistically signiﬁcant, however.

The results from adding the controls in X are reported in column (2).31 Attending
a retirement meeting now has a small, negative coefﬁcient estimate for both deﬁnitions of
saving, however both estimates remain statistically insigniﬁcant at conventional levels.32

The other coefﬁcient estimates in column (2) are qualitatively similar to those in

 

29 1 construct the family issues variable from a variety of responses, the most common being divorce, child
care, and care of a family member.
3° Those who gave retirement as the reason for leaving were classiﬁed as such no matter what other choices
they checked, as the use of pension money is speciﬁc to retirement income. Being disabled took
precedence over merely quitting or family issues, as it is describes more speciﬁc circumstances. Being
ﬁred is chosen when any of the others were also checked (except for retiring). Being ﬁred is also very
speciﬁc and may come with subsequent economic hardship. 1 choose those who gave family issues as a
reason only when the worker also quit his job.
3 ‘ Tables 1.5 and 1.6 include year dummies for when the distribution was received, as well as a dummy
ﬂagging respondents whose spousal income was set to zero due to refusal to interview or to answer the
income question.

The year dummies are statistically signiﬁcant, with an F statistic of 13.68 using all types of saving and
22.36 for retirement saving. The larger F statistic for speciﬁcations using retirement saving may be due to
the effect different pension and tax legislation had speciﬁcally on tax-deferred accounts.

38

previous pension distribution research. Those with larger distributions have a higher
probability of saving their distributions, at a rate of about 1/2 of one percent per $10,000
in the pension account, regardless of the deﬁnition of saving used. Those with college
degrees save their distributions 14 percent more Often than non-college graduates in all
types of saving vehicles, and 19 percent more often in retirement vehicles. Whether the
individual was age 40 or under matters only for retirement saving (estimated at just under
90 percent conﬁdence), with a 15 percent lower probability of saving in retirement
vehicles than older participants. Those with household labor income of greater than
$35,000 at the time of the distribution save their distributions more often than those with
lower income; 15 percent for all types of saving and 18 percent for retirement saving
only.

One interesting result lies in the estimates on the reason for job separation.
Estimation using all types of saving shows that those who left their job because Of
disability have an 18 percent lower likelihood of saving their distributions (signiﬁcant at
the ﬁve percent level) than those who quit their job. Those who left because of family
issues are also less likely to save their distributions. This estimate is not statistically
signiﬁcant at conventional levels, although the t statistic is well above 1.0. Estimates
using retirement saving, on the other hand, are much smaller in magnitude for the
disability and family issues variables, and none of the other reasons for job separation
affect the choice to save a distribution.

Leaving a job because of disability and family issues are life situations over
which the individual has little control. They are also situations in which the immediate

need for ﬁnances - be it health care costs for disability or increased living costs for taking

 

32 1 refer to conventional levels as statistical signiﬁcance of no more than at the 10 percent level.

39

care of a family member - may override the need to save for retirement, be it in general or
retirement vehicles. To test whether all four Of the dummies belong in equation (1), I test
their joint signiﬁcance and obtain an F statistic of 1.82 with a p-value of 0.12. Although
the estimate is not statistically signiﬁcant at conventional levels, the relatively low p-
value provides moderate evidence that circumstances at the time of distribution receipt do
have an effect on the use of pension distributions.

Given that the estimates thus far Show that retirement education has no effect on
pension preservation, I must address an issue in the data that may affect this result. The
HRS does not give the year the individual attended the retirement meeting, but it does
give the year he separated from his job. As a result, some workers could have taken the
class after making the choice on how to use their distribution. If this is the case, then the
coefﬁcient estimate on the retirement meeting variable does not measure what is intended
in this analysis - the effect of retirement classes on pension preservation. The estimates
may show no effect of retirement education on pension preservation, when in fact an
effect does exist in the relevant population.

In order to investigate the possibility of the sequence of events affecting the
results, I estimate equation (1) using a sub sample of 299 individuals from my sample of
93 8, all of whom were not employed at the time of the interview. The HRS differentiates
between two types of individuals in the data: respondents that are currently working
(CW) and those who are currently not working (CNW).33 Both types state the year they
left their last job (and hence made their choice regarding the use of their pension
distribution). Approximately one-third of the sample is CNW. Of those CNW, 80

percent of those who attended a retirement meeting attended it at an employer. Since

40

these CNW do not have a current job, but attended the meeting at an employer, they must
have taken the class before they received their distribution upon separation from this
previous employer.34 This fact in mind, I re-estimate equation (1), only this time using
only those CNW who took their class from their employer. I then compare the results to
those for the full sample.

Tables 1.7 and 1.8 report the results for both deﬁnitions of saving. Estimates
using the whole sample are also included for comparison. I report estimates without
control variables added to the model in columns (1) and (3). As is reported in Table 1.5,
all respondents have a 5 percent increased probability of saving their distributions in all
types of saving if they took a class, compared to 3 percent for the CNW. When looking
at only retirement vehicles, all respondents have a 4 percent increased probability of
saving their pension if they attended a meeting, while the CNW experience a 3 percent
increased probability. Both estimates from the sub sample, like the full sample, are not
statistically signiﬁcant. Columns (2) and (4) report the results when adding controls. As
is shown for the full sample, the estimates for the CNW become negative and remain
small and statistically insigniﬁcant. While the full sample estimates are -0.04 for all
types of saving and -0.07 for retirement saving, the CNW estimates are -0.07 and -0.05,
respectively. The similarity of estimates from the two samples leads me to conclude that

the lack of information regarding the sequence of events is not driving the results in

Tables 1.5 and 1.6.

 

33 They are currently not working for various reasons such as retirement, disability, unemployment, etc.

3‘ Since I am only using the most recent distribution in the analysis, the possibility of the CNW respondent
taking an employer-sponsored class from an employer previous to the one from which he received the
distribution is not a concern.

41

Selection

The decision to attend a retirement meeting may be motivated by the same
unobserved saving tastes that affect the decision to save a pension distribution. In fact,
Bemheim and Garrett (1996) ﬁnd that the decision to use employer-sponsored retirement
education is negatively correlated with unobserved saving tastes. If this is in fact the
case, then the coefﬁcient on R in equation (1) is underestimated, and all parameter
estimates are biased and inconsistent.

To address the possibility that DC participants select into retirement meetings
based upon omitted saving characteristics, I include a set of proxy variables to control for
unobserved tastes. The vector, P, in equation (1) contains these proxies. The HRS is a
rich source from which to choose proxy variables. 1 construct variables from measures
that control for bequest intent, mortality risk, planning horizon, uncertainty about the
future and liquidity constraints, and risk aversion. These variables control for a wide
variety of personal characteristics that affect saving behavior in general. However, since
it is not possible to be certain if additional unobservables exist after I add the proxy
variables to the model, I must assume that no further omitted factors are incorporated in
the error term that are correlated with either the observables or the proxy variables in
order to obtain consistent estimates.

I base my choice of variables to include in P on two factors. First, a good proxy
variable must be correlated with innate saving characteristics of the individual, rather
than characteristics that are the result of taking a retirement class. One way to reduce the
probability of reverse causality is to use stock, rather than ﬂow, variables. Stock

measures such as current net wealth take a relatively long time to accumulate. Yearly

42

saving measures, on the other hand, are accumulated in short time periods and can change
signiﬁcantly from year to year. These measures are more likely to represent temporary
saving, which would be more likely than net wealth to reﬂect tastes for saving that result
from taking a class.

I also look at the theoretical framework put forth by HLP (1998) to decide which
proxy variables to include in P. By augmenting an extended version of a life cycle model
developed by Hurd (1989a, 1990b, 1997)35 to pension distributions speciﬁcally, HLP
identify speciﬁc saving characteristics that affect the pension preservation choice. They
then construct an empirical model testing the relationship of these characteristics to
pension preservation using data from the 1992, 1994 and 1996 waves of the HRS. The
following paragraphs include descriptions of the proxy variables I use in my estimation,
many of which are also used by HLP in their empirical model.

Bequest intent is measured using a binary variable that equals one if the
respondent said he intends to leave a large bequest, zero otherwise. I quantify
expectations about mortality in a binary variable constructed from the respondent rating
the likelihood that he will live until age 85. This variable represents the respondent’s
own estimation of his mortality risk, and equals one if he feels there is at least a 50

percent chance he will live to be age 85. The fact that this question asks for the

 

35 HLP augment a version of the life cycle that includes bequest motive (Hurd 1989a, 1990b), uncertain
health and medical expenditures for both members of a couple (Lillard and Weiss 1997), as well as using a
couple-level utility function in addition to an individual utility function (Hurd 1997). Following Yaari
(1965), they use a time-separable utility function, separable in consumption and bequests, where the only
uncertainty is the date of death. Couples and individuals have initial bequeathable wealth with a non-
negativity constraint, rights to pensions, and access to annuity streams such as Social Security. The model
does not include a choice whether to work or not, hence, it is applicable to respondents only after they
reach retirement or become disabled.

A life cycle model estimating the amount one would be willing to pay to annuitize bequeathable wealth
is also incorporated into the analysis. Together these two models lend support to an empirical model with a
trinomial choice dependent variable: annuitize the distribution, save it, or consume it. However, because

43

respondent’s own assessment of longevity (versus estimates based on standard longevity
scales) is important: how long the individual himself expects to live affects his rate of
future discount. Expecting a shorter life - and hence - a shorter retirement (if one at all)
could mean a shorter time horizon for planning and subsequent low amount of saving,
and a lower probability of saving a pension distribution.36

1 construct two sets of dummy variables to control for individual planning
horizon. The ﬁrst set includes three binary variables that represent the ﬁnancial planning
period most important to the respondent. I incorporate the responses into three different
time periods: a few months to a year, a few years to ten years, and more than ten years.
The second set includes four dummy variables that I construct from responses to a
question asking how much the individual thinks about retirement. The answers given by
the respondents are classiﬁed into thinks much, some, little, or not at all about retirement.

I control for uncertainty about the future and liquidity constraints by including a
continuous variable for total net worth, measured in 1993 dollars. Those with lower net
worth have fewer private resources for retirement and may be more uncertain of future
ﬁnancial security. In addition, lower net worth may mean more debt, resulting in a lower
likelihood of securing future credit.

To control for risk aversion, I construct a binary variable that equals one if the

individual is risk averse, and equals zero otherwise.37 Risk averse individuals may have a

greater taste for precautionary saving, thus being more likely to save their distributions to

 

the number of respondents in the data choosing to annuitize is so small, annuitization is incorporated into
the choice described as saving the distribution, and the model becomes a binomial choice model.
36 Unless, of course, the individual is planning on using the pension to cover medical costs. I attempt to
control for this in the variable indicating leaving the job because of disability.
37 . . . . . . . . .

This measure IS constructed from a question regardmg an Income gamble. The Indrvrdual Is asked
whether he would take a new job that had a ﬁﬂy percent chance of doubling his income and a ﬁfty percent
chance of cutting it in half. I label him as risk averse if he chooses not to take the job.

44

insure themselves against uncertain events, particularly those in old age.

I report the estimates on the retirement meeting variable after adding the proxy
variables in Tables 1.5 and 1.6, column (3). Table 1.5 shows that the coefﬁcient estimate
remains negative and decreases by a small amount - about 1 percentage point to -0.05 -
but continues to be statistically insigniﬁcant when looking at all types of saving. When
using the narrow deﬁnition of saving in Table 1.6, the estimate is also negative and
decreases by a slightly greater amount (2 percentage points) to -0.09. The estimate is still
statistically insigniﬁcant at conventional levels, but with at statistic of 1.48.

The increase in magnitude (in absolute value) of the estimates with the addition of
the proxy variables suggests downward bias in the initial estimates on the retirement
meeting variable. However, with such small changes in estimates that are statistically
equal to zero, it is difﬁcult to determine with any amount of certainty whether selection

bias is in fact affecting the estimates.

The Effect of Retirement Meetings on Different Subgroups

If it is true that all DC participants react similarly to retirement education (and
there is no selection bias), then the estimates in column (2) of Tables 1.5 and 1.6
represent the effect of retirement education on pension preservation for the population of
individuals who had a DC plan in a previous job. Because it is possible that these
estimates are being driven by the different effects retirement classes have on select
groups (as is suggested from the tabulations in Table 1.4), I estimate equation (1) and
include the interaction terms (R*c), (R*y) and (R*f). I estimate the model with and

without proxy variables.

45

Table 1.9 reports the effect of retirement meetings on saving a distribution by age,
education and gender. 1 report the estimates for all types of saving in column (1) and for
retirement saving only in column (3). Workers age 40 and under have a 28 percent
higher38 probability of saving their distribution in all types of savings if they attended a
meeting (estimated with 90 percent conﬁdence), but there is no effect on investment in
retirement vehicles only. These results are not surprising considering the stage in the life
cycle of these workers. Younger individuals, many who are not yet planning seriously
for retirement, attend a retirement class and realize the importance of saving for
retirement. But because of the decreased liquidity that is characteristic of tax-deferred
vehicles, they are more likely to choose to save their distributions in all types of saving
vehicles than in retirement accounts only.39 Estimates show no effect of retirement
classes on the saving of distributions for individuals over age 40.

College graduates who attend meetings have a lower probability of saving their
distributions than college graduates who do not attend a meeting, regardless of the
deﬁnition of saving used. Those with a college degree have a 16 percent smaller
likelihood of saving their distribution in all types of savings vehicles, and an 18 percent
smaller probability of re-investing the money in a retirement account. Both estimates are
statistically signiﬁcant at conventional levels. There is no effect of retirement education

on the use of a pension distribution for non-college graduates.

 

3" I calculate the effect of retirement education for those age 40 and under by evaluating the linear
combination: orj + a)” + (c)(o2j + (f) (or,- , where c and f are the sample means for college graduates and
females, respectively. I calculate the effects for the other ﬁve groups in a similar manner.

39 In order to see if the preference for general saving vehicles is driving the strong effect of the estimates
using the broad deﬁnition of saving, 1 deﬁne a new dependent variable measuring re-investment in general
saving vehicles only, which is equal to the difference between the variables for all types of saving and

- retirement saving only. The same speciﬁcation is estimated, only using this new dependent variable. The
estimate on the retirement meeting variable is 0.35, with a robust standard error of 0. 17 (not reported in
tables).

46

The negative coefﬁcient estimates for college graduates appear on the surface to
be unintuitive, especially considering existing evidence suggests that retirement
education has a positive, albeit relatively small, effect on ﬂow savings and net wealth
measures.40 A smaller effect of retirement education on saving distributions for college
graduates would follow from the evidence that college graduates possess more ﬁnancial
knowledge, hence gaining less from a retirement class. However, the fact that the effect
of retirement education is negative means that college educated individuals are actually
more likely to spend their distributions after taking a class.

Perhaps the explanation for the results for college graduates lies within the
difference between deciding to save a pension distribution and to save in general. As
was shown in the ﬁrst section of the paper, the most widely offered topics in retirement
education are description of the pension plan (88 percent) and estimation of income
needed for retirement (73 percent). Impact of pre-retirement withdrawals on retirement
income is the least covered topic, with only 39 percent of ﬁrms including this topic in
their retirement education agenda (EBRI, 1995). Because consumption of pension
distributions has become a concern to private as well as government organizations only
recently, an even smaller percentage of ﬁrms most likely covered the importance of
pension preservation in their educational material if the individual took the retirement
class in the 19805 or early 19905.

A 1996 EBRI Issue Brief shows how workers may react to knowledge gained
from retirement education. Employees whose retirement education included estimating
the amount needed for retirement had the highest average equity allocation (42 percent),

in comparison with any other topics covered. This is attributed to the surprise many

 

‘0 See Bemheim and Garrett (1996).

47

employees feel when they realize how much they actually need to save for retirement.
This realization may result in either increased pension contributions or a reallocation of
the pension portfolio to more risky assets in hopes of increasing long-run rates of return.

The EBRI report also notes that two of the lowest average equity holdings are for
retirement education that covers explanation of the company pension plan (40 percent)
and the impact of pre-retirement withdrawals (38 percent). In the report the authors
propose that upon learning that distributions are available before retirement, employees
treat their funds more as a short-run investment and decrease the aggressiveness of their
holdings (EBRI, 1996). And if employees learn through retirement education classes that
they have access to their pensions before retirement, they may also decide to spend them
upon job separation.

This may explain why an individual may choose to spend his pension after taking
a class, but it does not explain why college graduates in particular are more vulnerable.
EBRI (1996) reports that the more ﬁnancially knowledgeable the individual, the more
likely he is to have retirement savings outside of Social Security and employer pensions
set aside for retirement.“ As is noted in Section 2, evidence shows that those with
college degrees are more ﬁnancially knowledgeable than non-college graduates. Hence,
college graduates should have more private savings set aside than non-college graduates.
With calculating retirement income and a description of the pension plan being two of the
most common educational topics covered, the college graduate may ﬁnd he has
overestimated his income needs and does not need to save his pension distribution. With

additional private savings set aside, he may feel he can afford it more so than the less

48

wealthy, non-college graduate.

Table 1.9 shows that females are estimated to have a 15 percent lower probability
of saving their distribution in retirement vehicles if they attend a meeting, with no effect
on all types of saving. Males do not use their distributions differently if they attend a
meeting, regardless of the deﬁnition of saving used. A negative effect for females than
for males would be logical, given that many females, especially those of the age bracket
studied in the HRS, are often the secondary earners in the household. Females may
consider preservation of their pension of secondary importance, and learning that they
can spend the money may result in an increased incidence of non-preservation.

Colmnns (2) and (4) report the estimates after adding proxy variables to the
speciﬁcations. Adding the proxy variables decreases the estimates for college educated
individuals by 2 percentage points, to -0.18 using the broad deﬁnition of saving. When
looking at saving distributions in retirement vehicles only, the estimate also decreases,
from -0.18 to -0.21. Both estimates remain statistically signiﬁcant. Workers age 40 and
under who take a retirement class save their distribution in all types of saving 24 percent
more often, down from 28 percent when the proxies are omitted. This 4 percentage point
drop results in the estimate for younger individuals to no longer be statistically signiﬁcant
at conventional levels. The rate at which retirement education effects female saving in
retirement accounts does not change with the addition of the proxy variables - the
estimate remains at -0.15.

