<%BANNER%>

Essays in Retirement Plans and Mutual Funds

MISSING IMAGE

Material Information

Title:
Essays in Retirement Plans and Mutual Funds
Physical Description:
1 online resource (92 p.)
Language:
english
Creator:
Sardarli, Sabuhi H
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Business Administration, Finance, Insurance and Real Estate
Committee Chair:
Ryngaert, Michael David
Committee Co-Chair:
Naranjo, Andy
Committee Members:
Houston, Joel F
Hamersma, Sarah Ellen

Subjects

Subjects / Keywords:
401k -- flows -- retirement -- spanning -- style
Finance, Insurance and Real Estate -- Dissertations, Academic -- UF
Genre:
Business Administration thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
In this study, I examine two different but related topics in finance.  The first part of the study examines the impact of defined contribution (DC) retirement plan investors on the performance-flow relationship of mutual funds.  Investors who hold mutual funds primarily through their DC plans are often deemed less sophisticated and less active in managing their DC accounts than retail mutual fund investors.  Considering a significant rise in the size of DC plans and mutual funds’ weight within these plans, I examine the effect of the percentage of a fund’s total net assets coming from DC money on the performance-flow relationship of mutual funds. I find that a higher proportion of a mutual fund’s assets held in these plans decreases the sensitivity of the fund’s flows to past performance. This effect is observed for both recent good performing and bad performing funds,although the size of this effect is larger for recent poor performers. I also find that while flows are negatively affected by return volatility for funds with low DC plan ownership, this impact decreases as the ratio of a fund’s assets held by DC plans increases.  The second part of the study focuses on efficiency tests on retirement plan menus.  I illustrate how testing for efficiency in 401(k) plan offerings depends on the benchmarks used.  I also show that, for specific benchmarks, spanning can be attained without optimization. However, based on return style analysis, I show that, relative to these benchmarks, performance can be significantly improved through optimization.  My results are consistent with a recent study that suggests that smaller 401(k) menus are objectively more efficient and with a large literature that has noted participants' subjective propensity to prefer smaller options in decision making.  Here I make a strong case for plan sponsors to recommend specific funds to help with plan efficiency.  Plan participants can then adopt allocations with which they are comfortable.  This approach contrasts with recent recommendations for default strategies and educational programs targeting participants who are viewed as the main source of plan inefficiency.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Sabuhi H Sardarli.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: Ryngaert, Michael David.
Local:
Co-adviser: Naranjo, Andy.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-08-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0045908:00001

MISSING IMAGE

Material Information

Title:
Essays in Retirement Plans and Mutual Funds
Physical Description:
1 online resource (92 p.)
Language:
english
Creator:
Sardarli, Sabuhi H
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Business Administration, Finance, Insurance and Real Estate
Committee Chair:
Ryngaert, Michael David
Committee Co-Chair:
Naranjo, Andy
Committee Members:
Houston, Joel F
Hamersma, Sarah Ellen

Subjects

Subjects / Keywords:
401k -- flows -- retirement -- spanning -- style
Finance, Insurance and Real Estate -- Dissertations, Academic -- UF
Genre:
Business Administration thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
In this study, I examine two different but related topics in finance.  The first part of the study examines the impact of defined contribution (DC) retirement plan investors on the performance-flow relationship of mutual funds.  Investors who hold mutual funds primarily through their DC plans are often deemed less sophisticated and less active in managing their DC accounts than retail mutual fund investors.  Considering a significant rise in the size of DC plans and mutual funds’ weight within these plans, I examine the effect of the percentage of a fund’s total net assets coming from DC money on the performance-flow relationship of mutual funds. I find that a higher proportion of a mutual fund’s assets held in these plans decreases the sensitivity of the fund’s flows to past performance. This effect is observed for both recent good performing and bad performing funds,although the size of this effect is larger for recent poor performers. I also find that while flows are negatively affected by return volatility for funds with low DC plan ownership, this impact decreases as the ratio of a fund’s assets held by DC plans increases.  The second part of the study focuses on efficiency tests on retirement plan menus.  I illustrate how testing for efficiency in 401(k) plan offerings depends on the benchmarks used.  I also show that, for specific benchmarks, spanning can be attained without optimization. However, based on return style analysis, I show that, relative to these benchmarks, performance can be significantly improved through optimization.  My results are consistent with a recent study that suggests that smaller 401(k) menus are objectively more efficient and with a large literature that has noted participants' subjective propensity to prefer smaller options in decision making.  Here I make a strong case for plan sponsors to recommend specific funds to help with plan efficiency.  Plan participants can then adopt allocations with which they are comfortable.  This approach contrasts with recent recommendations for default strategies and educational programs targeting participants who are viewed as the main source of plan inefficiency.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Sabuhi H Sardarli.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: Ryngaert, Michael David.
Local:
Co-adviser: Naranjo, Andy.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-08-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0045908:00001


This item has the following downloads:


Full Text

PAGE 1

1 ESSAYS IN RETIREMENT PLANS AND MUTUAL FUNDS By SABUHI H SARDARLI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 3

PAGE 2

2 201 3 Sabuhi H Sardarli

PAGE 3

3 To my family without whom this would never be possible

PAGE 4

4 ACKNOWLEDGMENTS I would like to thank my parents for always making any sacrifice necessary to make sure that my sister and I get best education possible. Without their love, support and patience, this work would never be possible I thank my Committee Michael Ryngaert ( co c hair) Andy Naranjo ( co c hair) Joel Houston and Sarah H a mersma for their significant cont ributions as well as countless hours of support regarding my dissertation. I would like to thank my fellow f inance Ph D students especially Thomas Doellman, for their friendship and comrade ry.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF FIGURES ................................ ................................ ................................ .......... 6 LIST OF TABLES ................................ ................................ ................................ ............ 7 ABSTRACT ................................ ................................ ................................ ..................... 8 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 10 Overview of Chapter 2 ................................ ................................ ............................ 11 Overview of Chapter 3 ................................ ................................ ............................ 12 2 THE IMPACT OF DEFINED CONTRIBUTION INVESTMENTS ON THE PERFORMANCE FLOW OF MUTUAL FUNDS ................................ ..................... 14 Relevant Literature ................................ ................................ ................................ 18 Data Description ................................ ................................ ................................ ..... 23 The Effect of DC Investors on Performance Flow Relationship .............................. 27 Univariate Analysis ................................ ................................ ................................ 27 Multivariate Analysis ................................ ................................ ............................... 29 The Effect of Tax loss Selling ................................ ................................ ................. 39 Ch apter 2 Concluding Remarks ................................ ................................ .............. 41 3 EFFICIENCY AND STYLE IN 401(K) PLANS ................................ ......................... 51 Data ................................ ................................ ................................ ........................ 56 Spanning Tests ................................ ................................ ................................ ....... 61 Style Perspective ................................ ................................ ................................ .... 66 Chapter 3 Conc luding Remarks ................................ ................................ .............. 69 4 CONCLUSION ................................ ................................ ................................ ........ 77 APPENDIX: COMPARISON OF RESULTS TO SIALM, STARK, AND ZHANG ............ 80 LIST OF REFERENCES ................................ ................................ ............................... 88 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 92

PAGE 6

6 LIST OF FIGURES Figure page 2 1 Performance Flow Relationship by DC Ratio. ................................ .................... 43

PAGE 7

7 LIST OF TABLES Table page 2 1 Summary Statistics ................................ ................................ ............................. 44 2 2 Pa st Performance and Fund Flows ................................ ................................ .... 45 2 3 The Effect of DC Ratio on the Non linear Perfo rmance Flow Relationship ........ 45 2 4 The Effect of DC Ratio on the Performance Flow Relationship while Contro lling for Institutional Money ................................ ................................ ...... 47 2 5 The Effect of DC Ratio on the Performance Flow Relationship while Controlli ng for Fund Distributions ................................ ................................ ....... 48 2 6 The Performance Flow Relationship Comparison between Low DC, Mid DC and High DC Funds ................................ ................................ ............................ 49 2 7 The Effect of DC Ratio on the Performa nce Flow Rela tionship and Tax loss Selling ................................ ................................ ................................ ................. 50 3 1 Plan Level Descriptive Statistics ................................ ................................ ......... 71 3 2 Plan Menu Option s Descriptive Statistics ................................ ........................... 72 3 3 Benchmark Indices ................................ ................................ ............................ 73 3 4 Spanni ng Test with 8 Benchmark Indices ................................ ........................... 74 3 5 Spanning T est with 13 Benchmark Indices. ................................ ........................ 75 3 6 Style Exposure of Retirement Investors ................................ ............................. 76 A 1 The Effect of Performance Variable on DC Flow vs. Non DC F low Performance Relationship ................................ ................................ .................. 84 A 2 The Effect of Performance Variable on DC Flow vs. Non DC Flow Performance Rela tionship Excluding Index Funds ................................ ............. 86

PAGE 8

8 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ESSAYS IN RETIREMENT PLANS AND MUTUAL FUNDS By Sabuhi H Sardarli August 201 3 Chair: Michael Ryngaert Co Chair:Andy Naranjo Major: Business Administration In this study I examine t wo different but related topics in finance. The first part of the study examines the impact of defined contribution (DC) retirement plan investors on the performance flow relationship of mutual funds. I nvestors who hold mutual funds primarily through their DC plans are often deemed less sophisticated and less active in managing their DC accounts than retail mutual fund investors. Considering a significant I examine the performance flow relationship of mutual funds. I find that a higher proportion of a past performance. This effect is observed for both recent good performing and bad performing funds, although the size of this ef fect is larger for recent poor performers. I also find that while flows are negatively affected by return volatility for funds with low DC increases. The second part o f the study focuses on efficiency tests on retirement plan menus. I illustrate how testing for efficiency in 401(k) plan offerings depends on the benchmarks

PAGE 9

9 used. I also show that, for specific benchmarks, spanning can be attained without optimization. H owever, based on return style analysis, I show that, relative to these benchmarks, performance can be significantly improved through optimization. My results are consistent with a recent study that suggests that smaller 401(k) menus are objectively more e fficient and with a large literature that has noted participants' subjective propensity to prefer smaller options in decision making. Here I make a strong case for plan sponsors to recommend specific funds to help with plan efficiency. Plan participants can then adopt allocations with which they are comfortable. This approach contrasts with recent recommendations for default strategies and educational programs targeting participants who are viewed as the main source of plan inefficiency.

PAGE 10

10 CHAPTER 1 INTRODUCTION In this study, I examine two different but related topics in finance: the impact of retirement investors on mutual fund flows and efficiency and style tests of retirement plan menus The first part of the study, Chapter 2 investigates how the performance flow relationship of a mutual fund changes as the proportion of its assets held by defined contribution (DC) plans increases. D espite the dramatic growth in the assets held by these retirement plans, the literature has largely ignored their impact on mutual fund characteristics. Using survey data by Pensions & Investments, I am able to calculate the proportion of funds assets owne d through DC retirement plans. Specifically, I test to see whether the performance flow relationship of a mutual fund changes as this proportion is increasing. In the second part of the study, Chapter 3 I examine the efficiency and styles exposure of 4 01(k) retirement plans. I contribute to the fragmented literature on 401(k) plan menu efficiency. My more recent and rich sample allows me to test for spanning in retirement plans without sample bias issues. I am able to address this issue by utilizing a unique database consisting of 7,991 401(k) plans provided by BrightScope Inc. Effective saving for retirement is the result two different processes: selection of efficient funds into the plan by plan sponsors and plan administrators, and selection of fu nds made by the plan participant. This chapter focused on the first process which is a necessary condition to make sure that retirees have an opportunity to create well diversified portfolio that spans returns that are attainable from widely accepted benc hmark indices. I also analyze the impact of optimization on efficiency testing and style exposure within retirement plans.

PAGE 11

11 Overview of Chapter 2 Assets owned by employer sponsored defined contribution plans such as 401(k) and 403(b) plans have been drama tically growing in the recent past. The proportion of these assets invested in open end mutual funds has also grown dramatically. Despite these trends, the literature on the performance flow relationship of mutual funds has largely ignored these investor s. Making the question more interesting is the fact that the retirement investors are often deemed to be less sophisticated, or at least less active in the wealth management in comparison to retail investors. In this chapter I examine whether defined con tribution plan investors affect the sensitivity of fund f lows to recent past performance. Due to these fundamental differences between DC retirement plan investors and retail mutual fund investors, I hypothesize that a higher concentration of DC investmen ts in a particular mutual fund I find that the sensitivity of flows to past performance decreases with the ratio of assets owned by DC plans. This effect holds for funds both in th e low and high end of the performance spectrum although the effect is much larger in size for low performing funds. In other words, DC investors are both less likely to pull money from recently poor performing funds (as funds migrate from bad to worse in rank) than retail investors, and are less likely to move to funds with recent good performance (as funds migrate from good to very good in rank). Additionally, I find that as the DC ratio of the fund increases, the sensitivity of the flows to fund return volatility decreases. Relative to low DC ratio funds, funds with high DC ratios continue observing investors direct higher fund flows to them, even in cases of poor return performance and higher volatility. My findings suggest that DC retirement account participants indeed tend to be more

PAGE 12

12 passive investors exhibiting lower sensitivity to both returns and fund risk. Results are consistent with DC investors being less sophisticated and making infrequent adjustments to their investment choices which would r esult in less sensitive fund flows. Overview of Chapter 3 employee sponsored defined contribution plans. Ultimately, the efficiency of these plans depend on the performance and costs of mutual funds offered within plans. In this ch apter I assess the efficiency of 401(k) plan offerings on the basis of their spanning properties and find the evaluation to vary significantly based on the benchmark used. I also show that if plan sponsors or providers steer participants toward funds that are included in mean variance optimal allocations, spanning occurs irrespective of the benchmark considered. In other words, I find that although it may seem counterintuitive, limiting the fund choices in retirement plans to the subset that is part of a n optimized portfolio increases the chances of spanning. This implies that plan participants can benefit from information on this subset which can be easily provided by plan administrators. Even if the participant chooses to engage in naive diversificati are better off to choose the smaller subset rather than full fund choices available. In addition, participants are in fact more likely to follow styles and/or choose the so idence suggesting that such allocation is a reasonable benchmark upon which one might be able to refine into more optimal strategies, I then compare returns styles for both optimal and naive strategies and found them to differ drastically, suggesting that despite the reasonable performance

PAGE 13

13 for the pursuit of optimal strategies. My results imply that plan administrators need to take more active role in providing guida nce to plan participants in their investments decisions. More specifically, my recommendation is for the plan administrators to encourage participants to consider only the funds picked by the optimization and let them allocate according to their preferenc conduct their own optimization analysis to limit their choices to the optimal set of funds, that information needs to be provided by the plan administrators. Though my recommenda tion differs from earlier educational and default optimal strategies, it is consistent with individual subjective preference for lesser decision making choices and with recent studies suggesting more efficient plans with lesser options.

PAGE 14

14 CHAPTER 2 THE IMPACT OF DEFINED CONTRIBUTION INVESTMENTS ON THE PERFORMANCE FLOW OF MUTUAL FUNDS The propensity of investors to increase or decrease investments into actively managed mutual funds has long been a topic of interest in academic research. A principa l finding in this literature dating back to Sirri and Tufano (1998) is that investors do increase allocations to better performing mutual funds but that the relationship between recent fund performance and fund flows is convex. Investors are less sensitiv e to poor and middling performance, but are much more likely to commit significantly more net fund flows to the highest performing funds. While this is not particularly surprising, it is not clear this behavior is entirely rational in that numerous studie s find only modest correlations between past and future fund performance. In fact, it is low performing funds that are most likely to persist in their poor performance. 1 Hence, this could be viewed as performance chasing. Whatever the wisdom of empirica l investor behavior, the convex relationship between fund flows and performance has an impact on how mutual funds are managed. The greater the convexity of this relationship, the greater is the incentive to engage in risk taking if managers seek to maximi ze assets and associated fees in what approximates a winner take all outcome. 2 The literature on fund flows and performance has noted factors that might mitigate the strength and convexity of the flow performance relationship. For instance, older funds ha ve less sensitivity to performance, because new performance information 1 See Carhart (1997). 2 See Ippolito (1992), Chevalier and Ellison (1997), Goetzmann and Peles (1997), Lynch and Musto (2003) for performance flow convexity; Chevalier and Ellison (1997) and Brown, Harlow, and Starks

PAGE 15

15 is not likely to change their long standing perceptions. 3 Additionally, pension funds for defined benefit plans and mutual funds that are primarily held by institutional investors are also less apt to be engaged in return chasing behavior relative to retail funds. 4 One area of increasing importance that has not been explored extensively to sponsored defined contrib ution (DC) plans on the fund performance flow relationship. These plans predominantly include tax deferred 401(k) plans for private sector employees, 403(b) plans for non profit institution employees, 457 plans for government employees, and Keogh plans fo r self employed individuals. The importance of studying the effect of these plans on mutual fund flows is driven primarily by the growing position of these investors in the global mutual fund industry and by the characteristics of the average investor in these plans. According to the Investment Company Institute 2012 Fact Book, 5 over 60 percent of households in the U.S. participate in employer sponsored defined contribution plans, with plan assets totaling more than $4.5 trillion as of the end of 2011. W hile assets in these plans have increased five fold since 1991, the percentage of plan assets invested in mutual funds has also increased substantially. While only 15 percent of these assets were invested in mutual funds at the end of 1991, this percentag e increased to 55 percent by the end of 2011. These trends have resulted in DC retirement plans in the United States constituting close to 10.5 percent of total worldwide mutual fund assets ($23.8 trillion) in 2011. Given the fact that DC 3 See Huang, We i, and Yan (2007). 4 See Del Guercio and Tkac (2002) and James and Karceski (2006). 5 Investment Company Institute is the national association of U.S. investment companies. 2012 Fact Book can be accessed at http://www.ici.org/pdf/2012_factbook.pdf.

