Investment Management Firms

The Selection and Termination of Investment Management Firms by Plan Sponsors* Amit Goyal Goizueta Business School 1300 Clifton Road Emory University Atlanta, GA 30322-2710 (404) 727-4825 Email: Amit_Goyal@bus.emory.edu Sunil Wahal WP Carey School of Business Arizona State University PO Box 873906 Tempe, AZ 85287-3906 (480) 965-8755 Email: Sunil.Wahal@asu.edu May 2006 * We are indebted to Allison Howard and Nick Mencher at Invesco, Robert Stein at Mercer Investment Consulting, and Keith Arends at iisearches for assistance with data issues and to the Goizueta Business School and the Q-Group for financial support. Jesse Ferlianto, Peter Left, Margaret Petri, Marko Sventina, and Fridge Vanzyp provided research assistance. We thank an anonymous referee, Roberto Barontini, George Benston, Mark Carhart, Chris Geczy, Charles Hadlock, Larry Harris, Narasimhan Jegadeesh, Kevin Johnson, Ananth Madhavan, Ed Rice, and seminar participants at HKUST, the University of Washington, Goldman Sachs, Arizona State University, UC – Irvine, Barclays Global Investors, Boston College, the European Finance Association Meetings in Moscow, the American Finance Association meetings in Boston and the Mitsui Life Symposium at the University of Michigan for helpful comments and suggestions. The Selection and Termination of Investment Management Firms by Plan Sponsors Abstract We examine the selection and termination of investment management firms by plan sponsors (public and corporate pension plans, unions, foundations, and endowments). We build a unique dataset that comprises hiring and firing decisions by approximately 3,500 plan sponsors over a 10-year period from 1994 to 2003. Our data represent the allocation of over $636 billion in mandates to hired investment managers and the withdrawal of $108 billion from fired investment managers. Plan sponsors hire investment managers after large positive excess returns up to three years prior to hiring. However, this return chasing behavior does not deliver positive excess returns thereafter; on average, post-hiring excess returns are indistinguishable from zero. Plan sponsors frequently, but not always, terminate investment managers after underperformance, but the excess returns of these managers after being fired are frequently positive. Using a matched sample of firing and hiring decisions, we find that if plan sponsors had stayed with fired investment managers, their excess returns would be larger than those actually delivered by newly hired managers. 1. Introduction Allen (2001) argues that financial institutions matter for asset pricing and laments the lack of attention to their behavior. Despite this clarion call, academic research has focused on two types of institutions, banks and mutual funds. There are good reasons for this. Banks have been a historically important component of the economy, and mutual funds are a relatively new but sizeable channel for retail investors to participate in capital markets. In addition, good data for both these types of institutions are widely available, permitting researchers to tackle issues with precision. However, another category of institutions, plan sponsors and institutional asset managers, is equally if not more important. At the end of 2003, there were 47,391 plan sponsors in the United States (corporate and public retirement plans, unions, endowments and foundations), responsible for delegating investment of $6.3 trillion to institutional investment managers (Money Market Directory, 2003). At the same time, there were 7,153 equity, bond and hybrid mutual funds with total assets of $5.4 trillion (Investment Company Institute, 2004). The sheer enormity of the assets under the jurisdiction of plan sponsors and their potential impact on asset prices are compelling reasons to examine their behavior. Moreover, the fact that the assets managed by many plan sponsors fund the retirement incomes of their beneficiaries makes studying their behavior important from a personal and public policy perspective. A comparison of institutional investment to the more widely studied retail marketplace provides some perspective. There are three basic streams to the retail investment / mutual fund literature: (a) investigations of performance, including persistence, (b) studies that examine the relationship between fund flows and returns, and (c) papers that examine investment choices made by individual investors. The general conclusion that emerges from these streams is that level of excess performance and the degree of persistence is weak and elusive, that the relationship between flows and returns is convex, and that retail investors make investment choices that can be construed as sub-optimal by some and simply noisy by others.1 A partial list of contributions in the literature on performance and persistence includes, Bollen and Busse (2005), Brown and Goetzmann (1995), Carhart (1997), Carhart et al. (2004), Daniel et al. (1997), Elton et al. (1992), Goetzmann and Ibbotson (1994), Grinblatt and Titman (1992), Grinblatt, Titman and Wermers (1995), Hendricks, Patel and Zeckhauser (1993), Ippolito (1989), Jensen (1968), Wermers (2000), and Zheng (1999). Fund flows and returns are studied by Chevalier and Ellison (1997), Gruber (1996), and Sirri and Tufano (1998). The third stream includes Barber and Odean (2000), Barber, Odean and Zheng (2003), and Odean (1998, 1999). This list of citations is certainly not comprehensive. Omissions are not willful and we offer our apologies to authors not cited. 