Unions can have either positive or negative effects on risk-adjusted returns in pension plans. On the positive side, a union can improve monitoring of pension advisors and asset managers. On the negative side, the union may sacrifice returns by making investments that promote union goals. This paper discusses how the structure of the pension plan affects the union's ability and willingness to sacrifice returns to promote union goals. Using panel data on over 38,000 pension plans drawn from IRS Form 5500 filings between 1988 and 2008, we find the lowest petforming plans are unionized multi-employer plans. Among defined contribution plans, the underperformance of multi-employer union plans disappears when the pension is controlled by individual participants. (JEL J32, J51)

I. INTRODUCTION

While private sector unionism has declined over the past 50 years, union pension funds continue to be an important source of retirement savings in the United States. In 2009, collectively bargained private pension plans controlled approximately $1.5 trillion in assets and covered 31.6 million people. (1) Relative to the universe of private pensions, the collectively bargained plans cover 24% of the participants and 28% of the assets.

Union leadership has a long history of encouraging its membership to leverage its control over these pension assets to promote union objectives. Unions encourage their members to invest pension assets in unionized companies and in projects or communities that employ union labor. More recently, union leadership has promoted the use of pension assets to initiate and support shareholder proposals for changes in corporate governance, and to pressure mutual funds on their proxy voting behavior.

Some commentators oppose the use of pension funds to promote union goals, arguing that

such activities reduce the risk-adjusted return on pension assets and violate the fiduciary requirements of the Employee Retirement Income Security Act of 1974 (ERISA). Union leadership argues that pro-union activities do not necessarily reduce risk-adjusted returns and could improve performance. (2)

This study discusses the various ways that unions could either enhance or harm risk-adjusted performance and empirically investigates union effects. While prior research

has focused on the effect of unions on pension performance, the most recent work is over 20 years old (Dorsey and Turner 1990). Since that study, the design of union and nonunion pensions has changed dramatically. In particular, there has been a pronounced shift from defined benefit (DB) to defined contribution (DC) plans. Also, among DC plans, participant direction has become much more popular. Our study discusses how these changes in plan design could alter the union's ability or incentive to sacrifice returns to promote union goals. Our empirical analysis of IRS Form 5500 filings over the past two decades shows that pension structure significantly alters the union effect on performance. While risk-adjusted returns are lower for multi-employer DB and DC plans, there is no evidence of a negative effect among participant-directed (PD) DC plans.

II. BACKGROUND

Unions have a long history of encouraging the use of pension assets to advance union goals. In June 1980, the American Federation of Labor and Congress of Industrial Organizations (AFL-CIO) announced that labor unions should become more actively involved in the administration of pension fund activities. (3) This pension activism comprises a wide range of activities, including selecting trustees, investing assets in unionized companies, making loans to support projects that favor union labor, and exercising voting rights on stock owned by the pension fund. Opponents question the legality of such activities since the ERISA requires that pension fiduciaries manage a plan for "the exclusive purpose of providing benefits to participants" (Section 404(1)(a)). In 1998, however, the Department of Labor provided an advisory opinion indicating that the fiduciary standards of ERISA do not preclude consideration of collateral benefits so long as "the investment offering the collateral benefits is expected to provide an investment return commensurate to alternative investments having similar risks." (4) Hence, union pension activism is acceptable so long as there is no reduction in the risk-adjusted return of the pension portfolio. Disagreement remains, however, about the impact of union pension activism on risk-adjusted returns. As discussed below, the effect on performance will depend on the specifics of the union activism.

Unions can promote their fund goals by investing in companies that employ union labor. This can be accomplished by direct purchases of a company's stock, but more recently, registered investment companies, insurance companies, and banks are designing investments that are diversified across a wide spectrum of union employers to offer greater risk diversification. For example, the Housing Investment Trust (HIT) and the Building Investment Trust (BIT) are bank-managed trusts that serve union pension plans. These trusts managed over $6 billion in assets in 2012 and are invested entirely in residential or commercial projects employing union labor. (5) Another example is the International Association of Machinists and Aerospace Workers (IAM) fund managed by State Street Group, which invests the majority of its assets in companies that either "(1) have entered into collective bargaining agreements with the IAM or affiliated labor unions or (2) are listed in the S&P 500 Index and have not been identified by the IAM as having nonunion sentiment." (6) Some banks attract union pension funds by offering "target CDs" that provide a guaranteed rate of return in exchange for a promise that the funds finance projects employing union labor. (7) One such agreement includes the purchase of a CD by a roofers' union pension fund with the agreement that low interest loans be provided for roofing projects performed by an approved union contractor.

Calabrese (1999) reports that over 80% of targeted investments by union pension funds in the 1990s were dedicated to financing union-built construction, and some evidence suggests the rising use of private debt and equity purchases to promote union goals. For example, the Union Labor Life Insurance Company (ULLICO) has a private equity fund that invests in small start-up firms in exchange for an agreement that the firm confer collateral benefits to the union--such as union neutrality or card check recognition.

Union goals can also be promoted with shareholder activism. Union-sponsored pension funds submitted 43% of shareholder corporate governance proposals in 2004 and the proposals vary in their objectives (Prevost, Rao, and Williams 2012). Some restrict executive compensation, other proposals place restrictions on takeover defenses, and still others require that union representatives be included on corporate boards or that managerial pay be tied to employee welfare.

Unions also pressure mutual funds to vote proxies in accordance with union objectives. Partly in response to union pressures, the Securities and Exchange Commission began to mandate disclosure of mutual fund proxy voting in 2003. In the same year, the AFL-CIO issued proxy voting guidelines for its union membership and began rating mutual funds based on their voting behavior (Cremers and Romano 2011). In 2005, the AFL-CIO went a step further by informing investment firms that organized labor would consider a firm's position on Social Security privatization when choosing mutual funds for pension investments (Furchtgott-Roth 2008).

The effect any particular type of union pension activism has on risk-adjusted performance will depend upon the specifics. Consider, for example, a policy that tilts investments toward the stock of companies that employ union labor. Hirsch (2007) concludes that unions, on average, lower accounting profits and the value of the firm. Lee and Mas (2012) find that a union election victory reduces a firm's stock value by about 10%. However, both studies also note that if financial markets are efficient, the decline in the stock price should reflect the expected effect of the union on the present value of all future profits, so that the expected return on stock in union and nonunion companies should be identical after the election result is made public. In this case, tilting the pension fund toward investments in unionized companies would not impact pension performance. On the other hand, if unions use pension assets to make loans below market rates for projects that employ union labor, the fund is sacrificing risk-adjusted returns to promote collateral benefits for union members.

Shareholder activism by a union pension fund could have positive or negative effects on risk-adjusted performance depending on the specific governance changes that the union proposes. For example, if the union promotes changes in corporate governance that increase union bargaining power and thus lowers corporate profits, there will be a reduction in risk-adjusted returns as the stock price falls to reflect a lower expected stream of future profits. If the change in corporate governance improves corporate performance (e.g., by improving corporate board structure or executive compensation rules), risk-adjusted performance would be enhanced.

Renneboog and Szilagyi (2009) and Prevost, Rao, and Williams (2012) show that union-sponsored proposals for changes in corporate governance can create either positive or negative stock price reactions depending on the nature of the proposal and the governance structure of the targeted firm. Agrawal (2012) finds that union voting behavior on corporate governance proposals varies depending on whether the targeted company employs union labor. For example, corporate governance proposals eliminating "poison pills" would make hostile takeovers easier and could improve stock performance. However, a hostile takeover could also result in layoffs for existing workers. Consequently, the union might favor elimination of poison pills when the firm has a nonunion workforce, but oppose elimination if it increases the risk of layoffs for union labor.