PAGE 16

16 investors have significantly increased their mark on the mutual fund industry as the academic literature on the performance flow relationship for mutual funds has developed, it is important to quantify the effect of DC investors on this relationship. Increasing the significance of this study is how the average DC investor differs from the retail and institutional investors studied in the current literature. DC investors are often believed to be less sophisticated, or at least less active, in their investment choices A typical DC plan investor is not actively choosing to enter the capital markets but rather is taking advantage of important benefits of employment: company contribution matching and the tax advantage. This might result in the average plan participant paying less attention to performance and/or simply keeping their flow contribution preferences are rarely altered after initial enrollment. 6 On the other hand, funds that are marketed to DC plan sponsors, and hence may already have high DC allocations, are known to get added to plans by sponsors when they perform well. 7 If certain funds with already high DC participation are on the hot funds to the DC plans this could result in a higher sensitivity of flows to high end performance than I otherwise would have expected. In general, however, I expect that 6 Madr ian and Shea (2001), Agnew, Balduzzi, and Sundn (2003) are studies that find strong inertia among 401(k) investors; Mitchell and Zeldes (1996) and Benartzi and Thaler (2001) show that average 401(k) make unsophisticated investment decisions in their retir ement accounts. 7 Elton, Gruber, and Blake (2007) find that plan administrators add funds with good recent performance and drop funds that have performed poorly.

PAGE 17

17 greater levels of presumably passive ex ante DC ownership in a particular mutual fund will weaken the performance flow sensitivity shown in the current literature. I use data from Pension examine whether fund performance drives fund flows to the same extent when a fund has a high level of DC holdings relative to cases where it does not. I find that the s assets held in DC plans, the less sensitive is performing and bad performing funds, although the size of this effect is larger for recent poor performers. At the mean, a one standard deviation increase in DC investment decreases the sensitivity of flows to recent poor performers by 31 percent and to recent good performers by 14 percent. Lower sensitivity of flows to performance cannot be explained by asymmetry in tax mot ivated selling between DC and retail investors. When I take a closer look at fourth quarter flows, where tax motivated selling would be more concentrated, I observe only modest differences in the effect of DC money. This further supports my hypothesis tha t it is the passive, potentially less sophisticated nature of the average DC investor that contributes to my findings. I also find that while flows are negatively affected by return volatility for funds with low DC plan ownership, this effect decreases as plans increases. Therefore, not only does higher percentages of DC money invested in a fund decrease the net investment flow sensitivity to past returns, it also diminishes the sensitivity to return volatility. I ar gue that the passive, and perhaps the less sophisticated nature of DC investors explains both these findings. To the extent that fund managers like stickier money to be invested, these attributes may make DC

PAGE 18

18 investments more desirable to fund managers. I also conclude that my findings are not driven by institutional money effect. This is particularly a concern because funds that are popular in 401(k) menus might also be popular with other institutional investors. Additionally, my findings continue to hold when I control for distributions paid by fund. performance flow relationship of mutual funds, I contribute to the rich literature on the flow sensiti vity of mutual funds taking a new trend in mutual fund industry into account. DC plans are increasingly getting larger and are increasingly investing more in equity mutual funds and their effect on the performance flow relationship is highly likely to onl y grow in importance. Because investments by DC retirement plans change the shape of performance flow relationship with respect to performance (less sensitivity) and to increase risk when their funds have a large fraction of their investments coming from DC investment vehicles. The remainder of the chapter is organized as follows. In next session I provide a brief summary of the background literature on the performan ce flow. The following section describes the data sources, sample details and calculation methodology for main variables. In the main part of the analysis, t he effect of DC investments on mutual fund flo ws is analyzed. Finally, I summarize the results i remarks Relevant Literature The positive link between recent mutual fund performance and fund flows is well documented. Ippolito (1992) builds a rational model in which a key outcome is that mutual fund investors will purchase good performers and he tests this prediction

PAGE 19

19 empirically. The findings confirm the model prediction with a positive performance flow relationship for mutual funds although he finds this relationship to be much stronger for better performing funds. Sirri and Tuffano (1998) take this analysis one step further by introducing non linearity to performance and conclude that the relationship is highly convex with negative performance not affecting flows at all. Additionally, they find some evidence for the negat investor dislike of volatility. There are a few theoretical explanations for this convex relationship. In a rational model framework, Berk and Green (2004) show that low performance pe rsistence, flow relationship is consistent with rational and competitive capital markets. Lynch and Musto (2003) argue that since past returns are less informative about future re turns for low performance periods (because of implied strategy shifting), investors do not have incentives to redeem their shares. Therefore, a failure to flee from poorly performing expectations of The convexity of the performance flow relationship also has implications for manager incentives. Since the highest performers enjoy a large influx of capital investments, asset managers have incentives for risk shifting behavior (Chevalier and Ellison (1997); Brown, Harlow, and Starks (1996)). Assuming investors pay attention to year end returns, poorly performing managers are more likely to increase portfolio risk at mid year with the hopes of achieving better results, while top performing managers are more likely to shift to safer positions so as not to endanger their performance

PAGE 20

20 ranking. They also document that as investor awareness has grown, incentives for have also increased. More recent papers in the literature have documented that the convexity of the performance flow relationship has been decreasing over time. Huang, Wei, and Yan (2007) argue that investors incur participation costs before purchasing m utual funds and that these costs are important factors in the performance flow relationship. Specifically, funds with low participation costs are more likely to attract investors in the medium performance range, resulting in increased flow sensitivity. I n contrast, funds with high participation costs will only attract investors if the performance is exceptional. This results in an overall flatter performance flow sensitivity for funds with low participation costs. They also conclude that since investor participation costs overall have been declining with more media coverage and advancements in technology, the convexity in the fund flow and performance relationship is less pronounced in the post 1990 time period. 8 A similar decline in the convexity is a lso observed in James and Karceski (2006), where they find positive and significant performance flow relationship within the poorest performing funds using a 1990 2001 sample period. They also find that flows to institutional share classes are less convex because investors in this class are more mutual fund investors, pension fund clients puni sh poorly performing funds (e.g., the worse the performance among the bottom 20 percent, the more fund outflows) and do 8 1990 tim e period.

PAGE 21

21 not disproportionally invest in recent top performing funds. Taken together, these last two studies are closely related to my chapter be cause they analyze the performance flow relationship with an understanding that the characteristics of investors can result in different flow sensitivity. The effect of DC investments on mutual fund characteristics has been analyzed by a few studies in a small but growing literature. Sialm and Starks (2011) examine characteristics. More specifically, they presume that mutual funds heavily owned by retirement accounts are ex pected to be less sensitive to tax efficient investment and distribution strategies. Consistent with this prediction, the authors find that funds with a higher percentage of DC investment have higher long term capital gain distributions, and they are less likely to sell losing positions and more likely to sell winning positions in exposure changes as the retirement investments in the fund increases. They find that as DC r atio of a fund grows, its manager also increases the market risk of portfolio investments although this increased risk exposure is uncompensated in future returns. My findings in this study are closely related to those found in Sialm, Stark, and Zhang (20 12). While in this chapter I analyze the effect of DC money on the performance total flow relationship, Sialm, Stark, and Zhang (2012) analyze the difference in the sensitivities of DC flows to performance and non DC flows to performance. Interestingly, they find that DC flows are more sensitive to performance measures when compared to non DC flows. Although at first glance my results may seem to at odds, it is important to point out that these two methodologies are answering

PAGE 22

22 two related but slightly dif ferent questions. Given a level of DC and non DC balance, their question answers whether shifts in DC flows or non DC flows are more performance driven. In this study I am asking whether total flow performance relationship is affected by DC ratio of a fu nd. I argue that this is the question more related to performance flow relationship literature to date, and it is the one that more directly analyzes the impact of retirement investments on this relationship. Additionally, other related implications that span from performance flow relationship, such as managerial risk shifting incentives, are more closely related to total flows. 9 However, to my knowledge, with the exception of Sialm, Stark, and Zhang (2012), the potential impact of DC retirement plans on the performance flow relationship of mutual funds has not been analyzed in the literature. Considering how rapidly this industry has grown and the fundamental differences between DC investors and retail investors, it is natural to examine whether DC flows can affect the characteristics of mutual fund flows and their response to performance. My contribution to the literature is two fold. First, I show that funds with greater DC investment tend to have less sensitivity of new net investment flows to perform ance and/or performance rank, particularly at the top and bottom of performance rankings. I also show that there is also less sensitivity of flows to return volatility among high DC ratio funds. Second, I document that the rise of defined contribution pla ns have changed the performance and fund flows dynamics. Any future research regarding performance flow relationship needs to account for DC investors and their characteristics. 9 Please see Appendix for more details in the differences between the two methodologies and for additional possible explanations for the differences between my results and those found in Sialm, Stark, and Zhang (2012).

PAGE 23

23 Data Description I DC plans with the help of survey data made available to me by Pensions & Investments. The survey is conducted annually and asks the mutual fund families surveyed to list their funds that have the greatest amount of assets held by DC plans as of December 31 st of the previous year. Firms are requested to include fund names, tickers, and the dollar amount of DC assets invested in each fund listed on the survey (this is done by fund type: domestic equity, domestic fixed income, international, balanced, target da te and money market funds). I obtained this annual survey data for the period of 1999 2011 to the ratio used by 10 I analyze the relationship of DC holdings on quarterly flow measures, but I canno t update my DC ratio every quarter. This means, for instance, that DC percentage holdings in the 2011 survey is used as the DC ratio for a fund at the end of 2010, and this measure is used to help explain fund flows in each of the four quarters of 2011. 11 The primary source for the mutual fund characteristics (other than DC holdings) is the CRSP Mutual Fund Database. This survivor free database provides monthly 10 Christoffersen and Simuti n (2012) also use Pensions & Investments survey to test whether mutual funds change their market risk exposure as the retirement money in the fund grows. 11 I use quarterly data because annual flow funds will be affected by contemporaneous, as well as lagg ed returns to a greater extent than would be the case for quarterly fund flows. This also allows me to use more observations for funds that disappear from a database midyear, which could result in more potential survivor bias in my analysis. In compariso n with annual calculation, quarterly calculation also has lower error related to the assumption of flows incurring at the end of the period, as described below.

PAGE 24

24 returns, total net assets (TNA), fund inception dates, management company names, expense ratios, portfolio turnover, and Lipper investment objectives. This database was merged with my DC holdings data using fund ticker symbols and fund names. While that most mut ual fund families in the survey data report DC assets at the fund level. Therefore, I am forced to aggregate all variables at the fund level for consistency. I do this by identifying multiple share classes that belong to the same fund with the help of Mo rningstar data which contains a fund level identifier for share classes. For funds that I failed to match to Morningstar, I use CRSP fund names to identify share classes that belong to the same fund. 12 Consistent with the literature, I focus exclusively o n equity mutual funds and drop index funds, funds that are closed to new investors, and, to avoid incubation bias, funds with beginning year TNA less than $10 million. 13 The following Lipper objective categories are represented in the sample: capital appre ciation, equity income, growth, growth income, mid cap, micro cap and small cap. Also consistent with the prior literature, I compute quarterly net fund flows according to Equation 1 1 which assumes flows occur at the end of the quarter: ( 1 1) 12 Although CRSP also has fund level identifier, its coverage and quality is inferior to Morning star identifier. For example, if I use the CRSP identifier there are 163 funds in my sample with DC Ratio over 1. However, with the Morningstar identifier there are 90 funds. Additionally, on average the CRSP identifier captures 3.78 classes per fund, w hile the Morningstar identifier captures 3.86 classes. 13 Documented by Evans (2010), newly initiated funds have upward biased returns. Sialm and Stark (2011) also apply a similar size filter.

PAGE 25

25 where is fund i returns for the current quarter. 14 This equation is the estimation of net new money flows after adjusting for the capital appreciation during the quarter. Since this estimation method can be greatly affec ted by fund mergers and splits, I truncate the upper and lower 2.5 percent tails of the flow data. 15 In addition to this truncation method, I also track mergers with the help of an identifier in CRSP and control for them with an additional test in the Appendix. Calculation methods for other variables are consistent with the literature for the purpose of comparison. Lagged performance is calcul ated by compounding raw monthly returns for the previous four quarters (one year). Return volatility is the standard deviation of the past 12 monthly returns. The previous literature also documents the importance of age and the total fees of the fund. A ge is the difference adding total expense ratios and one seventh of front end load fees (the average holding period for mutual funds documented in the literature is 7 yea rs). 16 The fund level fees are a value weighted average of share class fees, whereas age is based on the year of turnover in the previous year as reported in CRSP. Ca tegory flows are calculated by 14 Many studies conclude that the performance flow relationshi p is not affected by when the calculation is adjusted for flows to occur in the beginning or in the end of the period (e.g. Ippolito (1992), Sirri and Tuffano (1998)). 15 Similar methods are performed by Huang, Wei, and Yan (2007) and Sialm and Starks (2011 ). 16 Sirri and Tuffano (1998) find that redemption rate for equity mutual funds implies an average holding period of seven years.

PAGE 26

26 aggregating all fund flows within the same investment objective scaled by total net assets in that category. I of fund assets held in institutional class share s as classified by CRSP. After requiring full data availability for all variables, my sample includes 1,117 unique funds from 111 different mutual fund companies for 16,384 quarter fund observations for the period of 1999 2011. I should also note that the composition of firms surveyed from year to year is not the same. Therefore, some mutual fund companies might drop out of the sample completely after a given year, while some re enter at a later date. However, large well known companies and funds are con sistently in the sample every year. Table 2 1 highlights the characteristics of my sample. The mean percentage level of DC holdings is 23.82 percent and the median is 19.31 percent. The median level of total net assets is $1.03 billion. As expected, the distribution of mutual fund size is skewed with large funds driving the mean of TNA in the sample to be 3.7 times larger than the median. The average fund is 14 years old with 1.35 percent total fees annually. It should also be noted that average quarte rly flows in my sample is negative since my period covers two significant economic recession periods which negatively affected flows to equity mutual funds. Additionally, there has also been an investment ctively managed mutual funds in the last 15 years. 17 Table 2 1 also reports the sample differences based on the level of the DC Ratio. I specifically report descriptive statistics for observations where DC Ratios are below the 17 See Investment Company Institute 2012 Fact Book for equity flows during recession periods and the my sa mple period.

PAGE 27

27 25 th percentile and above the 75 th percentile, which corresponds to DC ratios of 8.66 percent and 33.59 percent, respectively. The mean DC ratios for these two groups of funds are 4.20 percent and 51.76 percent, respectively. According to this univariate comparison, there are no sta tistically significant differences in fund flows or raw returns between the low and high DC Ratio samples. Although there is a slight difference in fund age means, the medians are identical. However, funds with high DC ratios are more than double the siz e (at the median) of funds with low DC ratios. High DC ratio funds also have 31 basis points lower total fees and a higher concentration of institutional investors. Another interesting observation is that while there is no difference in raw returns, the volatility of these returns is higher for the high DC ratio group by 1 percentage point. Overall, these differences in fund characteristics between low and high DC ratio funds are very similar to those found in Sialm and Starks (2011) and in Sialm, Stark, and Zhang (2012). The Effect of DC Investors on Performance Flow R elationship Univariate Analysis As a first attempt to analyze the effect of DC ratio on the shape of the performance flow relationship, I graphically compare the flows as a function of perf ormance ranks created according to lagged raw returns. Every quarter, I rank funds in my sample according to their return over the past four quarters within their objective categories and assign them to 20 equally sized groups. Figure 2 1a visually repre sents mean quarterly flows for these groups while using all funds in my sample. I find that average quarterly flows for the poorest performing funds are negative 1 to 2 percent per

PAGE 28

28 quarter. 18 It is also clear that the highest performing funds enjoy strong capital inflows of about positive 3 percent per quarter. In the next two figures, I repeat the same visual exercise for funds with DC ratios which are below the 25 th percentile and above the 75 th percentile cutoffs. Figure 2 1b displays similar performance flow sensitivity for the low DC ratio group as in the previous figure, except for a steeper dip in flows in the left tail, indicating higher outflows for poor performing funds. While Figu re 2 1a depicts all funds in the sample as more consistent with the findings in Sirri and Tuffano (1998), more recent papers do find a relationship closer to the one depicted in Figure 2 1b for poor performers. The lowest performing funds (bottom 5 percen t) for the low DC ratio group has a loss of 3 percent of assets from flow changes, but the highest group has a gain of about 3 percent. This constitutes about a 6 percent spread. However, Figure 2 1c depicts quite a different story for the funds with a high DC ratio relative to the low DC ratio funds in Figure 2 1b. There is a much more muted sensitivity between lag one year returns and fund flows in the form of a much flatter line with the lowest performing funds (bottom 5 percent) losing less than 1 percent of assets due to net outflows while the top performers (top 5 percent) gain only 1.5 percent of assets due to net inflows. The sensitivity line is much more concentrated around zero when compa red to the two previous figures. A simple mean difference test performed on the data in Figures 2 1b and 2 1c yields statistically significant different slopes for the three lowest and the three highest groups at the 95 percent confidence level. Overall, the initial visual analysis hints at the fact that the percentage of net assets held by DC 18 Similar graphs by Sirri and Tuffano (1998) and Huang, Wei, and Yan (2007) also display negative flows for poorest performing fund groups.