The changes in the estimates for the college educated and younger individuals

move in different directions with the addition of proxy variables. For college educated

 

4' It is possible that those who are more ﬁnancially knowledgeable got that way because they (for some
other reason) chose to start saving for retirement. Bemheim (1998) rejects this line of causation by

49

individuals, the increase (in absolute value) of the estimates suggests the original
estimates are underestimated, while the decrease in the estimates for workers age 40 and
under suggests the original estimates are overestimated.

In addition to reporting the effects of retirement education separately by age,
education and gender, 1 look at the effects for groups each possessing different
combinations of these three characteristics. Table 1.10, columns (1) and (3) report OLS
estimates for eight different types of DC participants, each group with speciﬁc age,
education, and gender characteristics.

Three types of individuals show statistically signiﬁcant effects on pension
preservation from attending retirement meetings. Males age 40 and under without
college degrees save their distributions in all types of savings 41 percent more often when
they take a retirement class (signiﬁcant at the 5 percent level). There is no effect using
the more narrow deﬁnition of saving. The absence of any effect of retirement meetings
for this speciﬁcation is not surprising, given the tendency for younger participants to save
their distributions in general saving vehicles. In fact, when looking at the OLS estimates
for the other three groups with individuals age 40 and under, in two of the three the
pattern is similar: positive estimates for saving in all types of saving vehicles, and
negative estimates for saving in retirement vehicles only, although some of the
coefﬁcients are not statistically signiﬁcant at conventional levels. 42

Females who have a college degree save their distributions less often when they

attend a retirement meeting, regardless of age. The Older group has a lower probability of

 

estimating a model using macroeconomic knowledge as an instrument for ﬁnancial knowledge.

‘2 Because the dependent variable for retirement saving is equal to 0 if an individual spends his pension or
saves it in a general account, the estimate using retirement saving can be negative and the individual can
still save the pension in a general savings account.

50

saving their distribution if they attended a meeting (29 percent for all types of saving, 27
percent for retirement saving), with both estimates statistically signiﬁcant at the 5 percent
level. Those age 40 and under save their pensions 31 percent less often in retirement
vehicles and 34 percent less often when I include proxies in the speciﬁcation, however
the estimate is statistically signiﬁcant only in the speciﬁcation including the proxies.
Meetings have no effect for this younger group when using the broad deﬁnition Of

saving.

Estimation with Instrumental Variables

The addition of proxy variables in equation (1) is one method of addressing
possible selection into retirement education classes. Another method is to use
instrumental variable (IV) estimation. For IV estimation to provide consistent estimates,
the instruments must be correlated with the retirement meeting variable, R, but be
uncorrelated with the error term.

The preferred instrument for the choice to attend a meeting is whether the
individual was oﬂered a retirement class by their employer“, since existing evidence
suggests that the offering of employer-sponsored classes is motivated by employer choice
rather than employee demand.44 Unfortunately this measure is not available in the HRS.
Data on occupation and industry is available for the individual's previous job, which may

be correlated with the employer's decision to offer a meeting.45 Industry is uncorrelated

 

‘3 Although individuals who took retirement meetings not sponsored by their employer are included in the
sample, 70 percent of respondents did take their meeting at an employer. Both the OLS and 1V estimates
using the variables for attending a meeting (any type) and attending an employer-sponsored meeting are
qualitatively similar.

4“ See BBS (1996).

BBS (1996) found that employers offer education more often with low participation and contribution

51

with attending a meeting (p-value of 0.7), and so is not a good IV candidate. Occupation,
however, is highly correlated with attending a meeting (p-value of 0.006, F statistic of
2.75) and so could be used to instrument for the retirement meeting variable. IV
estimation requires the addition of a reduced-form equation for attending a meeting:

R=XB+er+v, (2)
where Z contains 10 binary variables indicating occupation at a previous job to be used as
instruments for attending a meeting.46

Whether or not occupation is correlated with the error term is questionable. If
choice of occupation is motivated by unobservables that also affect the decision to save a
distribution, then the IV estimates are inconsistent. It is possible that motivating factors
in choosing a job include attitude toward risk, planning horizon, or uncertainty about the
future. For example, those who prefer risk may choose employment as an investment
banker or an entrepreneur, while those who are more risk averse may prefer a more stable
occupation. Because of this possibility, I estimate the model ﬁrst by omitting the proxy
variables and then by including the proxies in the speciﬁcation.

I report the IV estimates for the effects of attending a meeting, with and without

proxy variables, in Table 1.11. In the ﬁrst row, I report the estimates for the effects of a

 

rates in their DC plans from low-compensated workers - in other words, as a remedial measure. Blue collar
and service workers tend to have lower income than white collar workers, so they may be more likely to be
offered a meeting. One problem with this line of reasoning exists, however. Since the makeup of white
collar, blue collar and service workers at each respondent's former employer is not known, I can not be
certain that just because an individual is a blue collar worker that the workforce at that employer is also
made up of mostly blue collar and service employees.

46 The HRS records occupation using the Census Bureau Master Codes, but condenses them into 17
different groups for data reporting purposes. The 17 codes could be described in terms Of three types of
occupations: white collar (i.e., managerial and professional occupations), blue collar (i.e., mechanical,
transport, and construction occupations), and service (i.e., public and food service occupations). Because
some of the codes contained under 20 observations, 1 condensed the 17 codes into 10 dummy variables
according to magnitude and sign of correlation with the retirement meeting variable. The resulting F
statistic testing joint signiﬁcance of the occupation variables is higher when using the 10 occupation

52

retirement class without accounting for variation among different groups. For all types
of saving, the IV estimate is ten times as large in magnitude as the estimate using OLS at
-0.40, with a robust standard error approximately 7 times as large as the OLS estimate.
The increase in the estimate and standard error is also large when estimating the model
using retirement saving vehicles only. IV estimates in column (3) Show that DC
participants who attend meetings save their distributions 41 percent less often than those
who do not attend a meeting. Estimates using either deﬁnition of saving are statistically
equal to zero.

Adding the proxy variables results in the estimates becoming substantially less
negative. Attending a retirement class decreases the probability of saving a distribution
in all types of savings by 11 percent, while reducing the likelihood of saving in retirement
vehicles by 15 percent. Standard errors remain extremely large in both speciﬁcations,
and neither estimate is signiﬁcant. Regardless of the inclusion of proxy variables, the
conclusion using IV estimation is the same: there is no effect of retirement education on
pension preservation.

The IV estimates measuring the effect of retirement classes by age, education, and
gender are imprecise, whether or not the proxy variables are included in the model. The
estimate for workers age 40 and under increases from 0.28 using OLS (with no proxies)
to 0.86 using IV estimation, with the standard error increasing from 0.16 to 0.55. Similar
patterns emerge when looking at college educated individuals and females. Estimates
rise in absolute value substantially, with an increase in standard error equal to or larger

than the increase in the estimate. Estimates that are signiﬁcantly different from zero

 

variables (F=2.8 without proxies in the speciﬁcation, 2.2 with proxies) than when using all 17 (F=l .6
without proxies, 1.4 without).

53

using OLS are no longer statistically signiﬁcant using IV estimation, whether proxy
variables are included or not. In addition, groups whose estimates show no effect of
retirement education using OLS - speciﬁcally those over age 40 and males - Show large,
statistically Signiﬁcant effects using IV estimation.

The huge IV estimates and standard errors, as well as the large difference between
the OLS and IV estimates, leads me to believe that IV estimation is not necessary. In
order to lend evidence to this conclusion, I perform Hausman tests, one using the IV
estimates without proxy variables in the speciﬁcation and one including the proxy
variables, for both deﬁnitions of saving. I do the test using the speciﬁcation without the
interaction terms.

P-values ranging from 0.35 to 0.72 indicate that I am unable to reject the null
hypothesis that the retirement variable is uncorrelated with the unobservables.47 With
these results, and the large differences between the OLS and IV estimates, the preferred

estimates are those using OLS.

6. CONCLUSION

This paper is the ﬁrst analysis to evaluate an additional means aside from tax
legislation in affecting the consumption of pension distributions. 1 ﬁnd that retirement
education substantially increases the probability of saving a distribution for participants
age 40 and under, while decreasing the probability of saving a distribution for college
graduates and females. These important differentials are concealed by estimating the

effect of retirement education on participants generally.

 

‘7 P-values are as follows: all types of saving, no proxies, 0.34; with proxies, 0.71; retirement saving, no
proxies, 0.37; with proxies, 0.72.

54

These results do not necessarily contradict the ﬁndings of Bemheim and Garrett
(1996), who estimate that retirement education has a positive effect on saving behavior. I
have proposed that one result of attending a retirement meeting may be to merely inform
the participant that he has the option to spend his distribution upon job separation. This
information in turn could increase the likelihood that certain groups of participants will
spend their distributions. Given the relatively small percentage of retirement education
that covers the topic of pension distributions in depth, it is likely that this information was
not accompanied by further explanations of the effects on retirement security of spending
distributions prematurely. More detailed data is needed to examine both the initial
knowledge of participants regarding pensions distributions, as well as the extent to which
retirement education programs are informing pension participants of the consequences of

spending their distributions.

55

REFERENCES

Andrews, Emily S., “Retirement Savings and Lump Sum Distributions,” Beneﬁts
Quarterly, 2: 1991, pp. 47-58.

Atkins, G. L., “Spend it or Save it? Pension Lump Sum Distributions and Tax Reform,”
Washington, DC: Employee Beneﬁt Research Institute, 1986.

Bassett, William F., Michael J. Fleming, and Anthony P. Rodrigues, “How Workers Use
401(k) Plans: The Participation, Contribution, and Withdrawal Decisions,” National Tax
Journal, 51(2): June 1998, pp. 263-89.

 

Bayer, Patrick, B. Douglas Bemheim, and John Karl Scholz, “The Effects of Financial
Education in the Workplace: Evidence from a Survey of Employers,” National Bureau of
Economic Research Working Paper no. 5655, July 1996.

Beller, Daniel J ., “Introduction and Summary of Major Findings,” in Trends in Pensions,
1992, edited by John A. Turner and Daniel J. Beller, 285-317. Washington, DC:
Government Printing Ofﬁce, 1992.

 

Bemheim, B. Douglas, "Comments and Discussion," Brookings Papers on Economic
Activity, (1): 1994b, pp. 152-166.

 

Bemheim, B. Douglas, "Financial Illiteracy, Education, and Retirement Saving," in
Living with Deﬁned Contribution Pensions: Remaking Responsibility for Retirement,
edited by Olivia S. Mitchell and Sylvester J. Schieber, 38-68. Philadelphia,
Pennsylvania: Pension Research Council, the Wharton School of the University of
Pennsylvania, and the University of Pennsylvania Press, 1998.

 

Bemheim, B. Douglas, “Personal Saving, Information, and Economic Literacy: New
Directions for Public Policy,” in Tax Policy for Economic Growth in the 19905,
Washington, DC: American Council for Capital Formation, 1994a, pp. 53-78.

 

Bemheim, B. Douglas, and Daniel M. Garrett, “The Determinants and Consequences of
Financial Education in the Workplace: Evidence from a Survey of Households,” National
Bureau of Economic Research Working Paper no. 5667, July 1996.

Bemheim, B. Douglas, and John Karl Scholz, "Private Saving and Public Policy,"
National Bureau of Economic Research Working Paper no. 4215, 1992.

Burman, Leonard E., Coe, Norma B., and William G. Gale, “Lump Sum Distributions

from Pension Plans: Recent Evidence and Issues for Policy and Research,” National Tax
Association Spring Symposium Meetings, May 1999.

56

REFERENCES (continued)

Chang, Angela E., “Tax Policy, Lump-Sum Pension Distributions, and Household
Saving,” National Tax Journal 49(2): June 1996, pp. 235-52.

 

Dickemper, Jennifer, and Paul Yakoboski, “Increased Saving but Little Planning: Results
of the 1997 Retirement Conﬁdence Survey,” Employee Beneﬁt Research Institute Issue
Brief no. 191, November 1997.

Employee Beneﬁt Research Institute, “Can We Save Enough to Retire? Participant
Education in Deﬁned Contribution Plans,” Employee Beneﬁt Research Institute Issue
Brief no. 160, April 1995.

Employee Beneﬁt Research Institute, “Participant Education: Actions and Outcomes,”
Employee Beneﬁt Research Institute Issue Brief no. 169, January 1996.

Engelhardt, Gary V., “Pre-retirement Lump Sum Distributions and Retirement Income
Security: Evidence from the Health and Retirement Survey,” Dartmouth College, Mimeo,
1999.

Fernandez, Phyllis A., “Preretirement Lump-Sum Distributions,” in Trends in Pensions
1992, edited by John A. Turner and Daniel J. Beller, 285-317. Washington, DC:
Government Printing Ofﬁce, 1992.

 

F ronstin, Paul et al., “Lump-Sum Distributions and Rollovers,” Employee Beneﬁt
Research Institute Databook on Employee Beneﬁts, eds. Deborah Holmes et al.,
Washington, DC: Employee Beneﬁt Research Institute, 1997.

 

 

Hurd, Michael D., "The Joint Retirement Decision of Husbands and Wives," in Issues in
the Economics of Aging, edited by David Wise, 231-254. Chicago, Illinois: University
of Chicago Press, 1990b.

 

 

Hurd, Michael D., "Mortality Risks and Bequests," Econometrica, 1989a, pp. 779-813.

 

Hurd, Michael D., "Mortality Risk and Consumption by Couples," RAND Working
Paper 206l-NIA, 1999.

Hurd, Michael D., Lee Lillard, and Constantijn Panis, "An Analysis of the Choice to
Cash Out Pension Rights at Job Change or Retirement," RAND Working Paper 1979-
DOL, October 1998.

Lillard, Lee, and Yoran Weiss, "Uncertain Health and Survival: Effects on End of Life
Consumption," Special Issue: Dynamic Structural Models, Journal of Business and
Economic Statistics, 15(2): 1997, pp. 254-268.

 

 

57

REFERENCES (continued)

Pensions and Investments. "Employees Getting More: Investment Education, Planning
Help on the Increase," January 23, 1995, p. 74.

 

Piacentini, Joseph S., "An Analysis of Pension Participation at Current and Prior Jobs,
Receipt and Use of Preretirement Lump-Sum Distributions, and Tenure at Current Job,"
Washington DC: Employee Beneﬁt Research Institute, 1990.

Poterba, James M., Steven F. Venti, and David A. Wise, “Lump-Sum Distributions From
Retirement Saving Plans: Receipt and Utilization,” in Inquiries in the Economics of
Aging, edited by David A. Wise, 85-105. Washington, DC: National Bureau of
Economic Research, 1998.

 

Scott, Jason S., and John B. Shoven, “Lump-Sum Distributions: Fulﬁlling the Portability

Promise or Eroding Retirement Security?” Employee Beneﬁt Research Institute Issue
Brief no. 178, October 1996.

Smith, James P, "Marriage, Assets, and Savings," RAND Working Paper 105 5-NIA,
March 1995.

United States Congress, Joint Committee on Taxation, “General Explanation of the Tax
Reform Act of 1986,” Washington, DC: Government Printing Ofﬁce, 1987.

Yaari, M., "Uncertain Lifetime, Life Insurance, and the Theory of the Consumer,"
Review of Economic Studies, 32: 1965, pp. 137-150.

 

Yakoboski, Paul, “Large Plan Lump-Sums: Rollovers and Cashouts,” Employee Beneﬁt
Research Institute Issue Brief no. 188, August 1997.

Yakoboski, Paul, “Preservation and Annuitization in Retirement Saving Accounts,”
Employee Beneﬁt Research Institute Notes 19(1), January 1998, pp. 1-3.

58

TABLE 1.1
Summary Statistics

 

 

M Standard Deviation
Attended a retirement class 19 % 39 %
Retirement class was sponsored by employer 72 45
Female 49 50
Black/other 22 41
College degree is highest degree obtained 25 43
Married 75 44
Age 40 and under 15 36
Total household labor income >$35,000 59 49
Left job - family problems 12 32
Left job - unemployed 29 45
Left job - disabled 13 34
Left job - retired 17 38
Left job - quit 28 45
At least a 50 percent chance of living until age 85 35 48
Expects to leave large inheritance 47 50
Thinks much about retirement 29 46
Thinks some about retirement 24 43
Thinks little about retirement 17 37
Does not think at all about retirement 30 46
Short saving horizon (few months-year) 25 44
Medium saving horizon (few years-5 to 10 years) 64 48
Long saving time horizon (over 10 years) 11 31
Risk averse 71 43
Age at time of distribution 49 8.5
Family labor income at time of distribution $50,247 $53,047
Amount of distribution in 1993 dollars $33,797 $81,147
Number of dependents in household in 1993 1.5 1.1
Total net worth in 1993 $295,039 $687,818

 

Total number of Observations equals 640. Individuals who did not answer the
retirement meeting question or whose use of pension balances could not be classiﬁed in
the savings dependent variable are not included in the tabulations. Responses that could
not be classiﬁed include other uses (1 percent of total sample), lost pension (4 percent),
or use of distribution not available (12 percent).

Tabulations for retirement class sponsored by employer are conditional on taking a
class, and are calculated from 121 observations. Variables for age, married, labor
income, and amount of distribution refer to the time period in which the respondent
separated from his job. Labor income and amount of distribution were initially recorded
in nominal dollars for the year of separation, and are expressed here in 1993 dollars.

59

TABLE 1.2
Uses of Pension Distribution Upon Job Separation
Expressed in Percentages

 

 

 

Use of Pension Percent of Full Sample
Saved in all types of savings vehicles 69 %

Saved in a retirement vehicle 60

Spent 31

Saved in non-retirement account

Paid down debt 3

Left in old employers account 30

Transferred to new employer 3

Rolled into an annuity 5

Rolled into/invested in an IRA 22

 

Total number of observations equals 640. Rows 1 and 2 report the percentages of
respondents who saved their distributions, using each deﬁnition of saving. Rows 3
through 9 report the percentage of respondents who used their distribution in each
speciﬁc way, and adds up to 100 percent.