Unions could enhance pension performance by improving monitoring of pension fund sponsors, service providers, or money managers. GAO (2009) and Turner and Muir (2012) underscore the prevalence of conflicts of interest for parties involved in pension plan administration. For example, pension advisors who are compensated by the mutual fund industry with commissions could steer the pension plan toward high expense funds that pay higher commissions. The expense fees in pension funds can be quite substantial. Purcell and Topoleski (2009) explain that the median fee for both money management and administration in a sample of 401(k) plans is 0.72% of assets, with significant variation from a rate of 0.35% at the 10th percentile and 1.72% at the 90th percentile, across plan types.

While plans are required by law to disclose most of the fees that plan participants pay, the information is provided piecemeal, making comparisons across plan options difficult (GAO 2006). In a DB plan, the employer has an incentive to keep fees low because any shortfall in return comes at its expense. In a DC plan, the fees are passed through to the participants, and the employer has little incentive to keep fees low without pressure from the participants. Without a union, it may not be cost effective for any single worker to spend the time and effort necessary to assure that money managers keep expenses to a minimum. With a union, pension participants are able to act collectively, reduce the free rider problem with monitoring, and potentially improve monitoring of the pension fund manager. (8) There is, however, the question of whether agency problems for a pension trustee are greater when a trustee is appointed by union leadership rather than management.

A. The Importance of Plan Design

While the union has a variety of ways to leverage pension assets to promote union goals, several pension design features could affect the union's ability or incentive to pursue activist policies. These design features include (1) whether the pension is a single- or multi-employer plan, (2) whether the pension is a DB or DC plan, and (3) whether the plan is trustee-directed (TD) or participant-directed (PD).

In collectively bargained single-employer plans, the firm chooses the pension manager and the union is not guaranteed any control over the pension assets, though a small number of unions have used the collective bargaining process to negotiate for joint-trusteeship between union and management (Cook 2002). Our research was unable to find information on how many unions have negotiated joint-trusteeship for single-employer plans.

Unions have the greatest control over investments in multi-employer (also called Taft-Hartley) pension plans where the plan covers workers from multiple employers. Such plans are most common where workers are not attached to a particular firm, for instance those employed in the building trades and trucking industries. The Taft-Hartley Act requires that such plans be jointly trusteed by union and management. Moreover, many observers believe that unions have had effective control over multi-employer plan management. (9)

Consistent with the premise that unions have little influence on single-employer plans, Dorsey and Turner (1990) found that unions had no impact on the performance of single-employer plans, but that unionized multi-employer plans underperformed. However, they also found that the inferior performance of the multi-employer plan existing in the late 1970s disappeared during the 1980s. They suggest that the shortfall in performance was eliminated when the Department of Labor issued advisories in 1980-1981 warning that an explicit sacrifice of earnings for nontraditional objectives was a violation of ERISA fiduciary requirements. (10)

Several pension design features can influence a union's incentive or ability to use pension activism to pursue union goals. For example, whether the pension is a DB or DC plan could be important. With a DB plan, the employer is liable for any shortfall in returns. With a DC plan, the employer is liable only for contributions to the pension, and any shortfall in returns reduces the employee's account balance at retirement. Consequently, management should be less willing to sacrifice returns in a DB plan. The reverse is true for the union membership. Therefore, if the employer controls the assets, union activities that reduce performance should face less resistance in DC than DB plans, although unions may be willing to offer wage concessions in exchange for greater control over assets in DB plans. If the union has control over assets (as in a multiemployer plan), it should be more willing to sacrifice returns in a DB plan.

Another factor that potentially affects how much union pension funds sacrifice returns to promote union goals is whether a DC plan is TD or PD. In a TD DC plan, each participant receives shares of a common portfolio in proportion to his or her contributions and investment earnings. In PD plans, the pension sponsor chooses the list of investment vehicles offered by the plan, but each individual decides how much to invest in the various investment options. Among DC plans, there has been a dramatic move from TD to PD plans over the past 20 years with the percentage of DC participants who manage some or all of their pension assets rising from 15% to 86%. (11)

A switch from TD to PD plans should reduce union pension activism if workers believe that it reduces risk-adjusted returns. As a group, the membership may believe that the benefits of the pension activism (e.g., increased employment of union labor) exceed the cost (reduced pension returns). Hence, with collective decision making, the union may decide that the collateral benefits to the group outweigh any loss in returns. With participant direction, each worker makes an individual decision based on a comparison of individual benefits and costs of activism. Because the vast majority of collateral benefits associated with any individual worker's investments spill over to other workers, each worker will ignore many of the benefits that accrue to the group at large. That is, many of the collateral benefits from activism are like a public good to the union membership. Switching from trustee to participant direction introduces a free rider problem and may limit pension investments in activities that generate a public good. (12)

In review, union influence could have positive, negative, or zero effects on risk-adjusted returns. Targeting investments that generate collateral benefits for union members could reduce risk-adjusted returns, but it could have zero effect if the investments are made in a competitive and efficient asset market. Shareholder activism by unions could have either positive or negative effects on performance depending on the nature of the governance proposals. Finally, unions could improve performance with additional monitoring of pension fund managers.

The union influence over pension investment practices may differ depending on the specifics of the plan type. A union's ability to exercise control over pension assets is greatest in multiemployer plans where they are directly involved in management of the pension. Employers may allow unions to exert greater influence in DC plans, which shift the cost of any underperformance to the workers. On the other hand, unions may be more willing to sacrifice returns in a DB plan, as it comes at the expense of the employer unless the employer is able to extract wage concessions in exchange for giving the union greater control. If a DC plan is PD, the union membership is more likely to act in their individual interest (instead of the union's interest) when deciding whether to invest in a pro-union project that reduces risk-adjusted returns. Participant direction of DC plans should therefore reduce any negative effect of unions on pension performance.

III. DATA AND EMPIRICAL METHODOLOGY

To investigate the effect of unions on pension fund performance, we draw data from pension plan filings of IRS Form 5500 for private sector plans between 1988 and 2008. These data contain information on asset values, investment income, and administrative and asset management expenses that can be used to estimate an annual rate of return. Because plans with less than 100 participants fill out an abbreviated form and some are excluded from the public release database, our analysis focuses only on plans with 100 or more participants. We also exclude Employee Stock Ownership Plans (ESOP) because they are not designed to be well-diversified portfolios, and present special issues when calculating the rate of return. (13) To estimate risk-adjusted returns, we need panel data on pension returns and restrict the sample to plans that have at least 10 years of data. The resulting sample includes 38,895 pension plans, of which 6,773 are collectively bargained plans. Because our sample excludes small plans, ESOPs, and plans with fewer than 10 observations, our 2008 data cover only 3.5% of all private sector pension plans. Nevertheless, the 2008 plans in our sample are large and cover 46% of the active participants and 56% of the assets for private plans in that year.

To estimate the rate of return on pension plans in the IRS Form 5500 data, we use the approach described by McCarthy and Turner (1989). The rate of return is calculated as investment income (dividends, interest, and both realized and unrealized capital gains) divided by assets at the beginning of the year. (14) This is the same methodology used by the Employee Benefits Security Administration in its annual summary of pension returns from the IRS Form 5500.

The rate of return estimates are net of any expenses charged by external money managers (e.g., master trusts, registered investment companies, or insurance companies) and also subtract any administrative, advisory, or asset management expenses explicitly charged to the pension plan. The IRS Form 5500 data may understate administrative expenses because an employer could absorb some of these expenses and fail to report them. (15)

Figure 1 compares the annual net of expense rate of return of nonunion with single- and multiemployer union plans over our sample period (1988-2008). The value-weighted return on all New York Stock Exchange (NYSE), American Stock Exchange (AMEX), and National Association of Securities Dealers Automated Quotation System (NASDAQ) stocks is presented as a benchmark. (16) The average annual returns on union and nonunion plans track each other closely. Average pension returns are highly correlated with stock market returns, but there is less volatility in pension returns. This is to be expected as most pension funds have a significant share of assets in bonds. For example, in 2009, 39% of assets in pensions sponsored by the S&P 1500 firms were in bonds (Hartsohn 2010).