1 1 In the institutional realm, the streams are rivulets. Lakonishok, Shleifer and Vishny (1992) provide the first investigation of performance and persistence. They persuasively argue that there are significant conflicts of interest in the money management industry and use proprietary data to examine the performance of 769 all-equity funds run by 341 investment managers.2 They paint a bleak picture of performance and argue, “[that] when all is said and done, we doubt that an industry that has added little if any value can continue to exist in its present form.” Coggin, Fabozzi and Rahman (1993) use proprietary data to study a sample of pension fund managers and find that they have limited skill in selecting stocks. Christopherson et al. (1998) find evidence of persistence among institutional equity managers using conditional methods and Busse et al. (2006) find that persistence exists in domestic equity, international equity and fixed income portfolios. Del Guercio and Tkac (2002) and Heisler, Knittel, Neumann and Stewart (2004) initiate investigations in the second stream by examining the relation between asset flows and returns. They find that excess (as opposed to raw) returns are the relevant metric for the flow-performance relationship in the institutional arena. The third stream, the actual investment choices by plan sponsors, is hitherto dry and of significant economic importance. Ex ante, one might expect that the level of expertise of plan sponsors in delegating assets to institutional investment management firms is higher than that of individual investors picking retail mutual funds. Whether this expertise generates excess returns or not, is ultimately an empirical question. Our paper is the first to tackle these issues in the institutional marketplace.3 Plan sponsors have certain investment goals and, working under self or externally imposed restrictions, allocate assets across classes in an attempt to achieve their goals. Within each asset class, mandates of specific dollar amounts are then delegated to investment Lakonishok et al.’s (1992) prescience regarding conflicts of interest is noteworthy. Twelve years later, the Securities and Exchange Commission launched an investigation into the manner in which pension consultants recommend investment managers to plan sponsors (Wall Street Journal, January 13, 2004, Wall Street Journal, October 25, 2004 and New York Times, March 21, 2004). A central theme in this investigation is to determine if consultants recommend certain investment managers to plan sponsors because they receive direct or indirect payments from investment managers (Wall Street Journal, November 9, 2004). 3 Institutions are more likely to be marginal traders than individual investors in most markets; consequently, their impact on asset pricing could be substantial. This is eloquently described by Cornell and Roll (2004) who argue “… consumption decisions, whether to buy a television or take a vacation are made by consumers. The decision to buy IBM or Intel is delegated”, and develop a simple yet elegant delegated agent asset-pricing model. 2 2 management firms, to be invested in a particular investment style. The raison d’être of a plan sponsor is then two fold: (a) to conduct asset allocation, and (b) to hire managers to deliver benchmarked returns, monitor, and if necessary fire, underperforming investment managers.4 It is this second task, viz., the hiring and firing of investment managers by plan sponsors that we focus on in this paper. To achieve our goal, we compile a unique database of hiring and firing decisions using the “Tracker” database from Mercer Investment Consulting, the “iisearches” database provided by Institutional Investor, and electronic searches in numerous trade publications. Our hiring sample consists of 9,214 decisions by 3,591 plan sponsors between 1994 and 2003. A total of $636 billion is delegated in these hiring decisions. Consultants advise in the selection of investment managers in approximately 64 percent of all hiring decisions. Plan sponsors may employ consultants if they lack sophistication or because they do not possess economies in evaluation and selection of investment managers. Consistent with this, we find that smaller plan sponsors are more likely to employ consultants. Executives that manage sponsor assets may also employ consultants to provide some measure of protection from adverse publicity and public criticism in the case of poor performance. We find that plan sponsors that are more sensitive (resistant) to headline risk are more (less) likely to use consultants. We examine the performance of hired investment managers using a variety of methods. We compute benchmark-adjusted cumulative excess returns, excess buy and hold returns, information ratios and calendar time alphas from factor models up to three years before and after hiring. All measurement methods show that for domestic equity and fixed income mandates, pre-hiring returns are positive, large and statistically significant, but that post-hiring returns are statistically indistinguishable from zero. For example, the three-year pre-hiring (post-hiring) cumulative excess return for domestic equity is 12.0 percent (0.7 percent). For international equity mandates, however, both pre and post-hiring excess returns are positive and large (15.5 percent three years prior to hiring and 9.8 percent three years thereafter). 4 Although we frequently refer to “investment managers”, our unit of analysis throughout the paper is the investment management firm, not individuals at these firms. 3 There is tremendous heterogeneity among plan sponsors in incentives created for plan executives, the identity of residual claimholders, size, capabilities, goals and restrictions on the investment process, and asset allocation strategies. Although it is almost impossible to neatly categorize plan sponsors into groups, we attempt to capture this heterogeneity on several of the dimensions described above. We find that large plan sponsors generate positive post-hiring excess returns, suggesting that size allows sponsors to develop expertise in managing the selection process. The goals and restrictions of plan sponsors, at least in part, manifest themselves in asset allocation. An index that measures the aggressiveness of asset allocation is positively correlated with excess returns. There are also complex interaction effects between the type of plan sponsor, its size and the aggressiveness of asset allocation. For instance, endowments