60

TABLE 1.3
Percentage of Individuals who Attended a Retirement Meeting
and Saved their Pension Distributions, by Group

 

 

 

 

Group Attended Saved in All Saved in Retirement
a Meeting Types of Vehicles Vehicles Only
(1) (2) (3)
Age received
distribution
<=age40 14% 58% 36%
> age 40 20 71 64
At least a 30 80 74
college degree
No college 15 66 56
degree
Female 18 63 51
Male 20 75 69

 

Total number of observations equals 640. Column (1) reports the percentage
of the sample that attended a retirement class. Columns (2) and (3) report the
percentage of individuals that saved their pension distributions, depending on
the deﬁnition of saving used.

61

_o>o_ 88:2: 3 :m :58me7. ...
65. 83:2: m :m Emomﬁwmm .1.

dogmas: no Co 3:: 3 Own :2: 83:28 8 on: 26:: :2: $2: 3380: we? 8:2:me 0:: :8» 0:: :8 Se: mammmﬁ: 3:88:

owe 9 3m :0: 80: ow :26 use :35: 98 9: Own 08:: mo 8:85: BE. .43: S: 8:: 2:355: :o .585: :03 8:23 maﬁa:

:o 5358 S: H: @532: £93380 32: o: ooze—m: wEEOMmSg ago—QED 58:8 5:? >308 wﬁ>m2 wows—2: .waram 2:25:33:
Amv 55:00 i=8 E588 mwﬁgm Eocow m E magma. :0 Bot :32: megs: 8:22: .mEBNm :0 89C. =< A C 55:00

62

 

 

 

 

am : 3.- mm. 2.. 3.- mm. mm. 3&2: owe—BU
8&6:
:3 so. mm. a. B. S. K. as:8 02
wmm *2. S. R. *3. mm. mm. 232
N: m 8.. mm. 3. no.. mo. 6. 2::th
3% 8. we. no. 5. 2.. mm. ov own :o>O
:25:
8 o 3. em. .8. 3. we. as... 3. a?
owe mo. om . we. mo. we. mm. 3:22:88: =<
5 88 SN: 38 G: 3: a: :85
$20 a 38:»: mam—U < $20 a 33?: 330 <
2233030 oocobba 8 Z ED 89:82 cacaoba :02 ED 39:32
1 mo :38: Z wE>mm 2:25:33: wE>mm mo 39$. =<
22:88.: :: temmoaxm

maggot»: weave—z 83:25.3: .3 32:2:me commaom :0 83835
:2 waged.

TABLE 1.5
Effects of Retirement Education on Pension Preservation

Saving deﬁned as ALL TYPES OF SAVING

 

 

Dependent Variable: (1) (2) (3) (4) (5)
All Types of Saving
Attended a retirement 0.05 -0.04 -0.05 0.07 0.06
class (0.05) (0.06) (0.06) (0.10) (0.10)
Female -0.05 -0.07 -0.03 -0.04
(0.05) (0.05) (0.05) (0.05)
Black/other -0.10 -0.12* -0.10 -0.12*
(0.06) (0.07) (0.06) (0.07)
College degree 0.14” 0.14** 0.20" 0.20”
(0.05) (0.05) (0.06) (0.06)
Married -0.03 -0.04 -0.04 -0.05
(0.06) (0.06) (0.06) (0.06)
Amount of distribution 0.005” 0.005M 0.005“ 0.005“
(0.002) (0.002) (0.002) (0.002)
Age 40 and under 003 -0.04 -0.09 -0.10
(0.10) (0.10) (0.10) (0.11)
Household annual labor 0.15" 0.12M 0.16“ 0.13**
income is > $35,000 (0.05) (0.06) (0.05) (0.06)
Number of dependents -0.01 -0.01 -0.02 -0.02
(0.02) (0.02) (0.02) (0.02)
Left job - ﬁred/laid off 002 -0.02 -0.02 -0.02
(0.06) (0.06) (0.06) (0.06)
Leftjob - disabled -0.18** -0.14* -0.18** -0.14*
(0.08) (0.08) (0.08) (0.08)
Left job - retired -0.04 -0.06 -0.03 -0.05
(0.07) (0.08) (0.08) (0.08)
Leftjob - family issues 013 -0.14 -0.12 -0.13
(0.09) (0.09) (0.08) (0.09)
Retirement meeting -0.20* -0.22*
*college degree (0.11) (0.12)
Retirement meeting*age 0.35* * 0.31 *
40 and under (0.18) (0.18)
Retirement meeting -0. 16 -0.15
*female (0.11) (0.11)
Total net worth 0.00007 0.00006
(0.0003) (0.0003)
Plan to leave large 0.03 0.03
inheritance (0.05) (0.05)
At least 50 percent 009* 008*
chance of living until (0.05) (0.05)

age 85

 

63

 

TABLE 1.5 (continued)

Thinks much about 0.03 0.02
retirement (0.06) (0.06)
Thinks some about 0.03 0.02
retirement (0.07) (0.06)
Thinks little about -0.01 -0.03
retirement (0.08) (0.08)
Saving horizon 1-10 0.10* 0.09
years (0.06) (0.06)
Saving horizon over 10 0.01 0.02
years (0.08) (0.09)
Risk averse 0.01 0.005
(0.05) (0.05)
Constant 0.68** 1.14** 1.13M 1.23** 1.14“
(0.02) (0.16) (0.18) (0.12) (0.16)
R2 0.002 0.18 0.20 0.20 0.22
Observations 640 461 449 461 449

 

Dependent variable is whether the respondent saved his pension upon changing jobs.
Saving is deﬁned as putting the money in a retirement vehicle (old employer's account,
IRA, annuity, new employer's account), general saving vehicle, or paying off debt.
Variables for net worth and amount of distribution are in ten-thousand 1993 dollars.
Base cases are quit job, does not think at all about retirement, and saving horizon less
than a year. Year dummy variables for the year of the distribution are included in
columns (2)-(5).

Columns: (1) mean; (2) add control variables; (3) add proxy variables; (4) add
interaction terms to (2); (5) add proxy variables to (4).

Robust standard errors are in parentheses.

** signiﬁcant at 5 percent level
* signiﬁcant at 10 percent level

64

TABLE 1.6

Effects of Retirement Education on Pension Preservation

Saving Deﬁned as RETIREMENT SAVING

 

 

Dependent Variable: (1) (2) (3) (4) (5)
Retirement Saving
Attended a retirement 0.04 -0.07 -0.09 0.09 0.06
class (0.05) (0.05) (0.06) (0.09) (0.09)
Female -0.09** -0.11** -0.06 -0.08
(0.05) (0.05) (0.05) (0.05)
Black/other -0.13** —0.15** -0.13** 016”
(0.06) (0.07) (0.06) (0.07)
College degree 019’” 0.18** 0.24" 0.23"
(0.05) (0.06) (0.06) (0.06)
Married -0.01 -0.01 -0.02 ~0.02
(0.06) (0.06) (0.06) (0.06)
Amount of distribution 0.006” 0.005M 0.005“ 0.005**
(0.002) (0.002) (0.002) (0.002)
Age 40 and under -0.15 -0.16* -0.16 -0.17
(0.10) (0.10) (0.10) (0.10)
Household annual 0.l8** 0.15M 0.18” 0.16“
labor income > $35,000 0.05) (0.05) (0.05) (0.05)
Number of dependents -0.03 -0.03 -0.03 -0.04*
(0.02) (0.02) (0.02) (0.02)
Left job - ﬁred/laid off 0.002 0.01 0.001 0.009
(0.06) (0.06) (0.06) (0.06 )
Left job - disabled -0.04 0.01 -0.03 0.02
(0.08) (0.08) (0.08) (0.08)
Leﬁ job - retired 0.009 0.02 0.01 0.02
(0.07) (0.08) (0.08) (0.08)
Leﬁ job - family issues -0.08 -0.07 -0.08 -0.07
(0.08) (0.08) (0.08) (0.08)
Retirement meeting -0. 17* -O. 19*
*college degree (0.11) (0.11)
Retirement meeting*age -0.03 -0.06
40 and under (0.20) (0.21)
Retirement meeting -0. 19* -0.15
*female (0.11) (0.11)
Total net worth -0.00006 -0.00008
(0.0003) (0.0003)
Plan to leave large 0.05 0.05
inheritance (0.05) (0.05)
At least 50% chance of 0.07 0.07
living until age 85 (0.05) (0.05)

 

65

 

TABLE 1.6 (continued)

(1) (2) (3) (4)

Thinks much about 0.12”“
retirement (0.06 )
Thinks some about 0.10
retirement (0.06)
Thinks little about 0.02
retirement (0.07)
Saving horizon 1-10 0.04
years (0.06)
Saving horizon over 10 0.02
years (0.08)
Risk averse 0.09”

(0.05)
Constant 0.59** 1.23** 1.06** 1.36”

(0.02) (0.16) (0.16) (0.12)

R2 0.001 0.27 0.30 0.28
Observations 640 461 449 461

(5)
012*
(0.06)
0.10
(0.06)
0.01
(0.07)
0.03
(0.06)
0.02
(0.08)
0.08*
(0.05)
1.18**
(0.15)
0.31
449

 

Dependent variable is whether the respondent saved his pension upon changing jobs.
Saving is deﬁned as putting the money in a retirement vehicle only. See Table 1.5 for

additional notes.
Robust standard errors are in parentheses.

* * signiﬁcant at 5 percent level
* signiﬁcant at 10 percent level

66

TABLE 1.7
Effects of Retirement Education on Pension Preservation
Full Sample Versus Currently Not Working (CNW)
Saving deﬁned as ALL TYPES OF SAVING

 

 

67

Full Sample Currently Not Working
(CNW)

Dependent Variable: ( 1) (2) (3) (4)

All Types of Saving Mean Add Controls Mean Add Controls
Attended a retirement 0.05 -0.04 0.03 -0.07
class (0.05) (0.06) (0.09) (0.10)
Female -0.05 -0.13

(0.05) (0.08)
Black/other -0. 1 0 -0.10
(0.06) (0.10)
College degree 014“ 0.20*
(0.05) (0.11)
Married -0.03 -0.007
(0.06) (0.09)
Amount of distribution 0.005" 0.009"
(0.002) (0.004)
Age 40 and under -0.03 0.38
(0.10) (0.48)
Household annual labor 0.15** 0.16*
income > $35,000 (0.05) (0.09)
Number of dependents -0.01 -0.02
(0.02) (0.04)
Left job - ﬁred/laid off -0.02 0.09
(0.06) (0.17)
Left job - disabled -0.l8** 0.13
(0.08) (0.18)
Leﬁ job - retired -0.04 0.20
(0.07) (0.17)
Left job - family issues -O.l3 -0.22
(0.09) (0.18)
Constant 0.68** 1.14** 0.58** -0.004
(0.02) (0.16) (0.03) (0.19)

 

TABLE 1.7 (continued)

(1) (2) (3) (4)
R2 0.002 0.18 0.0005 0.31
Observations 640 461 251 165

 

Dependent variable is whether the respondent saved his pension upon changing jobs.
Saving is deﬁned as putting the money in a retirement vehicle (old employer's account,
IRA, annuity, new employer's account), general savings vehicle, or paying off debt.
Amount of distribution is in ten-thousand 1993 dollars. Labor income is in 1993 dollars.
Base case for job separation dummies is quit job. Year dummy variables for the year of
the distribution are included in columns (2) and (4).

Robust standard errors are in parentheses.

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

68

TABLE 1.8

Effects of Retirement Education on Pension Preservation
Full Sample Versus Currently Not Working (CNW)

Saving deﬁned as RETIREMENT SAVING

 

 

69

Full Sample Currently Not Working
(CNW)
Dependent Variable: (1) (2) (3) (4)
Retirement Saving Mean Add Controls Mean Add Controls
Attended a retirement 0.04 -0.07 0.03 -0.05
class (0.05) (0.05) (0.09) (0.10)
Female —0.09** -0.13
(0.05) (0.08)
Black/other -0.13 * * -0.07
(0.06) (0.10)
College degree 019’” 0.21 **
(0.05) (0.11)
Married 00] 0.03
(0.06) (0.10)
Amount of distribution 0.006** 0.009”
(0.002) (0.004)
Age 40 and under - 0.15 0.38
(0.10) (0.47)
Household annual labor 0.18** 0.17*
income > $35,000 (0.05) (0.09)
Number of dependents -0.03 -0.02
(0.02) (0.04)
Left job - ﬁred/laid off 0.002 0.10
(0.06) (0.17)
Left job - disabled -0.04 0.16
(0.08) (0.18)
Leﬁ job - retired 0.009 0.20
(0.07) (0.17)
Left job - family issues -0.08 -0.20
(0.08) (0.18)
Constant 0.59** 1.23M 0.57** -0.03
(0.02) (0.16) (0.03) (0.19)

 

TABLE 1.8 (continued)

(I) (2) (3) (4)
R2 0.001 0.27 0.0002 0.32
Observations 640 461 251 165

 

Dependent variable is whether the respondent saved his pension upon changing jobs.
Saving is deﬁned as putting the money in a retirement vehicle only. See Table 1.7 for
additional notes.

Robust standard errors are in parentheses.

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

70

TABLE 1.9

by Age, Education, and Gender

 

 

OLS Estimates of the Effect of Retirement Education on Pension Preservation

 

 

 

All Types of Saving Retirement Saving
Group OLS OLS OLS OLS
With Proxies With Proxies
(1) (2) (3) (4)
All -0.04 -0.05 -0.07 -0.09
respondents (0.06) (0.06) (0.05) (0.06)
Age 40 0.28* 0.24 -0.08 -0.13
and under (0.16) (0.16) (0.19) (0.20)
Over age 40 -0.06 -0.07 -0.05 -0.06
(0.06) (0.06) (0.06) (0.06)
College -0.16* -0.18* -0.18** -0.21**
degree (0.09) (0.09) (0.08) (0.09)
No college 0.04 0.04 -0.008 -0.02
degree (0.07) (0.07) (0.09) (0.07)
Female -0.09 -0.09 -0.15** -0.15**
(0.08) (0.08) (0.08) (0.08)
Male 0.07 0.05 0.04 0.0006
(0.08) (0.08) (0.08) (0.09)

 

Number of observations: 461

Robust standard errors are in parentheses

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

71

TABLE 1.10
OLS Estimates of the Effect of Retirement Education on Pension Preservation
for those with Combinations of Age, Education, and Gender Characteristics

 

 

 

 

 

 

All Types of Saving Retirement Saving
Group OLS OLS OLS OLS
With Proxies With Proxies
(1) (2) (3) (4)

No college, 0.07 0.06 0.09 0.06
>40, male (0.10) (0.10) (0.09) (0.09)
No college, -0.09 -0.08 -0.10 -0.09
>40, female (0.09) (0.10) (0.09) (0.09)
No college, 0.41 * 037* 0.06 -0.007
<=40, male (0.18) (0.18) (0.22) (0.23)
No college, 0.25 0.22 -0.14 -0.16
<=40, (0.17) (0.17) (0.19) (0.20)
female

College, -0. l 3 -O.15 -0.08 -0. 12
>40, male (0.09) (0.10) (0.09) (0.10)
College, -0.29* -0.30* -0.27* -0.27*
>40, female (0.11) (0.12) (0.11) (0.11)
College, 0.05 0.01 -0.31 -0.34*
<=40, (0.20) (0.20) (0.19) (0.20)
female

College, 0.21 0.16 -0.11 -0.19
<=40, male (0.20) (0.20) (0.21) (0.23)

 

Number of observations: 461

Robust standard errors are in parentheses
** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

72

TABLE 1.11

IV Estimates of the Effects of Retirement Education on Pension Preservation
by Age, Education, and Gender

 

 

 

 

 

All Types of Saving Retirement Saving
Groue m m m 11
With Proxies With Proxies
(1) (2) (3) (4)
All -0.40 -0.1 l -0.41 -0.15
respondents (0.43) (0.41) (0.42) (0.40)
Age 40 0.86 0.87 0.35 0.36
and under (0.55) (0.55) (0.53) (0.53)
Over age 40 -0.65* -0.52 -0.57* -0.28
(0.33) (0.42) (0.32) (0.41)
College -0.62 -0.50 -0.62 -0.45
degree (0.54) (0.53) (0.48) (0.47)
No college -0.33 -0.23 -0.36 -0.28
degree (0.34) (0.34) (0.34) (0.33)
Female -0.07 0.02 -O.21 -0.02
(0.41) (0.41) (0.40) (0.39)
Male -0.75** -0.62* -0.65* -0.62*
(0.37) (0.37) (0.36) (0.35)

 

Number of observations: 461
Robust standard errors are in parentheses

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

73

Chapter 2: INVESTMENT CHOICE IN PRIVATE PENSIONS:
THE EFFECTS OF RETIREMENT EDUCATION
ON ASSET ALLOCATION

INTRODUCTION

As 401(k) pension plans become more common in the workplace of many
Americans, so does the worker's choice on how to invest his pension assets. In 1993, 91
percent of participants in thrift and saving plans in medium and large ﬁrms could choose
how to invest their DC plan assets (Department of Labor (DOL), 1993). Plans that allow
participant choice are often called self-directed plans, because the worker is given a range
of options from which he can choose how to allocate the assets. The proportion of self-
directed plans is likely to continue to rise, as 62 percent of those who contribute their
own salaries to plans report they prefer to make their own investment decisions
(Employee Beneﬁt Research Institute (EBRI), January 1996).

Most previous research on self-directed plans has focused on the demographic
and ﬁnancial characteristics of participants and how these relate to their portfolio
decisions (Poterba and Wise, 1998; Goodfellow and Schieber, 1997). Identifying the
asset allocation choice for different groups of individuals is certainly important when
evaluating the investment decisions of workers, given that the choice of how to allocate
assets and the level of saving both affect the balance available at retirement. Returns on a
portfolio with mostly low-risk assets will differ considerably from a diversiﬁed portfolio
or one with predominately risky assets, resulting in different levels of pension income for

a given amount of contribution.