PAGE 29

29 plans is affecting the performance flow sensitivities, although multivariate analysis is needed to control for other factors that could be affecting fund flows. Mul tivariate Analysis While testing the performance flow relationship, previous studies have found that it is important to control for fund size, age, total fund fees, turnover, historical return e category. As a starting point, I analyze the performance flow relationship linearly and then introduce the effect of DC ratio. Specifically, I regress quarterly mutual fund flows on the buy and hold raw return over the four quarters leading up to the q uarter for which the fund flows are calculated, the 4 Factor Carhart alpha measured using lagged 36 month returns 19 and the additional control variables mentioned previously. This specification helps me to differentiate between absolute returns and rank r eturns, represented by raw and risk adjusted returns. Quarterly time fixed effects are also added to control for any cross sectional correlations in the observations because of overall flow fluctuations over time. The results are reported in column (1) o f Table 2 2. I find that a 1 percent higher raw return is rewarded by an annualized 0.60 percentage point increase in fund flows while an additional 1 percent in risk adjusted performance is rewarded with a 1.58 percentage point increase in annual fund fl ows. 20 These findings are qualitatively similar to those found in Del Guercio and Tkac (2002) where a similar analysis is conducted for mutual funds and is contrasted with pension funds. Additionally, consistent with prior literature, I find that capital flows into mutual funds increase with fund category flow levels and 19 At least 20 months return data was required to calculate the alpha returns. Monthly historical Marke t, http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. 20 Coefficient estimates are multiplied by four to estimate annualized flow effects.

PAGE 30

30 decreases with age, turnover, and total fees. Fund size is statistically insignificant in this specification which is likely caused by the fact that the survey contains mainly large funds and the impact of size has become insignificant. I also find that size is strongly correlated with fund age (univariate correlation coefficient of 0.489). When I remove age from my specifications, fund size strongly negatively affects fund flows. Throug hout the chapter I employ two methods of analyzing the influence of DC ratio on the performance flow relationship. First, I introduce DC Ratio as a continuous variable; and second, I percentage of as th percentile, and zero otherwise. As mentioned previously, this corresponds to funds with a DC ratio above 33.59 percent. The dummy variable is a concession to the possibility of some error in variable in measu ring DC ratios by just focusing on the group with very high ratios. Following the first method, I interact DC Ratio as a continuous variable with past performance and report the results in column (2) of Table 2 2. The last column reports the same regression but with the DC dummy variable. Based on the results in column (2), I get a positive coefficient on flow sensitivity to both raw returns and alpha. I also see that the interaction of DC ratio and lagged return is negative and statistically sign ificantly different from zero at the 1 percent level. The sign on the alpha interaction is also negative, but not statistically significant. Based on column (2) figures, I find that at the mean, a one standard deviation increase in DC investments (from a DC ratio of 24 percent to a DC ratio of 44 percent) weakens the relationship between past raw performance and fund flows by 7.5 percent. 21 Using the 21 0.164 + ( 0.057)(0.4360) = 0.139 sensitivity is 7.5 percent lower than 0.164 + ( 0.057)(0.2382) = 0.150

PAGE 31

31 high DC ratio dummy variable in column (3), I find that the interaction of the DC dummy and both raw and r isk adjusted returns are negative although the latter is not statistically significant. Using the DC dummy, coefficient estimates suggest that these funds have 8 percent less sensitivity of flows to alpha returns and 14 percent less sensitivity to lagged one year raw returns. 22 These initial findings suggest that funds with higher DC fund exposure have a weaker relationship between fund performance and its flows even after sponse to returns declines. I next turn to analyzing the effect of DC ratio on flows in a non linear model because the effect of performance on returns has been found to be non linear in the extant literature. For that, I employ a piecewise linear regress ion first proposed by Sirri and Tuffano (1998), and introduce the DC Ratio variable as well as its interaction with performance variables in the following regression: ( 1 2) where , and are performance ranks created based on lagged four quarter raw returns within each objective category. More specifically, , and are performance variables within the lowest, the middle three and the highest quintiles respectively, and are defined as 22 Mutual fund investors paying attention to risk adjusted returns might seem surprising at a first glance, but Del Guercio and Tkac (2002) find that the relationship between alpha returns and flows exist because of Morningstar ratings that are included in fund summary documents. Our result about alpha returns suggests that DC investors do not pay attention these ratings as much as retail investors.

PAGE 32

32 , respectively, where is the lagged 12 months raw returns rank within fund objective category quarterly. The purpose of the , and variables is to ascertain if there is any sensitivity to fund flows given the fund has a better ranking within the group of low, middle, or high performance groups. For this rank variable, I use all mutual funds on the CRSP tape with quarterly returns to construct the rank. Lastly, is a vector of control variables including aggregate fund category flow, fund age, fund size, total fees, turnover and return volatility. 23 Table 2 3 reports coefficient estimates for OLS regressions with quarterly fixed effe cts. 24 In column (1), I run a base specification without DC Ratio and interactions, and find a well documented convex relationship between performance and mutual fund flows. Unlike Sirri and Tuffano (1998), however, I do find a positive and statistically significant relationship for the variation in rank within the lowest performance funds (i.e., the bottom quintile of performance rank). In other words, mutual fund investors do, on within the group of already poorly performing funds. This positive relationship for the lowest performers is also found by Huang, Wei, and Yan (2007) and James and Karceski (2006) when using data for the post 1990 period. Even with this positive slope finding for t he bottom 23 In untabulated regressions, I also include an interaction term between fund age and performance rank. I find that the fund flow is negatively impacted by the interaction of age and overall performance rank, consistent with the notion that less is learned from the performance of older funds in the eyes of investors that may be considering reallocation of funds based on prior year performance. Inclusion or exclusion of this interaction term does not have any important effect on my regression results. 24 I a lso repeat all my tests with Fama and MacBeth (1973) regressions which produce conservative standard errors by removing the independence assumption for observations that fall within the same year. More specifically, these estimates are created by running separate regressions within each year and reporting consolidated coefficient estimates and standard errors. Our results are virtually identical to those estimated by OLS regressions with quarterly fixed effects as reported in tables.

PAGE 33

33 performance group, I still observe that top performing funds enjoy 101 percent higher fund flows to a 1 percent higher return than low performing funds. 25 Therefore, while Sirri and Tuffano (1998) document the convex performance flow relationship starting with a flat line and then an upward slope toward the end of the performance scale, I find positive slopes at both ends of the curve with significantly higher slopes for the highest performing group of funds. Hence, convexity is still present, th ough to a lesser extent than that documented originally by Sirri and Tufano (1998). I also find that mutual fund increase in return volatility (annualized standard d eviation), fund flows decline by 0.1 percent annually. In column (2), I introduce the DC Ratio variable and its interaction with piecewise performance ranks. The results yield a negative effect of DC Ratio on performance sensitivity at all levels althou gh the sensitivity coefficient for the DC interaction term is statistically significant at 1 percent for the bottom 20 percent of performance, and at 10 percent for the top 20 percent of performance. When DC Ratio is introduced as a continuous variable in column (2), I find that when fund assets held by DC plans increase by one standard deviation at the mean, the sensitivity of quarterly flows to returns declines from 0.097 to 0.063, a 35 percent decrease for low performing funds. 26 In fact, the regression suggests that going from no DC holdings to a fund fully invested with DC funds causes the sensitivity to moving higher or lower in terms of rank in the low performance group of funds (bottom 20 percent) to be negligible. While the DC 25 The coefficient of High rank variable is 101 percent higher than that of Low rank variable. The difference between these two coefficient is statistically significant at 1 percent level. 26 0.138 + ( 0.173)(0.4360) = 0.0626 which is 35 percent lower than 0.138 + ( 0.173)(0.2 382) = 0.0968.

PAGE 34

34 ratio and performanc e rank interaction variable in the midrange of performance is negligible, the reduced sensitivity to rank is economically important and statistically significant at 10 percent within the highest performing group of funds. At the mean DC ratio level, one s performance to decline by 11%. 27 In column (3), I replace the DC ratio with a dummy variable that equals 1 if the th percentile (above 33.59 percent). Again, I get a negative and significant coefficient for the low and high performance ranks with the DC dummy statistically significantly different from zero at the one percent level. For funds rmance flow sensitivity is lower by 58 percent in column (3). 28 Overall, the results are consistent with a more muted response by DC investors to performance at both the high and low end, which suggests overall less sensitivity to performance, but still s ome element of convexity of the investment performance flow relationship. It should be noted that in both columns (2) and (3), DC Ratio by itself also has a positive and significant effect on overall fund flows. Mutual funds enjoy higher fund flows when t hey have a higher percentage of investments coming from DC plans. Specifically, a one standard deviation increase in DC plan assets results in 2.7 percent higher annual flows for the fund. This could be attributable to the constant flow of money that is received by investors in DC funds and also may be attributable to higher level DC funds (e.g., funds that cater to DC plan sponsors) obtaining new clients as 27 0.220 + ( 0.106)(0.4360) = 0.1738 which is 11 percent lower than 0.220 + ( 0.106)(0.2382) = 0.1948. 28 Sensitivity coefficient of 0.218 0.091=0.127 for funds with highest DC concentration is 58 percent lower than sensitivity of 0.218.

PAGE 35

35 more companies use 401(k) plans as opposed to defined benefit plans over the course of time in my sample. While most studies to date have largely focused on the effect of fund performance on its flows, very little attention has been given to the effect of the volatility of that performance on the flows. However, studies that include return volatility as a control variable find that return volatility has a significantly negative effect on flows. If my hypothesis about DC investors being passive (or less sophisticated about risk) is indeed true, then I should also observe that those investors pay less a ttention to volatility than average retail investors. Therefore, DC ownership of mutual funds should weaken the negative relationship between return volatility and fund flows. To test this prediction, I rep eat the regression in Equation 2 2 with the addit ion of the DC ratio variable interacted with volatility. Results in columns (4) and (5) indicate that while the effect of volatility on fund flows is negative and significant, a greater concentration of fund assets in DC plans does indeed act to significa ntly mitigate this the overall risk aversion of its investor base. Specifically, I find that a one standard deviation increase in DC assets from zero decreases the n egative sensitivity of flows to fund return volatility from 0.058 to 0.030, a 48 percent reduction. 29 Therefore, not only do I significantly less sensitive to the return volatility of the fund. 30 29 D ifference in sensitivities of 0.058 and 0.058 + (0.142)(0.1978) = 0.0299 30 This finding is also confirmed in Christoffersen and Simutin (2012) where they find that as DC ratio of the fund grows, market risk of investments also grows.

PAGE 36

36 It should be noted that in column (4), with the inclusion of the DC ownership and significant for the bottom performance rank but becomes more negative and more statistically significant for the high performance rank group. However, there is still an asymmetric impact of DC Ratio on performance between these two groups. While a one standard deviation increase in DC assets at the mean decreases the sensitivity of the performance flow relationship for low performing funds by 31 percent, it only decreases the sensitivity for highest performing funds by 14 percent. 31 However, the coefficient on DC ratio no longer has a statistically meaningful impac t on flows of funds. Perhaps, some of this was being transmitted through the volatility interaction. A plausible explanation for the asymmetric effect of DC holdings on the performance flow relationship is the choice of mutual funds that make it into the retirement account menu. The asset flows to mutual funds by DC plan investors is the outcome of a two step process. First, third party plan administrators commissioned by plan sponsors select funds from a vast universe of investment options to make up th e plan menu. Then, plan participants must decide how to allocate their contributions across these defined set of options. Therefore, fund flows coming into DC plans are affected by both plan sponsors and plan participants. Even if plan participants are indeed passive, if the plan sponsor exhibits return chasing behavior by choosing recent good performers to be included in the plan, DC plan flows will have similar return chasing characteristics as retail mutual fund investors. Elton, Gruber, and Blake (2 007) 31 The impact on Lo w performance rank is calculated as 1 [0.135 + ( 0.155) (0.4360)] / [0.135+( 0.155) (0.2382)] = 0.3126 while the impact on High performance rank is calculated as 1 [0.227 + ( 0.139) (0.4360)] / [0.227 + ( 0.139) (0.2382)] = 0.1418.

PAGE 37

37 find that plan sponsors are indeed more likely to add funds to a plan menu following recent superior performance. Therefore, the lower negative effect of DC ratio on the performance flow sensitivity on the high end relative to the lower end could be the outcome of the choices of funds included into the plan. It is also possible that the effect I am finding is actually the effect of institutional class shares which are catered to large institutional investors. This is particularly a concern because so me DC plans carry very large investment balances and therefore could gravitate toward institutional class shares to take advantage of lower fees. Even DC plans with low investments could also offer mutual funds that have large total net assets because of their non DC investor base. Since James and Karceski (2006) find that the performance flow relationship is weaker for institutional funds, it is possible that the DC Ratio variable is picking up the institutional fund effect and is not truly the passive i nvestor effect I am describing above. That would help explain my finding of a negative relationship between DC Ratio and the performance flow sensitivity only because DC money and institutional money tend to drift toward same mutual funds. To account for that possibility, I that are invested class level and classes are identified to be institutional if the investment minimum is larger than $100,000. Since my analysis is at the fund lev el rather than class level, instead, I calculate the ratio of net assets invested via share classes whose investment minimums are larger than $100,000. Results are reported in Table 2 4 where I find that while institutional money does decrease the sensiti vity of returns to flows, the negative effect of DC ratio on the performance flow relationship is preserved with similar

PAGE 38

38 economic and statistical magnitudes. I also continue observing the positive effect of DC ration on the volatility flow relationship. Overall, I conclude that the dampening effect of DC money on the performance flow relationship cannot be explained by the institutional money effect even if DC retirement plans gravitate toward institutional share classes or funds with large proportions of assets invested in institutional share classes. One important difference between investments in retail mutual funds and retirement 401(k) funds are the way fund distributions reach investors. While retail investors are less like to reinvest funds distrib are reinvested because taking distributions out from a 401(k) plan would constitute as a withdrawal which is costly for a retirement investor. Therefore, unlike most retail mutual fund investors, retirement inve stors automatically reinvest fund distributions. Therefore, if better fund performance means more distributions then the performance flow relationship might be stronger higher DC funds with more distributions since the money is more likely to stay in the fund. To control for such effect, I obtain all capital gains and income distributions paid by the fund from CRSP add them to create Total Distributions variable by scaling it by beginning fund assets. I include this variable in Table 2 5 specifications ( 1) and (2) and find that higher distributions in general translates to lower flows since distributions paid to investors is a form of outflow from the fund. However when I interact this variable with DC ratio in specifications (3) and (4), I find that as DC ratio increases the negative impact of distributions on flows are reversed because of reinvestments, although this interaction term is not statistically significant. It is also worth noting that my performance and volatility performance continue to hold while controlling for fund distributions.

PAGE 39

39 Finally, I employ a slightly different methodology to test the impact of DC ratio on the performance flow relationship of mutual funds. Instead of using interaction term s of DC ratio with performance measures, I divide my sample into three groups according to 32 In Table 2 6 Panel A, Low and High DC funds are identified as funds with lowest and highest quartile of DC ratio. These cutoffs in my data correspond to 8.66 percent and 33.59 percent, respectively. Mid DC funds are is the subsample of funds with DC ratio between these two figures. Next, I run separate performance flow relationship regressions for each subsample. Consistent with my previou s interaction term results, sensitivity coefficients on the performance variables is much smaller in size for High DC funds in comparison with those for Low DC funds. In other words, performance flow relationship is much weaker when a fund contains higher level of DC money. As alternative identification method, in Panel B I divide the sample in equally sized terciles according to DC money and redefine Low DC, Mid DC, and High DC funds. I continue finding similar results of lower sensitivities for higher DC money funds. The Effect of Tax loss Selling One important consideration while examining the performance flow relationship between retail investors and DC investors is the tax consequence of trades and the fundamental differences between these investors. Retail mutual funds investors, similar to stock market investors, are subject to taxes and also can enjoy tax deductions from capital losses while DC plans are tax deferred accounts and are not subject to these tax 32 This methodolo gy is also employed by Sialm, Stark, and Zhang (2012) although their findings are in the opposite direction. One important difference between their regressions and those in Table 6 is the granularity of the analysis. Sialm, Stark, and Zhang (2012) uses m onthly flow data for this regression while I continue my analysis in quarterly terms. Please see the Appendix for other possible reasons for the differing results.

PAGE 40

40 factors. 33 Therefore, while retail inve stors sometimes may have incentives to trade for tax benefits, particularly toward the end of a calendar year, DC investors do not. To rule out the possibility that tax passive nature, particularl y for performance rank funds, I run my main specification with special attention to the fourth quarter. If tax loss selling is indeed the reason behind my findings, I should observe that DC Ratio decreases the performance flow relationship of the poor est performing funds mostly in the last quarter. I should not observe this finding much during the first three quarters of the year. In Table 2 7, I repeat the specification in E quation 2 2 and add a dummy variable (Q4) equal to one for flows in the four th quarter along with interaction terms. First, I percent for poorly performing funds in the fourth quarter which is consistent with tax motivated selling for poor performers. Sensitiv ity for the middle and highest performing group is virtually equal during the first three quarters and the last quarter. More importantly, I observe that the negative effect of DC Ratio on the performance flow relationship is evident during all four quarters. The statistical and economic magnitude of this negative effect is comparable to those found in the earlier runs. I also find that all t hree triple interactions in this specification are statistically indistinguishable from zero: the effect of DC Ratio for the last quarter is negative, indeed most negative in the fourth quarter, but not statistically significantly different from the other DC variable performance rank interactions. Therefore, I conclude that the asymmetry in tax motivated trading between retail and DC investors is not the driving factor behind the 33 See Badrinath and Lewellen (1991), Odean (1998), and Grinblatt and Keloharju (2002).