and foundations that allocate assets aggressively generate positive post-hiring excess returns but public plans that allocate assets in this manner do not. We also find that mandate size is negatively related to post-hiring returns, which is consistent with diseconomies of scale in investment management. Finally, excess returns of consultant-supported hiring decisions are larger, indicating that consultants add value in the manager selection process. Our sample of investment manager terminations consists of 910 firing decisions involving almost $108 billion by 500 plan sponsors between 1996 and 2003. The number of termination decisions is substantially smaller than hiring decisions because data sources are geared towards assisting investment managers in obtaining new business, and because there is a natural disinclination to report terminations by both investment managers and plan sponsors. In addition, the increase in the assets under the jurisdiction of plan sponsors over time implies that the number of hiring decisions endogenously exceeds the number of firing decisions. The average cumulative excess return of fired investment managers from three years (one year) prior to the firing quarter is 0.6 (-1.0) percent, with a standard error of 1.3 (0.7) percent. The large standard errors reflect the heterogeneity in reasons for termination. In fact, out of 910 termination decisions, 32 percent are self-identified as being due to poor performance, 13 percent due to organizational reasons (personnel turnover or mergers), 12 percent due to asset reallocations by the plan sponsor, and the remainder unidentified. Not surprisingly, the prefiring returns for performance-based terminations are significantly negative. Three years (one 4 year) after the firing decision, the corresponding excess returns are positive, 4.3 (0.7) percent with a standard error of 1.3 (0.7) percent. Although a cross-sectional analysis of post-firing excess returns is hampered by small sample size, investment managers that are terminated for organizational reasons appears to deliver significant excess returns three years after termination. One might argue that hired managers’ post-hiring returns should not be judged against their own pre-hiring returns, and similarly, fired managers’ post-firing returns should not be judged against their own pre-firing returns. A more appropriate comparison might be between the post-hiring return of hired investment managers, and the post-firing return that would have been delivered by fired investment management firms. Since there are a multitude of complicated mechanisms by which firing and hiring decisions are coordinated, we build a sample of “round-trip” firing and hiring decisions manually. We identify 660 round-trip decisions between 1996 and 2003. For these decisions, pre-firing returns are negative and pre-hiring returns are positive. On average, post-firing returns are positive and in some cases, statistically significant. Post-hiring returns are statistically indistinguishable from zero.5 Although the opportunity cost of firing and hiring decisions (the post-firing return minus post-hiring return) is positive, the standard errors are large. The conclusion to be drawn from our work depends largely on one’s view of persistence in performance, and of the role of frictional costs of the hiring and firing process. Take, for example, hiring decisions that are necessitated by the growth of plan sponsor assets. For such decisions, and ignoring potentially non-trivial search costs, one may take the view that since post-hiring excess returns are statistically indistinguishable from zero, plan sponsors appropriately achieve their (benchmarked) objective. This is especially the case if there is little or no persistence in the performance of investment managers to begin with – in that case, hiring simply represents selection from a random time-series sample, implying that post-hiring returns should be zero.6 But Christopherson et al. (1998) and Busse et al. (2006) show that there is persistence in institutional portfolios over a one to two year period. If this is the case, then hiring investment managers could generate excess returns at some horizons, if timed correctly. These results mirror those of Odean (1998) in the retail setting. Odean finds that the excess returns on winning stocks sold by individual investors are larger than the excess returns on loser stocks that could be, but are not, sold. 6 This also implies that post-firing returns should also be zero, which is counterfactual at some return horizons. 5 5 For hiring decisions necessitated by the termination of incumbent investment managers, one has to judge the hired manager’s returns against the returns that the fired manager would have delivered, as well as frictional costs in moving portfolios. In such situations, one might argue that since pre-firing returns are negative and post-hiring returns are indistinguishable from zero, plan sponsors appropriately fire underperforming investment managers and hire managers that deliver their benchmark performance. In a world with no transaction costs and no opportunity losses, this is a zero net present value proposition, and an outcome that may be acceptable to the plan sponsor’s beneficiaries. However, our round-trip sample suggests that post-firing returns are larger than post-hiring returns. Addressing the issue of transaction costs is a more difficult problem, as there are no publicly available data on the costs of moving portfolios. The process of moving assets from the legacy portfolio of the fired investment manager to the target portfolio of the hired manager is frequently outsourced to “transition management firms” that attempt to minimize the costs associated with the transition. Estimates of transition costs by practitioners in the public press suggest that average costs range between 2 and 5 percent of the portfolio, with a standard deviation of 1 percent (see, for example, Proszek (2002), Bollen (2004) and Werner (2001)). Private