74

It is essentially this difference in returns - and ultimate retirement payouts - that
has recently concerned plan sponsors and policy makers. The Institute of Management
and Administration (1995) reports that in one recent survey two-thirds of employers were
concerned about the asset allocation decisions their employees were making. Although
some employers feel that workers are choosing overly risky portfolios, most are
concerned that participants are investing too conservatively. By comparing aggregate
deﬁned beneﬁt (DB) portfolio allocations with those of deﬁned contribution (DC) plans
(predominately 401(k) plans), DB plan portfolios hold substantially more of their assets
in equities than do 401(k) plans (Wyatt Company, 1994; Smith, 1993). Additionally,
when asked how they would choose pension investments knowing that their pension
beneﬁts would go down with investment losses and up with investment gains, 69 percent
of workers said they would prefer low-risk investments, even if the returns were also
lower (EBRI, 1993).

Whether participants are making the portfolio choices that are best for their
retirement security is difﬁcult to determine. Observable characteristics such as age and
income contribute to this decision and are observed to vary with asset allocation
(Goodfellow and Schieber, 1997). However, economic theory suggests that an
individual's degree of risk aversion also affects his portfolio choice, and this
characteristic is often difﬁcult to measure. Thus empirical analysis of the determinants of
asset allocation has not been able to determine if — and to what extent — individual risk-
taking behavior affects a worker's portfolio choice.

Choosing the optimal portfolio also requires substantial ﬁnancial knowledge.

How average returns differ between asset types, the trade off between risk and return, and

75

how returns are affected by inﬂation are just a few of the concepts involved in
understanding how to allocate assets optimally. Evidence suggests that many individuals
do not have the ﬁnancial knowledge to make sophisticated choices. The 1995 Retirement
Conﬁdence Survey (RCS) reports that many individuals do not know how historical rates
of return differ for various classes of investments. EBRI reports that only half of those
surveyed know that over the past 20 years government bonds have yielded a lower rate of
return than the US. stock market (EBRI, January 1996). Other examples of ﬁnancially
unsophisticated behavior include ignoring changes in expected future earnings when
choosing saving targets (Bemheim, 1994a) and the number of individuals that lose
money in ﬁnancial scams (Alliance Against Fraud in Telemarketing, 1992).

If individuals do not have the ﬁnancial knowledge needed to determine an optimal
portfolio allocation, education may increase their knowledge, leading to more efﬁcient
asset allocation decisions. An EBRI/Greenwald survey shows that of those reading
written material or attending retirement classes in the workplace, 44 percent said it led
them to change the allocation of their money within their 401(k) account (EBRI, 1995).
Employees also appear interested in ﬁnancial education: of those who participated in
some form of employer-sponsored retirement education program, 92 percent reported
reading the material (EBRI, 1995).

In this paper I examine the effects of retirement education - speciﬁcally retirement
classes - on asset allocation within a self-directed 401(k) pension plan. This is the ﬁrst
analysis to examine whether retirement education affects DC asset allocation decisions.

I include both employer-sponsored meetings and those offered by public and private

institutions in my analysis. Because individual characteristics may affect the way in

76

which participants respond to retirement education, I also analyze the effects of
retirement education for different types of individuals.

I use a sample of 1,236 workers- with 1,322 self—directed plans (each worker can
have more than one self-directed plan) - from the 1992 Health and Retirement Study
(HRS). The HRS is unique in that it also contains questions speciﬁcally designed to pick
up unobservable characteristics related to saving and risk-taking behavior often missing
from other data sets. The inclusion of variables such as degree of risk aversion and
expected longevity add additional insight into what inﬂuences individual portfolio
choices in private pensions.

The next section provides background information on self-directed plans,
individual portfolio choice, and retirement education. In Section 3, I summarize the
literature on asset allocation within pension plans and retirement education. I provide an
overview of the data in Section 4. Section 5 provides econometric evidence of the effect
of retirement education on 401(k) asset allocation. I use OLS regressions to estimate the
effects of retirement education on the percentage of equity in the pension portfolio, and
compare these values with ordered logit estimates. 1 then estimate these effects for
different types of individuals. I also address the possibility of selection bias in the model

and examine methods to address this issue. I offer concluding remarks in Section 6.

2. BACKGROUND
Pension plans can be categorized into two types: deﬁned beneﬁt and deﬁned
contribution plans. DB plans typically determine the amount of the beneﬁt using a

formula that includes years of service and/or salary. DB plans are established and funded

77

by the employer for the employee, and asset plan managers typically manage one account
for all employees.

The design of DC plans is quite different. Salary-reduction plans such as 401(k)
plans, Employee Stock Option Plans (ESOPS), and Simpliﬁed Employee Pension Plans
(SEPs) are all types of DC plans. Each employee has his own account into which he, and
often his employer, makes contributions. Currently, most plan sponsors allow DC
participants to make their own portfolio choices. According to a Bureau of Labor
Statistics survey, 91 percent of participants in thriﬁ and savings plans in medium and
large ﬁrms could choose how their contribution was to be invested, and 62 percent could
direct the investment of the employer contribution (DOL, 1993).

Participants choose their mix of assets from a range of investments as determined
by the plan sponsor. According to a 1993 Hewitt Associates survey of employers who
offered 401 (k) plans, employers offer on average 4.5 investment options for employee
contributions. The percent of employers offering only 1 to 2 options is quite small, at 7
percent, while 22 percent offer 6 or more options. Employees usually have the choice of
at least one equity fund (either growth and income equity, growth equity, or equity
index), one balanced fund (preset allocations of bonds and stocks) and one ﬁxed-income
fund (assets with low risk, such as Guaranteed Investment Contracts (GICs) 48 or
government treasury bills) (Yakoboski, 1995). There is evidence that the number and
characteristics of the options available affects the asset allocation mix (Galer et al., 1999).
Goodfellow and Schieber (1997) examine the allocations of participants by speciﬁc DC

plan. After taking into consideration such variables as the average age of the participant

 

48 GICs are contracts offered by insurance companies and guarantee a certain rate of return and payment of
principal and interest at a speciﬁc time.

78

and the age of the plan, signiﬁcant similarities remain in investment behavior between
those in the same plan.

If the employee is able to choose how the employer's contribution is invested, the
options are often more limited. The 1993 Hewitt data show that on average only 3.5
options are offered for the employer contribution, with 29 percent of employers offering
only 1 choice (Yakoboski, 1995). Often this choice is limited to employer stock,
particularly if the plan is organized as a proﬁt sharing or stock bonus plan (e.g., ESOP).
Although an extensive data base prepared by EBRI and the Investment Company Institute
(ICI) shows that less than one percent of plans mandate the employer contribution be
made in employer stock, most of these plans are large, resulting in 15 percent of
participant accounts and 25 percent of assets with the employer contribution
predetermined (Galer et al., 1999).

The introduction of the self-direction of 401 (k) plans lagged behind the
introduction of the 401(k) plans themselves, at least initially. Allowing employees to
choose their own investments can have legal implications, as plan sponsors can be held
liable for losses incurred by participants, even if these losses are the result of the
participant’s own decisions. In 1992 the DOL ﬁnalized section 404(c) regulations of
ERISA,49 which outlines characteristics of self-directed plans that may limit the
possibility of liability for plan sponsors.

Section 404(c) compliance is not mandatory. However, establishing a compliant

plan should provide a defense against charges that the participant’s ability to choose

 

‘9 The effective date of the ﬁnal regulations was January 1, 1994.

79

resulted in a ﬁduciary breach.50 Generally, compliance requires that participants be
allowed to “exercise independent control” over their accounts and be given at least three
investment alternatives with different risk/retum characteristics (employer stock cannot
be one of the three). Sponsors are allowed to restrict how often employees can change
their portfolios, however, they must offer enough ﬂexibility for employees to respond to
expected market changes in their investments.51 Sponsors are also required to provide
education about the plan and the investments available. Sponsors may include
information on risk versus return and other investment topics, but they must inform the
participants that they are responsible for the outcome of the investments they choose and

of any additional transaction fees that could be incurred for certain assets types.

Why Be Concerned About Individual Investment Choice in Pension Plans?

As DC plans become more common, more workers will be making decisions that
will affect their retirement security. In 1988, 26 percent of workers reported DC plans as
their primary plan type. By 1993, this percentage nearly doubled to 50 percent (EBRI,
1995). Increasingly, workers are depending upon their own contributions to an employer
retirement plan to support them in retirement. The 1999 RCS reports that while only 8
percent of current retirees depend on own-funded employer retirement plans, 32 percent

of current workers list these plans as their primary source of retirement income (Galer et

 

so For example, a participant who chooses his own investments and incurs losses may allege a breach has
occurred because the sponsor did not override his investment decisions.

5 ' Access Research (1995) reports that beginning in the late 1980’s, plan sponsors (particularly those
specializing in mutual funds) allowed daily valuations and trading for plan participants. Since that time
over 30 percent of 401 (k) plans allow daily valuations and trading, and over 50 percent of ﬁrms with over
1,000 employees include this feature.

80

al., 1999). How assets are allocated within the pension account can have signiﬁcant
effects on the pension funds available for retirement.52

Recently concerns have arisen in both private and government circles regarding
the increasing numbers of DC participants who are able to choose how their money is
invested. The primary concern is that workers are choosing overly conservative
portfolios that will not provide adequate returns for retirement security. Reports of
lower aggregate holdings of equities in DC plans than in DB plans have prompted much
of this concern. 1991 DOL tabulations of the assets held in private pension plans with
more than 100 participants show that DB plans held nearly twice as much equity53 as DC
plans (43 percent versus 27 percent) (DOL, 1995).54

Whether this comparison is informative is debatable. Portfolio managers make
their asset allocation decisions with different goals in mind than the individual investor.
Managers and individuals have different investment time horizons, which may cause their
willingness to accept risk to differ. An individual investor can have a fairly longtime
horizon if he is just starting his career, or a short one if he is approaching retirement. The
DB manager, on the other hand, maximizes the return for an inﬁnitely lived agent, the
plan sponsor. DB plan accounts also hold all assets in one account, allowing the presence

of fLmds for employees of different ages to decrease risk. Additionally, the government

 

52 Kritzman (1994) illustrates the difference in returns by comparing two alternatives for a $100,000
investment. An investor puts it all in either a risk-less asset with an annual rate of return of 3 percent or a
riskier asset with an annual rate of return of 10 percent with a 15 percent standard deviation. Alter 20
years, the risk-less asset is worth 8180,61 1, while the riskier asset will be worth between $180,651 and
$2,565,345.

53 The term equity usually refers to both own-company stock and outside stock. The ﬁgures reported by the
DOL deﬁne equity as outside stock only.

54 1989 Form 5500 ﬁlings show that DB plans hold two-thirds more assets in equities than DC plans do
(Wyatt Company, 1994). Greenwich Associates estimates that DC participants invest 39 percent of their
assets in equities, while DB plans invest more than half of their assets in equities (Smith, 1993).

81

insures DB assets through the Pension Beneﬁt Guaranty Corporation (PBGC); DC assets
are not insured.55

It can be misleading to draw conclusions about DC investments from the
aggregate data. Both Goodfellow and Schieber (1997) and Papke (1998) analyze
individual data and the choices participants make in self-directed plans. Papke estimates
that DC participants who are able to choose their investments actually allocate 15 percent
more of their balance to equities than do participants of DC plans without choice.56
Goodfellow and Schieber use a database of 36,000 participants in 24 DC plans and
examine the percentage of assets in equities according to characteristics such as age and
wage income. The aggregate numbers show a pattern of a relatively small percentage of
equities (28 percent) in DC plans. However, when they stratify the data by age they ﬁnd
that overall, investment in equities falls with age, an investment strategy most ﬁnancial
professionals advocate. Since older workers tend to have larger balances in their DC
accounts than younger workers, it follows that a smaller percentage of total DC assets
will be invested in equities.

Goodfellow and Schieber do offer some evidence, however, that certain groups of
participants are investing either too conservatively or too risky than ﬁnancial planners
would advocate for their stage in the life cycle. By examining percentages of each asset

type for potential investors of that asset in different age groups, they ﬁnd that 23-34

percent of participants age 30 and under hold 80 percent or more of their assets in ﬁxed-

 

” One could propose that the existence of insurance for DB assets may provide incentive for DB sponsors
to invest in overly risky assets. The incentive for ﬁrms to take such risks is probably not very strong.
Warshawsky (1997) reports that up until the early 1990s the PBGC was ﬁnancially volatile, causing
Congress to raise premiums paid by employers of DB plans in 1987 and again in 1989. Additionally,
penalties are imposed on the ﬁrm if the value of the assets in the plan fall below a minimum funding
requirement set by the PBGC.

82

income funds - a very conservative portfolio for this age group.57 Some groups of
participants also appear to expose themselves to excessive risk. Fourteen percent of those
ages 61 and older hold over 80 percent of their assets in domestic equity funds”, and 5
percent of this age group holds over 80 percent of their balances in company stock.
Additionally, 13-31 percent of those under 30 and nearly 10 percent of those 30 to 40
years of age hold 80 percent or more of their assets in company stock. Holding all or
most of a DC portfolio in company stock alone is an overly risky investment - even for
younger participants - as the collapse of the company would not only mean loss of
income for the participant, but loss of retirement security as well.

Will allowing individuals to choose their own investments result in lower
retirement wealth than if the investments are made by an investment manager? It is not
possible to answer this question until those who will depend on funds in self-directed
accounts enter retirement. Examination of the available evidence suggests that in general
individuals are not investing as conservatively as the aggregate data show. However,
there are groups of participants that are not making what ﬁnancial advisors would term

"sophisticated ﬁnancial decisions."

Retirement Education
Retirement education broadly consists of two types: education offered within the
workplace, and education offered privately by such institutions as community colleges

and private ﬁnancial planners. Retirement education in the workplace has risen

 

56 This estimate is assuming the plan is not a proﬁt sharing type, since proﬁt sharing plans often mandate
(or provide incentives for) investment in company stock.

57 If these participants have non-pension stock holdings designated for retirement, the overall portfolio may
not be that conservative. The data, however, does not provide information on outside stock holdings.

83

dramatically throughout the 19905: a 1994 survey by EBRI and Matthew Greenwald and
Associates reports that 73 percent of workers participating in a 401(k) plan said their
employer offered retirement education, versus 45 percent of employers of employees
reporting being retired in 1994 (EBRI, 1995).

Retirement education can come in many forms. Both private entities and
employers often use pamphlets, seminars, or individual ﬁnancial counseling as a means
to communicate their message. More recently interactive phone response systems and
software and Internet sites have been added to assist clients. Buck Consultants (1994)
reports that in the early 19905, the most common modes of communication for education
in the workplace were the summary plan description or other written material (53 to 68
percent of plans) and seminars or individual meetings (32 to 47 percent).

The frequency with which retirement education is offered in the workplace differs
depending on the plan sponsor. A 1993 survey of 401(k) plan sponsors shows that 25
percent of sponsors provide retirement education quarterly, 11 percent provide it
annually, and 20 percent provide it upon participant enrollment (EBRI, 1995). Whether
education is being offered frequently enough to have adequate impact upon asset
allocation is debatable. Financial planners report that short, frequent sessions on
appropriate asset allocation is the best way for employer education to make a difference
(Yakoboski, 1995). Because learning about such complex issues as risk tolerance and
purchasing power are difﬁcult to grasp - and recall - the frequency with which the
message is communicated is very important.

In order to increase the probability that participants will understand and

implement sophisticated asset allocation in their plans, the topics covered must also be

 

58 This may not be considered too risky if the relevant time horizon extends substantially past retirement.

84

relevant to the asset allocation question. A 1994 EBRI survey of self-directed 401(k)
sponsors shows that the most common topic covered is asset allocation, with 93 to 99
percent of plans (ranging from under 100 employees to over 10,000) providing this
information. Understanding risk and risk tolerance is covered by 84 to 99 percent of
plans, and estimating income needed for retirement is included in material for 94 to 99
percent of plans (EBRI, 1996). Estimating income needed for retirement is particularly
relevant to the asset allocation decision, as rate of return, amount of risk willing to be
assumed, and contribution rate must be determined simultaneously to pinpoint a desired
level of retirement security.

Employers may be more likely to offer retirement education with the 1992
ﬁnalization of section 404(c) regulations. As I described in the previous section,
compliant sponsors of self-directed plans must offer information about the plan that will
aide employees in choosing investment wisely. 404(c) guidelines offer certain legal
protections that may now provide incentive for employers to introduce self-direction, and
hence education, into the workplace. In fact, Buck Consultants (1994) reports that 50
percent of sponsors cited the release of 404(c) regulations as a factor in offering

education to their participants.

Why Might Retirement Education Affect Asset Allocation?

Determining if individuals would choose an optimal portfolio allocation without
the aid of retirement education is difficult, as there is not agreement on one economic
theory that describes how participants should choose their investments within their plan.

A number of theories have been proposed that follow a life cycle model and emphasize

85

choosing investments depending on the length of the saving time horizon. Bodie, Merton
and Samuelson (1992) propose that households should reduce equity holdings as they
age, suggesting that as the relevant saving time horizon decreases, equity and other risky
asset holdings should also decrease. Bodie (1995), on the other hand, uses a model of
option pricing and concludes that risk does not diminish the longer the stock is held.
Samuelson (1989, 1990) proposes that the increasing longevity of retirees means
investment in equities remains an option near and after retirement. The time horizon for
retirement saving should not stop at retirement, since individuals are living well past
current retirement age.

Kritzrnan (1994) uses a standard model of declining marginal utility and
increasing wealth to argue that risk aversion is so strong for some investors that the stage
in the life cycle does not matter. For these individuals, low-risk or risk-less assets will be
chosen regardless of time horizon. For this short-term risk aversion to carry over to the
long term, three conditions must be met: ﬁrst, risk aversion must be invariant to changes
in wealth; second, investors must believe that returns on risky assets are random; and
ﬁnally, the investor's future wealth must depend only on investment income.