PAGE 41

41 lower sensitivity of funds flows to performance. The ratio of DC money inves ted in a calendar year. Chapter 2 Concluding Remarks In this chapter I examine whether defined contribution plan investors affect the sensitivity of fund flows to recent past performance since these investors are believed to be more passive and less sophisticated. Despite the dramatic growth of these types of retirement pla ns and the mutual funds flows from these plans, the performance flow literature to date has largely ignored these investors. Due to fundamental differences between DC retirement plan investors and retail mutual fund investors, I hypothesize that a higher concentration of DC investments in a particular mutual fund will have an I find that the sensitivity of flows to past performance decreases with the ratio of assets owned by DC plans. This effe ct holds for funds both in the low and high end of the performance spectrum although the effect is much larger in size for low performing funds. In other words, DC investors are both less likely to pull money from recently poor performing funds (as funds migrate from bad to worse in rank) than retail investors, and are less likely to move to funds with recent good performance (as funds migrate from good to very good in rank). Additionally, I find that as the DC ratio of the fund increases, the sensitivity of the flows to fund return volatility decreases. Relative to low DC ratio funds, funds with high DC ratios continue observing investors direct higher fund flows to them, even in cases of poor return performance and higher volatility. Our findings sugge st that DC retirement account participants indeed tend to be more passive investors exhibiting lower sensitivity to both returns and fund risk. Results are

PAGE 42

42 consistent with DC investors being less sophisticated and making infrequent adjustments to their in vestment choices which would result in less sensitive fund flows. These characteristics make DC investors very valuable to fund managers because they are less likely to abandon the fund in the case of poor performance. When retail investors are pulling th eir money out of funds due to poor performance, DC investors are helping to soften the negative effect on flows. Overall, I provide convincing evidence that DC money changes the slope of the performance flows relationship in mutual funds at the low and hig h ends of the performance spectrum. Since the shape of this relationship also has implications on an unclear direction. While lower sensitivities to performance might weaken propensity to change fund risk, my year risk shifting behavior needs to be re analyzed empirically as the ratio of DC money of the fund increases

PAGE 43

43 A B C Figure 2 1 Performance Flow Relationship by DC Ratio F igures visually represent the performance flow relationship. Each quarter, mutual funds in the sample are assigned to 20 equally sized groups according to their past four quarter performance within investment objective category (horizontal axis).Vertical axis reports mean quarterly flows for each group. A) includes all funds in the sample, while B) and C) contain funds with lowest and highest DC plan funds, respectively

PAGE 44

44 Table 2 1 Summary Statistics This table reports the summary statistics for main variables in my sample from 1999 to 2011 All Funds Low DC Ratio Funds High DC Ratio Funds High Low DC Ratio Mean Median St. Dev. Mean Median Mean Median DC Ratio (%) 23.82 19.31 19.78 4.20 4.38 51.76 46.79 47.56 Fund Flow (quarterly, %) 0.59 1.60 6.45 0.37 1.71 0.48 1.23 0.11 (0.76) TNA (billion) 3.77 1.03 9.76 1.63 0.59 5.32 1.18 3.69 *** (16.19) Lag Return (%) 6.20 8.58 24.64 6.41 8.25 5.71 8.37 0.70 ( 1.31) Return Volatility (%) 17.92 16.67 8.81 17.48 16.23 18.49 17.41 1.01 *** (5.29) Age (years) 19.60 14.00 16.64 18.49 13.00 17.31 13.00 1.18 *** ( 3.48) Total Fees (%) 1.40 1.35 0.50 1.47 1.46 1.16 1.10 0.31 *** ( 31.14) Turnover (%) 81.14 64.00 73.36 83.48 64.00 76.07 58.00 7.41 *** (4.59) % Institutional 20.28 0.65 32.69 19.06 0.63 27.44 0.01 8.38 *** (9.96) Quarter fund Obs. 16,384 4,203 4,209 to a mutual fund manager survey. Fund Flow is calculated quarterly according Equation 2 1 in the text. and hold returns for the previous four quarters while Return Volatility is the annualized standard deviation of those monthly returns. Age is the differenc total expense ratios and one seventh of front during the previous year. This table also reports summar y statistics for subsamples of low and high DC ratio funds. These subsamples are determined according to lowest and highest quartile of DC Ratio variable (quartile cutoffs are 8.66% and 33.59%, respectively). The last column is the result of mean differe nce test between the low and high DC ratio subsamples. t stats are reported in the parenthesis. ***,**,* denote significance at 1%, 5% and 10%, respectively.

PAGE 45

45 Table 2 2 Past Performance and Fund Flows DC Ratio Variable Measured as (1) (2) Continuous (3) High DC Ratio Dummy Lag Return 0.151 (0.000) 0.164 (0.000) 0.156 (0.000) Lag Return* DC Ratio 0.057 (0.000) 0.021 (0.000) Alpha 0.396 (0.000) 0.397 (0.000) 0.402 (0.000) Alpha* DC Ratio 0.001 (0.989) 0.032 (0.319) DC Ratio 0.004 (0.271) 0.001 (0.537) Category Flow 0.260 (0.000) 0.260 (0.000) 0.257 (0.000) Log(1+Age) 0.009 (0.000) 0.008 (0.000) 0.009 (0.000) Log Lag TNA 0.001 (0.314) 0.001 (0.310) 0.001 (0.333) Total Fees 0.343 (0.007) 0.330 (0.013) 0.343 (0.010) Turnover 0.003 (0.000) 0.003 (0.000) 0.003 (0.000) Return Volatility 0.061 (0.000) 0.063 (0.000) 0.063 (0.000) Constant 0.031 (0.000) 0.030 (0.000) 0.031 (0.000) Obs. 7,914 7,914 7,914 Adj R 2 0.145 0.146 0.146 This table examines the effect of DC ratio on the relationship of raw and risk adjusted alpha returns with equity mutual fund flows. The dependent variable is Fund Flow, calculated quarterly according Equation 1 1 in the text. Lag Return is buy and hold returns for the prev ious four quarters while Alpha is abnormal return according to Carhart 4 factor model using previous 36 monthly returns. DC Ratio is the ratio of the s urvey. In column (2), it is represented as a continuous variable, and in column (3), it is a dummy variable indicating funds with DC ratios above 75th percentile of the sample. Category Flow is aggregate net flows for all funds in the same investment cat egory. Age is the difference between data year and Total Fees are estimated by adding total expense ratios and one seventh of front load fees. Turnov er is deviation of previous 12 monthly returns. Regressions are run with quarterly fixed effects and p values in the parenthesis are calculated wit h Newey West standard errors.

PAGE 46

46 Table 2 3 The Effect of DC Ratio on the Non linear Performance Flow Relationship DC Ratio Variable Measured as (1) (2) Continuous (3) High DC Ratio Dummy (4) Continuous (5) High DC Ratio Dummy Low 0.099 (0.000) 0.138 (0.000) 0.116 (0.000) 0.135 (0.000) 0.114 (0.000) Mid 0.068 (0.000) 0.071 (0.000) 0.069 (0.000) 0.069 (0.000) 0.069 (0.000) High 0.199 (0.000) 0.220 (0.000) 0.218 (0.000) 0.227 (0.000) 0.220 (0.000) Low* DC Ratio 0.173 (0.010) 0.075 (0.008) 0.155 (0.022) 0.061 (0.033) Mid* DC Ratio 0.010 (0.474) 0.005 (0.384) 0.004 (0.764) 0.005 (0.426) High* DC Ratio 0.106 (0.099) 0.091 (0.002) 0.139 (0.031) 0.095 (0.001) DC Ratio 0.034 (0.003) 0.016 (0.001) 0.004 (0.758) 0.005 (0.358) Category Flow 0.472 (0.000) 0.473 (0.000) 0.473 (0.000) 0.470 (0.000) 0.472 (0.000) Log(1+Age) 0.015 (0.006) 0.015 (0.000) 0.007 (0.003) 0.015 (0.000) 0.015 (0.000) Log Lag TNA 0.002 (0.014) 0.002 (0.013) 0.002 (0.000) 0.002 (0.000) 0.002 (0.008) Total Fees 0.285 (0.002) 0.320 (0.001) 0.328 (0.001) 0.326 (0.001) 0.327 (0.001) Turnover 0.004 (0.000) 0.004 (0.000) 0.004 (0.000) 0.004 (0.000) 0.004 (0.000) Return Volatility 0.025 (0.002) 0.026 (0.001) 0.027 (0.001) 0.058 (0.000) 0.036 (0.000) Return Volatility* DC Ratio 0.142 (0.000) 0.043 (0.000) Constant 0.029 (0.005) 0.018 (0.000) 0.013 (0.001) 0.010 (0.024) 0.011 (0.006) Obs. 16,384 16,384 16,384 16,384 16,384 Adj R 2 0.193 0.194 0.194 0.195 0.195 This table examines the effect of DC ratio on the relationship of returns with equity mutual fund flows non linearly. The dependent variable is Fund Flow, calculated quarterly according Equation 1 1 in text. Every quarter funds are ranked according to th eir performance in the last four quarters within their investment objective category (Lag Return Rank). Low, Mid, and High are performance variables within lowest, mi ddle three and highest quintiles, and are defined as , a mutual fund manager survey. In columns (2) and (4), it is represented as a continuous variable, and in columns (3) and (5), it is a dummy variable indicating funds with DC ratios above 75th percentile of the sample. Category Flow is aggregate net flows for all funds in the same investment cat egory. Age is th e ter end; and Total Fees are estimated by adding total expense ratios and one seventh of front portfolio level turnover during the previous year. Return Volatility is the annualized standard deviation of previous 12 monthly returns. Regressions are run w ith quarterly fixed effects and p values in the parenthesis are calculated with Newey West stan dard errors.

PAGE 47

47 Table 2 4 The Effect of DC Ratio on the Performance Flow Relationship while Controlling for Institutional Money DC Ratio Variable Measured as (1) Continuous (2) High DC Ratio Dummy Low 0.139 (0.000) 0.122 (0.000) Mid 0.074 (0.000) 0.072 (0.000) High 0.190 (0.000) 0.206 (0.000) Low* DC Ratio 0.164 (0.027) 0.094 (0.003) Mid* DC Ratio 0.017 (0.272) 0.010 (0.140) High* DC Ratio 0.009 (0.896) 0.054 (0.095) DC Ratio 0.006 (0.643) 0.017 (0.002) Category Flow 0.500 (0.002) 0.502 (0.002) Log(1+Age) 0.015 (0.000) 0.014 (0.000) Log Lag TNA 0.002 (0.000) 0.002 (0.000) Total Fees 0.066 (0.568) 0.052 (0.645) Turnover 0.004 (0.000) 0.003 (0.000) Return Volatility 0.052 (0.000) 0.030 (0.004) Return Volatility* DC Ratio 0.129 (0.000) 0.029 (0.145) % Institutional 0.010 (0.003) 0.009 (0.008) % Institutional* Lag Return Rank 0.007 (0.237) 0.006 (0.350) Constant 0.018 (0.002) 0.020 (0.000) Obs. 13,329 13,329 Adj R 2 0.195 0.194 This table examines the effect of DC ratio on the relationship of returns with equity mutual fund flows non linearly while controlling for institutional money invested in the fund. The dependent variable is Fund Flow, calculated quarterly according Equati on 1 1 in text. Every quarter funds are ranked according to their performance in the last four quarters within their investment objective category (Lag Return Rank). Low, Mid, and High are performance variables within lowest, middle three and highest qui ntiles, and are defined as , contribution plan investors according to a mutual fund manager survey. In column (1), it is represented as a continuous variable, and in column (2), it is a dummy variable indicating funds with DC ratios above 75th percentile of the sample. Category Flow is aggregate net flows for all funds in the same investment ratios and one seventh of front previous year. Return Volatility is the annualized standard deviation of previous 12 monthly returns. % nal class shares. Regressions are run with quarterly fixed effects and p values in the parenthesis are calculated with Newey West standard errors.

PAGE 48

48 Table 2 5 The Effect of DC Ratio on the Performance Flow Relationship while Controlling for Fund Distribu tions DC Ratio Variable Measured as (1) Continuous (2) High DC Ratio Dummy (3) Continuous (4) High DC Ratio Dummy Low 0.136 (0.000) 0.116 (0.000) 0.136 (0.000) 0.116 (0.000) Mid 0.069 (0.000) 0.070 (0.000) 0.069 (0.000) 0.070 (0.000) High 0.227 (0.000) 0.219 (0.000) 0.227 (0.000) 0.219 (0.000) Low* DC Ratio 0.158 (0.019) 0.075 (0.008) 0.157 (0.019) 0.075 (0.008) Mid* DC Ratio 0.004 (0.753) 0.005 (0.428) 0.005 (0.743) 0.005 (0.426) High* DC Ratio 0.136 (0.035) 0.093 (0.002) 0.137 (0.034) 0.093 (0.002) DC Ratio 0.005 (0.715) 0.021 (0.012) 0.004 (0.745) 0.012 (0.012) Category Flow 0.470 (0.000) 0.471 (0.000) 0.470 (0.000) 0.471 (0.000) Log(1+Age) 0.015 (0.000) 0.016 (0.000) 0.015 (0.000) 0.015 (0.000) Log Lag TNA 0.002 (0.000) 0.002 (0.000) 0.002 (0.000) 0.002 (0.000) Total Fees 0.320 (0.001) 0.304 (0.002) 0.320 (0.001) 0.304 (0.002) Turnover 0.004 (0.000) 0.004 (0.000) 0.004 (0.000) 0.004 (0.000) Return Volatility 0.058 (0.000) 0.038 (0.000) 0.058 (0.000) 0.038 (0.000) Return Volatility* DC Ratio 0.142 (0.000) 0.051 (0.004) 0.143 (0.000) 0.051 (0.004) Total Distributions 0.296 (0.000) 0.295 (0.000) 0.336 (0.000) 0.315 (0.000) Total Distributions* DC Ratio 0.225 (0.422) 0.111 (0.691) Constant 0.010 (0.040) 0.011 (0.040) 0.010 (0.042) 0.011 (0.004) Obs. 16,384 16,384 16,384 16,384 Adj R 2 0.197 0.196 0.197 0.196 This table examines the effect of DC ratio on the relationship of returns with equity mutual fund flows non linearly while controlling for distributions paid by the fund. The dependent variable is Fund Flow, calculated quarterly according Equation 1 1 in text. Every quarter funds are ranked according to their performance in the last four quarters within their investment objective category (Lag Return Rank). Low, Mid, and High are performance variables within lowest, middle three and highest quintiles, an d are defined as , contribution plan investors according to a mutual fund manager survey. In columns (1) and (3), it is represented as a continuous variable, and in columns (2) and (4), it is a dummy variable indicating funds with DC ratios above 75th percentile of the sample. Category Flow is aggregate net flows f or all funds in adding total expense ratios and one seventh of front turnover during the previous year. Return Volatility is the annualized standard deviation of previous 12 monthly returns. Total Distributions are the distributions paid by the fund d uring the quarter. Regressions are run with quarterly fixed effects and p values in the parenthesis are calculated with Newey West standard errors.

PAGE 49

49 Table 2 6 The Performance Flow Relationship Comparison between Low DC, Mid DC and High DC Funds Table 2 6. Panel A. (1) Low DC Funds (2) Mid DC Funds (3) High DC Funds Low 0.100 (0.000) 0.121 (0.000) 0.061 (0.012) Mid 0.069 (0.000) 0.069 (0.000) 0.064 (0.000) High 0.212 (0.000) 0.229 (0.000) 0.134 (0.000) Category Flow 0.362 (0.000) 0.458 (0.000) 0.640 (0.000) Log(1+Age) 0.012 (0.000) 0.012 (0.000) 0.014 (0.000) Log Lag TNA 0.001 (0.482) 0.002 (0.000) 0.001 (0.039) Total Fees 0.034 (0.876) 0.385 (0.003) 0.780 (0.000) Turnover 0.004 (0.001) 0.005 (0.000) 0.002 (0.263) Return Volatility 0.088 (0.000) 0.014 (0.199) 0.035 (0.031) Constant 0.002 (0.798) 0.022 (0.000) 0.009 (0.227) Obs. 4,037 8,275 4,072 Adj R 2 0.166 0.217 0.172 Table 2 6. Panel B (1) Low DC Funds (2) Mid DC Funds (3) High DC Funds Low 0.112 (0.000) 0.132 (0.000) 0.055 (0.010) Mid 0.069 (0.000) 0.065 (0.000) 0.069 (0.000) High 0.212 (0.000) 0.194 (0.000) 0.182 (0.000) Category Flow 0.349 (0.000) 0.468 (0.000) 0.621 (0.000) Log(1+Age) 0.012 (0.000) 0.013 (0.000) 0.012 (0.000) Log Lag TNA 0.001 (0.536) 0.003 (0.000) 0.001 (0.773) Total Fees 0.112 (0.539) 0.298 (0.054) 0.800 (0.000) Turnover 0.005 (0.000) 0.003 (0.014) 0.004 (0.007) Return Volatility 0.064 (0.000) 0.033 (0.015) 0.042 (0.002) Constant 0.004 (0.520) 0.030 (0.000) 0.006 (0.354) Obs. 5,421 5,505 5,458 Adj R 2 0.189 0.204 0.190 These tables compare the performance flow relationship between funds according to their DC money level. In Panel A, Low DC Funds are defined as funds with less than 25th percentile of DC money level (less than 8.66%), while High DC Funds hold higher than 75th percentile of DC money level (more than 33.59%). Mid DC Funds are the funds with DC money between these two figures. In Panel B, Low DC Funds are defined as funds with less than 33th percentile of DC money level (less than 12.17%), while High DC Fun ds hold higher than 67th percentile of DC money level (more than 27.31%). Mid DC Funds are the funds with DC money between these two figures. The dependent variable in all regressions is Fund Flow, calculated quarterly according Equation 1 1 in text. E very quarter funds are ranked according to their performance in the last four quarters within their investment objective category (Lag Return Rank). Low, Mid and High are performance variables within lowest, middle three and highest quintiles, and are def ined as , contribution plan investors according to a mutual fund m anager survey. In column (1), it is represented as a continuous variable, and in column (2), it is a dummy variable indicating funds with DC ratios above 75th percentile of the sample. Category Flow is aggregate net flows for all funds in the same invest ment ratios and one seventh of front load fees. previous year. Return Volatility is the annualized standard deviation of previous 12 monthly returns. Total Distributions are the distributions paid by the fund during the quarter. Regressions are run with quarterly fixed effects and p values in the parenthesis are calculated with Newey West standard errors.