estimates of all-in transition costs provided to us by an anonymous large transition management firm vary between 1.0 and 2.0 percent. This firm also indicates that transition costs are much higher for international, fixed income and small-cap transitions, and when the legacy and target portfolios are in different asset classes. Regardless of the actual magnitude, the size of this transition business, estimated by some observers to be almost $2 trillion annually, suggests that transaction costs are substantial. Finally, if frictions are important, then one would expect the return threshold for retention decisions (in which an incumbent manager is “rehired”) to be lower than for brand new hiring decisions. Consistent with this, we find that pre-retention excess returns are positive but lower than pre-hiring excess returns. Overall, in light of such large transactions costs and positive opportunity costs, our results suggest that the termination and selection of investment managers is an exercise that can be costly to plan beneficiaries. But the efficacy of the process appears to be correlated with various attributes of plan sponsors. Our results produce a set of stylized facts in an important 6 institutional setting that mirror those motivating the equilibrium model of Berk and Green (2004); selection and termination decisions are analogous to fund flows and are related to performance, and the data are suggestive of diseconomies of scale. Our analysis largely focuses on the costs of selection and termination, and we are unable to measure potentially important compensating benefits. For example, it may be that termination disciplines fired investment managers and causes them to improve returns in the future. It is also possible that termination disciplines incumbent (not-fired) as well as potential investment managers. Such benefits, if significant, can be regarded as the price of enforcing discipline. Our paper proceeds as follows. In section 2, we provide a brief description of the institutional marketplace and investment process. In section 3, we describe data sources and sample construction procedures. We present results on the selection of investment managers in section 4, and the termination of investment managers in section 5, and round-trip termination and selection decisions in section 6. Section 7 discusses robustness checks. Sections 8 and 9 provide a discussion and conclusion to the paper. 2. Institutional details In this section, we describe the institutional marketplace and the investment process followed by most plan sponsors. A more detailed description of the pension fund industry can be found in Fabozzi (1997), Lakonishok et al. (1992), Logue and Rader (1998), and Travers (2004). 2.1 The Institutional Marketplace There are basically two types of plan sponsors, those that manage retirement assets and those that manage non-profit assets. The former include corporate plans, public plans for employees at the city, county or state-level, single-employer plans and Taft-Hartley multiemployer plans for organized labor.7 The latter consist foundations and endowments, including those for universities. Retirement plans can be set up as defined benefit plans, defined Such plans are set up under Section 302(c) (5) of the Taft-Hartley Act, passed by Congress in 1947. Plan assets are jointly managed by a board of trustees representing labor and management. This is a sizeable market. Brull (2006) reports that 1,600 multiemployer plans had assets totaling $333 billion in 2002, and covered almost 10 million workers in 2005. He also reports that some 30,000 single employer plans reported assets of $1.6 trillion in 2002 and covered 34.6 million workers. 7 7 contribution plans, or both. In a defined benefit plan, beneficiaries receive a fixed set of payments upon retirement. The trustees of the plan are responsible for investing the beneficiaries’ contributions to ensure that future benefits can be paid. In defined contribution plans, beneficiaries receive variable payments upon retirement. The plan sponsor typically selects providers of various investment options (such as Vanguard or Fidelity) who then allow beneficiaries to directly invest their assets in various funds. Some plans sponsors offer both defined benefit and defined contribution plans. All plan sponsors share one common feature: the trustees of the plan are charged with the task of managing assets in the best interests of their beneficiaries. However, organizational structure and incentives can generate tremendous variation in behavior across plan sponsors. In corporate defined benefit plans, if the plan is overfunded, the excess funds belong to the corporation. This creates powerful incentives for the treasurer’s office (the trustee) to generate superior performance. If the plan is underfunded, the shortfall is a senior claim on the corporation in the event of bankruptcy. The Pension Benefit Guarantee Corporation (PBGC) insures the benefits (up to a statutory limit) if the corporation has insufficient assets to cover its obligations. Lakonishok et al. (1992) note that this structure produces a bias against passive investment management (since it reduces the potential power of the treasurer’s office), and against internal investment management (since it is easier to blame another organization for poor performance). In federal, state or local government pension plans, the residual claimant is the government authority (and ultimately the taxpayer), and the trustees of the plan are political appointees and/or bureaucrats. Such plans are not formally covered by ERISA. The incentives to generate superior performance are not as strong as those for corporate plans. Similarly, the residual claimants at single employer union plans are union members and the PBGC provides downside protection. Trustees are drawn from members. However, in multi-employer TaftHartley plans, if one employer files for bankruptcy, the shortfall is assumed by solvent companies remaining in the plan. Non-retirement