Regardless of the theory (or theories) that best describes investment behavior, if
individuals do not possess enough ﬁnancial and economic knowledge they will not make
optimal choices. While it is necessary for participants to be able to estimate how much
income they will need at retirement in order to determine what asset mix is optimal for
their needs, most Americans have never made this calculation. The 1997 RCS reports
that only 36 percent of workers in the United States have any idea of how much they will

need for retirement, and of these individuals, 24 percent cannot provide an actual ﬁgure

86

(Dickemper and Yakoboski, 1997). Workers are also ill informed regarding risk and
return tradeoffs. Just over half of all workers surveyed (56 percent) know that
historically, the probability of losing money from a diversiﬁed portfolio of stocks has not
risen the longer the investment is held (EBRI, 1996).

Participants admit they are not prepared to make such complex decisions.
Although 64 percent of participants in a DC plan would like to make their own
investment decisions (EBRI/Gallup Organization, 1994), only 26 percent of participants
in a Phoenix-Hecht study believe they are well qualiﬁed to make their own investment
decisions and only 8 percent of sponsors believe so (Phoenix-Hecht, 1994a). With the
array of knowledge that is needed to determine asset allocation, it is no surprise that most
workers do not feel they alone have the ability to make sound decisions.

Perhaps the reports of plan sponsors and employees themselves are the best
indicator of whether individuals are making optimal asset allocation decisions in the
absence of retirement education. Since most professional advisors advocate increasing
proportions of equity with age, plan sponsors use this advice as a benchmark with which
to compare employee investment choice. An EBRI survey of plan sponsors shows that
49 percent of sponsors feel that educational materials have a substantial effect on younger
participants investing less conservatively, while 35 percent feel there is a moderate effect
(EBRI, 1996). In a 1994 EBRI/Greenwald survey, 44 percent of employees said
retirement education materials led them to change the allocation of their money (EBRI,
1995). Likewise, in two separate studies of Continental Bank and Rite Aid Corporation,
a large proportion of employees reallocated their assets after using educational material.

Sixty-ﬁve percent of employees at Continental Bank who used an interactive software

87

package to estimate retirement needs and asset allocation mixes reported they plan to
change their investment allocation. Rite Aid Corporation targeted female clerks under
the age of 40 with a video educational campaign. When the campaign was initiated in
1991, 54 percent of these employees allocated their assets to the ﬁxed-income fund.
Figures reported shortly after 1994 show this percentage down to 37 percent (EBRI,
1995). If participants change their behavior after using retirement educational materials,

they must believe their original choices were not optimal.5'9

3. PREVIOUS RESEARCH
What Factors Affect DC Plan Asset Allocation?

Much of the previous work on asset allocation within DC plans examines how
characteristics such as age, income, gender and risk aversion affect participant investment
decisions. Bajtelsmit and VanDerhei (1997) examine the correlation between employee
characteristics and the likelihood of allocation to different asset types for participants in a
self- directed DC plan. They use a 1993 sample of 20,000 management employees from
a large United States employer.

Looking at the speciﬁcation measuring DC balances (they also include
speciﬁcations using current DC contributions), their estimates show that women are more
likely than men to invest in ﬁxed-income assets, but less likely to invest in employer
stock. Age also affects investment in ﬁxed-income assets - but in a non-linear fashion.
As participants age their allocation to ﬁxed-income investments decreases initially, but at
a speciﬁc age they begin to increase their allocations to these assets. Bajtelsmit and

VanDerhei also offer evidence of decreasing relative risk aversion with respect to total

 

59 - - . . . .
This IS true, of course, only if sound advrce rs given.

88

401(k) wealth. The likelihood of investing in equities and employer stock increases with
DC wealth, but at a decreasing rate. The fact that participants treat investment in
employer stock similarly to non-employer equity funds lends evidence to the hypothesis
that participants see employer stock as a risky investment.60

Poterba and Wise (1998) examine participant-level data for DC participants in
self-directed plans provided by the ﬁnancial service institution TIAA-CREF. TIAA
accounts typically are invested in ﬁxed-income assets; CREF accounts in equity. In
1993, 44 percent of assets were in TIAA accounts. Investment in these ﬁxed-income
ﬁmds increases with age, with 37 percent of assets of those under age 35 and 49 percent
of those ages 55 to 64 invested in TIAA. Percentage of assets in TIAA also falls with
income, with 53 percent of assets of those who earn $25,000 annually and 40 percent of
assets for those with incomes over $100,000 in TIAA.

Econometric analysis of the data shows that participants with income greater than
$100,000 and those with net worth over $250,000 allocate a smaller percentage of their
assets to TIAA. Females invest 4 percent more in TIAA accounts, while those with less
than 12 years of formal education invest 10 percent more of their assets in TIAA.
Higher-income and higher-wealth participants also invest less in TIAA - 5 to 7 percent
fewer DC assets for those with incomes over $100,000 are invested in the ﬁxed-income
fund, as are 4 percent fewer assets for those with higher net wealth.

The characteristics of DC plans may make a difference in the composition of
assets in the plan. While the aforementioned research examines only participants with

self-directed plans, Papke (1998) uses participants with and without choice to examine

 

6° There is some evidence that individuals may perceive company stock as low-risk and substitute it for
ﬁxed-income assets. See Galer et al. (1999).

89

whether there is a difference in pension portfolio allocation if a participant is able to
choose his own investments. She uses the 1992 National Longitudinal Survey (NLS) of
Mature Women to estimate the effect that being in a self—directed plan has on the
percentage of equities invested in the plan.61 She ﬁnds that those who are able to choose
their investments allocate 15 percentage points more to equities, given the plan is not a
proﬁt sharing plan. She ﬁnds that women in plans with choice choose to invest 5
percentage points less in equities than do men, however, this difference is not statistically
signiﬁcant. Investments in stock also decline with age, with a 2-percentage point
decrease in stock investment for each additional year of age. Papke's results seem to lend
evidence that DC participants actually invest less conservatively when they are able to
make their own portfolio decisions.

The majority of the existing literature on pension asset allocation decisions
describes how portfolio choice differs by characteristics of the participant. My analysis
adds to the literature by including controls, such as mortality risk and saving time
horizon, which are not included in previous studies. But more speciﬁcally, I test one way
in which participants may alter their portfolio allocation - via retirement education - that
provides evidence as to whether policies aimed at increasing individual ﬁnancial

knowledge in the United States are useful.

Does Retirement Education Affect Portfolio Choice? Evidence in the Literature
There is no previous research on retirement education outside of the workplace,
and more speciﬁcally, no analyses of how retirement education affects DC asset

allocation. However, there are two studies that address the relationship between

 

6' The 1992 NLS for Mature Women includes women and their spouses, ages 55-69 in 1992.
90

employer-sponsored retirement education and savings behavior, both inside and outside
the pension plan.

Bemheim and Garrett (1996) use a sample of 2,055 individuals, ages 30 through
48 in 1994. The survey is sponsored by Merrill Lynch and includes information on both
the availability and use of employer-sponsored education. The type of retirement
education offered (written materials or seminars) is not speciﬁed, so the analysis
estimates the effects of the typical retirement education program in the workplace.

Bemheim and Garrett estimate separately the effects of employers offering and
employees participating in retirement education on the household saving rate, retirement
saving rate, total wealth, retirement wealth, 401(k) participation rate, and 401(k)
balances. Workers whose employers oﬂer retirement education save 1.7 percent more
than those not offered education, and have $2,488 more in retirement wealth. Employees
who use retirement education materials save 2.2 percent more than those who do not,
while accumulating $2,176 more in retirement wealth. Additionally, those who use
retirement education are 19 percent more likely to participate in a 401 (k) plan
(conditional on eligibility) and have accumulated $2,826 more in the 401(k) account. All
estimates are statistically signiﬁcant (at the 10 percent level or lower).

Bemheim and Garrett also test the hypothesis that attending a retirement meeting
is correlated with unobserved saving tastes. They conclude that those with lower
propensities to save are more likely to be offered and use retirement education.

Bayer, Bemheim and Scholz (1996) use a 1993-1994 longitudinal survey of 1,100
employers selected by KPMG Peat Marwick to study the effects of employers offering

retirement education on participation and contribution rates in 401(k) plans. The data

91

allows them to examine the effects of both seminars and written material, as well as the
frequency in which education is offered. When estimating these effects for both highly
compensated (HC) and non-highly compensated (N HC) employees62, they ﬁnd that
participation rates are 11.5 percent higher for NHC employees and 10.5 percent higher
for HC employees when retirement education is offered, given that the seminars are
offered frequently.63

The studies done by Bemheim and Garrett (1996) and BBS (1996) provide
valuable information on the effects of retirement education on various forms of saving.
While how much is saved is a large factor in determining the ﬁnal retirement payout, the
choice of asset allocation is also a major factor in determining the ultimate worth of a
pension plan. Given that many Americans do not have the ﬁnancial knowledge with
which to make such a sophisticated decision, it is worthwhile to examine the effect of

retirement education on asset allocation choice within DC plans.

4. RETIREMENT EDUCATION AND ASSET ALLOCATION IN THE 1992 HRS
The 1992 HRS is the ﬁrst wave of an ongoing survey of 12,057 men and women,
most between the ages of 51-61 in 1992.64 Along with standard demographic and
ﬁnancial information, the survey also includes questions speciﬁc to retirement. Detailed
pension information and expectations about the retirement years make for a rich source of

data for analyzing retirement issues. My sample consists of the 1,236 DC participants

 

62 Highly compensated employees are deﬁned as having annual earnings of $100,000 or more, owning
more than 5 percent of the company, or having annual earnings of $65,000 or more if this amount is in the
top quintile of the ﬁrm’s salary distribution.

63 The offering of retirement education has no effect if the education is in the form of written materials, or
if seminars are offered occasionally.

6‘ Because spouses of respondents of this age group are also asked the same questions, only 9,673
individuals ﬁt into this age category.

92

who describe their current pension plan as self-directed. Because each individual can
have more than one plan, I analyze a total of 1,322 plans, with 1,155 participants having
only 1 plan, 76 having 2 plans, and 5 having 3 plans.

Along with the fact that each individual can have more than one plan, each family
can also have more than one plan. In the HRS, married respondents and their spouses are
both asked each question. Thus, if the spouse also has at least one self-directed plan,
there can be more than one plan per family. Of the 1,236 DC participants, 106
individuals (53 couples) have one or more plans, resulting in 1,030 family identiﬁers.
For 617 (60 percent) of these families, at least one of their self-directed plans is their
primary plan.

Although the HRS is a longitudinal study, I only use the 1992 wave. The
questions addressing retirement education are only asked during the 1992 survey.
Although using asset allocation information from subsequent waves would allow me to
exploit the beneﬁts of longitudinal analysis, information on attendance in retirement
classes after 1992 is not available. Given the acceleration of retirement education in the
19903, there is a high probability of undocumented participation for those who attended a
meeting after 1992.

I report summary statistics for the sample in Table 2.1. Twenty-nine percent of
DC participants attended a retirement class, with 79 percent of these seminars taken at the
workplace. The percentage of individuals with self-directed plans who attended a
meeting is signiﬁcantly higher than the full sample of the HRS, the sample of those with
DC plans, and those with DC plans without participant choice. Only 17 percent of the

full sample attended a meeting, with a larger percentage of all DC participants- 24

93

percent - attending. Only 17 percent of those with DC plans without self-direction
attended a meeting (not reported in tables).

The average age in the sample is 54 years. F orty-ﬁve percent of women have
self-directed plans, and 92 percent of participants have at least a high school degree. The
average expected time until retirement is 8.5 years.65 Seventy-one percent of individuals
have another employer-sponsored pension plan: either a DB plan in their current or past
job, another self-directed DC plan, a non self-directed DC plan, or a plan having
characteristics of both a DC and a DB plan.

Fourteen percent of the plans are deﬁned as either proﬁt sharing, stock option, or
ESOP plans. Because these types of plans are more likely to offer company stock and/or
mandate employer contributions be in company stock, I group these types of plans
together under the general heading proﬁt sharing plans. Identiﬁcation of these plans is
crucial to the analysis, as they may contain a larger percentage of equity that is not the

direct result of choice by the employee.66

Asset Allocation in the HRS

All individuals are asked if they have a pension plan with their current employer,
and if they do, whether the plan has characteristics of a DC plan, a DB plan, or both.
Thirty-ﬁve percent of those currently working are covered by an employer-sponsored

pension plan; 53 percent of these plans are described as having characteristics of a DC

 

6" The HRS asks participants in what year they expect to retire. Seven percent of participants said they will
never retire. Because I am unable to determine a retirement saving time horizon for these individuals in
terms of years, these responses are set to missing.

66 50 percent of proﬁt sharing plans in the sample are invested in mostly equities, compared to 26 percent of
plans not described as proﬁt sharing (not reported in tables).

94

plan.67 Fifty-six percent of DC plans (1,314 plans) are self-directed.

The HRS records how the participant chooses to allocate his DC assets by
assigning him one of three categories. These options include investing the balance
mostly or all in interest earning assets (hereafter, also called bonds), mostly or all in
equities (hereafter, also called stock), or split between the two (a diversiﬁed portfolio). In
Table 2.2, row 1, I show the percentage of participants in each of these three categories.
In order to avoid double-counting individual characteristics, I restrict those individuals
who have more than one self-directed plan to their primary self-directed plan. Out of
1,108 participants where the respondent also answered the retirement education question,
28 percent report investing their DC assets in mostly stocks, 27 percent choose mostly
bonds, and 45 percent hold a mixed portfolio. This distribution of assets is similar to
those reported by Papke (1998), whose sample from the NLS also reports asset allocation
in these same three categories and consists of participants of roughly the same age
group.68

Several participant characteristics affect DC asset allocation. As I discussed in
Section 2, stage in the life cycle, willingness to take risks, and even gender have been
proposed to affect an individual's investment choice. The HRS not only contains
standard demographic and ﬁnancial measures that allow econometric testing of these
theories, but also includes measures often omitted from other surveys that capture
retirement saving horizon and degree of risk aversion. Beginning in row 2 of Table 2.2, I

show participant allocation by characteristics that have been highlighted in the pension

 

'57 This percentage does not include those individuals who describe their pension as having characteristics
of both a DB and DC plan.

(’8 When making comparisons with Papke (1998), it must be kept in mind that while my study includes only
self-directed plans, Papke’s study includes both plans that are self-directed and those that are not.

95

allocation literature as being correlated with DC asset allocation.

Conventional wisdom in the professional ﬁnancial community states that the
longer the time horizon for investing, the larger proportion in equities one should hold.
This means that as one ages, the proportion of stock in the portfolio should decrease. The
tabulations in Table 2.2 do not show a pattern consistent with this belief. While 26-28
percent of participants in their pre-retirement years (under age 60) hold mostly equities,
29 percent of participants in or very close to their retirement years also hold mostly or all
assets in equities. A higher proportion of older individuals do hold all or mostly bonds,
however, but the pattern is not linear. While only 24-25 percent of those under age 55
hold mostly low-risk portfolios, 33 percent of those ages 55 to 60 and 28 percent of those
age 60 and over choose to do so. Although the HRS restricts the sample of participants to
those in the later stages of the life cycle, other studies using individuals of all ages report
similar patterns. Poterba and Wise (1998) use regression analysis to examine pension
asset allocation with participants of all ages and also ﬁnd that age is not an accurate
predictor of pension portfolio choice.

There is evidence that women invest more conservatively than men by investing
less often in equities (Jianakoplos and Bemasek (1994), Hinz et a1. (1997)). This pattern
is also evident in the HRS data. Thirty—one percent of males report they invest in mostly
equities, while only 23 percent of women do. This difference is statistically signiﬁcant,
with a p-value of 0.003. The gender difference remains when dividing the groups further
by marital status. Lower percentages of both married and unmarried women invest in
equities (25 and 19 percent, respectively) than do married and unmarried males (32 and

26 percent, respectively) (not reported in table). A larger percentage of females than

96

males choose portfolios that include both stocks and bonds, however, with 47 percent of
women and 41 percent of men choosing to diversify their portfolios (statistically
signiﬁcant with a p-value of 0.07).

Both participants without high school degrees and individuals with lower incomes
show patterns of more conservative investing. While 28 percent of participants with high
school degrees invest mostly in equities, 25 percent without a degree choose mostly
stocks, but this difference is not statistically signiﬁcant. The difference between the
percentage of participants who invest in mostly bonds is quite striking: 39 percent of
those without a degree choose mostly bonds, while only 26 percent of those with a degree
make this choice. The difference in this allocation type is statistically signiﬁcant at the
ﬁve percent level (p-value equal to 0.02).

Asset allocation also varies by family income. While 21 percent of participants
who earn less than $25,000 annually invest their DC assets in mostly stocks, 33 percent
of individuals whose family income is greater than $100,000 per year invest in this
manner. This difference is statistically signiﬁcant with a p-value of 0.05. Poterba and
Wise (1998) use econometric analysis and conclude that lower-income, less-educated
workers allocate smaller portions of their DC assets to equities.

In Table 2.2, I also examine DC asset allocation according to how many years the
respondent expects until he retires. This measure is valuable for capturing retirement
saving horizon and may represent stage in the life cycle more appropriately than does
age, since workers may choose to retire at different ages and save accordingly.69 In

accordance with traditional ﬁnancial advice, the HRS data shows a general pattern

 

69 If workers save for retirement by looking past the retirement date to post-retirement years, then this
measure may not be as accurate.

97

toward more conservative investing as years until retirement decreases. Twenty-two
percent of participants who have 11 or more years until retirement invest in mostly
bonds, while 33 percent of those with less than 5 years until retirement choose mostly or
all low-risk assets (estimated with a p-value of 0.002). The pattern in investment in
mostly equities ﬂuctuates very little, however, as one gets closer to retirement. Twenty
seven percent of those with 5 to 8 years until retirement invest in mostly stocks, and this
proportion only rises 2 percentage points to 29 percent for those with 8 to 11 years until
retirement. Twenty-nine percent of participants with either less than 5 years until
retirement or with 11 or more years invest in mostly equities.