One concern with our panel data set is that it could create a survivor bias in estimated rates of return. Ex ante, we have no strong priors on the direction of any survivorship bias as plans could have fewer than 10 years of data for a number of reasons that are unlikely correlated with pension performance. For example, plans that started after 1999 would be unable to generate 10 years of data by the end of our sampling period, thus necessarily excluded from our data. Also, plans that existed prior to 2000 could have been terminated if there was a merger of firms, a bankruptcy of the sponsoring firm, or if one plan was replaced by another (e.g., a DB is replaced by a DC.) (17)

To determine whether survivor bias is a problem, we estimate a regression of net pension returns as a function of whether it survives the requirement for 10 years of data, controlling for plan characteristics and fixed year effects. The results indicate that the difference between the average net return in the survivor sample and full sample (the "survivor bias") is +13 basis points for nonunion plans, +12 basis points for single-employer union plans, and +16 basis points for multi-employer union plans. (18) Consequently, controlling for plan characteristics and year effects, restricting the sample to plans with 10 or more years of data, results in slightly higher returns for all three plan types, but the effect on differential performance across plan types is negligible. (19)

Table 1 provides the sample characteristics for nonunion (NonU), single-employer union (SingleU), and multi-employer union (MultiU) plans. Over the sample period, NonU plans had lower average returns and higher within-plan volatility than either SingleU or MultiU plans. The average annual rate of return (net of expenses) was 6.8%, 8.0%, and 6.8% for NonU, SingleU, and MultiU plans, respectively. The within-plan standard deviation of net returns was 12.3%, 11.0%, and 9.0%. On the basis of these simple comparisons, one might conclude that NonU plans underperform relative to unionized plans. However, this would ignore several important factors. First, union and nonunion plans are not in the sample for the same years. The returns need to be adjusted for which years of data are used to calculate the rate of return. To make this adjustment, we estimated a regression of net returns with dummies for plan type (SingleU and MultiU) and added year fixed effects. The results imply that, compared with NonU plans, SingleU plans have a slightly higher return (0.3%) and a slightly higher variance of returns (.03%); MultiU plans have a lower return (1.0%) and a higher variance of returns (.05%). (20) Second, union plans have many more participants than nonunion plans. Since there are scale economies in pension administration with respect to both the number of participants and assets per participant, this should, ceteris paribus, result in superior performance among the unionized plans. (21) Third, economies of scope in the administration of pension plans could give SingleU plans an advantage relative to the NonU or MultiU, since SingleU employers are the most likely to offer more than one pension plan. (22) Fourth, both SingleU and MultiU plans are approximately three times more likely than nonunion plans to be DB plans, which are more expensive to administer than DC plans. (23) Union plans are also less likely to be PD or have a 401(k), profit sharing, or money purchase feature. To the extent that such features impact the administrative costs of the plan, this could result in differential performance by union status. Finally, union plans could take on a very different risk profile than nonunion plans. Any comparison of returns across plans requires an adjustment for differential risk exposure in the plans.

To determine how unions affect risk-adjusted return performance, we use a version of the Fama-French (FF) model (Fama and French 1993). We include four risk factors for stocks and two risk factors for bonds. This methodology is a common way to examine the risk and return features of a stock or mutual fund. (24) The essence of the approach is described by the regression below where excess pension returns are regressed on a constant and six FF factors:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where i indexes the pension plan and t indexes time. The dependent variable is the annual return on the pension portfolio (net of expenses, measured in percentage points) in excess of the "risk free rate" measured as the average of the corresponding annualized 1-month U.S. Treasury bill rates. The control variables include six FF factors. The first four represent stock market risk factors: VWRF is the value-weighted returns on all stocks traded on the NYSE, AMEX, and NASDAQ in excess of the risk free rate (the annualized yield on 1-month U.S. Treasury bills); SMB (small minus big) is the difference in returns on small-versus large-cap portfolios; HML (high minus low) is the difference in returns on value and growth portfolios; MOM (momentum) represents the difference in returns on portfolios formed on size categorized as "winners" and "losers" based on returns in the prior 2-12 months. (25) The remaining two factors represent bond risk. TERM is the difference in annual returns on 20-year government bonds and 1-month treasury bills. DEF is the difference between the annual returns on 20-year government bonds and a portfolio of 20year AAA and AA corporate bonds. (26) Because the pension plan reporting year does not necessarily coincide with the calendar year, all risk factors are calculated to match the 12-month period corresponding to the plan's IRS Form 5500 filing.

In the FF model, the coefficients on the FF factors reflect the exposure to risk factors in the market. For example, if a pension is entirely invested in the overall market portfolio (as measured by VWRF), [[beta].sub.1] would equal 1; and [[beta].sub.2] through [[beta].sub.6] would equal 0. If instead the portfolio was invested entirely in 1-month U.S. Treasury bills, [[beta].sub.1] through [[beta].sub.6] would equal zero. If there were no administrative, record keeping, or investment fees for managing the pension, [alpha] would equal zero in both cases. To the extent that there are such fees, [alpha] will be negative. Idiosyncratic risk is captured in the residual. In theory, idiosyncratic risk is not rewarded with higher expected returns. (27) A plan's risk-adjusted net returns is measured by [alpha].

The FF regression is estimated separately for each pension plan and by union status using OLS. Because there are few observations per plan (minimum of 10, maximum of 21), the coefficient estimates may be estimated imprecisely making it difficult to accurately test for differences in plan behavior by union status. The large number of plans, however, potentially mitigates this problem and, as seen below, the vast majority of FF regressions have statistically significant explanatory power, which reveals statistically significant differences in the parameters by union status.

The estimated FF coefficients are used to examine the impact of unions on pension performance. First, using a procedure similar to that of Carhart's (1997) and Busse, Goyal, and Wahal's (2010) to analyze the performance of mutual fund and institutional investment managers, the union's effect on risk-adjusted performance is estimated by regressing each plan's measure of risk-adjusted performance including the year-specific residual from the FF equation ([[alpha].sub.i] + [e.sub.it]), on a dummy variable, indicating whether the plan is collectively bargained. Our measure of performance varies across time for a given plan because the residual from the first stage equation is added to [[alpha].sub.1]. This allows us to estimate whether control variables that vary over time for a given plan (or across plans) systematically explain its variation in performance over time. The standard errors for the coefficients in the risk-adjusted return regression are corrected for clustering of residuals by plan and heteroskedastic errors. Because union status might correlate with other variables impacting performance, we add controls for other pension characteristics that might affect performance and year dummies that might account for changes in performance owing to regulatory or technological changes potentially affecting administrative or asset management expenses.

IV. RESULTS

A. FF Regressions

A separate FF regression is estimated for each of the 38,895 pension plans in the panel data set. The means of the estimated coefficients are presented in Table 2 along with the percentage of times that each coefficient is statistically different from zero at the .05 level. (28) The number of regressions corresponds to the number of pension plans in the panel data set (38,895). The average of the regression [R.sup.2] values is .88. The excess return on the market equity index is the FF factor that has the greatest explanatory power. The average coefficient is .52, which is statistically different from zero at the .05 level for 93% of pension plans. The fact that the coefficient is substantially below one is consistent with Figure 1, which shows that the return on the average pension plan tracks the market index but is less volatile. The other FF factors have smaller coefficients (in absolute value) and are less frequently statistically significant. A test of the joint hypothesis that all the coefficients on the six factors equal zero is rejected for 95% of pension plans. Overall, despite the small sample sizes for each pension plan, the regressions generally have statistically significant explanatory power.