PAGE 50

50 Table 2 7 The Effect of DC Ratio on the Performance Flow Relationship and Tax loss Selling DC Ratio Variable Measured as (1) Continuous (2) High DC Ratio Dummy Low 0.131 (0.000) 0.117 (0.000) Low* Q4 0.020 (0.011) 0.002 (0.931) Mid 0.073 (0.000) 0.069 (0.000) Mid* Q4 0.007 (0.473) 0.002 (0.780) High 0.202 (0.000) 0.206 (0.000) High* Q4 0.065 (0.131) 0.044 (0.175) Low* DC Ratio 0.134 (0.089) 0.068 (0.021) Low* DC Ratio* Q4 0.115 (0.450) 0.022 (0.454) Mid* DC Ratio 0.024 (0.134) 0.009 (0.210) Mid* DC Ratio* Q4 0.050 (0.121) 0.012 (0.397) High* DC Ratio 0.068 (0.356) 0.075 (0.026) High* DC Ratio* Q4 0.140 (0.353) 0.053 (0.438) DC Ratio 0.029 (0.031) 0.015 (0.002) DC Ratio* Q4 0.140 (0.581) 0.006 (0.426) Q4 0.010 (0.178) 0.008 (0.097) Category Flow 0.513 (0.000) 0.513 (0.000) Log(1+Age) 0.144 (0.000) 0.144 (0.000) Log Lag TNA 0.002 (0.000) 0.002 (0.000) Total Fees 0.334 (0.001) 0.333 (0.001) Turnover 0.004 (0.000) 0.004 (0.000) Return Volatility 0.116 (0.028) 0.017 (0.021) Constant 0.017 (0.001) 0.013 (0.001) Obs. 16,384 16,384 Adj R 2 0.198 0.198 This table examines the effect of DC ratio on the relationship of returns with equity mutual fund flows with a special attention to fourth quarter flows. The dependent variable is Fund Flow, calculated quarterly according Equation 1 1 in text. Every quar ter funds are ranked according to their performance in the last four quarters within their investment objective category (Lag Return Rank). Low, Mid, and High are performance variables within lowest, middle three and highest quintiles, and are defined as , plan investors according to a mutual fund manager survey. In column (1), it is represented as a continuous variable, and in column (2), it is a dummy variable indicating funds with DC ratios above 75th percentile of the sample. Q4 is a dummy variable equal to 1 the quarter is fourth, zero otherwise. Ca tegory Flow is aggregate net flows for all funds in the same investment category. Age is the difference previous quarter end; and Total Fees are estimated by adding total expense ratios and one seventh of front is the annualized standard deviation of previous 12 monthly returns. Regressions are run with quarterly fixed effects and p values in the parenthesis are calculated with Newey West standard errors.

PAGE 51

51 CHAPTER 3 EFFICIENCY AND STYLE IN 401(K) PLANS Defined contribution plans, such as those provided by employers under section 401(k) of the U.S. tax c ode, figure prominently in the retirement planning of millions of working Americans. These 401(k) plans also offer a significant number of individuals access to the financial markets through their contributions and allocation decisions. By the end of the year 2011, more than 67% of the $4.53 trillion in defined contribution assets were invested in 401(k) plans. 1 Sponsoring employers face the task of providing 401(k) plans that would efficiently enable their employees the opportunity to reach all potentia l (but risky) outcomes, depending on individual employee risk appetites. On the other hand, the availability of such opportunities may not be effectively exploited by plan participants. This limitation may be due to a number of factors, such as bias, beh avioral inertia, framing or lack of financial literacy, as evidenced by a number of studies (cf. Benartzi and Thaler (2001), Agnew et al. (2003), Duflo and Saez (2002), Van Rooij et al. (2007) among many others). 2 One might be inclined to think that 401 (k) plans with larger option menus are more likely to be efficient. Though there is no universal definition of efficiency, two definitions have been particularly popular in academic circles. One is based on the notion of spanning and its objective is to assess the extent to which the mean variance efficient frontier generated by the funds in a plan is similar to that obtained from the 1 20 12 Investment Company Institute Fact Book 2 Other examples of this literature include Liang and Weisbenner (2002), Ameriks and Zeldes (2004), Madrian and Shea (2001), Hancock (2002), Lusardi and Mitchell (2007), and Huberman and Sengmueller (2004).

PAGE 52

52 index funds in a given benchmark. (cf. DeRoon and Nijman (2001) for a review.) The second popular view on efficiency is c aptured via the Sharpe ratio and has been used by Goldreich and Halaburda (2013). They show that, under certain mild conditions, the Sharpe ratio decreases as the number of funds in a plan increases. This objective result is separate from the behavioral preference for smaller menus. In some way it conflicts with the claim made by Tang et al. (2010) according to which the number of options is less important than the range of funds provided. In fact, Tang et a l. (2010) argue that most plan sponsors offer adequate options but plan participants are the main source of inefficient pension investments. They then suggest that educational programs or default investment strategies be pursued. Our view is that educati onal programs are generally rarely followed through by the vast majority of plan participants. Furthermore, these are unlikely to be comfortable with un intuitive and uneven fund recommendations resulting from optimization. In contrast, my ultimate concl usion is to support plan sponsors in recommending a subset of funds which the participants can be free to allocate, even naively according to the so Among the recent studies that evaluate the spanning efficiency of 401(k) plans, Elto n et al. (2006) and Tang et al. (2010) refer to the same benchmark consisting of eight market indices. The respective proportions of plans found to span relative to this benchmark in Elton et al. (2006) and Tang et al. (2010) differ drastically and this d ifference may be explained by the more recent and substantially larger data set of

PAGE 53

53 Tang et al. (2010). 3 With an even more comprehensive data set, my objectives in this chapter are threefold. I first observe that there may be more than one reasonable benc hmark to use for the spanning test. Second, given that these spanning tests are joint sign tests on the intercept for regressions where the predictors are the excess returns of the funds in a plan and the dependent variables are those of benchmark indices I should account for the impact of collinearity. Consequently, I perform spanning tests using (i) the same benchmark indices as those of Elton et al. (2006) and Tang et al. (2010) with two variants, one where I employ all the funds in a plan and another where I use only those that are picked in a mean variance optimization; and (ii) a different, style driven benchmark with two variants as in (i). This benchmark choice is partly predicated on the preponderance of return style analysis for fund performanc e evaluation and also on behavioral attributes of the average investor. Specifically, it is unlikely that the average plan participant performs any kind of optimization but, rather, is allocation across funds. This plan participant is more likely to pick funds based on fund description or style and, in fact, there is considerable evidence in support of style level momentum in individual portfolio allocations (cf. Barberis an d Shleifer (2003) and Wahal and Yavuz (2013) and references therein). Our study shows that if I pick the same benchmark as Elton et al. (2006) and Tang et al. (2010) and consider all the funds with each plan, then 81% span, as compared with 53%, for the f ormer, and 97%, for the latter. On the other hand, if only the optimal funds are included, then 96% span based 3 In co mparison with Elton et al. (2006), Tang et al. (2010) analyze spanning of retirement plans in more recent period and include only Vanguard sponsored 401(k) plans which are known to be low cost and well diversified.

PAGE 54

54 on my data and according to the same benchmark. Furthermore, using a set of 13 indices that is typically employed in the style analysis literat ure, 4 I find that when funds in a plan that is mean variance optimal is included, then 99.7% of all plans span. The main drivers for my two variants, with and with out optimization, are partly related to fiduciary responsibility, on the one hand, and to individual behavior in the presence of choice proliferation, on the other. Another reason for my two variants is due to the highly likely presence of collinearity am ong plan funds, thus reducing their number to only those that optimize should help reduce this phenomenon. While both plan sponsor and third party administrator (TPA) undertake some actions to shoulder the burden of fiduciary responsibility, it remains unc lear whether either would go so far as to make specific recommendations within a plan. 5 With my spanning focus on the optimal set of funds within a plan, I hopefully offer some basis for making certain recommendations, especially given that there is consi derable evidence that individuals prefer to choose among fewer alternatives. 6 Furthermore, when given these fewer choices, or if they pick a subset out of a larger offering, individual participants tend to (2001), Iyengar et al. (2004) and Iyengar and Kamenica (2010)). Tang et al. (2010) further assess the impact of the deviation of the risk adjusted performance across all participants relative to the mean variant efficient portfolios and 4 Both sets of benchmark indices are fu lly described in my spanning section below. 5 Third party administrator is a generic term used in the 401(k) industry for the financial institution charged with administering the plan. These financial institutions provide administrative services and are often relied on by the plan sponsor to provide expertise when choosing the investment menu for the plan. 6 See Tversky and Shafir (1992), Iyengar and Lepper (2000), and Boatwright and Nunes (2001)

PAGE 55

55 conclude that while plans may be spanning, individual portfo lio construction is overwhelmingly inefficient. Their recommendation is then to support strategies targeting behavioral change including improved default strategies and educational programs. However, behavioral change is difficult to achieve (see, e.g., Iyengar and Kamenica (2010)). In fact, the estimation issues that arise in the course of mean variance optimization have led some to question the practicality of mean variance efficient eally be a viable strategy (see DeMiguel et al. (2009) and references therein). Tang et al. (2010) conclude that individual investors are inefficient by comparing their realized returns my emphasis) be optimally attai ned for the same post and the other ex ante. In contrast, DeMiguel et al. (2009) compare Sharpe ratios for strategies on an out of sample basis and find that gai ns from optimal diversification can be more than offset by estimation error. They find that for parameters calibrated to U.S. stock market data, for a portfolio of 25 assets, for example, the required estimation window is 3000 months. Though they refrain heuristic, they strongly urge using it as a benchmark for comparison purposes. In this vein, I variance optimal portfolio for each of the pla ns in my sample. Using a Mahalanobis type of distance to account for estimation volatility, my results indicate that there is a In this chapter, I contribute to the fragm ented literature on 401(k) plan menu efficiency. My more recent and rich sample allows me to test for spanning in retirement

PAGE 56

56 plans without sample bias issues. Additionally, I document that details in spanning tests, however small they may seem, can affec t results. Therefore, I carefully describe my steps in spanning methodology which might be helpful for future studies in this literature. Additionally, I find that although it may seem counterintuitive, limiting the fund choices in retirement plans to th e subset that is part of an optimized portfolio increases the chances of spanning. This implies that plan participants can benefit from information on this subset which can be easily provided by plan administrators. Even if the participant chooses to eng are better off to choose the smaller subset rather than full fund choices available. The remainder of the chapter is organized as follows. The next section describes my data sources and sample d etails. In the following I describe my methodology of spanning tests in greater details and present resul ts using my sample. T he effect of optimization on the style exposure of retirement plans is also studied at end of the analysis. Finally, I summari ze the results Data My primary dataset is provided by Brightscope, Inc., an independent information provider of retirement plan ratings and investment analytics to plan participants, sponsors, asset managers, and advisors. These information services are important because they provide individuals with resources to better educate themselves about investment decision in a somewhat opaque retirement plan market. Brightscope provides data on retirement plan quality such as overall plan rating, company matching generosity, and plans costs which ultimately may help potential employees to evaluate retirement benefits before joining a company.

PAGE 57

57 defined contribution (DC) plans, such as 401(k) and 403(b) plans. The specific dataset provided to me is a cross sectional snapshot of plans at the end of 2007, and it contains over 25,000 DC plans. Items included in the data that are important to my analysis include: plan menu investment fund options, plan size (net a ssets), individual fund balances, fund expense ratios, administrative costs, and plan sponsor and service provider information. For my analysis in this study I by using the Department of Labor codes, and I eliminate any supplementary plans offered by the same plan sponsor. 7 This initial sample includes 17,386 DC plans. I further require full return data availability for every mutual fund within a plan in 2004 2008 period from either CRSP Mutual Fund Database or Morningstar Direct Database. Because of this data constraint my final sample used in the analysis consists of 7 ,991 DC plans. However, my sample is significantly larger and richer than previous studies that have analyzed retirement plan menu efficiency. For instance, Elton et al. (2006) analyze a relatively small sample of 417 plans and while Tang, et al. (2010) analyze a larger sample of 1,003, all of these plans are administered by one of the top mutual fund 7 I do this because I I simply an

PAGE 58

58 companies in the industry Vanguard. 8 My data covers both publicly traded and private companies of all sizes which hire many types of TPAs. 9 Table 3 1, Pa nel A provides comparative descriptive statistics for the initial and the final sample. The average plan in my initial sample has $22.1 million in total net assets and offers 22.1 funds as investment options. 10 In contrast, the average plan in my final sa mple is larger than the average plan in the initial sample with $31.8 million in total assets and contains 18.2 fund options in its investment menu with a balance of $1.9 million per fund. For the majority of my sample I can also identify the plan third p arty administrator (TPA), the financial institution in charge of designing and servicing the retirement plan. TPA categories used in my analysis can be found in Panel B of Table 3 1, along with plan size characteristics across these different categories. Mutual fund families represent a heavy majority of the subsample where I can identify the TPA. This is consistent with the overall retirement plan market where mutual fund companies hold a majority of the market share. In my sample, plans administered b y investment banks, large commercial banks, asset management advisory firms, and mutual fund companies are considerably larger than plans administered by small/regional commercial banks, 401(k) services companies, and insurance firms. This is not particul arly surprising since 8 for profit firms. Sample used in Tang et al (2010) is provided by Vanguard, a company that is well kno wn to provide low cost and well diversified portfolios with heavy preference to index funds. 9 I categorize TPAs in my sample into one of seven categories: mutual fund families, large/small (greater than or less than $50 billion in assets) commercial bank s, insurance companies, asset management advisory companies, investment banks, and 401(k) services companies 10 The average plan size in my initial sample is very comparable to the average plan size ($25.2 million) er of information on 401(k) plan.

PAGE 59

59 the clientele of the latter groups are most likely to be smaller firms with fewer participants and lower retirement plan balances. 11 In Table 3 2, I provide more details on the types of funds offered in investment menus of plans in the sample. As summarized in Panel A, an average retirement plan in my sample offers 12.3 domestic equity funds, 1.8 domestic bond funds, 1.9 international funds and 0.6 low risk investments such as money market funds, stable value funds, guaranteed investme nt contracts, or annuities (MSGA). 12 Not surprisingly, almost all retirement plans include at least one domestic equity fund while 97% of plans offer at least one domestic bond fund and one international fund in their investment menus. Additionally, about 60% of plan menus contain at least one MSGA, while 4% of plans in the sample also include company stock as one of investment choices. 13 Plan participants in my sample, on average, direct 68% of their retirement wealth to domestic equity funds, 9% to domes tic bond funds, 14% to international funds and 9% to MSGA options. When company stock is offered in the DC plan, the average plan participant also invests 13% of plan assets in the stock. 14 Further, Panel B reports the average number of unique Lipper Inve stment Objective categories covered by plans in the final sample. On average, 13 different objective categories are represented in plan menus. 11 One exception is the insurance company group. One drawback of my dataset is the fact that insurance firms are underrepresented due to common use of propriety funds that do not exist in my return data sources (CRSP or Mo rningstar). 12 In my analysis I only focus on mutual funds in the plan and exclude company stock and all MSGA options. 13 Interestingly, retirement participants of about 92% of plans in my sample have also borrowed against their retirement wealth, on averag e 8.3% of their plan balance. 14 Although this figure seems quite high, often there are special incentives in investing in company stock for employees.