As the choice of which assets to invest in involves risk and return tradeoffs, the
willingness to take risks - particularly ﬁnancial risks - affects asset allocation. One way
to measure risk aversion is to examine individual choice given different risks and returns.
The HRS includes a question designed to capture degree of risk aversion. The
respondent is asked to consider a scenario in which he can take an equally good job with
a 50 percent probability of doubling his income and differing probabilities of decreasing
his income by half. I quantify degree of risk aversion for each participant based upon the
probability of loss of income he is willing to accept.

The last rows of Table 2.2 stratify DC asset allocation by degree of risk aversion.
Twenty-nine percent of participants who are the least willing to accept risk (or who are
extremely risk averse) are invested mostly in equities, compared to 33 percent of
participants who have the highest tolerance for risk (or those with low risk aversion).
The relationship between risk aversion and percent in equities is not linear. While those

participants at the two extremes of risk aversion appear to conform to the expected

98

pattern of trading risk for return, a smaller percentage of individuals who have high
amounts of risk aversion (26 percent) invest in mostly equities than do those with an
extreme degree of aversion (29 percent). The differences between each level of risk

aversion and extreme and low degrees of risk aversion are not statistically signiﬁcant.

Retirement Education in the HRS

The HRS is the only public data source that contains data on both retirement
education and asset allocation within pension plans. All respondents were asked two
questions about retirement education: Have they ever attended any meetings on
retirement or retirement planning? Of those who answered yes, they are then asked if
that meeting was organized by the respondent's or his spouse's employer.

In Table 2.3, column (1), I report the proportion of participants who attended a
retirement class, by individual characteristics. In columns (2)-(4), I report the proportion
of individuals who choose each type of asset allocation (taken from Table 2.2). Notice
that participants without a high school degree and also those with lower income not only
have lower participation rates in retirement meetings, but also tend to invest more
conservatively in their pension portfolios. The difference in retirement meeting
attendance is more apparent when looking at income levels: while 21 percent of
participants with incomes under $25,000 attend meetings, 32 percent of individuals with
incomes over $100,000 attend meetings.70 This difference is statistically signiﬁcant, with
a p-value of 0.10.

When comparing retirement meeting attendance and asset allocation in Table 2.3

99

it must be emphasized that there is no reason to believe that taking a retirement class will
increase holdings of equity in the portfolio. Nor is there evidence that more conservative
investing necessarily is sub optimal for the participant, or necessarily results in lower
retirement wealth. I merely suggest that if those with lower education levels or income
are investing more conservatively because they lack ﬁnancial knowledge, and their
decisions are in fact not optimal, then increasing attendance in retirement classes may
increase their well-being.

I have presented tabulations in Tables 2.2 and 2.3 that show patterns of asset
allocation and retirement meeting attendance by speciﬁc characteristics of DC
participants in self-directed plans. In the following section I use econometric techniques
to examine whether retirement education affects the asset allocation decision, for
participants generally as well as for those individuals with varying time horizons and

degrees of risk aversion.

5. ECONOMETRIC RESULTS ON THE EFFECTS OF TAKING A
RETIREMENT CLASS ON DC ASSET ALLOCATION

In this section I describe the model I use in the econometric analysis and report
the results. I estimate the model using OLS and ordered logit. The model is as follows

(omitting individual subscripts):

A = x0+ 01R + (R*T)(0 + (R*V)¢ +Pw + a, (1)

where A is a trinomial variable equal to 100 if the plan is invested mostly or all in

 

7° Increased attendance could of course be the result of more of an interest in saving, which has been shown
to be positively correlated with education and income. This selection issue will be addressed further in the
next section.

100

equities, equal to 50 if investment is mixed between equities and bonds, and equal to zero
if investment is mostly or all in bonds. The independent variable of interest, R, is a
dummy variable equal to one if the respondent attended a retirement meeting, and zero
otherwise. X contains the observed variables that are thought to affect the asset
allocation decision. The vectors (R*T) and (R*V) contain interactions of the retirement
meeting variables with the dummy variables for years until retirement and degree of risk
aversion, respectively. The variables for years until retirement and degree of risk
aversion are included in X. P is a vector of proxy variables to control for unobserved
individual saving tastes.

Since the HRS responses on pension asset allocation are not reported as
continuous percentages, it is necessary to express the dependent variable, A, as a
trinomial choice variable. Simply choosing numbers for this variable and estimating the
model using ordered logit is one possible estimation technique. However, using logit will
only give estimates of the probability of each response occurring, and because the
estimates are not expressed as partial effects, simple interpretation of the estimates is
limited to examination of sign and statistical signiﬁcance.

By using a linear model I am able to interpret the estimates as percentage change
in equity. 1 follow Papke (1998) by assigning each asset allocation response 100 (mostly
or all equities), 50 (a mix of equities and bonds), or 0 (all bonds). I choose these values
to simplify the interpretation of the coefﬁcient estimates as percent change in equities.
Assigning extreme portfolio allocations will likely result in some mismeasurement of the
true asset mix, particularly if the proportion of different assets varies widely within each

category. This possibility of bias due to mismeasurement must be kept in mind when

101

interpreting the results. For a complete discussion of structuring the dependent variable
in this manner, see Papke (1998).

I estimate equation (1) using both OLS and ordered logit, without interaction
terms or proxy variables. The OLS results are presented in Table 2.4 and the ordered
logit estimates in Table 2.5. The logit estimates are not partial effects, and so magnitudes
in columns (1) through (5) cannot be compared to the corresponding columns of OLS
estimates. However, the direction and statistical signiﬁcance of the logit estimates are
similar to those using OLS.

The OLS standard errors in Table 2.4 are robust to heteroskedasticity, and
because each participant and family can have more than one self-directed plan, standard
errors are also corrected for individual and family-level correlation. I report the simple
relationship between meetings and asset allocation without any control variables in
column (1). DC participants choose to invest 1.79 additional percentage points in stock if
they attended a retirement meeting. This effect is not statistically signiﬁcant, however.

I add demographic, income and net worth variables to the speciﬁcation and report
these results in column (2). In column (3), I add additional ﬁnancial variables as well as
variables that describe the pension plan. The estimate on attending a meeting drops
slightly to 1.28 in column (2) and then rises to 3.16 (with a t-statistic of 1.1) in column
(3), but is not statistically signiﬁcant in either speciﬁcation.

Many of the estimates for other explanatory variables are consistent with previous
asset allocation research. In column (2), the estimate on age shows that participants
invest almost one-half a percentage point less in stock for each additional year (with a p-

value of 0.8), although the effect becomes very small and statistically insigniﬁcant in

102

column (3).71 Whether participants actually alter their investments in stocks and bonds as
they age is not clear from the existing literature. While Goodfellow and Schieber (1997)
and Papke (1998) ﬁnd decreasing equity investment with age, Hinz et a1 (1997) do not
ﬁnd age to be a factor in portfolio choice. It should be noted that the estimate for age in
this model does not generalize to the total population of DC participants in self-directed
plans, as the HRS restricts its sample to those in the pre-retirement years.

Income and net worth also affect DC asset allocation. As seen in column (2),
those with incomes over $100,000 invest 10.41 percentage points more in stock than do
those with incomes under $25,000, although this effect decreases signiﬁcantly with the
additional controls in column (3). Net worth affects DC investment choice in both
speciﬁcations. As shown in column (3), those with $100,000 to $250,000 in net worth
invest 9.66 percentage points more in stock than those with less than $50,000 in net
worth, while participants with net worth between $250,000 and $500,000 choose 12.8
percentage points more in stock (p-values of 0.025 and 0.13, respectively). Net worth
variables are estimated jointly with a p-value of 0.04.

Gender affects asset allocation decisions. As is seen in column (3), females invest
more conservatively than males, allocating 6.45 percentage points less to stocks. This
effect holds even after controlling for income and degree of risk aversion.

Existing evidence of a gender effect on DC asset allocation is mixed. Hinz et al.
(1997) ﬁnd that females invest less of their assets in equities after controlling for

demographic and ﬁnancial variables. Jianakoplos and Bemasek (1994), on the other

 

7' Not surprisingly, the addition of the variable measuring years until retirement in column (3) contributes
signiﬁcantly to the change in the estimate on age. Estimating the speciﬁcation in column (3) without the
years until retirement variables produces an estimate of -0.35 on age, with a p—value of 0.23 (not shown in
table).

103

hand, do not ﬁnd a gender difference after controlling for demographic and wealth
characteristics.

Because the HRS is particularly concerned with retirement issues, it provides
detailed data on all pension plans and expectations about retirement. Speciﬁcally, it
provides details about the respondent's self-directed plan as well as other DC and DB
plans, both from current and past employers. I include a binary variable in the
speciﬁcation in column (3) that controls for the existence of either a DB plan from a
current or past employer, another self-directed plan, a DC plan that does not offer choice,
or a plan with characteristics of both DB and DC plans. Individuals with other pension
plans may be more willing to invest in riskier assets if they have additional pension
security. Although I ﬁnd that participants with another plan invest 3.3 percentage points
more in equity, the estimate is not statistically signiﬁcant.

One characteristic of the self-directed plan that affects pension allocation is
whether the plan is of the proﬁt sharing type. Participants with plans of this type (recall
I group all ESOP, stock option, and proﬁt sharing plans under this heading) have a
signiﬁcantly higher percentage in stock than those with other types of DC plans. These
plans hold 18.02 percentage points more stock than other types of plans, estimated with a
p-value of 0.0. This result is not surprising given the fact that proﬁt sharing plans often
consist of mandated employer matching contributions in company stock.

Including a variable in the model for non-pension equity holdings may be
controlling for one of two things. Either the participant has a taste for stocks or he
diversiﬁes across his pension, non-pension and wealth holdings. If this equity variable is

positively correlated with the percentage of equity in a DC plan, then it is most likely

104

controlling for a taste for stock. If a negative relationship is observed, however, then
individuals may be choosing their investments within the pension plan by diversifying
across all their asset holdings. Recent work by Bodie and Crane (1997) and Weisbenner
(1999) ﬁnds no evidence of diversiﬁcation across pension and non-pension portfolios.

Papke (1998), likewise, ﬁnds no evidence of diversiﬁcation. She ﬁnds a positive
relationship between outside stock holdings and the percentage of equity in the pension
plan. She reports that for every $1,000 in outside stock holdings, the participant holds
about one-third of a percentage point more in stocks. I ﬁnd positive, although smaller,
effects. For every $1,000 held in outside equity, the participant invests 0.02 of a
percentage point more in stocks in his DC plan (one-ﬁfth of a percentage point for every
$10,000 in outside stock).72

One of the quantitatively largest effects on percent invested in equities is the
presence of an employer match to the employee contribution. If the employer provides a
match, the participant invests 15.16 percentage points more in stock than if there is no
match, estimated with a p-value of 0.0. The reason for this effect is not immediately
apparent, as this relationship has not been observed in other DC asset allocation
literature.

Participants with employer matches may invest more in stock if the employer

match is mandated to be in company stock. Galer et a1. (1999) report that employees in

 

72 It can be argued that those who take a class may use knowledge they learn in class and change their
outside stock holdings. If this is the case then the outside stock variable is not picking up a taste for stocks.
It is doubtful that the estimate on the outside stock variable is representing any other effect besides the
correlation between outside and pension stock allocations. First of all, the outside stock variable is the total
amount of equity accumulated in all accounts, not a ﬂow of investments in stock for only the current year.

In addition, taking out the outside stock variable from the regression in column (3) has only a minimal
effect on the retirement education variable. When I include the outside stock variable in the speciﬁcation,
as 1 have done in column (3), the estimate on the retirement meeting variable is 3.16, with a standard error

105

plans with mandated employer contributions put a higher proportion of their own
contributions in company stock than do employees in companies without the mandate.
Unfortunately, I do not have information on whether the match is mandated to be in
company stock. 1 do know, however, whether a plan is of the proﬁt sharing type. Proﬁt
sharing type plans have a greater likelihood than other types of plans of requiring
company stock matching contributions. In order to test the null hypothesis that
participants with mandated employer matches do not increase their DC equity holdings
more than those without mandated matches, I interact the proﬁt sharing term with the
employer match variable and include the resulting interaction in the speciﬁcation in
column (3). The coefﬁcient estimate on this interaction term is statistically insigniﬁcant
(with a p-value of 0.50) (not reported in table), leading me to conclude that mandated
employer matching is likely not the reason for employer matches increasing the
percentage of equity in the DC plan.

An alternative explanation may have to do with how the participant perceives his
investments when his employer contributes funds. Employees make sophisticated asset
allocation decisions if they treat both their contributions and those from their employer
equally, since employer contributions are part of in-kind compensation. However,
employees may perceive the employer contribution as a windfall, and invest in higher-
risk assets as a result. By treating the employer match as one might treat another person's

money, the participant is willing to hold more high-risk equities in his portfolio.

 

of 3.02. Eliminating the stock variable increases the retirement meeting variable by 0.12 of a percentage
point to 3.28, with the same standard error.

106

Selection

An individual may choose to attend a retirement class for several reasons. He
may feel the need to begin preparing for retirement, leading him to seek out either a
ﬁnancial advisor or a class at work. Alternatively, he may realize he does not have the
necessary knowledge to make educated ﬁnancial decisions and solicit the advice of
someone more knowledgeable. Or his employer may simply require that he attend a
class.73 It is thus reasonable to assume that the choice to attend a class may be motivated
by unobserved factors that also affect the asset allocation decision. If so, then the
coefﬁcient estimate on R in equation (1) and all other estimates are biased and
inconsistent.

To address the possibility of selection bias, I include a set of proxy variables to
control for unobserved tastes. The vector, P, in equation (1) contains these proxies. The
HRS is a rich source from which to choose proxy variables. Along with measures of risk
aversion, retirement planning horizon, and future liquidity constraints (the net wealth
variables) that I include as basic controls in the model, I also construct variables that
control for mortality risk and general saving time horizon. In order to obtain consistent
estimates, I must assume that no other omitted factors in the error term are correlated
with either the observables or the proxy variables.

One's own assessment of mortality risk can affect one's investment choice.
Expectations of a shorter life - and hence a brief retirement (if one at all) - shortens the
retirement saving horizon. The HRS asks the respondent to rate the chance he or she will
live to be to age 85. I quantify the responses by constructing a binary variable equal to

one if the individual feels there is at least a 50 percent chance he will live to be age 85. If

107

participants follow the traditional ﬁnancial advice of increasing investment in risky assets
the longer the time horizon, then the probability of a longer life should increase the
percentage allocated to equity.

In addition to measuring the retirement saving horizon with the years until
retirement variable, I construct three dummy variables that describe general saving
horizon." The respondent is asked which time horizon is most relevant to his and his
family's saving plans: the next few months, next year, next few years, next 5-10 years, or
longer than 10 years. I incorporate the responses into three different time periods: a few
months to a few years, ﬁve to ten years, and more than ten years. The effect of general
time horizon on pension equity allocation depends upon whether the individual general
and retirement saving horizons coincide, as well as whether the participant sees his DC
plan as a vehicle for other types of saving besides retirement saving.

The presence of a safety net, such as Social Security, may also affect the asset
allocation choice. Respondents in the HRS are asked to rate the probability that Social
Security beneﬁts will decrease in the next ten years. Since the majority of individuals in
the survey are between the ages of 51 and 61, the next 10 years coincides with their
retirement years. The HRS groups the responses on a scale from 0 to 10, with 0 being
absolutely no probability of decreased beneﬁts. I group the responses into three
categories: those reporting a zero to 40 percent probability of decreased beneﬁts, those
believing there is a 40 to 70 percent chance, and those who are 70 to 100 percent sure
beneﬁts will decrease. The 1995 RCS reports that those who believe Social Security will
provide little, if any, beneﬁts for their retirement also expect to rely more on employer-

sponsored retirement plans (EBRI, January 1996). There could be two effects on pension

 

73 Unfortunately the HRS does not report whether the class was voluntary or mandatory.

108

asset allocation for individuals: they could feel the need to increase future balances
within their pensions and invest more in equities. Alternatively, they could invest more
in less risky ﬁxed-income instruments for fear of losing all or most of their retirement
security. Theoretically, the direction of the effect of lowering Social Security beneﬁts on
pension asset allocation is ambiguous.

It is also important to identify one other important factor that may affect both the
decision to attend a meeting and the portfolio choice: ﬁnancial knowledge.
Unfortunately, I am unable to directly measure the ﬁnancial knowledge of each
individual. Note, however, that two measures already included in X in equation (1) may
serve as proxies for this measure: formal education and amount of outside stock
holdings. Bemheim and Scholz (1992) ﬁnd that those with more formal education are
more likely to show evidence of sophisticated ﬁnancial planning. Likewise, individuals
who have non-pension stock holdings may have acquired ﬁnancial knowledge through
making decisions regarding these holdings, or they could have possessed knowledge
initially that led them to seek out and invest in equities.74

Table 2.4, column (4) reports the estimates for the speciﬁcation including the
proxy variables. Each proxy variable estimate is statistically insigniﬁcant, and an F test
of the joint signiﬁcance of all proxy variables produces a small F statistic of 0.32, with a
p-value of 0.90. The addition of the proxy variables leads to an increase from 3.16 to
3.87 in the retirement education variable, with a p-value of 0.20 and a t-statistic of 1.3.

Whether retirement education is exogenous in the model is unclear, even after the

addition of the proxy variables. The addition of the proxies increases the estimate on

109

retirement education, suggesting downward selection bias in the original model.
However, the proxies themselves are statistically insigniﬁcant, and the relative magnitude

of the change in the retirement education variable is not very large.