The results of the FF regressions suggest that SingleU plans have the highest average risk-adjusted performance (a). The average estimate of a is -.13 for SingleU, -.44 for NonU, and -.71 for MultiU. The coefficients on the market index suggest that MultiU plans have the least exposure to equities while NonU plans have the highest. MultiU plans have the greatest exposure to the term and default factor. The greater exposure to the term and default factors is consistent with the fact that, among the plans that report asset holdings, MultiU plans hold a larger share of assets in government and corporate bonds. (29) The fact that risk exposure differs across plan type could reflect differences in the preferences of the members. For example, if the workers in MultiU plans are older or more risk averse, on average, they would likely prefer a larger share of assets to be invested in bonds. Alternatively, it might be that union control over assets affects the taste for risk--though we do not have a theory for why union trustees would be more or less conservative than their members. Unfortunately, we do not have data on the characteristics of plan members and are unable to investigate the underlying reasons for differential risk exposure across plan types. Nevertheless, the systematic differences in risk exposure highlight the importance of risk-adjusting returns to compare performance across plans.

Figure 2 plots the estimated kernel density distributions for risk-adjusted net returns ([alpha]) by union status and shows that the lower mean of risk-adjusted returns for union plans is not the result of outliers in the data. The observed performance differentials could be due to the factors correlated with unionism. Regression analysis is required to isolate the effect of unionism.

B. Regression Analysis of Excess Returns

Table 3 presents several specifications analyzing the determinants of risk-adjusted net pension returns ([alpha]). (30) With the exception of the single-employer and multi-employer union plan dummies, all variables are measured as deviations from means so that the intercept represents the mean value of [alpha] for the nonunion sample holding other plan characteristics equal to the mean for all plans.

Besides year and industry controls, specification (1) contains a dummy variable for single-employer union plans and another for multi-employer union plans. Compared with NonU plans, SingleU plans outperform nonunion plans by 19 basis points, and the difference is statistically significant at the .01 level of significance. MultiU plans underperform by 22 basis points. This simple specification, however, does not adjust for several factors that might affect administrative and money management costs.

To control for plan differences that might influence performance, we first estimate separate regressions for DB and DC plans (specifications 2 and 3) without any controls for plan characteristics. Among DC plans, SingleU plans outperform the NonU plans by 48 basis points; the performance of MultiU plans is not significantly different from the performance of NonU plans. For DB plans, the performance of neither SingleU plans nor MultiU plans is significantly different from the performance of NonU plans.

These simple comparisons do not control for the fact that union and nonunion plans' differences could account for the performance differentials. In specifications (4) and (5), we add controls for number of participants, assets per participant, the age of the plan, year and industry dummies, and whether the employer offers other plans. (31) The estimated effects of these controls are generally consistent with expectations. For example, the positive effects of per capita assets and the positive but diminishing effects of the number of participants suggest scale economies in both dimensions. Also, the better performance of older plans could reflect start-up costs and/or improved efficiencies from experience.

As discussed previously, unions can exercise the greatest influence on investments in multiemployer plans, which is where we expect to find the largest union effect. In single-employer plans, unions have limited ability to implement activist policies since the employer makes the investment decisions. The union could, however, improve performance by strengthened monitoring of pension fund managers. We also predict that a MultiU DB plan will be more willing to sacrifice returns as underperformance comes at the expense of the employer unless the employer is able to extract wage concessions to compensate for the loss of performance.

The pattern of results in specifications (4) and (5) are consistent with our hypotheses. First, MultiU plans underperform relative to NonU plans, and the difference is statistically significant at the .01 level. Moreover, the underperformance is worse among DB plans (76 basis points) than DC plans (36 basis points).

In contrast to the findings for MutliU plans, we find no significant performance differential between SingleU and NonU plans. Among DC plans, SingleU plans have a slight performance edge (14 basis points) over NonU plans, but the effect is statistically insignificant at the .10 level. Among DB plans, the SingleU plans slightly underperform (4 basis points), but this effect is not statistically significant at the .10 level either.

Overall, the empirical results imply that SingleU plans do not underperform, whereas MultiU plans do, particularly among DB plans. There are three competing explanations for the underperformance of MultiU plans: (1) they are more expensive to administer; (2) MultiU plans sacrifice returns to promote union goals; or (3) there are greater agency problems in MultiU plans causing a sacrifice of returns for the trustees' own personal gains. An inability to accurately measure administrative expenses with IRS Form 5500 data precludes determining whether higher administrative expenses are the source of underperformance. Administrative expenses are inaccurately measured, as asset management practices vary. For example, if a plan invests its funds with a mutual fund or insurance company, the administrative expenses for managing the underlying funds would simply be subtracted from the funds' returns and will not appear as expenses on the IRS Form 5500. On the other hand, a pension fund that hires an investment advisor to manage its funds will show the explicit charges for the investment management services on the IRS Form 5500. To further complicate matters, some employers may absorb part of the administrative costs associated with the pension plan by providing office space, bookkeeping staff, and other expenses that do not reduce net returns. A MultiU plan is not affiliated with any single-employer, thus less likely to have an employer absorb administrative costs.

C. The Role of Participant Direction

While we are unable to provide a direct test of the relative importance of competing explanations for the underperformance of MultiU plans, the rapid growth of participant direction in pension plans could shed some light. In a TD pension, all participants hold shares in a common pool of assets controlled by the pension manager. The pension manager frequently employs investment advisors to assist with asset allocation decisions. In a PD plan, the plan sponsor chooses a list of investment options and each participant decides how to allocate his or her own assets. In most cases, PD plans consist of offerings from a registered investment company (mutual funds) or an insurance company.

A pension switching from TD to PD could affect performance in several ways. First, the asset manager must now manage a separate account for each participant rather than one common account, which increases expenses. Conditional on plan size, the effect of PD on expenses is unlikely to differ for union and nonunion plans and thus should not affect the relative performance of union and nonunion plans. Second, PD could make it easier for workers to monitor pension performance. Because most PD plans offer a list of mutual funds, it is easy to track performance and compare it to benchmarks. Tracking performance could be much more difficult in a TD plan that might hold individual stocks or bonds. Once again, any improvement in the ability to monitor from PD should have similar effects on union and nonunion plans and should not affect their relative performance. Finally, PD could reduce the union's ability to control assets since individual plan members decide how to invest their assets. This shift in control over asset allocation decisions could alter plan performance, depending on who selects better investments. If the union is willing to sacrifice returns to improve performance, PD should improve performance of union relative to nonunion plans.

To examine the effect of PD on risk-adjusted returns, we restrict our sample to DC plans and estimate risk-adjusted return regressions separately for PD and TD plans. Because data on asset allocations used in subsequent regressions are not available for the period 1988-1989, the sample is restricted to plan year observations from 1990 forward. In the first two specifications, the models control for all the plan characteristics used in Table 3. With the exception of the dummies for SingleU and MultiU, all variables are measured as deviations from means. As a result, the intercept represents the estimate of risk-adjusted performance for a NonU plan with plan characteristics matching that for the average pension plan.

Among PD plans, there is no significant difference in the risk-adjusted performance of NonU, SingleU, and MultiU plans. That is, PD completely eliminates the underperformance of multi-employer pension plans. Among TD plans, there is no significant difference in risk-adjusted net returns for NonU and SingleU plans, but MultiU plans underperform by 50 basis points. Another interpretation of the estimates is that a switch from TD to PD has little effect on risk-adjusted performance for NonU and SingleU plans, but substantially improves performance for MultiU plans. Since a switch from TD to PD is likely to have similar effects on administrative expenses for both union and nonunion plans, we take this as evidence that the forfeiture of union control over investment decisions with a movement to PD eliminates negative effects on performance.

One possible explanation for the impact of PD on performance is the significant asset management differences between PD and TD. Lakonishok, Shleifer, and Vishny (1992) argue that the mutual fund industry outperforms institutional pension fund managers because of agency problems in pension administration. Moreover, TD plans are much more likely to rely on institutional pension fund managers than PD plans. One explanation is that TD plans have a common pool of money to manage for all participants, whereas PD plans must have separate accounts that allow for differing participant asset allocations. The requirements of PD give the mutual fund and insurance industry a cost advantage relative to internal money management, as these industries use technology designed for managing individual accounts that bring significant scale economies in fund administration.