PAGE 60

60 For purposes of spanning and style analysis I use two sets of benchmark funds. I obtain all return data for the se benchmarks from DataStream. Table 3 3 lists both sets of benchmark indices and provides descriptive statistics on monthly returns over the analysis period. Despite some changes in management company affiliations, the first set of benchmark are identic al to sets used by related spanning literature, including Elton et al. (2006) and Tang et al. (2010). In the first set, Barclays Capital Aggregate Bond Index, Credit Suisse High Yield Bond Fund, and Citigroup World Government Bond Non US$ Index capture re turns of fixed income securities; Russell 1000 Growth, Russell 1000 Value, Russell 2000 Growth, and Russell 2000 Value indices captures returns of large, mid and small equities, and MSCI EAFE Index provides international exposure. In the second set, each of investment categories are represented with more benchmark indices. This set is well used in the style analysis literature, including Sharpe (1992). In this set, fixed income benchmarks are Barclays Government Intermediate, Barclays United States Aggr egate Long Government/Credit, Barclays Investment Grade: Corporates, Barclays US Agency Fixed Rate MBS, Citigroup World Government Bond Index World 5+ Year Non USD; equity benchmarks are Standard p 400/ Citigroup Value; 600/ Citigroup Growth, and finally, international benchmarks are MSCI Europe, MSCI Pacific and S&P IFCI Emerging Market Index. As summarized in Table 3 3, average month returns on most benchmark indices was negative in the 2004 2008 period due to the financial crisis of 2008. 15 15 Average monthly returns are positive of all indices in both sets if year 2008 returns are removed from

PAGE 61

61 Finally, for portfolio optimization in both spanning and style analysis I estimate expected fund returns with the help of Fama French 3 factor model and obtain historical Spanning Tests Suppose a 401(k) plan consists of a set of funds. To assess the efficiency of this plan relative to a benchmark of index funds, we want to determine whether the mean variance frontier associated with the funds coincides with that generated with the augmented set of funds. In other words, the funds frontier of the funds. Note that this spanning notion is based on the first two moments only and differs from a similarly labeled spanning in the context of derivatives pricing, where it applies to every state of the world not just in the mean and variance sense. Formally, denote by and the expected returns vectors of, respectively, the K plan funds and the N benchmark indices. Also define the corresponding covariance matrices , where subscripts R and r refer to the funds and benchmark indices, respectively. Then the covariance matrix across the assets is defined as ( 2 1) with the superscript used for matrix transposition. A mean variance efficie nt portfolio is efficient across the assets if it solves the optimization problem ( 2 2) where is the vector of expected returns across assets, is the coefficient of risk aversion, is the ( long unit vector, and the Lagrange the descriptive statistics.

PAGE 62

62 multiplier is the interpreted as the expected return on the global minimum variance portfolio. The plan funds are mean variance spanning if the optimal portfolio is of the form ( 2 3) where is associated with funds and is an dimensional vector of zeros, and satisfies the first order condition ( 2 4) for all values of and Solving for above yields ( 2 5) where is dimensional unit vector. Substituting this solution in the last rows of the first order conditions above, and after rearranging terms, yields ( 2 6) where defined as These deterministic equations can be used to specify regression models to empirically test for spanning by replacing the expected returns with their corresponding realized returns. Regression based spanning tests go back to Huberman and Kandel (1987). However, DeRoon et al. (2001) were the first to account for the explicit inclusion of short sales constraints that are integral to fund investing in 401(k) plans. With the excess returns of the plan funds over the risk free rate as predictors and those of the benchmark indices as dependent variables, they show that spanning occurs if all intercepts are jointly not statistically significantly positive. Testing for this non positivity constraint involves a Wald test where the test statistic is a mixture of distributions, for

PAGE 63

63 which critical values have long be en known to only be obtainable via Monte Carlo simulation (see e.g., Wolak (1989)) or via their upper and lower bounds (Kodde and Palm (1986)), with the latter often crude. Specifically, the regression form is ( 2 7 ) where is the e xcess return on month for the of the indices of the reference benchmark, is an loading vector and is the vector of benchmark excess returns. A positive intercept is then interpreted as the additional risk adjusted mean return that thanks to the short sale constraint, would be attained through additional funds in the plan. Thus the null hypothesis for the spanning test is where is the transpose of the vector ( ), 0 is the vector of 0 elements, and the ineq uality is element wise. The corresponding test statistic is given by (c.f., Perlman (1969), Kodde and Palm (1986) and Wolak (1989)): (2 8 ) where is the OLS estimate for in (1) above, is the inverse of the covariance matrix for the components of and the superscript is for transposition. The critical value at level satisfies (2 9 ) where is a random variable following the distribution with degrees of freedom, and the weight is the probability that the optimal solution for (2) has exactly negative elements (out of ). I should recall here that is a random variable as it depends on the statistics (random variables) and The

PAGE 64

64 determination of rests on that of these weights, a challenging task that is usually performed via Monte Carlo simulation. Kodde and Palm (1986) provide upper and lower bounds for which apparently are relied upon in both Elton et al. (20 06) and Tang et al. (2010). 16 However, at the 5% level, for an 8 index benchmark, the lower and upper bounds for the critical value are, respectively, 2.706 and 14.853, thus consisting of a substantial range. In fact, with my data, and for the same benchm ark with 8 indices, more than 60% of the cases fall within these bounds and for which no decisions can be made regarding rejection of the null. 17 I thus resolve to estimating , via Monte Carlo simulation. Following Wolak (1989), I generate a sa mple of 1000 values from a multivariate normal distribution with mean zero and covariance matrix Replacing in (2) by a sample vector , leads to a minimizing solution In turn, these lead to estimating by the proportion out of vectors that have exactly negative entries. The critical value is then determined as the root of the equation below: 18 (2 10) I performed my spanning tests on two variants of my data, with one using all the funds available in each plan, and the other restricted to only the funds that were part of the optimal portfolio (in a mean variance sense) within each plan. For this mean variance optimization, I considered two types of esti mators for the intra plan fund 16 See Tang et al. (2010) footnote on p. 1078. 17 These cases can go as high as 93% with alternative benchmark set of 13 indic es. 18 I

PAGE 65

65 covariance matrix. The first is based on the standard MLE estimate, and the other is based on the shrinkage method of Ledoit and Wolf (2004), one of the many available to reduce estimation error (cf. DeMiguel et al. (2009)). For my spanning test, I also used two benchmark types: (i) the same eight (or their close equivalent since their studies) as in Elton et al. (2006) and Tang et al. (2010), and (ii) an alternative with 13 indices in Sharpe (1992) ( I also used for my return s style analysis in the next section). As displayed in Table 3 4, relative to the same benchmark used by Elton et al. (2006) and Tang et al. (2010), and instead of their respective values of 53% and 97%, I find that 81% of plans span. If, on the other ha nd, participants are steered toward picking funds that are mean variance optimal within a plan, then spanning occurs around 96 or 97 percent of the time. Incidentally, the choice of covariance estimation method in the optimization procedure appears not to matter at this global level but leads to significantly different fuzzy cases (critical value falling within the lower and upper bounds of Kodde and Palm (1986)). On the other hand, should my alternative benchmark of 13 indices be used, spanning occurs at a much higher rate and is not affected by whether all funds or only optimal funds are picked, as illustrated in Table 3 5. 19 With both sets of indices it is clear from the results that optimization increases the number of plans that span in my sample. If the TPA and/ or the plan sponsor would choose to offer guidance on the set of funds in the plan menu that are part of the mean variance optimized portfolio then a plan participant is more likely to create an investment portfolio that spans a set of benc 19 Although results reported in Tables 4 and 5 are at 5% significance level, I also repeat my analysis with 10% significance level and find qualitatively similar results.

PAGE 66

66 plan sponsors are known to be reluctant in offering specific investment recommendations in order to avoid a full fiduciary responsibility on behalf of the participants. 20 Style Perspective Individual participants ar e unlikely to engage in optimizing their allocations in their 401(k) plans, or anywhere else for that matter. More likely, during initial portfolio setup they will chase winners or be influenced by fund style descriptions (cf. Barberis and Shleifer (2003) ; Wahal and Yavuz (2013)). Fund descriptions however are often vague because of their limitations to just a handful of words (such as growth, value, balanced, income, etc.). In addition, their managers tend to regularly readjust their holdings, thus indi cating that fund returns do not necessarily fit the style description provided in the first place. The style analysis approach of Sharpe (1992) is widely used in practice for the classification and performance evaluation of mutual funds (see e.g., Chan et al. return on cash and a set of equity classes. Their corresponding regression coefficients are constrained to be non negative and to sum up to one. They thus represent t he for performance evaluation. For my implementation I used the same asset 13 indices as for the spanning tests with an addition of US One month T bill. For each pl an, I determined style vectors for two portfolios, one mean I 20 Highlighted in a recent Government Accountability Office (GAO) report http://www.gao.gov/new.items/d11119.pdf>.

PAGE 67

67 highlighted earlier the difficulty of reliably implementing mean variance optimization, pointing to a study (DeMiguel et al. (2009)) even q uestioning its out of sample efficacy large amount of data required for accurate parameter estimation, which is essential to a stable implementation (cf. references i n DeMiguel et al. (2009) for an overview). Their the promising and efficient optimization procedure devised by Brandt et al. (2009), which relies on constructing strateg ies based on a given benchmark allocation. I therefore endeavor to compare the return styles of the optimal and naive allocations in this spirit. Table 3 6 contains select percentiles of the loading, as well as their differences, for both strategies acros s all plans in my data. The most striking disparity between the two strategies tends to occur along the value dimension. Whereas, the equity and some fixed income f unds, the optimal strategy style loads on large, mid, and small capitalization equity funds, with significant skewness towards small equity funds. The style vectors thus summarized are further analyzed to account for the inherent variability in their estim ation. I thus use the following Mahalonobis distance: (2 11) where and are the style vectors for the naive and optimized allocations, respectively, and is the covariance matrix of The involvement of the inverse is to diminish the influence of large variability in the difference The regression used is

PAGE 68

68 (2 12 ) where is the portfolio excess return at time with standing for in one case and in the second, is the corresponding excess return vector for the asset classes, with element wise and This regression is run separately for the optimized and naive portfolios, and thus cannot generate a direct estima te for the covariance of To estimate the latter, I ran the following two regressions: (2 13) with element summing to zero, and (2 14) with the same restriction and an intercept pinned at zero. The Mahalonobis distance can also be viewed as a Wald statistic following a distribution with 14 degrees of freedom. At the 5% level I found that for 6,698 plans out of 7,991 plans, the style vectors are statistically different when using the first specification, and the same result obtains for 6,710 plans when using the second specification with pinned intercept. My results show that optimization within retirement plans shifts the style exposu re of the investment portfolio, and this shift is statistically significant for more than 80% of the plans in the sample. Specifically, the optimization strategy creates higher exposure to value equity funds. In contrast, the nave strategy is heavily ex posed to large value caps and mid cap growth, with a significant exposure to Europe equity and non negligible exposure to small cap growth, U.S. real estate, and U.S. government short term borrowing. The optimal strategy is consistent with empirically lon g held view of the superiority of value strategies (see e.g., Lakonishok et al. (1994)). On the other hand, the nave strategy is likely a reflection of the data sample period.

PAGE 69

69 C hapter 3 Concluding Remarks Employer sponsored 401(k) plans play a determinan t role in the retirement of millions of Americans and affect financial markets in significant fashion. The main aspect of their efficiency has centered on (i) whether plans offered funds that would enable participants to reach all available risk return op portunities, and (ii) whether plan participants would indeed take advantage of these opportunities. In this chapter I assess the efficiency of 401(k) plan offerings on the basis of their spanning properties and find the evaluation to vary significantly b ased on the benchmark used. I also show that if plan sponsors or providers steer participants toward funds that are included in mean variance optimal allocations, spanning occurs irrespective of the benchmark considered. Though spanning is a property of f unds available for investments within plans, it is generally acknowledged that plan participants are not likely to optimize their allocations to take advantage of available risk reward opportunities. Even if offered educational opportunities, participants are unlikely to follow through with optimizing strategies and are not likely to be comfortable with optimal yet unintuitive default strategies. Furthermore, parameters estimates needed for optimization are notoriously unstable and inaccurate. In additio n, participants are in fact more likely to follow styles and/or choose the so that such allocation is a reasonable benchmark upon which one might be able to refine into more optimal strategi es, I then compare returns styles for both optimal and naive strategies and found them to differ drastically, suggesting that despite the reasonable strong reason for the pursuit of optimal strategies.

PAGE 70

70 My results imply that plan administrators need to take more active role in providing guidance to plan participants in their investments decisions. More specifically, my recommendation is for the plan administrators to e ncourage participants to consider only the funds picked by the optimization and let them allocate according to their conduct their own optimization analysis to limit thei r choices to the optimal set of funds, that information needs to be provided by the plan administrators. Though my recommendation differs from earlier educational and default optimal strategies, it is consistent with individual subjective preference for l esser decision making choices and with recent studies suggesting more efficient plans with lesser options.

PAGE 71

71 Table 3 1. Plan Level Descriptive Statistics Table 3 1. Panel A. Beginning Sample N = 17,386 Mean Standard Deviation 10th percentile 25th percentile Median 75th percentile 90th percentile 22,100 111,000 1,157 2,534 5,894 13,700 34,200 Average Fund Size 1,228 8,456 63 133 313 734 1,865 Number of Fund Options 22.13 15 11 15 19 26 32 Final Sample N = 7,991 Mean Standard Deviation 10th percentile 25th percentile Median 75th percentile 90th percentile 31,827 136,058 1,141 2,684 6,697 17,172 52,676 Average Fund Size 1,891 8,564 77 180 426 1,059 3,110 Number of Fund Options 18.21 6.78 10 13 17 23 28 Table 3 1. Panel B. Number of Plans Mean Standard Deviation Min Median Max Mutual Fund Families 2,340 60,089 272,790 37 13,448 9,641,714 Asset Management Advisory 519 40,030 204,915 49 6,557 2,602,408 Investment Banks 183 91,400 355,811 14 14,628 4,040,556 Large Commercial Banks 725 41,197 139,039 70 8,508 2,261,397 Small/Regional Comm. Banks 203 12,821 33,027 10 4,756 349,804 Insurance Firms 435 15,002 32,758 147 5,826 470,023 401(k) Services Companies 362 18,860 62,782 15 4,091 5,825,942 TPA Unknown 3,224 13,183 52,622 3 4,090 1,146,085 Panel A reports descriptive statistics for the size of plans and fund balances in the overall sample provided by BrightScope, Inc. These only include 401(k) plans that me to eliminate a large number of plans from the main analysis, Panel A also reports descriptive statistics for the final sam ple. Panel B reports descriptive statistics on plan size by the type of company administering the 401(k) plan i.e. third party administrator (TPA).

PAGE 72

72 Table 3 2 Plan Menu Options Descriptive Statistics Table 3 2. Panel A. Number of Fund Options (unconditional) Number of Fund Options (conditional) % of Plans Assets Held in (unconditional) % of Plans Assets Held in (conditional) % of Plans Containing at least one Mean St. Dev. Mean St. Dev. Mean St. Dev. Mean St. Dev. Domestic Equity Funds 12.36 6.01 12.37 6.00 68.18% 13.50% 68.24% 13.35% 99.92% Domestic Bond Funds 1.82 1.16 1.88 1.13 8.97% 19.72% 9.25% 19.96% 96.98% International Funds 1.86 1.16 1.92 1.13 14.36% 9.63% 14.82% 9.43% 96.90% Money Market/ Stable Value/ GIC/ Annuity (MSGA) 0.63 0.57 1.06 0.29 8.80% 11.77% 14.88% 11.99% 59.12% Company Stock 0.04 0.21 1.06 0.23 0.51% 3.95% 12.97% 15.36% 3.93% Participant Loans 7.67% 9.32% 8.32% 9.43% 92.17% Table 3 2. Panel B. Mean Standard Deviation 10th percentile 25th percentile Median 75th percentile 90th percentile 13.21 6.03 6 9 12 17 22 The panels report information related to the final sample used in the analysis sections. Panel A provides frequency data on the different types of long with ratio of plan assets directed to ve categories.

PAGE 73

73 Table 3 3 Benchmark Indices Monthly Excess Returns Gross Net of Fees Mean St.Dev. Mean St.Dev. Set 1 MSCI EAFE Index 0.04% 4.73% 0.03% 4.73% Barclays Capital Aggregate Bond Index 0.14% 1.35% 0.12% 1.35% Credit Suisse High Yield Bond Fund 0.25% 2.94% 0.29% 2.94% Citigroup World Government Bond Non US$ Index 0.26% 2.37% 0.25% 2.37% Russell 1000 Growth 0.45% 4.05% 0.47% 4.05% Russell 1000 Value 0.24% 3.75% 0.26% 3.75% Russell 2000 Growth 0.28% 5.62% 0.30% 5.62% Russell 2000 Value 0.10% 4.98% 0.11% 4.98% Set 2 Barclays Government Intermediate 0.17% 0.94% 0.19% 0.94% Barclays United States Aggregate Long Government 0.25% 2.91% 0.27% 2.91% Barclays Investment Grade Corporates 0.56% 2.09% 0.57% 2.09% Barclays US Agency Fixed Rate MBS 0.21% 0.93% 0.19% 0.93% Non US$ 0.33% 2.58% 0.32% 2.58% Standard and Poor's 500 / Citigroup Value 0.52% 3.87% 0.53% 3.87% Standard and Poor's Mid cap 400 / Citigroup Value 0.26% 4.73% 0.28% 4.73% Standard and Poor's Mid cap 400 / Citigroup Growth 0.23% 4.87% 0.25% 4.87% Standard and Poor's Small cap 600 / Citigroup Value 0.18% 4.93% 0.20% 4.93% Standard and Poor's Small cap 600 / Citigroup Growth 0.09% 5.00% 0.11% 5.00% MSCI Europe Index 0.22% 4.95% 0.22% 4.95% MSCI Pacific Index 0.12% 4.84% 0.12% 4.84% S&P IFCI Emerging Market Index 0.52% 7.31% 0.52% 7.31% This table lists the set of benchmark indices used in spanning and style analysis, and it provides descriptive statistics on monthly returns for the analysis period (2004 2008). Set 1 indices are consistent with Elton et al. (2006), Tang et al. (2010), while Set 2 indices are more aligned with those used in Sharpe (1992).