plans such as endowments and foundations do not receive any protection from the PBGC and do not have a residual claimant per se. Cash outflows for endowments and foundations have more of a discretionary element to them than retirement plans. If a foundation’s performance is weak, it can (to some extent) lower 8 distributions and curtail charitable activity whereas a retirement system has to fulfill its cash outflow obligations. Incentives are also provided by the market for human capital. Superior performance in managing the investment process can generate substantial opportunities. This appears to be the case, especially for endowments, where even though the residual claimant is not well-defined, executives that manage the investment process effectively generate significant human capital.8 2.2 The Investment Process The above discussion suggests that the goals of a plan sponsor are influenced by the structure of claims and the nature of payouts; to some extent, the plan sponsor categories capture these effects. The investment process followed by plan sponsors is designed to achieve those goals. Typically this process begins with an investment policy statement drafted by the investment committee, often spearheaded by a chief investment officer. The investment policy statement describes the goals of the plan sponsor, the roadmap for reaching those goals, and any restrictions on the investment process. For example, the statement typically contains broad recommendations about allocations across asset classes. For retirement plans, this is done in conjunction with actuarial projections of cash outflows (based on employees’ ages, retirement patterns etc.). If a plan sponsor changes its actuarial projections as part of an asset-liability study (often conducted by a consultant), it may revise the long-term asset allocation recommendations. In addition to asset allocations, the investment policy statement also contains prescriptions about the degree to which assets should be actively versus passively managed, and the extent to which assets may be managed internally rather than delegated to external portfolio managers. The restrictions imposed on the investment process by the investment committee (and the resulting asset allocations) may originate from a desire to control risk and return profiles, but are also influenced by the funding ratio (the ratio of assets to projected liabilities) as well as political considerations. Restrictions take a variety of forms. They can influence the quantity and quality of asset classes available. For instance, certain asset classes (such as hedge funds or real estate) Two well known examples of this are David Swensen of the Yale University Endowment and Jack Meyer of the Harvard University Endowment. 9 8 may be excluded or capped at a particular percentage of total assets. There may also be restrictions on specific securities to be held within qualified asset classes. Quality restrictions, for example, might involve excluding “sin” stocks or including only dividend paying securities. The practical consequence of this is that investment management firms have to adjust their composite portfolios to meet specific plan sponsor restrictions. Under such circumstances, mandates are invested in separate accounts which are very similar to composite portfolios but accommodate restrictions. Plan sponsor size also generates variation in the investment process across plan sponsors. Larger plan sponsors likely benefit from economies of scale in generating information and managing the investment process. In addition, large plan sponsors have an advantage in that they may be allowed preferential access to certain funds because they can provide large amounts of capital; most investment management firms have minimum investment requirements that small plan sponsors may not be able to meet. 2.3 The Hiring and Firing Process Once broad asset allocations have been established, manager searches begin. The plan sponsor puts out an RFP (request for proposals) and may retain a consultant to assist it in the search. The process involves screening investment managers who provide investment products in the mandate stated by the plan sponsor. The mandate can be either broad (e.g., domestic equity) or narrow (e.g., small-cap equity value), depending on the specificity of the investment policy statement and the peculiarities of the plan sponsor. The list of candidate managers is then culled based on relative performance. The list is further trimmed with written questionnaires and interviews, and the investment committee or trustees make a final choice. For an investment manager, being part of the initial list of managers is a critical hurdle. As a result, most organizations voluntarily provide information to various databases that record performance and other characteristics. Such databases are produced by independent organizations, such as iisearches (affiliated with Institutional Investor publications) or Nelson’s Directories (affiliated with Thomson Financial), as well as by pension consultants such, as Mercer Investment Consulting. A list of common databases is contained in Travers (2004). 