Our data confirm the greater reliance of PD plans on external money managers. We estimate that the average share of assets managed "externally" (i.e., by trusts, registered investment or insurance companies) is 56% for TD and 86% for PD plans. For multi-employer union pensions, the difference is more pronounced: 35% for TD and 84% for PD plans.

To determine whether the type of asset management explains the positive PD effect on MultiU plan performance, we add controls for the share of the fund's assets managed by registered investment or insurance companies and the share of assets held in common/collective trusts, pooled separate accounts, master trusts, or 103-12 investment entities ("trusts"). The omitted asset category reflects assets managed directly by the pension fund. The results of this specification are in the last two columns of Table 4.

Consistent with Lakonishok, Shleifer, and Vishny (1992), the share of assets managed externally has a positive effect on risk-adjusted returns for both PD and TD plans. A 10 percentage-point increase in the share of assets managed by registered investment or insurance companies increases performance by around 4 basis points. An increase in the share of assets in trusts by 10 percentage points improves performance by 2-3 basis points. Nevertheless, controlling for the share of assets managed externally does not eliminate the underperformance of multi-employer plans when they are TD.

We provide several tests for robustness of our empirical results. Following Busse, Goyal, and Wahal (2010), we consider three other variations of the model presented in Table 4. First, we exclude the momentum factor and reestimate the FF regressions to yield an alternative estimate of risk-adjusted performance. Second, we estimate value-weighted regressions of risk-adjusted performance where value is measured by plan assets with and without controls for the momentum factor. All of the models control for the same list of variables as in the last two specifications of Table 4. A summary of the coefficients of interest under these alternative specifications is given in Table 5.

Most of the earlier results are robust to the alternative specifications; others are not. For example, among DB plans, MultiU plans under-perform relative to NonU plans; SingleU plans do not underperform. Among DC plans, the underperformance of MultiU plans remains in three of the four specifications. The exception is the value-weighted specification that does not adjust performance for momentum where the performance of MultiU relative to NonU plans is actually worse with participant direction. Hence, when returns are not adjusted for the momentum factor, a comparison of weighted and unweighted results implies that only small MultiU plans benefit from a switch to PD--large MultiU plans actually experience a decline in risk-adjusted performance. In fact, if we restrict our analysis to plans with assets below the median for DC plans (approximately $5 million), the result that participant direction eliminates underperformance of MultiU plans is robust to all four specifications considered.

An obvious question is how one should interpret the inconsistency in the results regarding the effect of participant direction. The answer is that there is a systematic relationship between the loading on the momentum factor, plan size, and participant direction. In separate analyses, we find that the importance of the momentum factor (measured by the coefficient on momentum in the FF regressions) rises with plan assets among TD plans, but falls with size among PD plans. As a result, a failure to control for the momentum factor causes the estimate of risk-adjusted performance of large MultiU plans to rise among TD plans, but fall among the PD.

In order to benefit from the momentum factor, a plan can invest in assets that recently outperformed the market and sell assets that recently underperformed. Apparently, the likelihood of following such a strategy rises with size among TD plans, but falls with size among PD plans. One possible explanation for the pattern in TD plans could be scale economies in trading costs which make large plans more likely to adjust their portfolio on the basis of recent performance. These scale economies would be less applicable in PD plans where individual participants are making the decisions on individual portfolios. We admit that this is speculative and that we do not have a good explanation for why, among PD plans, the smaller MultiU plans are more likely to take advantage of momentum. Better data on trading behavior and the details of the assets held could shed light on this.

D. Are the Effects Large Enough to Matter?

While our empirical results suggest that unions have statistically significant effects on pension performance, the question of whether the effects are large enough to cause concern among union members or policy makers remains. To address this, we examine the effects of union status on retirement wealth among workers in DC plans.

To provide some sense of the magnitude of the effects, we estimate that in 2008, the average union member with a DC plan had annual contributions of $3,700. The average real rate of return (net of expenses) on union DC plans over our entire sample period was 4.1%. If real contributions for the typical worker hold constant over a 30-year career and earn a 4.1 % real rate of return, the worker would accumulate real wealth of $211,015 by retirement.

Using the above as a benchmark, we estimate the effects of MultiU pensions on retirement wealth. For example, we estimate that MultiU DC plans underperform by 36 basis points relative to nonunion plans (Table 3, specification 4). If the underperformance of MultiU plans was eliminated, workers could increase pension wealth at retirement by 6.2% (approximately $13,000) without any increase in annual contributions.

As an alternative illustration, consider a MultiU DC plan with and without participant direction. Our unweighted estimates correcting for momentum (Table 4) suggest that participant direction is associated with a 33 basis-point improvement in performance. This would result in 5.7% more wealth (approximately $12,000) for workers at retirement in the benchmark situation detailed above. In addition to improved returns, PD allows workers to adjust their individual portfolios to their personal and potentially age-specific risk tolerances.

V. SUMMARY AND CONCLUSIONS

This paper describes the numerous ways that unions can affect pension performance and how pension plan design might alter the effect. We investigate several hypotheses. First, if there is a union effect on performance, we expect it to be largest in multi-employer union plans where unions have the greatest control over assets. Second, multi-employer union plans should be more willing to sacrifice returns in a DB than a DC plan, as their employers are responsible for any shortfall in a DB plan's performance while union membership absorbs the shortfall in performance of a DC plan. This premise, however, assumes that employers are unable to extract wage concessions in exchange for giving the union greater control over pension assets. Finally, among DC plans, a change from trustee to participant direction should lessen the union membership's sacrifice of returns to promote union goals. With trustee direction, leadership controls the asset allocation decisions. With participant direction, membership controls the decisions individually and should be less willing to sacrifice returns for union goals because of a free rider problem.

Our empirical results support most of our predictions. Multi-employer plans afford unions the greatest control over pension assets, which also manifest the greatest underperformance relative to nonunion plans. Also, among multi-employer union plans, performance relative to nonunion plans is worse in DB than DC plans. While the underperformance of multi-employer plans could be due to higher administrative expenses or greater agency problems, evidence that a switch to participant direction improves the relative performance of multi-employer union DC plans suggests that union control over investment decisions is part of the reason that multi-employer plans underperform. If higher administrative expenses are the reason that multi-employer plans underperform, a switch to participant should not eliminate that disadvantage. The fact that a switch to participant direction improves the relative performance of multi-employer union plans suggests the participants do a better job of choosing investments than the union trustees. We should emphasize, however, that this result is most robust for small plans. For large plans, the result is sensitive to how risk-adjusted returns are calculated.

Our overall conclusion is that union control over investment decisions have no significant influence on the performance of single-employer plans, and negative or neutral effects on performance in multi-employer union plans, depending on pension design. On the basis of these results, we expect that if DB plans continue to decline in popularity and participant direction continues to grow among DC plans, any negative effect of unions on pension performance will diminish over time. (32)

REFERENCES

Agrawal, A. K. "Corporate Governance Objectives of Labor Union Shareholders: Evidence from Proxy Voting." Review of Financial Studies, 25(1), 2012, 187-226.

Agrawal, A., and G. N. Mandelker. "Large Shareholders and the Monitoring of Managers: The Case of Antitakeover Amendments." Journal of Financial and Quantitative Analysis, 25, 1990, 143-61.

Bennett, J.,\ and M. Johnson. "Union Use of Employee Pension Funds: Introduction and Overview." Journal of Labor Research, 2, 1981, 181-90.

Bikker, J., and J. DeDreu. "Operating Costs of Pension Funds: The Impact of Scale. Governance, and Plan Design." Journal of Pension Economics and Finance, 8(1), 2009, 63-89.

Busse, J., A. Goyal, and S. Wahal. "Performance and Persistence in Institutional Investment Management." Journal of Finance, 65, 2010, 765-90.

Calabrese, M. "Building on Past Success: Labor-Friendly Investment Vehicles and the Power of Private Equity." Working Paper, Center for National Policy, 1999.

Carhart, M. M. "On Persistence in Mutual Fund Performance." Journal of Finance, 52, 1997, 57-82.