PAGE 74

74 Table 3 4 Spanning Test with 8 Benchmark Indices. Outside of Bounds Fuzzy Total Span No Span Span No Span Span No Span Without Optimization 2,217 1,064 4,265 443 6,482 1,507 27.75% 13.32% 53.39% 5.55% 81.14% 18.86% Optimization method covariance matrix) 4,518 156 3,166 149 7,684 305 56.55% 1.95% 39.63% 1.87% 96.18% 3.82% Optimization (MLE estimate covariance matrix) 3,606 173 4,129 81 7,735 254 45.14% 2.17% 51.68% 1.01% 96.82% 3.18% This table summarizes spanning test results at 5% significance level with 8 indices: MSCI EAFE Index; Barclays Capital Aggregate Bond Index; Credit Suisse High Yield Bond Fund; Citigroup World Government Bond Non US$ Index; Russell 1000 Growth; Russell 1000 Value; Russell 2000 Growth; c falls within the upper and lower bounds of the critical value provided by Kodde and Palm (1986). For these cases the critical value was estimated via Monte Carlo simulation following Wolak (1989) and 1000 simulation runs. Results correspond to two vari ants of my data (i) without performing mean variance optimization at plan level and allowing all funds within the plan to be included in the spanning test, and (ii) performing mean variance optimization and keeping funds with positive weights in the optimi zed portfolio. Mean variance optimization is also performed in two different estimators for the intra plan fund covariance matrix. and Wolf (2004), whil e the last row results employ standard MLE estimate of the matrix. N=7,991

PAGE 75

75 Table 3 5 Spanning Test with 13 Benchmark Indices. Outside of Bounds Fuzzy Total Span No Span Span No Span Span No Span Without Optimization 3,037 110 4,718 124 7,755 234 38.01% 1.38% 59.06% 1.55% 97.07% 2.93% Optimization method covariance matrix) 509 11 7,454 15 7,963 26 6.37% 0.14% 93.30% 0.19% 99.67% 0.33% Optimization (MLE estimate covariance matrix) 637 8 7,330 14 7,967 22 7.97% 0.10% 91.75% 0.18% 99.72% 0.28% This table summarizes spanning test results at 5% significance level with 8 indices: US One month T bill; Barclays Government Intermediate; Barclays United States Aggregate Long Government / Credit; Barclays Investment Grade : Co rporates; BarCap US Agency Fixed Rate MBS; Citigroup World Government Bond Index World 5 + Year Non Value; Growth; S Citigroup cases where the Wald statistic falls within the upper and lower bound s of the critical value provided by Kodde and Palm (1986). For these cases the critical value was estimated via Monte Carlo simulation following Wolak (1989) and 1000 simulation runs. Results correspond to two variants of my data (i) without performing m ean variance optimization at plan level and allowing all funds within the plan to be included in the spanning test, and (ii) performing mean variance optimization and keeping funds with positive weights in the optimized portfolio. Mean variance optimizati on is also performed in two different estimators for the intra covariance through transformation proposed by Ledoit and Wolf (2004), while the last row results employ standard M LE estimate of the matrix. N=7,991

PAGE 76

76 Table 3 6 Style Exposure of Retirement Investors Style exposure with optimized weights Style exposure with 1/N weights Difference Mean Q1 Median Q3 Mean Q1 Median Q3 Mean Q1 Median Q3 US One month T bill 1.003 0.001 0.001 0.001 5.591 2.242 4.913 8.605 4.588 7.315 3.797 1.504 Barclays Government Intermediate 0.167 0.001 0.001 0.001 3.483 0.001 0.001 5.599 3.316 5.422 0.001 0.001 Barclays United States Aggregate Long Government 1.666 0.001 0.001 0.001 2.072 0.001 1.042 3.639 0.405 2.359 0.001 0.001 Barclays Investment Grade Corporates 1.360 0.001 0.001 2.130 3.84 0.001 2.236 6.825 2.48 6.458 0.954 0.001 Barclays US Agency Fixed Rate MBS 0.916 0.001 0.001 0.001 7.271 0.001 5.966 11.768 6.355 10.627 5.344 0.001 Bond Index World 5+ Y Non US$ 0.701 0.001 0.001 0.001 0.813 0.001 0.001 0.815 0.112 0.135 0.001 0.001 Standard and Poor's 500 / Citigroup Value 17.667 0.001 11.243 0.001 24.579 20.11 24.904 29.162 6.912 21.625 13.046 1.983 Standard and Poor's Mid cap 400 / Citigroup Value 17.666 0.001 14.659 27.411 0.572 0.001 0.001 0.001 17.095 0.001 14.277 29.116 Standard and Poor's Mid cap 400 / Citigroup Growth 7.725 0.001 2.52 29.813 24.398 20.402 24.342 28.158 16.673 24.276 19.442 10.452 Standard and Poor's Small cap 600 / Citigroup Value 30.32 2.624 30.121 12.391 2.346 0.001 0.001 3.651 27.974 2.624 26.491 47.872 Standard and Poor's Small cap 600 / Citigroup Growth 8.767 0.001 0.001 51.126 7.076 3.528 6.965 10.223 1.691 6.614 1.022 7.397 MSCI Europe 6.25 0.001 1.698 13.553 12.411 10.416 12.527 14.551 6.16 11.762 8.22 1.978 MSCI Pacific 1.895 0.001 0.001 11.161 3.619 2.561 3.508 4.573 1.725 3.777 2.515 0.432 S&P IFCI Emerging Market Index 3.897 0.001 0.764 2.233 1.931 0.001 1.152 2.666 1.965 1.039 0.001 3.457 ch column reports mean, quartile 1, median, and quartile 3 loads for each benchmark index according to the allocation weights (in % terms). The last column reports simi lar summary statistics for the difference between the style exposures between the two strategies. For 6,698 plans out of 7,991 pl ans the difference in style exposures is statistically different according to Mahalonobis distance test (slight variation of the test yields 6,710 plans to be different)

PAGE 77

77 CHAPTER 4 CONCLUSION In this study, I examine two different but related topics in finance: the impact of retirement investors on mutual fund flows and efficiency and style tests of retirement plan menus. Both topics are important for both retirement plan and mutual fund liter ature In first part of the study, Chapter 2 I examine whether defined contribution plan investors affect the sensitivity of fund flows to recent past performance since these investors are believed to be more passive and less sophisticated. Despite the d ramatic growth of these types of retirement plans and the mutual funds flows from these plans, the performance flow literature to date has largely ignored these investors. Due to fundamental differences between DC retirement plan investors and retail mutu al fund investors, I hypothesize that a higher concentration of DC investments in a particular I find that the sensitivity of flows to past performance decreases with th e ratio of assets owned by DC plans. This effect holds for funds both in the low and high end of the performance spectrum although the effect is much larger in size for low performing funds. In other words, DC investors are both less likely to pull money from recently poor performing funds (as funds migrate from bad to worse in rank) than retail investors, and are less likely to move to funds with recent good performance (as funds migrate from good to very good in rank). Additionally, I find that as the DC ratio of the fund increases, the sensitivity of the flows to fund return volatility decreases. Relative to low DC ratio funds, funds with high DC ratios continue observing investors direct higher fund flows to them, even in cases of poor return perform ance and higher volatility. My findings suggest that DC retirement account participants indeed tend to be more

PAGE 78

78 passive investors exhibiting lower sensitivity to both returns and fund risk. Results are consistent with DC investors being less sophisticated and making infrequent adjustments to their investment choices which would result in less sensitive fund flows. Overall, in this chapter, I provide convincing evidence that DC money changes the slope of the performance flows relationship in mutual funds at the low and high ends of the performance spectrum. Since the shape of this relationship also has implications dencies, DC ratio may also affect these behaviors, although in an unclear direction. While lower sensitivities to performance might weaken propensity to change fund risk, my may offset that effect. Th year risk shifting behavior needs to be re analyzed empirically as the ratio of DC money of the fund increases. In Chapter 3, I focus on efficiency and style exposure of e mployer sponsored 401( k) plans. These plans play a determina nt role in the retirement of millions of Americans and affect financial markets in significant fashion. The main aspect of their efficiency has centered on (i) whether plans offered funds that would enable participants to reach all available risk return o pportunities, and (ii) whether plan participants would indeed take advantage of these opportunities. In this chapter I assess the efficiency of 401(k) plan offerings on the basis of their spanning properties and find the evaluation to vary significantly based on the benchmark used. I also show that if plan sponsors or providers steer participants toward funds that are included in mean variance optimal allocations, spanning occurs irrespective of the benchmark considered. Though spanning is a property of funds available for investments within plans, it is generally acknowledged that plan participants are not likely to optimize their allocations to take

PAGE 79

79 advantage of available risk reward opportunities. Even if offered educational opportunities, participant s are unlikely to follow through with optimizing strategies and are not likely to be comfortable with optimal yet unintuitive default strategies. Furthermore, parameters estimates needed for optimization are notoriously unstable and inaccurate. In additi on, participants are in fact more likely to follow styles and/or choose the so such allocation is a reasonable benchmark upon which one might be able to refine into more optimal strateg ies, I then compare returns styles for both optimal and naive strategies and found them to differ drastically, suggesting that despite the reasonable strong reason for the pursuit of optimal strategies. My results imply that plan administrators need to take more active role in providing guidance to plan participants in their investments decisions. More specifically, my recommendation is for the plan administrators to encourage participants to consider only the funds picked by the optimization and let them allocate according to their conduct their own optimization analysis to limit the ir choices to the optimal set of funds, that information needs to be provided by the plan administrators. Though my recommendation differs from earlier educational and default optimal strategies, it is consistent with individual subjective preference for lesser decision making choices and with recent studies suggesting more efficient plans with lesser options.

PAGE 80

80 APPENDIX COMPARISON OF RESULTS TO SIALM, STARK, AND ZHANG (2012) This study is directly related to Sialm, Stark, and Zhang (2012) although I employ a slightly different methodology to answer a related question. Sialm, Stark, and Zhang (2012) are testing to see whether flows that span from DC sources or from non DC sour ces are more responsive to performance measures. To answer this question they calculate growth of DC money and non DC money held by a fund according to below two equations: (A 1) (A 2) Further, authors regress these two flow variables against performance rank variables along with traditional set of controls. They find that the coefficients on the performance variables for DC flow regression are higher than those for Non DC flow regression, implying that the growth of DC money is more responsive to performance than the growth of inflows from the retail investors. It is worth noting that an analysis with this methodology ca n only be conducted at the annual level because the survey of DC money is collected once a year. 1 Additionally, it should also be noted that Sialm, Stark, and Zhang (2012) define the performance rank variable slightly differently than the previous studie s in the performance flow literature. In my study, I rank the historical performance of the fund in 1 Please also note that this methodology creates much larger DC flow values for funds with low levels of DC money.

PAGE 81

81 comparison to all available funds with the same investment category in CRSP Mutual Fund database, a method consistent with studies such as Sirri and Tuffan o (1998), Huang, Wei, and Yan (2007), and James and Karceski (2006). This method of comparing a fund to the universe of available funds while controlling for its investment objective is important because each fund is competing for flows and is usually bei ng evaluated within its risk exposure, captured by the investment objective categories. Sialms, Stark, and Zhang (2012), however, construct performance rank measures that compare the fund against other funds that in the sample (as opposed to universe of f unds) and without any attention to investment objective. I believe that this method might drive some of their results in favor of more sensitivity with DC flows. To test this prediction, I make an attempt to replicate their results first, and then change the performance rank measure definition. Another important difference between the sample used in my study and that in Sialm, Stark, and Zhang (2012) is the exclusion of index funds. As suggested in the literature, since flows to index funds are more lik ely to be driven by macroeconomic events rather than the performance on behalf of the fund manager, I choose to eliminate index funds in my analysis while Sialm, Stark, and Zhang (2012) do not. With the hopes of replicating their results with my data, I al so conduct similar analysis without exclusion of index funds and report my findings in Table A 1 Panel A, specifications (1) (3). Although my sample periods are slightly different 2 I find a similar pattern of results as in Sialm, Stark, and Zhang (2012) where DC flow is more sensitive to past performance variables when compared to Non DC flow. However, I do not 2 Our sample period covers 1999 2001, while Sialm, Stark, and Zhang (2002) analyze 1996 2009 period in their study.

PAGE 82

82 reach quite the same levels of statistical significance as reported in their Table 3. Next, in specifications (4) (6) I change the performance r ank variables to be calculated as within sample and within investment objective category. With this alternative performance definition I find that sensitivity of highest performance to DC vs. Non DC flows actually becomes opposite, implying that Non DC fl ows are more sensitive to the performance of High. I further change the definition of performance rank variables to be within universe of funds and without attention to investment objective. As reported in Panel B specifications (1) (3), now both Low and High performance variables become more sensitive to Non DC flows in comparison with DC flows. Lastly, in specifications (4) (6), I employ the most commonly used performance definition in the literature: within universe of funds and within investment obje ctive category. Similar change of direction is observed for highest performing group, where the Sialm, Stark, and Zhang (2012) findings do not hold. I also drop all index funds to repeat these regressions and find very similar results as reported in Tabl e A 2. Findings in this appendix may shed light on somewhat puzzling results that are found in Sialm, Stark, and Zhang (2012). It seems that a specific choice of definition for fund performance measures is affecting the results found in their study. Howe ver, it is important to note that since I do not have their exact same sample, I cannot confirm this prediction definitively as I fail to replicate the statistical significance in their results. Fund Mergers One particular concern in the performance flow literature has been mergers between mutual funds. Because flows are not directly observable and have to be estimated by using E quation 1 1 in my study, fund mergers can create large errors in flow estimations. The most common way of decreasing effects of mergers in the

PAGE 83

83 literature has been truncation of observations where there are large flow outliers. In the main analysis of this chapter I follow the literature in this regard and eliminate observations that fall in the top and bottom 2.5% of flow data. Alternatively, one could use CRSP fund merger identifier to correct or to control for fund mergers. With the help of this identifier, I found 353 fund mergers in my sample and have created a variable called Fund Merger Ratio, defined as assets of the ac quired fund divided by assets of the acquirer fund. 3 I control for this variable in my main regressions in Table A 3 where I find that my results are robust to fund mergers. It should also be noted that Fund Mergers Ratio has a positive but not statistic ally significant effect on fund flows. This may be due to the overwhelming number of zeros in this variable because there are very few mergers in comparison to my overall sample. Additionally, instead of controlling for Fund Merge Ratio variable, I also use it as a criterion to truncate the data with high values of merge ratio and find that my results are also robust 3 For example, if Fund A with 50 million assets merges into Fund B with 100 million and seizes to exist, then Fund Merge Ratio for Fund B in the same period would be 0.5.