Employer and Non-Employer Sponsored Retirement Classes

Performing instrumental variable (IV) estimation would be the next logical step in
addressing the potential endogeneity of the retirement education variable. However, an
instrumental variable that satisﬁes the necessary conditions for consistent estimation does
not present itself. As an alternative, I examine whether attending an employer-sponsored
class is motivated by saving tastes. 1 specify a model in which the retirement meeting
variable, R, is replaced by two different variables, depending on whether the individual

took the class from his employer or not:

A = XB+ 5E + 0N +Pw + u. (2)

A, X, and P contain the same variables as in equation (1). E is equal to 1 if the individual
took an employer-sponsored class, 0 otherwise. N is equal to 1 if the individual took a
non-employer—sponsored (henceforth known as outside) class, 0 otherwise.

Existing literature on retirement education focuses on employer-sponsored
programs. There are several reasons for this. These programs are more homogenous in
content and structure than the array of available private ﬁnancial planners and community

college classes. In addition, private retirement education usually requires payment and

 

7" Whether the individual has any non-pension stock or not is perhaps a better proxy than amount of stock.
I evaluate the model replacing only the amount of stock with a binary variable expressing whether the
individual has outside stock holdings and ﬁnd no statistical difference in the estimates.

110

time outside of work, which are in smaller supply to low-income workers. Employers
educating their employees about saving and retirement may be a way for ﬁnancial
education to reach those who need to save for retirement the most.

I can reasonably assert that attending an outside retirement class has a greater
likelihood of being motivated by individual saving tastes than do employer-sponsored
classes, because the individual must be motivated by some force to seek out the class in
the ﬁrst place.75 Employers, on the other hand, often require these classes in conjunction
with entering a DC plan or offer them conveniently to employees during work hours.

Starting with the assumption that attending an outside meeting is not exogenous, I
use equation (2) to test if effects of employer-sponsored meetings on DC asset allocation
are statistically equal to the effects of outside meetings. I report the retirement education
coefﬁcient estimates in the ﬁrst and second rows of Table 2.6, and the difference between
the two in the third row. Each column uses the same variables as in columns (1)-(4) of
Table 2.4. The difference is not statistically signiﬁcant in any of the speciﬁcations,
including the model with the proxy variables. Therefore, I cannot reject the null
hypothesis that the estimates are statistically the same.

I cannot conclude from this comparison that the retirement meeting variable in
equation (1) is not endogenous, as I am simply comparing two broad types of retirement
education. However, this result does lend evidence to the conclusion by Bemheim and
Garrett (1996) that using employer-sponsored education may be motivated by

unobservable individual characteristics that also affect saving behavior. As more

 

75 It could be argued that those with particular tastes for saving demand education from their employers.
Bayer, Bemheim and Scholz (1996) examine the motivation behind employers offering retirement
education and conclude that it is low participation by lower-paid workers - not employee demand for
classes - that is the prime motivator.

111

employers decide to offer retirement education, its effectiveness in changing behavior is
likely to be an ongoing topic of study - one in which employee motivations for

attendance must be taken into consideration.

The Effect of Retirement Education by Individual Characteristics

Participants may react to retirement education differently depending on individual
characteristics. Individuals with different time horizons may follow traditional advice
and allocate their assets depending upon their stage in the life cycle. Likewise,
participants with different degrees of risk aversion may take ﬁnancial advice and tailor
their investments to the amount of risk and return they are willing to accept.76 In order to
test these differences econometrically, I estimate model (1) and include the interaction
terms R*T and R*V. I estimate the model with and without proxy variables.

In Table 2.7, I report OLS estimates of the effect of retirement meetings on
percent invested in equities, by years until retirement and degree of risk aversion. I
report the estimates using OLS without proxy variables in column (1), and with proxy
variables in column (2).77 Those with a high degree of risk aversion allocate 20 to 21
percentage points more to equity if they attend a retirement class, with p-values of 0.006
and 0.004 for OLS with and without proxies, respectively. The estimates on the other
measures of risk aversion and years until retirement do not appear to follow any

particular pattern, and all are statistically insigniﬁcant.

 

76 Participants could also respond differently to retirement education based upon characteristics such as
gender, education, and income. I estimate many different speciﬁcations using different interaction terms,
and do not ﬁnd adding any other interactions to be of additional explanatory value (not reported in tables).
77 The effect of retirement education for the variable less than 5 years until retirement is calculated by
evaluating the linear combination: or + to. + (N00. + (N2)¢2 +(N3)¢3 , where NJ- is the sample mean of the
corresponding degree of risk aversion variable. I calculate the effects for the other 7 groups in a similar

112

In addition to reporting the effects of retirement education separately by years
until retirement and degree of risk aversion, I look at the effects for groups, each
possessing different combinations of these characteristics. In Table 2.8, column (1), I
report OLS estimates for eight different types of DC participants, each group having a
different combination of time horizon and risk aversion characteristics. Because I will
discuss only the results for those with extreme and high degrees of risk aversion, and
because the estimates for the eight groups that each have either a moderate or low
aversion to risk are all statistically insigniﬁcant, I omit these groups from the table.

All 4 groups in Table 2.8 that have high degrees of risk aversion increase their
holdings of stock if they took a retirement class. All coefﬁcients are estimated with a p-
value of at the most 10 percent, with the majority of the estimates statistically signiﬁcant
at less than the 5 percent level. A general pattern of stock investment consistent with life
cycle predictions is evident if the individual has a high degree of risk aversion, although
the effect is not linear. As is seen in column (1), those expecting to retire in 8 or more
years invest from 20 to 28 percentage points more in equity if they attend a meeting,
while those with up to 8 years until retirement invest approximately 16 to 18 percent
more in equity.

These effects are quite large for those with less than 8 years until retirement. It is
possible that individuals are looking past the actual time of retirement in their saving
horizons. Also, employers are only recently realizing that retirement age is becoming
more variable, as some employees are choosing to retire early while others choose to

work into their late 603 and 703. Temeoy (1995) reports that in the past, speciﬁc

 

manner, only using the mean value of each years until retirement variable when evaluating the effect of
degree of risk aversion.

113

retirement counseling at employers was done at the age of 55 or older, but more recently
many employers are starting retirement counseling around age 50.

Why are only those with high levels of risk aversion - and not those with extreme
levels - responding to retirement education? One explanation may be related to the
curvature of each group's utility function. In Section 3, I discuss Kritzrnan's (1994)
theory on why certain individuals will not invest in a risky asset, no matter how long their
time horizon. The theory proposes that for certain individuals with extremely concave
utility functions, the potential disutility associated with losing even a little wealth is
greater than the potential gains that could be had from more risk and larger returns.

It may be the individuals with extreme levels of risk aversion that display
behavior similar to that described in Kritzrnan's model. Regardless of any knowledge
pertaining to holding period and risk learned at a retirement class, the participant does not
want to hold risky assets. On the other hand, those with relatively less aversion to risk -
or those labeled as having high levels of risk aversion - respond to information at a class

concerning holding risky assets and choose to hold more equities.

6. CONCLUSION

This paper is the ﬁrst study to evaluate the effects of taking a retirement class on
asset allocation within a self-directed DC plan. I ﬁnd that retirement education
substantially increases the percentage invested in equities for those with a high level of
risk aversion, regardless of the individual retirement time horizon. I ﬁnd no effect of
retirement education on participant equity choice generally. Because the HRS only
surveys individuals ages 51 to 61, these results do not generalize to all DC participants

with choice in their pension plans.

114

It is pertinent to point out certain factors that may affect the results and
implications of this analysis. First of all, the HRS does not report the year in which the
participant took the class. Hurt (1998) reports that classes began to consist of more
detailed investment information beginning only in the late 1980s. The preliminary
release in 1987 and ﬁnal release in 1992 of section 404(c) guidelines also may have
affected the content of classes over time. Before guidelines were speciﬁed to restrict
legal liability for employers, asset allocation information may have been minimized to
avoid being interpreted as investment advice.

Characteristics of the organization offering the class, or even of the instructor, can
also vary signiﬁcantly. Although asset allocation and risk/return issues are among the
topics most often covered in an employer-sponsored class (EBRI, 1995), Goodfellow and
Schieber (1997) ﬁnd signiﬁcant variation in asset allocation for participants with similar
characteristics among different plans. This suggests that perhaps there are other
characteristics of classes besides content, such as the method by which the information is
transmitted, that may differ across classes.

Bias in the estimates due to selection is another possible - although not probable -
factor. Although my estimates on the retirement education variable do increase with the
addition of the proxy variables, the increase is small and the estimate is statistically
insigniﬁcant in each speciﬁcation. The proxy variables are also largely jointly
insigniﬁcant, suggesting they do not explain much (if any) variation in pension portfolios.
Hence the existing evidence suggests that selection bias is not affecting the retirement
education estimates. However, without further means to investigate the possibility of

bias (such as IV estimation), 1 can only come to this conclusion with a moderate amount

115

of certainty.

Lastly, the content of the HRS data and the structure of the questions regarding
both retirement education and DC asset allocation may affect the results. The HRS does
not report the number of investment choices the participant has to choose from, or what
those choices are. VanDerhei et al. (2000) report that the options offered to the employee
affect the allocation he chooses. The HRS also does not identify whether the employer
match is mandated in company stock. I am able to identify those plans that have a greater
probability of containing this mandate (by ﬂagging those plans of the proﬁt sharing type),
but not all plans of this type have this feature.

Perhaps the most likely reason for the lack of effect of retirement classes on asset
allocation is the manner in which the HRS presents the question regarding pension asset
allocation to its survey participants. If individuals decide to rearrange their portfolios
after taking a class, small shifts, rather than large movements, of assets are more likely.
Thus the appearance of no shift in assets for those who attended a meeting may be a
function of the inability of the model to pick up small changes in investment patterns.
The fact that the model may not be able to pick up these small changes does not mean the
results tell us nothing about retirement education and asset allocation, however. Rather,
the fact that there is no effect in this model speaks to what retirement education does not
do for the average participant - and that is cause large shifts in equity within the pension

account.

116

REFERENCES

Access Research, Inc., “Participant Attitudes and Behavior,” Report prepared for
members of the Society of Professional Administrators and Recordkeepers. Windsor:
Access Research, 1995.

Alliance Against Fraud in Telemarketing, Top Emerging Scams of 1992. Washington,
DC: Alliance Against Fraud in Telemarketing, Winter 1992.

 

Bayer, Patrick, B. Douglas Bemheim, and John Karl Scholz, “The Effects of Financial
Education in the Workplace: Evidence from a Survey of Employers,” National Bureau of
Economic Research Working Paper no. 5655, July 1996.

Bemheim, B. Douglas, "Personal Saving, Information, and Economic Literacy: New
Directions for Public Policy," in Tax Policy for Economic Growth in the 19903,
Washington, DC: American Council for Capital Formation, 1994a, pp. 53-78.

 

Bemheim, B. Douglas, and Daniel M. Garrett, “The Determinants and Consequences of
Financial Education in the Workplace: Evidence from a Survey of Households,” National
Bureau of Economic Research Working Paper no. 5667, July 1996.

Bemheim, B. Douglas, and John Karl Scholz, "Private Saving and Public Policy,"
National Bureau of Economic Research Working Paper no. 4215, 1992.

Bodie, Zvi, "On the Risk of Stocks in the Long Run," Financial Analysts Journal, 51:
May/June 1995, pp. 18-22.

 

Bodie, Zvi, and Dwight B. Crane, "Personal Investing: Advice, Theory, and Evidence,"
Financial Analysts Journal, November/December 1997, pp. 13-23.

 

Bodie, Zvi, Merton, RC, and W.F. Samuelson, "Labor Supply Flexibility and Portfolio
Choice in a Life Cycle Model, " Journal of Economic Dynamics and Control, 16(3-4):
July-October 1992, pp. 427-49.

 

Buck Consultants, 401(k) Plans, Employer Practices and Policies, A Survey Report -
September 1994. New York: Buck Consultants, Inc., 1994.

 

 

Dickemper, Jennifer, and Paul Yakoboski, “Increased Saving but Little Planning: Results
of the 1997 Retirement Conﬁdence Survey,” Employee Beneﬁt Issue Brief no.191,
November 1997.

Employee Beneﬁt Research Institute, The 1999 Retirement Conﬁdence Survey (RCS)
Summary of Findings. Washington, DC: Employee Beneﬁt Research Institute, 1999.

 

 

117

REFERENCES (continued)

Employee Beneﬁt Research Institute, “Can We Save Enough to Retire? Participant
Education in Deﬁned Contribution Plans,” Employee Beneﬁt Research Institute Issue
Brief no. 160, April 1995.

Employee Beneﬁt Research Institute, “Participant Education: Actions and Outcomes,”
Employee Beneﬁt Research Institute Brief no. 169, January 1996.

Employee Beneﬁt Research Institute, Public Attitudes on Investment Preferences .
Washington DC: Employee Beneﬁt Research Institute, 1993.

 

Employee Beneﬁt Research Institute, "Worker Investment Decisions: An Analysis of
Large 401(k) Plan Data," Employee Beneﬁt Research Institute Issue Brief no. 176,
August, 1996.

Employee Beneﬁt Research Institute/Gallup Organization Inc., "Public Attitudes on
Investment Preferences," Employee Beneﬁt Research Institute Report no. G-61,
November 1994.

Galer, R., Quick, C., Rea, J. and Jack VanDerhei, "401(k) Plan Asset Allocation, Account
Balances, and Loan Activity," Employee Beneﬁt Research Institute Issue Brief no. 205,
January 1999.

Goodfellow, GR, and SJ. Schieber, "Investments of Assets in Self-Directed Retirement
Plans," in Positioning Pensions for the Twenty-ﬁrst Century, edited by MS. Gordon,
O.S., Mitchell, and M.M. Twinney, 67-90. University of Pennsylvania Press, 1997.

 

Hinz, Richard P., McCarthy, David D. and John A. Turner, "Are Women More
Conservative Investors? Gender Differences in Participant-Directed Pension
Investments," in Positioning Pensions for the Twenty-ﬁrst Century, edited by MS.
Gordon, O.S., Mitchell, and M.M. Twinney, 67-90. University of Pennsylvania Press,
1997.

 

Hurt, Ronald D., "The Changing Paradigm of 401(k) Plan Servicing," in Living with
Deﬁned Contribution Pensions: Remaking Regonsibility for Retirement, edited by
Olivia S. Mitchell and Sylvester J. Schieber, 193-207. Philadelphia, Pennsylvania:
Pension Research Council, the Wharton School of the University of Pennsylvania, and
the University of Pennsylvania Press, 1998.

 

Institute of Management and Administration, Deﬁned Contribution Plan Investing.
Washington, D.C., April 25, 1995.

 

J ianakopolos, Nancy and Alexandra Bemasek, "Are Women More Risk Averse?" Paper

presented at the Western Economics Association annual meeting, San Diego, California,
June 1994.

118

REFERENCES (continued)

Kritzman, Mark, "What Practitioners Need to Know About Time Diversiﬁcation,"
Financial Analysts Journal, January/February 1994, p. 15.

 

Oppenheimer Funds, Restoringthe Dream of a Comfortable Retirement. New York:
Oppenheimer Funds Distributor, 1994.

 

Papke, Leslie B, "How Are Participants Investing Their Accounts in Participant-Directed
Individual Account Pension Plans?" Michigan State University Working Paper, 1998.

Phoenix-Hecht, Plan Sponsor Survey: Large Deﬁned Contribution Plans Respondent
Report. Research Triangle Park, NC: Phoenix-Hecht, 1994a.

 

Poterba, James M., and David A. Wise, "Individual Financial Decisions in Retirement
Saving Plans and the Provision of Resources for Retirement," in Privatizing Social
Security, edited by Martin F eldstein, 85-105. Chicago: University of Chicago Press,
1998.

 

Samuelson, Paul A., "Asset Allocation Could Be Dangerous to Your Health," Journal of
Portfolio Management, 16: Spring 1990, pp. 5-8.

 

Samuelson, Paul A., "The Judgment of Economic Science on Rational Portfolio
Management: Indexing Timing and Long-Horizon Effects," Journal of Portfolio
Management, 16(1): Fall 1989, pp. 4-12.

 

 

Smith, Roger, "DC Plans Accepted, Assets Growing...What's the Problem?" Pension
World, February 1993, pp. 28-30.

United States Department of Labor, Pension and Welfare Beneﬁts Administration,
Abstract of 1991 Form 5500 Annual Reports. Private Pension Plan Bulletin No. 4,
Washington, DC: United States Government Printing Ofﬁce, Winter 1995.

 

United States Department of Labor, Bureau of Labor Statistics, “Employee Beneﬁts in

Medium and Large Firms,” Washington, DC: United States Government Printing
Ofﬁce, 1993. '

VanDerhei, Jack, Holden, Sarah and Carol Quick, "401(k) Plan Asset Allocation,
Account Balances, and Loan Activity in 1998," Employee Beneﬁt Research Institute
Issue Brief No. 218, February 2000.

Warshawsky, Mark J ., "Funding of Deﬁned Beneﬁt Plans," in Positioning Pensions for
the Twenty-ﬁrst Century, edited by MS. Gordon, O.S., Mitchell, and M.M. Twinney,
107-138. University of Pennsylvania Press, 1997.

 

 

119

REFERENCES (continued)

Weisbenner, Scott, "Do Pension Plans with Participant Investment Choice Teach
Households to Equitize?" Dissertation. Massachusetts Institute of Technology, 1999.

Wyatt Company, Wyatt Insider. October 1994, p. 8.

 

Yakoboski, Paul, “Participant-Directed Retirement Plans Today and Critical Issues for
Tomorrow,” in When Workers Call the Shots: Can They Achieve Retirement Security?
edited by Dallas L. Salisbury, 9-22. Washington, DC: Employee Beneﬁt Research
Institute, 1995.