Cohen, L., A. Frazzini, and C. Malloy. "The Small World of Investing: Board Connections and Mutual Fund Returns." Journal of Political Economy, 116, 2008, 951-79.

Conte, M. A. "ESOP Rates of Return." Report to U.S. Department of Labor, Pension and Welfare Benefits Administration, 1994.

Cook, R. "The Case for Joint Trusteeship of Pension Plans." Working USA, 6(10), 2002, 26-55.

Cremers, M., and R. Romano. "Institutional Investors and Proxy Voting: The Impact of the 2003 Mutual Fund Voting Disclosure Regulation." American Law and Economics Review, 13, 2011, 220-68.

Dorsey, S., and J. Turner. "Union-Nonunion Differences in Pension Fund Investments and Earnings." Industrial and Labor Relations Review, 43, 1990, 542-55.

Entine, J. "Pension Fund Politics: Do Union 'Social Investing' Schemes Benefit Their Workers?" American Enterprise Institute for Public Policy Research, Speeches and Government Testimony, July 2008.

Even, W. E., and D. A. Macpherson. "Growth of Participant Direction in Defined Contribution Plans." Industrial Relations, 42, 2010, 190-208.

Fama, E. F., and K. R. French. "Common Risk Factors in the Returns on Stocks and Bonds." Journal of Financial Economics, 33, 1993, 3-56.

Furchtgott-Roth, D. "Pension Plans: How Secure are Union Members' Investments?" Report for Hudson Institute, 2008.

GAO. "Changes Needed to Provide 401(k) Plan Participants and the Department of Labor Better Information on Fees." 2006.

--. "Conflicts of Interest Can Affect Defined Benefit and Defined Contribution Plans." GAO-09-503T, 2009.

Ghilarducci, T., and K. Terry. "Scale Economies in Union Pension Plan Administration: 1981-1993." Industrial Relations, 38, 1999, 11-17.

Gustman, A. L., and T. Steinmeier. "The Stampede toward Defined Contribution Pension Plans: Fact or Fiction?" Industrial Relations, 31, 1992, 361-69.

Hartsohn, A. Quarterly Pensions Update. New York: Mercer, 2010.

Hirsch, B. T. "What Do Unions Do for Economic Performance?," in What Do Unions Do? A Twenty Year Perspective, edited by J. T. Bennett and B. E. Kaufman. New Brunswick, NJ: Transaction Publishers, 2007, 193-237.

Hirsch, B. T., and B. A. Morgan. "Shareholder Risk and Returns in Union and Nonunion Firms." Industrial and Labor Relations Review, 47, 1994, 302-18.

Hustead, E. "Living with Defined Contribution Pensions," in Living with Defined Contribution Pensions: Remaking Responsibility for Retirement, edited by O. Mitchell and S. Schieber. Philadelphia: University of Pennsylvania Press, 1998, 166-77.

Investment Company Institute. Inside the Structure of Defined ContributionJ401(k) Plan Fees: A Study Assessing the Mechanics of What Drives the "All-In" Fee. Washington, DC: Investment Company Institute, 2009.

Ippolito, R. A. "Explaining the Growth of Defined Contribution Plans." Industrial Relations, 34, 1995, 1-20.

Kruse, D. L. "Pension Substitution in the 1980s: Why the Shift toward Defined Contribution?" Industrial Relations, 34, 1995,218-41.

Lakonishok, J., A. Shleifer, and R. W. Vishny. "The Structure and Performance of the Money Management Industry," in Brookings Papers on Economic Activity. Microeconomics. Washington, DC: The Brookings Institution, 1992.

Lee, D., and A. Mas. "Long-Run Impacts of Unions on Firms: New Evidence from Financial Markets, 1961-1999." Quarterly Journal of Economics, 127, 2012, 333-78.

McCarthy, D. D., and J. A. Turner. "Pension Rates of Return in Large and Small Plans," in Trends in Pensions, edited by J. A. Turner and D. J. Beller. Washington, DC: U.S. Department of Labor, Pension and Welfare Benefits Administration, 1989, 235-86.

Mitchell, O. S., and E. S. Andrews. "Scale Economies in Private Multiemployer Pension Plans." Industrial and Labor Relations Review, 34, 1981, 286-98.

Morningstar. Ibbotson SBBI2009 Classic Yearbook: Market Results for 1926-2009. Chicago, IL: Morningstar, Inc., 2010.

Northrup, J. P, and H. R. Northrup. "Union Divergent Investing of Pensions: A Power, Non-Employee Relations Issue." Journal of Labor Research, 2, 1981, 191-208.

Prevost, A. K., R. P. Rao, and M. A. Williams. "Labor Unions as Shareholder Activists: Champions or Detractors?" Financial Review, 47, 2012, 327-49.

Purcell, P, and J. Topoleski. "401 (k) Plans and Retirement Savings: Issues for Congress." Congressional Research Service Report 7-5700, 2009.

Renneboog, L., and P. G. Szilagyi. "The Success and Relevance of Shareholder Activism through Proxy Proposals." Working Paper, European Corporate Governance Institute, 2009.

Shleifer, A., and R. W. Vishny. "Large Shareholders and Corporate Control." Journal of Political Economy, 94, 1986, 461-88.

Tonks, I. "Performance Persistence of Pension-Fund Managers." Journal of Business, 78, 2005, 1917-42.

Turner, J., and D. Muir. "The Market for Financial Advisers." Unpublished manuscript, 2012.

U.S. Department of Labor. Employee Benefits Security Administration. Private Pension Plan Bulletin: Abstract of 2009 Form 5500 Annual Reports, December 2011. Accessed on February 24, 2014 http://www.dol.gov/ebsa/publications/form5500datare search.html#statisticalsummaries. 2014.

Vernuccio, F. V. "Your Retirement or Our Political Agenda." Competitive Enterprise Institute. Unpublished manuscript, 2009.

(1.) See U.S. Department of Labor (2011).

(2.) See, for example, Vernuccio (2009) or Entine (2008).

(3.) See Bennett and Johnson (1981) and Northrup and Northrup (1981) for details on union efforts to gain control over pension funds in the 1980s and the potential conflict between these efforts and federal pension regulations contained in the Employee Retirement Income Security Act of 1974.

(4.) Advisory opinion 98-04A issued by the Pension and Welfare Benefits Administration can be found at http://www. erisaadvisoryopinions.com

(5.) For details on asset holdings and objectives, see www.aflcio-hit.com and www.aflcio.bit.com

(6.) See http://mypension.iamnpf.org/media/13784/IAM_ National_Investment_Option_Summary.pdf

(7.) Calabrese (1999) provides specific examples of such agreements.

(8.) This argument is similar to the belief that institutional ownership of a corporation improves monitoring because large equity positions make research and monitoring expenditures more cost effective. Thus, for example, corporations with many small owners may be monitored less closely than those with a few large institutional owners. See. for example, Agrawal and Mandelker (1990), or Shleifer and Vishny (1986).

(9.) See Dorsey and Turner (1990) and the cites contained therein.

(10.) An alternative explanation for the underperformance of union pension funds is that union plans tilt equity investments toward unionized companies and the stock returns were lower for unionized companies in the late 1970s and early 1980s, but were similar for union and nonunion firms both before and after these time periods (Hirsch and Morgan, 1994).

(11.) See Even and Macpherson (2010) for evidence on the causes and consequences of the shift toward PD plans.

(12.) While participant direction could reduce investments in union projects when risk-adjusted returns are reduced, this same logic does not apply equally to shareholder activism as proxy voting rights are held by the pension trustees or their investment managers regardless of whether or not the plan is PD.

(13.) See Conte (1994) for a discussion of the issues with calculating the rate of return on ESOP plans.

(14.) To be precise, the annual rate of return is calculated as net income from assets divided by invested assets. The net income from invested assets is defined as the sum of interest, dividends, rents, royalties, net realized gain or loss on sale or exchange of assets, other income, unrealized appreciation or depreciation of assets, and net investment gain from trusts minus unrealized appreciation or depreciation of buildings and depreciable property used in plan operations. Investable assets are defined to reflect assets at the beginning of the year plus additional purchases of assets through the year with the assumption that all purchases are made midyear.