PAGE 84

84 Table A 1. The Effect of Performance Variable on DC Flow vs. Non DC Flow Performance Relationship. Table A 1. Panel A (1) DC Flow (2) Non DC Flow (3) Difference (4) DC Flow (5) Non DC Flow (6) Difference Low 0.692 (0.016) 0.369 (0.012) 0.323 (0.307) 0.804 (0.004) 0.378 (0.010) 0.426 (0.166) Mid 0.257 (0.000) 0.243 (0.000) 0.014 (0.851) 0.343 (0.000) 0.296 (0.000) 0.047 (0.555) High 0.954 (0.013) 0.722 (0.000) 0.232 (0.534) 0.168 (0.564) 0.516 (0.002) 0.349 (0.240) Category Flow 0.734 (0.000) 0.544 (0.000) 0.189 (0.373) 0.922 (0.000) 0.670 (0.000) 0.251 (0.247) Log(1+Age) 0.160 (0.000) 0.059 (0.000) 0.101 (0.000) 0.159 (0.000) 0.057 (0.000) 0.102 (0.000) Log Lag DC TNA 0.136 (0.000) 0.021 (0.002) 0.158 (0.000) 0.135 (0.000) 0.023 (0.001) 0.158 (0.000) Log Lag Non DC TNA 0.115 (0.000) 0.049 (0.000) 0.163 (0.000) 0.111 (0.000) 0.053 (0.000) 0.164 (0.000) Total Fees 2.754 (0.455) 1.826 (0.367) 0.928 (0.789) 0.794 (0.831) 0.703 (0.726) 0.092 (0.979) Turnover 0.038 (0.012) 0.002 (0.889) 0.040 (0.001) 0.038 (0.012) 0.002 (0.903) 0.040 (0.002) Return Volatility 0.207 (0.286) 0.346 (0.002) 0.139 (0.473) 0.242 (0.189) 0.355 (0.000) 0.112 (0.551) Constant 0.435 (0.000) 0.325 (0.000) 0.110 (0.330) 0.393 (0.000) 0.304 (0.000) 0.089 (0.417) Obs. 3,571 3,571 3,571 3,571 3,571 3,571 Adj R 2 0.141 0.112 0.103 0.141 0.121 0.104 Performance Rank Defined In within sample within sample and within objective category

PAGE 85

85 Table A 1. Continued Table A 1. Panel B (1) DC Flow (2) Non DC Flow (3) Difference (4) DC Flow (5) Non DC Flow (6) Difference Low 0.213 (0.541) 0.307 (0.043) 0.093 (0.799) 0.742 (0.004) 0.522 (0.010) 0.220 (0.166) Mid 0.492 (0.000) 0.326 (0.000) 0.166 (0.052) 0.299 (0.000) 0.234 (0.000) 0.065 (0.555) High 0.670 (0.074) 0.833 (0.000) 0.163 (0.674) 0.859 (0.564) 1.122 (0.002) 0.263 (0.240) Category Flow 0.732 (0.000) 0.548 (0.000) 0.184 (0.388) 0.924 (0.000) 0.685 (0.000) 0.239 (0.247) Log(1+Age) 0.162 (0.000) 0.060 (0.000) 0.102 (0.000) 0.155 (0.000) 0.053 (0.000) 0.102 (0.000) Log Lag DC TNA 0.136 (0.000) 0.021 (0.002) 0.157 (0.000) 0.137 (0.000) 0.021 (0.001) 0.158 (0.000) Log Lag Non DC TNA 0.114 (0.000) 0.049 (0.000) 0.163 (0.000) 0.112 (0.000) 0.052 (0.000) 0.164 (0.000) Total Fees 2.852 (0.439) 1.806 (0.375) 1.045 (0.763) 0.875 (0.831) 1.161 (0.726) 0.286 (0.979) Turnover 0.040 (0.007) 0.001 (0.992) 0.041 (0.001) 0.036 (0.012) 0.003 (0.903) 0.039 (0.002) Return Volatility 0.223 (0.275) 0.356 (0.001) 0.133 (0.517) 0.226 (0.189) 0.345 (0.000) 0.119 (0.551) Constant 0.455 (0.000) 0.304 (0.000) 0.151 (0.215) 0.364 (0.000) 0.246 (0.000) 0.118 (0.417) Obs. 3,571 3,571 3,571 3,571 3,571 3,571 Adj R 2 0.128 0.096 0.102 0.145 0.132 0.103 Performance Rank Defined In within universe of funds within universe of funds and within objective category These tables examine the effect of performance variables definition on the DC flow performance and Non DC flow performance relationships. In both panels DC flow and Non DC flow are calculated according to Sialm, Stark and Zhang (2012), and also are expla ined in the Appendix. The fundamental difference between two panels is the way Low, Mid, and High performance variables are calculated. In Panel A, ev ery quarter funds are ranked according to their performance in the last four quarters within the sample with and without attention to investment objective category (Lag Return Rank). In contrast, in Panel B, funds are ranked according to their performance in the last four quarters within the universe of available funds with and without attention to investme nt objective category (Lag Return Rank). Further, Low, Mid, and High performance variables are calculated as , respectively. Category Flow is ion ta l expense ratios and one seventh of front andard deviation of previous 12 monthly returns. Regressions are run with quarterly fixe d effects and p values in the parenthesis are calculated with Newey West standard errors

PAGE 86

86 Table A 2 The Effect of Performance Variable on DC Flow vs. Non DC Flow Performance Relationship Excluding Index Funds. Table A 2. Panel A (1) DC Flow (2) Non DC Flow (3) Difference (4) DC Flow (5) Non DC Flow (6) Difference Low 0.662 (0.027) 0.345 (0.019) 0.318 (0.334) 0.836 (0.005) 0.376 (0.012) 0.459 (0.154) Mid 0.285 (0.000) 0.260 (0.000) 0.025 (0.760) 0.375 (0.000) 0.318 (0.000) 0.057 (0.495) High 0.920 (0.021) 0.689 (0.001) 0.231 (0.551) 0.081 (0.791) 0.499 (0.003) 0.418 (0.180) Category Flow 0.778 (0.000) 0.532 (0.000) 0.247 (0.268) 0.982 (0.000) 0.662 (0.000) 0.320 (0.160) Log(1+Age) 0.164 (0.000) 0.054 (0.000) 0.111 (0.000) 0.166 (0.000) 0.053 (0.000) 0.113 (0.000) Log Lag DC TNA 0.145 (0.000) 0.023 (0.003) 0.168 (0.000) 0.144 (0.000) 0.025 (0.001) 0.169 (0.000) Log Lag Non DC TNA 0.125 (0.000) 0.056 (0.000) 0.181 (0.000) 0.122 (0.000) 0.060 (0.000) 0.182 (0.000) Total Fees 3.508 (0.440) 2.414 (0.289) 1.094 (0.795) 2.180 (0.638) 2.037 (0.382) 0.144 (0.973) Turnover 0.040 (0.010) 0.001 (0.953) 0.039 (0.002) 0.040 (0.011) 0.001 (0.941) 0.039 (0.003) Return Volatility 0.182 (0.366) 0.336 (0.003) 0.153 (0.448) 0.214 (0.266) 0.340 (0.001) 0.126 (0.523) Constant 0.435 (0.001) 0.358 (0.000) 0.077 (0.551) 0.389 (0.004) 0.340 (0.000) 0.049 (0.706) Obs. 3,284 3,284 3,284 3,284 3,284 3,284 Adj R 2 0.147 0.123 0.110 0.147 0.133 0.111 Performance Rank Defined In within sample within sample and within objective category

PAGE 87

87 Table A 2. Continued Table A 2. Panel B (1) DC Flow (2) Non DC Flow (3) Difference (4) DC Flow (5) Non DC Flow (6) Difference Low 0.206 (0.571) 0.306 (0.048) 0.101 (0.790) 0.682 (0.021) 0.430 (0.004) 0.252 (0.437) Mid 0.511 (0.000) 0.339 (0.000) 0.172 (0.052) 0.343 (0.000) 0.279 (0.000) 0.064 (0.418) High 0.675 (0.084) 0.826 (0.000) 0.151 (0.709) 0.789 (0.031) 1.090 (0.000) 0.301 (0.420) Category Flow 0.777 (0.000) 0.532 (0.000) 0.245 (0.273) 0.987 (0.000) 0.681 (0.000) 0.306 (0.176) Log(1+Age) 0.166 (0.000) 0.054 (0.000) 0.112 (0.000) 0.163 (0.000) 0.051 (0.000) 0.113 (0.000) Log Lag DC TNA 0.145 (0.000) 0.023 (0.003) 0.168 (0.000) 0.145 (0.000) 0.024 (0.001) 0.164 (0.000) Log Lag Non DC TNA 0.124 (0.000) 0.056 (0.000) 0.180 (0.000) 0.112 (0.000) 0.059 (0.000) 0.181 (0.000) Total Fees 3.593 (0.430) 2.439 (0.288) 1.154 (0.784) 1.965 (0.668) 1.406 (0.532) 0.559 (0.895) Turnover 0.041 (0.007) 0.003 (0.850) 0.039 (0.002) 0.040 (0.010) 0.001 (0.928) 0.038 (0.003) Return Volatility 0.196 (0.354) 0.341 (0.003) 0.145 (0.499) 0.206 (0.290) 0.335 (0.001) 0.129 (0.515) Constant 0.452 (0.001) 0.334 (0.000) 0.118 (0.394) 0.406 (0.002) 0.319 (0.000) 0.087 (0.509) Obs. 3,284 3,284 3,284 3,284 3,284 3,284 Adj R 2 0.134 0.105 0.109 0.151 0.148 0.110 Performance Rank Defined In within universe of funds within universe of funds and within objective category These tables examine the effect of performance variables definition on the DC flow performance and Non DC flow performance relationships without any index funds in the sample. In both panels DC flow and Non DC flow are calculated according to Sialm, Stark and Zhang (2012), and also are explained in the Appendix. The fundamental difference between two panels is the way Low, Mid, and High performance variables are calculated. In Panel A, every quarter funds are ranked according to their performance in the last four quarters within the sample with and without attention to investment objective category (Lag Return Rank). In contrast, in Panel B, funds are ranked according to their p erformance in the last four quarters within the universe of available funds with and without attention to investment objective category (Lag Return Rank). Further, Low, Mid, and High performance variables are calculated as , respectively. Category Flow is aggregate net flows for all funds in the same investment category. Age is the difference d Total F ees are estimated by adding total expense ratios and one seventh of front year. Return Volatility is the annualized standard deviation of previous 12 monthly returns. Regress ions are run with quarterly fixed effects and p values in the parenthesis are calculated with Newey West standard errors.

PAGE 88

88 LIST OF REFERENCES Agnew, J., P. Balduzzi, and A. Sunden (2003). Portfolio choice and trading in a large 401(k) plan. American Economic Review 93(1), 193 215. Agnew, Julie, Pierluigi Balduzzi, and Annika Sundn, 2003, Portfolio choice and trading in a large 401(k) plan, American Economic Review 93, 193 215. Ameriks, J. and S. Zeldes (2004). How do household portfolio sha res vary with age? Working paper. Badrinath, S., and Wilbur Lewellen, 1991, Evidence on tax motivated securities trading behavior, Journal of Finance 46, 369 382. Barberis, N. and A. S hleifer (2003). Style investing Journal of Financial Economics 68, 161 199. Benartzi, S. and R. H. Thaler (2001). Nave diversification strategies in defined contribution savings plans. American Economic Review 91(1), 79 98. Benartzi, Shlomo, and Richard H. Thaler, 2001, Naive diversification strategies in defined contribution saving plans, American Economic Review 91, 79 98. Berk, Jonathan B., and Richard C. Green, 2004, Mutual fund flows and performance in rational markets, Journal of Political Economy 112, 1269 1295. Boatwright, P. and J. C. Nunes (2001). Reducing assortment : an attribute based approach Journal of Marketing 65(3), 50 63. Brandt, M.W., P. Santa Clara, and R. Valkanov (2009). Parametric portfolio policies: Exploiting characteristics in the cross section of equity returns. Review of Financial Studies 22(9), 3411 3447. Brown, Keith C., W. V. Harlow, and Laura T. Starks, 1996, Of tournaments and temptations: An analysis of managerial incentives in the mutual fund industry, Journal of Finance 51, 85 110. Brown, S. and W. N. Goetzmann (1997). Mutual fund style. Journal of Financial Economics 43, 373 399. Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57 82. Chan, L. K. C., H. L. Chen, and J. Lakonishok (2002). On mutual fund investment styles. Review of Financial Studies 15(5), 1407 3447. Chevalier, Judith A., and Glenn Ellison, 1997, Risk taking by mutual funds as a response to incentives, Journal of Political Economy 105, 1167 1200.

PAGE 89

89 Christoffersen, Susan E. K., and Mikhail Simutin, 2012, Risk taking and retirement inve sting in mutual funds, Working Paper Del Guercio, Diane, and Paula A. Tkac, 2002, The determinants of the flow of funds of managed portfolios: Mutual funds vs. pension funds, Journal of Financial and Quantitative Analysis 37, 523 557. DeMiguel, V., L. Gar lappi, and R. Uppal (2009). Optimal versus naive diversification: How inefficient is the 1/n portfolio strategy? Review of Financial Studies 22(5), 1915 1953. DeRoon, F and T. Nijman (2001). Testing for mean variance spanning: A survey. Journal of Empirica l Finance 8, 111 155. DeRoon, F., T. Nijman, and B. Werker (2001). Testing for the mean variance spanning with short sales constraints and transaction costs: The case of emerging markets. Journal of Finance 56(2), 721 742. Duflo, E. and E. Saez (2002). Par ticipation and investment decisions in a retirement Journal of Public Economics 85(1), 121 148. Elton, E., M. Gruber, and C. Blakel (2006). The adequacy of investment choices offered by 401(k) plans. Journal of P ublic Economics 90, 1299 1314. Elton, Edwin J., Martin J. Gruber, and Christopher R. Blake, 2007, Participant reaction ant the performance of funds offered by 401(k) plans, Journal of Financial Intermediation 16, 249 271. Evans, Richard B., 2010, Mutual fu nd incubation, Journal of Finance 65, 1581 1611. Fama, Eugene, and James MacBeth, 1973, Risk, return and equilibrium: Empirical tests, Journal of Political Economy 81, 607 636. Goetzmann, William N., and Nadav Peles, 1997, Cognitive dissonance and mutual fund investors, Journal of Financial Research 20, 145 158. Goldreich, D. and H. Halaburda (2013), When smaller menus are better: Variability in menu setting ability. Management S cience (to appear). Grinblatt, Mark, and Matti Keloharju, 2001, What makes investors trade?, Journal of Finance 56, 589 616. Hancock, J. (2002). Eighth defined contribution plan survey. John Hancock Financial Services. John Hancock, Boston. Holden, S., M. Hadley, and S. Lutz (2010). The economics of providing 401(k) plans: Services, fees, and expenses, 2010. ICI Research Perspective 17(4), 1 32.

PAGE 90

90 Huang, Jennifer, Kelsey D. Wei, and Hong Yan, 2007, Participation costs and the sensitivity of fund flows to pas t performance, Journal of Finance 62, 1273 1311. Huberman, G. and P. Sengmueller (2004). Performance and employer stock in 401(k) plans. Review of Finance 8, 403 443. Huberman, G. and S. Kandel (1987). Mean variance spanning. Journal of Finance 42, 873 88 8. Ippolito, Richard A., 1992, Consumer reaction to measures of poor quality: Evidence from the mutual fund industry, Journal of Law and Economics 35, 45 70. Iyengar, S. and E. Kamenica (2010). Choice proliferation, simplicity seeking, and asset allocation Journal of Public Economics 94(1), 530 539. Iyengar, S. and M. Lepper (2000). When choice demotivating: can one desire too much of a good thing? Journal of Personality and Social Psychology 79(6), 995 1006. Iyengar, S., G. Huberman, and W. Jiang (2004). How much choice is too much: determinants of individual contributions in 401K retirement plans. In: Mitchell, O.S., Utkus, S. (Eds.), Pension Design and Structure: New Lessons from Behavioral Finance. Oxford University Press, Oxford, pg. 83 95. James, Ch ristopher, and Jason Karceski, 2006, Investor monitoring and differences in mutual fund performance, Journal of Banking and Finance 30, 2787 2808. Kodde, D. and F. Palm (1986). Wald criteria for jointly testing equality and inequality restrictions. Econometrica 56, 1243 1248. Journal of Finance Ledoit, O. and M. Wolf (2004). Honey, i shrunk the sample covariance matrix. Journal of Portfol io Management 110 119. Liang, N. and S. Weisbenner (2002). Investor behavior and the purchase on company stock in 401(k) plans the importance of plan design. Finance and Economics Discussion Series 2002 36 Working Paper Lusardi, A. and M. Olivia (200 7). Baby boomer retirement security: the roles of planning, financial literacy, and housing wealth. Journal of Monetary Economics 54(1), 205 224. Lynch, Anthony W., and David K.Musto, 2003, How investors interpret past fund returns, Journal of Finance 58, 2033 2058. Madrian, B. and D. Shea (2001). The power of suggestion: inertia in 401(k) participation and savings behavior. Quarterly Journal of Economics 116, 1149 1187.

PAGE 91

91 Madrian, Brigitte C., and Dennis F.Shea, 2001, The power of suggestion: Inertia in 401( k) participation and savings behavior, Quarterly Journal of Economics 116, 1149 1187. Mitchell, Olivia S., and Stephen P. Zeldes, 1996, Social security privatization: A structure for analysis, American Economic Review (Papers and Proceedings) 86, 363 367. Odean, Terrance, 1999, Do investors trade too much?, American Economic Review 89, 1279 1298 Perlman, P. E. (1969). One side problems in multivariate analysis. Annals of Mathematical Statistics 40, 549 567. Sharpe, W.F. (1992). Asset allocation: management style and performance measurement. Journal of Portfolio Management Winter 7 19. Sialm, Clemens, and Laura T. Starks, 2011, Mutual fund tax clienteles, forthcoming in Journal of Finance. Sialm, Clemens, Laura T. Starks, and Nanjiang Zhang 2012, Defined co ntribution pension plans: sticky or discerning money?, Working Paper Sirri, Erik R., and Peter Tuffano, 1998, Costly search and mutual fund flows, Journal of Finance 53, 1589 1622. Tang, N., O. Mitchell, G. Mottola, and S. Utkus (2010). The efficiency of sponsor and participant portfolio choices in 401(k) plans. Journal of Public Economics 94, 1073 1085. ter Horst, J. R., T. E. Nijman, and F. A. deRoon (2004). Evaluating style analysis. Journal of Empirical Finance 11, 29 53. Tversky, A., and E. Shafir (1 992). Choice under conflict: the dynamics of deferred decision. Psychological Science 3(6), 358 361. Van Rooij, M. C., C. J. Kool, and H. M. Prast (2007). Risk return preferences in the pension domain: are people able to choose? Journal of Public Economics 91(3 4), 701 722. Wahal, S. and D. Yavuz (2013). Style investing, co movement. And return predictability. Journal of Financial Economics 107, 136 154. Wolak, F. (1989). Testing inequality constraints in linear econometric models. Journal of Econometrics 4 1, 205 235.

PAGE 92

92 BIOGRAPHICAL SKETCH Originally from Azerbaijan, Sabuhi graduated from Pittsburg State University with B.B.A degrees in e conomics and management After also earning his M. B.A. degree from University of Missouri Columbia he worked as a financial analyst in Sprint Co. In the fall of 2007 he joined the PhD program in f inance at the University of Florida After completing his PhD in the summer of 2013 Sabuhi will join the finance department at Kansas State University as an assistant professor of finance.