10 Plan Sponsor Magazine (2003) reports that approximately 60 percent of all searches employ consultants. Since different plan sponsors conduct manager searches that are correlated in time and investment mandate, pension consultants can reap economies in information gathering that plan sponsors cannot. To the extent that larger plan sponsors make more hiring/firing decisions, they may be less likely to employ consultants. As mentioned earlier, plan sponsors may also employ a consultant to shield themselves from adverse publicity associated with negative outcomes from hiring decisions. Once an investment management firm has been hired, its performance is generally monitored on a quarterly basis. If performance relative to a benchmark deteriorates over consecutive evaluation horizons, the firm may be put on a “watch list”. If performance improves, the firm is removed from the watch list. Continued deteriorate in performance may result in the firm’s contract being terminated. If the firm is terminated, the assets are held in place, “parked” in an index fund, or transferred to newly hired investment manager’s portfolio by a transition organization. Large investment houses, such as State Street Global Advisors and Barclays Global Investors, provide such transition management services, the aim of which is to minimize the frictional loss in transitioning between the legacy and target portfolios. Aside from performance, there are at least two other reasons why an investment management firm may be terminated. The plan sponsor may view the superior performance of the investment manager’s portfolio as being directly attributable to a particular individual. If such an individual(s) leaves the firm, the plan sponsor may decide to terminate its relationship with the investment management firm. For example, in 1996 the two principal partners of Apodaca-Johnston Capital Management separated to start their own investment management firms. As a result, the Los Angeles County Employees Retirement Association terminated its contract with the firm. Alternatively, if the plan sponsor decides to make broad changes in asset allocation, it may terminate investment managers in mandates that are downsized. Hiring of investment managers also takes place for several reasons. The replacement of a fired manager or an increase in asset allocation to a particular mandate can trigger hiring. Additionally, if the size of the plan sponsor’s asset base increases, it may hire new investment managers rather than increase allocations to existing managers. 11 3. Data Sources and Sample Construction 3.1 Selection and Termination Data We obtain data on the selection and termination of investment managers from three different sources: the “Tracker” database developed by Mercer Investment Consulting, the “iisearches” database created by Institutional Investor Publications, and electronic searches of articles published in Pensions and Investments (P&I). The Tracker and iisearches databases are used by investment management firms to market their services to plan sponsors. These sources provide: the name of the plan sponsor, the type of the plan sponsor, the name of the investment manager hired, the name of the consultant(s), the type and amount of the investment mandate, and a hiring date. Although similar in spirit, the two databases differ in four key ways. First, the Tracker database provides a hiring date, whereas iisearches updates information in its records after the hiring, and replaces its hire date with the date the data were last updated. This dating convention is important for matching records and for identifying the quarter in which hiring takes place. Second, the Tracker database does not record the termination of investment managers. The iisearches database does record parallel information on investment managers that are fired, but the firing data are sparse and record only single matching firing and hiring decisions. For example, if a plan sponsor fired two investment managers (A and B) and hired three (X, Y and Z), the data base records X as hired to replace A, Y to replace B, and Z as only a hiring. Thus, a sample of firing decisions (A and B) and a sample of hiring decisions (X, Y and Z) can be constructed, but a sample of round-trip firing/hiring decisions cannot. We return to round-trip decisions later in the paper. Third, iisearches provides a column containing textual information about the hiring/firing that can help in identifying the reason for the termination. Here again, the data are sparse – only some records contain textual information. As a result, we use manual searches in trade journals to fill in the gaps. Fourth, the Tracker database contains data from 1994 through 2003 whereas the iisearches database starts in 1995. We also perform electronic searches for articles in P&I, a widely used and respected source of weekly information for this industry. It reports on searches and terminations by major plan sponsors, often providing contextual information that is not recorded in the Tracker or 12 iisearches databases. We perform keyword searches of all issues of P&I between 1996 and 2003 using the following phrases: “hiring”, “firing”, and “termination”. We then read these articles and manually record the same data elements as Tracker and iisearches. In all, we read approximately 2,700 pages to record these data. We remove all non-U.S. plan sponsors from each of these databases and discard observations where the hiring (or firing) concerns custodians or record keepers. We also remove observations for employee-directed (defined contribution) retirement plans since such hiring decisions are not for a specific mandate and it is not clear how much money will flow in. This results in 15,940 hiring observations from Tracker, 11,537 hiring observations from iisearches and 1,184 observations from P&I. We use these data sources to create as comprehensive a sample as possible and to crosscheck information. To eliminate duplicates, we first create master files that uniquely identify different permutations and spellings of plan sponsor, investment manager and consultant names. We then splice the datasets together, from which we identify duplicates observations as those in which the same plan sponsor hires/fires the same investment manager within 90 days of each other. Where data sources disagree on hiring/firing dates, we regard the earliest recorded date as being accurate. We use this convention because iisearches updates its date field as it collects more information on the hiring/firing. Similarly, when data sources disagree on other aspects of the hiring/firing, we use a reasonable algorithm to determine the final value for the field (for instance, taking the minimum value of the mandate amount). In situations where the data sources disagree on the investment mandate, we treat the mandate as unknown. The intersection of these three databases produces 19,975 hiring decisions and 1,737 firing decisions. 