(15.) Both Mitchell and Andrews (1981) and Dorsey and Turner (1990) note that administrative expenses are rarely reported in single employer plans because the firm frequently absorbs the costs as part of their everyday operations. However, when pension funds are externally managed (e.g., registered investment companies, insurance companies, or master trusts), there are typically explicit charges for administration of the plan, and investment management fees are subtracted from the returns on the funds that they manage.

(16.) The returns on the overall market are drawn from Kenneth French's website at http://mba.tuck.dartmouth. edu/pages/faculty/ken.french/Data_Library/f-f_factors.html. The sample used to generate the plot of annual returns is restricted to the 81% of pension plans whose filing year matches the calendar year. There are few observations in most of the other months, which reduce the precision of the estimated mean for those months.

(17.) If pensions hold some share of assets as company stock, bankruptcy or mergers could be related to pension performance. However, our elimination of ESOP plans reduces this potential source of survivor bias.

(18.) In an OLS regression of returns on a survivor dummy and other controls using all plans (survivors and nonsurvivors), the regression line passes through the mean so that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where [bar.r] is the mean return for all plans, [bar.S] is the mean of a dummy variable indicating whether a plan is a survivor, and [bar.X] is a vector containing the means of the plan characteristics and year effects. Controlling for plan characteristics and year effects, the difference between the mean return of a sample with only survivors ([bar.S] = 1) and the entire sample is [[alpha].sub.1](1 - [bar.S]). This logic is applied separately for each type of plan by introducing interaction terms between dummies for plan type (nonunion, union single employer, union multiemployer) and the survivor dummy.

(19.) The comparison of survivor and non-survivors is conditioned on the plan characteristics and controls for year effects. Without controlling for these differences in plan characteristics, the survivor bias appears larger as, for example, the survivors tend to be larger and older plans that have performance advantages for reasons we discuss below. Also, as any plan started after 2000 is excluded from the survivor group, it includes fewer observations from this time period, which had below average returns.

(20.) All of the differences between NonU and either SingleU or MultiU plans are statistically significant at the .01 level. To compare variance of returns across plan types, we regressed the squared-residual from the net return regression with year fixed effects and plan type dummies for SingleU and MultiU.

(21.) Investment Company Institute (2009) provides a good review of the factors affecting administrative costs and shows economics of scale in administration with respect to balance per participant and number of participants. See also Mitchell and Andrews (1981), Ghilarducci and Terry (1999), and Bikker and De Dreu (2009) for evidence on scale economies in the administration of pension plans.

(22.) The fact that an employer offers more than one pension plan does not necessarily mean that the same workers are covered by multiple plans. For example, a given employer may offer two pension plans and these two plans may cover the same workers or two entirely different groups of workers. Regardless of whether the plans cover the same workers or different workers, there is the potential to take advantage of economies of scope in administration.

(23.) Hustead (1998) shows that administrative expenses are higher in DB than in DC plans. He also provides evidence that regulatory reforms have driven up the cost of DB relative to DC plans over time and may partially account for the shift from DB to DC plans, particularly at small firms.

(24.) For recent examples employing this methodology to mutual and pension funds, see Tonks (2005) and Cohen et al. (2008).

(25.) The construction of these variables and the data are available from Kenneth French's website at http:// mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_ Library/f-f_factors.html

(26.) The data source for all bond returns is Morningstar (2010).

(27.) This approach to measuring risk-adjusted performance is very similar to that used by Dorsey and Turner (1990) for examining pension performance, except that we use multiple risk factors reflecting different types of stock and bond market risk, whereas they use a single factor based on a mixed stock and bond market index.

(28.) The r-statistics used to determine statistical significance for a given plan's regression coefficients are compared to critical values from the f-distribution with (n - 7) degrees of freedom, where n represents the number of observations for that plan and are calculated using robust standard errors.

(29.) Among pension plans that hold less than 25% of their assets in master trusts (where there is no breakdown provided on asset types held), the share of assets in government and corporate bonds is 41.3% for MultiU plans, 21.8% for SingleU plans, and 22.3% for NonU plans.

(30.) The reported t-statistics are based upon robust standard errors corrected for clustering of residuals by pension.

(31.) The regressions include year and industry dummies to allow for the fact that risk-adjusted performance might differ systematically across years and the cost of administration could differ across industries because of different work force

characteristics or seasonal variability in the timing of contributions.

(32.) For research regarding the shift from DB to DC plans, see Gustman and Steinmeier (1992), Kruse (1995), and Ippolito (1995).

WILLIAM E. EVEN and DAVID A. MACPHERSON *

* We thank Stuart Dorsey, Darren Grant, Barry Hirsch, Shu-Chin Huang, David Shrider, and John Turner for helpful comments.

Even: Department of Economics, Miami University, Oxford, OH 45056. Phone 513-529-2865, Fax 513-529-6992, E-mail evenwe@miamioh.edu

Macpherson: Department of Economics, Trinity University, San Antonio, TX 78212. Phone 210-999-8112, Fax 210 999-7255, E-mail David.Macpherson@trinity.edu

TABLE 1

Sample Means for IRS Form 5500 Pension Plan Data, 1988-2008

                                                   Single-     Multi-
                                                   Employer   Employer
                                All     Nonunion    Union      Union

Union                          17.3%        0%       100%       100%
Return net of expenses          6.9%      6.8%       8.0%       6.8%
Standard deviation of          11.9%     12.3%      11.0%       9.0%
    net returns
  Within plans                 11.6%     12.0%      10.6%       8.7%
  Between plans                 3.1%      3.2%       3.7%       3.0%
Participants                   1,841     1,411      2,973      5,123
Per capita assets ($ 1000s)     27.1      27.0       25.1       30.5
Percent assets managed         69.8%     72.2%      77.9%      31.9%
  externally
Defined benefit                30.0%     22.0%      71.7%      63.4%
Defined contribution plan      70.0%     78.0%      28.3%      36.3%
Participant-directed-          49.5%     56.7%      19.9%       9.2%
  defined contribution plan
401 k plan                     54.9%     63.3%      22.3%       4.6%
Profit sharing plan            60.0%     69.2%      22.4%       8.2%
Money purchase plan             5.4%      4.1%       3.7%      21.5%
Offer other DB plan            22.0%     18.9%      60.4%       5.3%
Offer other DC plan            31.9%     30.0%      67.3%       6.4%
Number of plans               38,895    32,461      4,119      2,654
Number of plan year          538,565   445,266     53,107     40,192
  observations
Number of observations         13.8       13.7       12.9       15.1
  per plan

Notes: The sample is based on IRS Form 5500 data for private
sector plans, with 100 or more participants between 1988 and
2008. The sample excludes plans with less than 10 years of data,
ESOP plans, and plans with missing data on union status or other
key variables.

TABLE 2
Summary Statistics from FF Regressions, 1988-2008

                                                    Single-     Multi-
                                                    Employer   Employer
                                 All     Nonunion    Union      Union
                                Plans     Plans      Plans      Plans

Alpha (in percentage points)    -0.43     -.44       -.13       -.71
                               (18.1%)   (17.9%)    (17.6%)    (20.2%)
Coefficients on FF factors
Market-risk free                 0.52      0.54       0.48       0.38
Small-large cap                (93.0%)   (93.8%)    (89.4%)    (89.5%)
                               -.0006     0.0008     0.004      -.021
                               (23.1%)   (23.0%)    (25.7%)    (21.3%)
High-low book-to-market         -.066     -.067      -.064      -.065
                               (27.7%)   (27.9%)    (25.6%)    (28.3%)
Momentum                        0.026     0.029      0.009      0.009
                               (20.5%)   (21.2%)    (18.4%)    (15.5%)
Term                            -.087     -.111      -.002      0.062
Default                        (27.1%)   (27.6%)    (22.4%)    (27.1%)
                                -.155     -.185      -.106      0.122
                               (27.0%)   (27.6%)    (22.4%)    (27.1%)
Percentage of regressions       95.1%     95.5%      92.2%      94.6%
  with coefficients on
  FF factors jointly
  significant at .05 level
Average of regression           0.88       0.88       0.88       0.85
  [R.sup.2]
Number of pension plans        38,895     32,461     4,119      2,654
Number of observations         538,565   445,266     53,107     40,192
Observations per pension        13.8       13.7       12.9       15.1
  plan

Notes: The numbers presented represent the average of the
estimated FF coefficients across the pension plans. The dependent
variable is excess pension return net of expenses measured in
percentage points. In parentheses are the percentages of plans
for which the corresponding coefficients are statistically
different from zero at the .05 level. The statistical
significance is based upon robust standard errors corrected
for clustering of residuals by plan and the t-distribution.