 

120

TABLE 2.1

Summary Statistics

 

 

 

 

 

Variable Mean Standard Plans/
Deviation Participants
Percent stock in pension plan 0.51 0.38 1187
Attended a retirement meeting 0.29 0.45 1108
Attended an employer-sponsored meeting 0.79 0.41 319
Attended a non employer-sponsored meeting 0.21 0.41 319
Female 0.46 0.50 1108
Black/other 0.16 0.36 1107
Married 0.82 0.38 1108
Age 54 4.9 1108
High school degree 0.92 0.27 1108
Income < $25K 0.06 0.24 1108
Income between $25 and $50K 0.29 0.45 1108
Income between $50 and $100K 0.46 0.50 1108
Income over $100K 0.19 0.39 1108
Net worth less than $50K 0.15 0.36 1108
Net worth between $50 and $100K 0.18 0.38 1108
Net worth between $100 and $250K 0.38 0.48 1108
Net worth between $250 and $500K 0.19 0.40 1108
Net worth over $500K 0.10 0.29 1108
Has proﬁt sharing plan 0.14 0.35 1160
Amount of non-pension equity $29,356 $106,522 1108
Amount in plan $52,335 $111,755 924
Has other DC or DB pension ~ 0.71 0.45 1108
Years expected until retirement 8.5 5.3 1039
Years in plan 7.7 6.5 1174
Employer matches a percentage of contribution 0.68 0.46 1041
Extreme degree of risk aversion 0.64 0.48 1098
High degree of risk aversion 0.14 0.35 1098
Moderate degree of risk aversion 0.11 0.31 1098
Low degree of risk aversion 0.11 0.31 1098
Has at least 50 percent chance of living until age 0.35 0.48 1099
85
Financial planning period a few months-few years 0.50 0.50 1093
Financial planning period 5-10 years 0.39 0.49 1093
Financial planning period greater than 10 years 0.11 0.31 1093
High expectation that Social Security beneﬁts will 0.55 0.50 1101

decline in next 10 years

 

121

 

TABLE 2.1 (continued)

 

Variable Mean Standard Plans/
Deviation Participants
Low expectation that Social Security beneﬁts will 0.18 0.38 1101

decline in next 10 years

 

Total number of plans in the sample is 1,322, with 1,236 participants in 1,130
families. Variables describing each plan (percent stock, proﬁt sharing plan, amount in
plan, years in plan, employer matches) are calculated using the total number of plans. All
other variables describe the participant and are calculated from the number of
participants. In both cases, observations are restricted to those individuals that reported
how DC assets are allocated and if they attended a retirement meeting.

Statistics for type of meeting are conditional on having attended a meeting. The
variable has other DB or DC plan describes those participants who have at least one DB
plan from current or past employment, other self-directed DC plans, non-self-directed DC
plans or plans that have both DC and DB characteristics.

122

TABLE 2.2
Proportion of Participants with Each Type
of Asset Allocation, by Individual Characteristics

Stocks Mixed Bonds Number of

 

 

Plans
GI’OUP ( 1) (2) (3) (4)
All plans .28 .45 .27 1108
Age
<50 years .26 .50 .24 167
50-55 years .28 .47 .25 523
55-60 years .27 .40 .33 327
>=60 years .29 .43 .28 91
Female .23** .48* .29 505
Male .31* * .42* .27 603
At least a high school .28 .46 .26** 1020
degree
No high school degree .25 .36 .39** 88
Income
<$25K .21 * * .42 .37 70
$25-50K .25 .44 .31 315
$50-100K .28 .46 .26 514
>=$100K .33 * * .45 .22 209
Years until retirement
<5 years .28 .39** .33** 260
5-8 years .27 .45 .28 233
8-11 years .29 .43 .28 219
>=ll years .28 .50** .22** 327
Risk aversion
Extreme degree .27 .44 .29 704
High degree .24 .50 .26 155
Moderate degree .30 .49 .21 121
Low degree .32 .42 .26 1 18

 

Participants not indicating how their assets are allocated or those not answering the
retirement education question are excluded. Columns (1)-(3) report the proportion of
participants who invested in mostly stocks, mixed, or mostly bonds. For the 7 percent of
participants with more than one self-directed plan, I use only the allocation for the
primary self-directed plan.

Statistical signiﬁcance for income refers to the difference between less than $25K
and equal to or more than $100K, and for years until retirement it is the difference
between less than 5 years and 11 years and over.

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

123

Table 2.3
Proportion of Participants Who Attended a Retirement Meeting
and Respective Asset Allocation, by Individual Characteristics

Attended Stocks Mixed Bonds Number of

 

 

 

 

a Meeting Plans
GrOUQ (1) (2) (3) (4) (5)
All participants .29 .28 .45 .27 l 108
Age
<50 years .23** .26 .50 .24 167
50-55 years .27 .28 .47 .25 523
55-60 years .33 .27 .40 .33 327
>=60 years .40** .29 .43 .28 91
Female .27 .23 .48 .29 505
Male .31 .31 .42 .27 603
At least a high school .30 .28 .46 .26 1020
degree
No high school degree .23 .25 .36 .39 88
Income
<$25K .21* .21 .42 .37 70
$25-50K .27 .25 .44 .31 315
$50-100K .31 .28 .46 .26 514
>=$100K .31* .33 .45 .22 209
Years until retirement
<5 years .41 ** .28 .39 .33 260
5-8 years .31 .27 .45 .28 233
8-11 years .23 .29 .43 .28 219
>=11 years .23** .28 .50 .22 327
Risk aversion
Extreme degree .29 .27 .44 .29 704
High degree .26 .24 .50 .26 155
Moderate degree .27 .30 .49 .21 121
Low degree .35 .32 .42 .26 118

 

Participants not indicating how their assets are allocated or those not answering the
retirement education question are excluded. I report the proportion of the sample that
attended a retirement class in Column (1). Columns (2)-(4) report the proportion of
participants who invested in mostly stocks, mixed, or mostly bonds. For the 7 percent of
participants with more than one self-directed plan, I include the allocation for the primary
self-directed plan.

Statistical signiﬁcance for income refers to the difference between less than $25K
and equal to or more than $100K, for years until retirement it is the difference between
less than 5 years and 11 years and over, and for age it refers to the difference between
less than 50 years old and equal to or greater than 60 years old.

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

124

TABLE 2.4
Percentage Point Increase in Stock if Attended a Retirement Class, by Plan

 

 

 

OLS Estimates
Dependent Variable: Stocks, ( 1) (2) (3) (4) (5)
Bonds, Mix
Attended a retirement meeting 1.79 1.28 3.16 3.87 2.70
(2.43) (2.46) (3.16) (3 .03) (9.42)
Female -5.93** -6.45** -6.47** -6.28**
(2.37) (2.89) (2.94) (2.87)
Black/other 1.21 5.18 5.26 4.45
(2.98) (3.70) (3.70) (3.64)
Married -0.72 -3.56 -3.91 -3.82
(3.19) (3.78) (3.81) (3.81)
Age -0.43* -0.06 -0.08 -0.009
(0.24) (0.37) (0.37) (0.37)
High school degree 5.00 8.72 8.48 8.59
(4.6) (5.48) (5.49) (5.50)
Income between $25 and $50K 3.56 2.48 2.06 3.16
(5.22) (7.44) (7.50) (7.38)
Income between $50 and $100K 5.72 7.43 7.45 7.93
(5.39) (7.53) (7.57) (7.45)
Income over $100K 10.41* 5.60 6.25 5.96
(5.94) (8.28) (8.31) (8.18)
Net worth between $50 and 672* 7.38 7.48 7.20
$100K (3.88) (4.76) (4.81) (4.74)
Net worth between $100 and 5.40* 9.66** 9.77** 10.04“
$250K (3.55) (4.31) (4.35) (4.33)
Net worth between $250 and 7.67* 12.80“ 13.20** 12.72**
$500K (4.14) (5.13) (5.20) (5.19)
Net worth over $500K 2.23 0.16 0.14 0.88
(5.12) (6.92) (7.03) (7.08)
Has proﬁt sharing plan 18.02** 18.39** 18.26**
(4.08) (4.09) (4.11)
Amount of non-pension stock 0.19 0.21* 0.20
holdings (0.12) (0.12) (0.12)
Balance in account 0.21 0.20 0.23
(0.15) (0.14) (0.14)
Has DB or other DC plan 3.30 3.05 4.02
(3.30) (3.35) (3.33)
Years until retirement < 5 -4.54 -5.16 -3.40
(4.54) (4.64) (5.41)

 

125

 

TABLE 2.4 (continued)

(1) (2)
Years until retirement between 5
and 8
Years until retirement between 8
and 11
Years in plan

Employer offers match

Extreme degree of risk aversion
High degree of risk aversion
Moderate degree of risk aversion

Has at least 50 percent chance of
living until age 85

70-100 percent chance Social
Security beneﬁts will decrease in
next 10 years

40-70 percent chance Social
Security beneﬁts will decrease in
next 10 years

Financial planning period 5-10
years

Financial planning period greater
than 10 years

Retirement meeting*years until
retirement <5

Retirement meeting*years until
retirement between 5 and 8
Retirement meeting*years until
retirement between 8 and 11
Retirement meeting*extreme
degree of risk aversion
Retirement meeting*high degree
of risk aversion

Retirement meeting*moderate
degree of risk aversion

(3)
-451
(4.31)
-0.15
(3.86)
-031
(0.22)
15.16**
(3.01 )
-5.17
(5.48)
-6.26
(4.68)
2.78
(5.80)

(4)
4.51
(4.31 )
0.81
(3.89)
-0.36
(0.23)
14.41**
(3.04)
-6.19
(4.74)
-4.22
(5.52)
3.13
(5.84)
1.52
(2.81)
-2.80
(3.64)

-139
(4.03)

-2.19
(3.01)
-1.75
(4.85)

(5)
-3.50
(5.06)
-1.48
(4.46)
-033
(0.23)
15.32**
(3.00)
-6.20
(5.94)
-1032
(6.75)
5.04

(11.42)
-10.62
(12.62)

 

126

 

TABLE 2.4 (continued)

(1) (2) (3) (4) (5)
Constant 50.66” 61.18** 30.13 34.5 27.18
(1.34) (15.20) (22.70) (23.02) (22.95)
R2 0.0005 0.02 0.113 0.115 0.123
Number of plans 1187 1186 763 757 763
Number of participants 1 108 1 107 727 722 727

 

(1) mean regression; (2) add demographic and ﬁnancial variables; (3) add other
control variables; (4) add proxy variables; (5) add interaction terms to (3). Amount of
non-pension stock holdings and balance in account are expressed in $10,000 1993
dollars.

P-values for F tests of joint signiﬁcance: column (3): income, 0.50; net worth, 0.04;
risk aversion, 0.13; years until retirement, 0.58; column (4): all proxies, 0.90; column (5):
interaction terms for degree of risk aversion, 0.07; interaction terms for years until
retirement, 0.59

Robust standard errors are corrected for family and individual-level correlation.

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

127

Ordered Logit Estimates

TABLE 2.5
Percentage Point Increase in Stock if Attended a Retirement Class, by Plan.

 

 

 

Dependent Variable: Stocks, (1) (2) (3) (4) (5)
Bonds, Mix
Attended a retirement meeting 0.09 0.07 0.14 0.18 0.08
(0.12) (0.12) (0.16) (0.16) (0.51)
Female -0.30** -0.33** -0.33** -0.33**
(0.12) (0.16) (0.16) (0.16)
Black/other 0.05 0.24 0.25 0.21
(0.15) (0.20) (0.20) (0.20)
Married -0.04 -0.16 -0.18 -0.18
(0.16) (0.21) (0.22) (0.21)
Age -0.02* -0.004 -0.005 -0.001
(0.01) (0.02) (0.02) (0.02)
High school degree 0.27 0.47 0.46 0.46
(0.22) (0.30) (0.30) (0.30)
Income between $25 and $50K 0.19 0.08 0.05 0.11
(0.26) (0.38) (0.39) (0.38)
Income between $50 and 0.30 0.35 0.34 0.38
$100K (0.26) (0.39) (0.39) (0.39)
Income over $100K 0.53* 0.27 0.30 0.29
(0.30) (0.43) (0.43) (0.43)
Net worth between $50 and 0.33* 0.34 0.35 0.34
$100K (0.19) (0.26) (0.26) (0.26)
Net worth between $100 and 0.32* 0.45* 0.46* 0.48**
$250K (0.18) (0.23) (0.24) (0.24)
Net worth between $250 and 0.38* 0.58** 0.61 ** 0.58**
$500K (0.21) (0.27) (0.27) (0.28)
Net worth over $500K 0.10 -0.19 -0.20 -0.16
(0.25) (0.37) (0.38) (0.38)
Has proﬁt sharing plan 0.99** 1.02** 1.02**
(0.23) (0.23) (0.23)
Amount of non-pension stock 0.0016* 0002* 0.002*
holdings (0.0009) (0.0009) (0.0009)
Balance in account 0.0011 0.0011 0.0012
(0.0008) (0.0008) (0.0008)
Has DB or other DC plan 0.18 0.16 0.22
(0.17) (0.18) (0.18)
Years until retirement < 5 -0.20 -0.23 -0.13
(0.24) (0.24) (0.28)

 

128

 

TABLE 2.5 (continued)

Years until retirement between
5 and 8
Years until retirement between

8 and 11
Years in plan

Employer offers match
Extreme degree of risk aversion
High degree of risk aversion

Moderate degree of risk
aversion

Has at least 50 percent chance
of living until age 85

70-100 percent chance Social
Security beneﬁts will decrease
in next 10 years

40-70 percent chance Social
Security beneﬁts will decrease
in next 10 years

Financial planning period 5-10
years

Financial planning period
greater than 10 years
Retirement meeting*years until
retirement <5

Retirement meeting*years until
retirement between 5 and 8
Retirement meeting*years until
retirement between 8 and 11
Retirement meeting* extreme
degree of risk aversion
Retirement meeting*high
degree of risk aversion

(1)

(2)

(3)
-0.19
(0.22)
0.01
(0.21)
-0.02
(0.01)
0.81“
(0.16)
-029
(0.28)
-034
(0.24)
0.16
(0.30)

(4)
-019
(0.22)
0.07
(0.21)
-002
(0.01)
0.78**
(0.16)
-034
(0.24)
-0.24
(0.28)
0.17
(0.30)
0.07
(0.15)
-0.17
(0.20)

-009
(0.22)

-0.1 1
(0.16)
-0.06

(0.24)

(5)
-0.16
(0.26)
-0.07
(0.24)
-002
(0.01)
0.82**
(0.16)
-O.36
(0.29)
-0.58*
(0.34)
0.28

 

129

 

TABLE 2.5 (continued)

(1) (2) (3) (4) (5)

Retirement meeting*moderate -0.55
degree of risk aversion (0.66)
Number of plans 1 187 1 186 763 757 763
Number of participants 1108 l 107 727 722 727

 

(1) mean regression; (2) add demographic and ﬁnancial variables; (3) add other control
variables; (4) add proxy variables; (5) add interaction terms to (3).

Standard errors in parentheses.

** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

130

TABLE 2.6
Effects of Attending an Employer-Sponsored Class or
an Outside Retirement Class on Asset Allocation, by Plan

 

Type of Meeting (1) (2) (3) (4) Number

 

 

of Plans
Employer- 1.19 1.12 3.20 3.80 271
sponsored (2.72) (2.75) (3.23) (3.24)
Non employer— 4.14 2.40 5.41 6.83 73
sponsored (4.46) (4.50) (5.93) (6.0)
Difference 2.95 1.27 2.21 3.04 -----

(4.89) (4.91) (6.32) (6.36)
Speciﬁcations in columns (1)-(4) correspond to the same columns in Table 2.4,

except for the replacement of the more general retirement education variable with
employer and non employer-sponsored education variables.

(1) mean regression; (2) add demographic and ﬁnancial variables; (3) add other

control variables; (4) add proxy variables.

Robust standard errors in parenthesis. Standard errors are corrected for family and
individual-level correlation.
** signiﬁcant at the 5 percent level

signiﬁcant at the 10 percent level

131

TABLE 2.7

Percentage Point Increase in Stock if Attended a Retirement Class
by Years until Retirement and Degree of Risk Aversion, separately

 

 

 

mp E OLS with Number
m LPLZLHE
(1) (2) (3)
Years until retirement < 5 0.64 1.80 278
(5.61) (5.64)
Years until retirement 5 to 8 -0.74 -0.59 258
(5.95) (6.00)
Years until retirement 8 to 11 10.86 10.41 236
(6.83) (6.93)
Years until retirement > 11 3.32 3.95 349
(5.32) (5.41)
Extreme degree of risk aversion 1.32 1.60 751
(3.73) (3.80)
High degree of risk aversion 20.40** 21 .62** 170
(7.44) (7.49)
Moderate degree of risk aversion -7.92 -6.68 130
(8.92) (8.98)
Low degree of risk aversion 2.70 2.11 123
(8.69) (8.17)

 

Robust standard errors in parenthesis. Standard errors are corrected for family and

individual-level correlation.
** signiﬁcant at the 5 percent level
* signiﬁcant at the 10 percent level

132

TABLE 2.8
Percentage Point Increase in Stock if Attended a Retirement Class
by Years until Retirement and Degree of Risk Aversion, together

 

 

 

Group O_L_S OLS with Plans
Proxies

(1) (2) (3)
Extreme degree of risk aversion / less than 5 years until -1.36 -0.35 188
retirement (5.99) (6.04)
Extreme degree of risk aversion / 5 to 8 years until -2.75 -2.75 178
retirement (6.23) (6.30)
Extreme degree of risk aversion / 8 to 11 years until 8.80 8.26 146
retirement (7.08) (7.1 8)
Extreme degree of risk aversion / greater than 11 years 1.32 1.79 194
until retirement (6.03) (6. 13)
High degree of risk aversion / less than 5 years until 17.71 ** 19.68” 39
retirement (8.95) (8.99)

High degree of risk aversion / 5 to 8 years until retirement 16.33* 17.28* 31
(9.04) (9.07)

High degree of risk aversion / 8 to 11 years until retirement 27.93** 28.28“ 33
(9.94) (10.03)

High degree of risk aversion / greater than 11 years until 20.39** 21 .82** 57

retirement (8.37) (8.45)

 

Robust standard errors in parenthesis. Standard errors are corrected for family and
individual-level correlation.

** signiﬁcant at the 5 percent level

* signiﬁcant at the 10 percent level

133

"llllllllllllllllllllf