3.2 Returns and Asset Size Data We obtain return information from Mercer’s Manager Performance Analytics database. This database contains quarterly returns (gross of fees) on approximately 9,000 products offered by 1,200 investment managers for the period 1981 to 2004. These are “composite” returns for unrestricted portfolios. The actual returns earned by a plan sponsor may differ slightly from these composite returns if the plan imposes significant restrictions on the portfolio. Mercer also 13 provides multiple benchmark return indices appropriate for each product category. For example, for the small-cap product category, Mercer provides 13 different benchmark indices. The correlation coefficients between these different indices are generally very high. Therefore, we select one index for each product category that we believe best describes the investment objective of that category. A list of each product category and the chosen index, along with a brief description, is provided in Table A1. We obtain asset information from the Money Market Directory of Investment Managers. This database contains the investment management firms’ name and the total assets under management in each year from 1996 to 2003. 3.3 Asset Allocation Data We obtain information on asset allocations for plan sponsors from two sources. P&I surveys the largest 1,000 corporate and public retirement plans in each year and records information on broad asset allocations in the following general categories: domestic equity, domestic fixed income, international equity, international fixed income, cash, private equity, real estate, mortgages and “others” (including distressed debt, oil and gas, timber etc.). These data also contain the percentage of assets in that are indexed and that are managed internally. There are several important qualifications to these data. First, they include only retirement plans and specifically exclude endowments, foundations, unions and insurance plans. Second, prior to 1996, only the largest 200 plan sponsors are surveyed. Third, the asset class categories and gradations change over time. For example, in some years, only allocations to equity, rather than domestic and international equity, are recorded. Similarly, allocations to private equity are not recorded until later in the time series. We supplement these data with hand-collected information from Nelson’s Directory of Plan Sponsors (2005). Nelson’s coverage of plan sponsors is better in that it includes endowments, foundations and union plans. However, its gradation of asset classes is not as fine as P&I. A more important issue is that a time series of allocation information is not available; we only observe allocations at the end of our sample period. 3.4 Sample Construction 14 We match the hiring/firing database with the return data in two steps. We first match the names of investment management firms across the two databases. We use Nelson’s Directory of Investment Managers (2004), the Money Market Directory of Investment Managers and Plan Sponsors (2004), and Internet searches to ensure that acquisitions of investment management firms are correctly accounted for in both databases. We match information on the investment mandate from the hiring/firing database to one of the products in the returns database. This process results in a loss of some data for three reasons. First, Mercer’s return database may not have returns for a particular hired/fired investment management firm. Second, Mercer’s return database may not have returns for the mandate for which the investment manager was hired or fired. This is often the case for “alternative asset” mandates that include venture capital and private equity. Third, we remove passive mandates from our sample since investment managers for these mandates are most likely selected for their ability to provide low cost passive exposure rather than beating a particular benchmark. Sometimes, mandate information in the hiring/firing database is available only at a broad level while the returns are available at a refined level. For instance, a hiring record may indicate that XYZ Investment Partners was hired for a large cap equity mandate. Our returns database may record return information for XYZ Investment Partners for large-cap growth, large-cap value and large-cap core products. In such situations, we use an equally weighted average return across all the relevant products and match it to the investment mandate. We perform all our tests without this averaging and note that it does not affect our conclusions. The intersection of the two databases produces a sample of 9,214 hiring decisions by 3,591 plan sponsors. These hiring decisions involve 604 investment managers hired to manage a total of $636 billion between 1994 and 2003. The firing database consists of 910 decisions by 500 plan sponsors between 1996 and 2003. These decisions involve the withdrawal of $108 billion from 247 investment managers. 4. The Selection of Investment Managers 4.1 Sample Distribution 15 Panel A of Table 1 describes the distribution of hiring decisions organized by the type of plan sponsor. Of the 9,214 hiring decisions, 21 percent (1,978) originate from corporate plan sponsors. The average size of such sponsors is $1.6 billion and the average mandate is for $41 million. Public plans represent a larger proportion of the sample (41 percent or 3,735 observations), are over ten times larger than their corporate counterparts ($10.9 billion) and have substantially larger mandates ($120 million). Endowments and foundations are smaller than corporate and public plans with an average size of only $503 million. Not surprisingly, their average mandate size is also smaller at only $23 million. Single and multiemployer union plans represent almost 10 percent of the sample and their average mandate is for $32 million. The miscellaneous category includes 843 hiring decisions by insurance plans, trusts and anonymous defined