TABLE 3
Determinants of Risk-Adjusted Returns Regressions, 1988-2008

                                        (1)        (2)        (3)
                                        All         DC        DB

Constant                               -0.435     -0.532    -0.122
                                      (-25.53)   (-28.16)   (-3.19)
Union single employer                  0.192      0.476     -0.0437
                                       (3.68)     (5.27)    (-0.60)
Union multi-employer                   -0.219     -0.080    -0.566
Plan assets per capita in $10,000     (-3.86)    (-0.95)    (-1.37)

Participants in 10.000s

(Participants in 10,000s) (2)
Plan age

Employer offers other defined
  benefit
Employer offers other defined
  contribution
Year dummies included                   Yes        Yes        Yes
Industry dummies included               Yes        Yes        Yes
Observations                          538,565    377,139    161,426
p values for hypothesis tests:
  Equal coefficients on union          0.000      0.000      0.000
    single and union multi-employer
  Equal coefficients on union            --       0.000
    single-employer in DB and DC
  Equal coefficients on union            --       0.000
    multi-employer in DB and DC
[R.sup.2]                              0.061      0.071      0.054

                                       (4) (a)       (5)
                                         DC          DB

Constant                               -0.508      -0.108
                                      (-27.02)    (-2.719)
Union single employer                   0.142      -0.0360
                                       (1.519)    (-0.476)
Union multi-employer                   -0.357      -0.757
Plan assets per capita in $10,000     (-3.859)    (-7.600)
                                       0.00649    0.000841
                                       (2.648)     (0.195)
Participants in 10.000s                0.0405      0.00285
                                       (6.609)     (1.062)
(Participants in 10,000s) (2)         -0.000191   -1.08e-06
Plan age                              (-5.859)    (-0.114)
                                       0.0354      0.0148
                                       (21.47)     (6.973)
Employer offers other defined           0.562      -0.112
  benefit                              (12.33)    (-1.649)
Employer offers other defined           0.338      0.00160
  contribution                         (8.239)    (0.0268)
Year dummies included                    Yes         Yes
Industry dummies included                Yes         Yes
Observations                           377.139     161,426
p values for hypothesis tests:
  Equal coefficients on union           0.000       0.000
    single and union multi-employer
  Equal coefficients on union           0.138
    single-employer in DB and DC
  Equal coefficients on union           0.003
    multi-employer in DB and DC
[R.sup.2]                               0.086       0.056

Notes. The coefficients are from a regression of the estimated
alpha plus the residual from the FF equations as the dependent
variable. The r-statistics in parentheses are based upon robust
standard errors corrected for clustering of residuals by plan.

(a) Column (4) also includes controls for 401 (k) plan, profit
sharing plan, and other money purchase plan.

TABLE 4
Determinants of Risk-Adjusted Returns, by Participant
Direction Status, for DC Plans, 1990-2008

                                       Participant-   Trustee-
                                         Directed     Directed

Constant                                  -0.328       -0.333
                                         (-13.31)     (-8.063)
Union single-employer                     0.108        0.128
                                         (1.060)      (0.821)
Union multi-employer                      -0.118       -0.504
                                         (-0.891)     (-4.030)
Share of assets invested with
  registered investment or insurance
  companies
Share of assets invested in trusts

Sample size                              262,731       95,210
[R.sup.2]                                 0.101        0.076

                                       Participant-   Trustee-
                                         Directed     Directed

Constant                                  -0.354       -0.291
                                         (-13.68)     (-7.144)
Union single-employer                     0.115        0.116
                                         (1.122)      (0.746)
Union multi-employer                      -0.107       -0.443
                                         (-0.809)     (-3.506)
Share of assets invested with             0.453        0.424
  registered investment or insurance     (5.920)      (5.704)
  companies
Share of assets invested in trusts        0.282        0.220
                                         (3.579)      (2.739)
Sample size                              262,731       95,210
[R.sup.2]                                 0.101        0.077

Notes: Coefficients are from a regression of the estimated
alpha plus the residual from the FF equations as the dependent
variable. T-statistics are in parentheses and are based upon
robust standard errors corrected for clustering of residuals
by plan.

(a) The sample period is 1990-2008. Share of assets invested
in trusts is defined as the sum of assets in common/collective
trusts, pooled separate accounts, master trusts, and 103-12
investment entities divided by total assets. All models also
include controls for industry, year, plan assets per capita in
10,000s, participants (in 10,000s) and its square, plan age,
whether employer offers other defined contribution plan, whether
the employer offers other defined benefit plan, 401(k) plan,
profit sharing plan, and other money purchase plan.

TABLE 5
Union Effect on Risk-Adjusted Performance
with Alternative Specifications (a)

                                                 Single-     Multi-
                                                 Employer   Employer
                                     Intercept    Union      Union

FF estimates excluding momentum factor, unweighted regressions
DB plans only                          0.047      -0.093     -0.689
                                      (1.82)      (1.77)     (8.80)
DC plans only                         -0.003      0.025      -0.809
                                      (0.24)      (0.44)    (11.53)
Participant-directed DC plans only     0.184      0.066      -0.557
                                      (11.02)     (1.09)     (6.11)
Trustee-directed DC plans only         0.012      -0.118    -0.866 (b)
                                      (0.43)       (1.19)      (8.93)

FF estimates including momentum factor, value-weighted regressions
DB plans only                          0.042      0.026      -1.082
                                      (0.24)      (0.43)    (-2.69)
DC plans only                         -0.017      0.427      0.261
                                      (-0.19)     (1.04)     (0.72)
Participant-directed DC plans only    -0.070      0.708      0.578
                                      (-0.58)     (1.58)     (1.21)
Trustee-directed DC plans only         0.253      -0.411    -0.563 (c)
                                      (1.83)     (-0.92)     (1.35)

FF estimates excluding momentum factor, value weighted regressions
DB plans only                          0.079      -0.099     -1.256
                                      (4.52)     (-0.22)    (-2.95)
DC plans only                          0.591      0.383      -0.805
                                      (9.80)      (1.27)    (-3.14)
Participant-directed DC plans only     0.701      0.597      -0.929
                                      (6.26)      (1.80)     (2.26)
Trustee-directed DC plans only         0.487      -0.513    -0.329 (d)
                                      (5.08)      (1.78)    (-1.39)

Notes: Coefficients are from a regression of the estimated alpha
plus the residual from the FF equations as the dependent variable.
T-statistics (in parentheses) are based upon robust standard errors
corrected for clustering of residuals by plan. Weighted regressions
use value of plan assets as weights.

(a) The sample period is 1990-2008. All models include controls for
industry, year, per capita assets, number of plan participants and
its square, plan age, whether employer offers other defined
contribution plan or defined benefit plan, and dummies indicating
whether the plan is a 401 (k), profit sharing, or money purchase
plan.

(b) Coefficient is significantly different from that for
participant-directed plans at .05 level.

(c) Coefficient is significantly different from that for
participant-directed plans at .10 level.

(d) Coefficient is not significantly different from that for
participant-directed plans at .10 level.