benefit plans. The variation is size is important because there may be economies of scale in the evaluation and selection of investment managers. Larger plan sponsors may be more sophisticated in asset allocation and in selection of investment managers. Retaining a consultant to assist in the evaluation and selection process can overcome this size disadvantage. Panel B shows frequency and size distribution for hiring decisions in which consultants were employed versus those in which no consultant was engaged. Both the average and median size of plan sponsors employing consultants is smaller than those that did not employ consultants. Panel C shows the time series distribution of hiring decisions. For the most part, the data appear to be relatively evenly distributed across time. However, there is a drop in hiring decisions starting in 2001 and continuing thereafter. This is because a large number of hiring decisions in this later period are in alternative assets (such as private equity partnerships, venture capital, real estate ventures etc.) for which we have no returns, and therefore disappear from our sample. We confirm this by examining the time series distribution of our original hiring sample, prior to its intersection with the returns database; the time series distribution of the original sample is indeed more uniform. Before RFP’s can be issued and an investment management firm hired, a plan sponsor must create an asset allocation plan that incorporates its investment goals and restrictions. Unfortunately, to our knowledge, there is no database of these goals, restrictions and/or 16 investment policy statements. In addition, the variation across plans is so enormous that a parsimonious characterization is not only difficult but would likely loose the very richness that makes it meaningful. Fortunately, the goals and restrictions in investment policy statements are realized (at least to some degree) in asset allocations, which are observable. As mentioned earlier, we obtain information on asset allocations by plan sponsors from P&I and Nelson’s Directory of Plan Sponsors. Because P&I surveys only the largest 1,000 plan sponsors (and in some periods, the largest 200 sponsors), data coverage is much better for large plan sponsors. In addition, there are many gaps in the time series because a plan sponsor may reach the size cutoff in one year and not in another year. We attempt to minimize the impact of these issues by also collecting data from Nelson’s Directory but allocation data are still only available for plan sponsors in approximately one-third of all hiring decisions. Panel A of Table 2 shows average asset allocations for the five different types of plan sponsors. Since our data sources provide different and not always consistent classifications of assets, we collapse all allocation information into five asset classes: domestic equity, fixed income, international equity, alternative assets (buyout funds, venture capital and hedge funds), and other assets (balanced, GICs, cash, real estate, timber, oil & gas etc.). Allocations to fixed income generate a more predictable stream of cash flows than those to equity. Therefore, plan sponsors that need to pay retirees might make higher allocations to fixed income than equity. Consistent with this, public and union plans allocate 34.7 and 38.4 percent of their assets respectively to fixed income portfolios compared with endowments that only allocate 26.3 percent. By this metric, allocations by corporate plans are relatively aggressive, allocating 47.5 percent of their assets towards domestic equity and only 26.4 percent to fixed income. Allocations to international equity portfolios are quite high from corporate and public plans (almost 11 percent for both), particularly compared to unions that invest only 2 percent of their assets in international equity. Corporate plan and endowment allocations to alternative assets are also high, but surprisingly, allocations from unions plan are also significant. Panel A also reports the percentage of assets that are indexed and managed internally by plan sponsors. Since these data elements are only available from P&I, the sample does not match that for asset allocation. In the available subsample, public plans manage a significant 17 proportion of their assets internally (13.1 percent) and also pursue indexation policies (20 percent), consistent with the increase in indexation reported by Lakonishok et al. (1992). In contrast, union plans rarely index and, not surprisingly, never manage their own assets. Interpretation of asset allocations is complicated by the fact that allocations are mechanically related (they must sum to one) and endogenously correlated. To get a cleaner sense of the “aggressiveness” of asset allocation, we create a simple allocation index that is the average of the allocation to equity (both domestic and international), alternative assets, nonindexed assets and externally-managed assets. For plan sponsors without data on indexation or externally managed assets, the average is computed only from available data elements. Although such an index is far from perfect and inevitably noisy, it is likely to be correlated with what we would really like to capture: the investment goals and restrictions in the investment policy of a plan sponsor.9 The allocation index is highest for corporate plan sponsors (0.63). This is again consistent with the idea that corporate plan sponsors can be more aggressive in asset allocation because they are the residual claimant. Somewhat surprisingly, endowments have the lowest allocation index. Panel B of Table 2 shows the distribution of hiring decisions by asset class. The vast majority of all hiring decisions, almost 55 percent, are for domestic equity mandates. We also subcategorize domestic equity mandates using a conventional size and growth-value grid. Although we do not display those results, the data still show a rich distribution across the grid. Most subcategories contain approximately 5 percent of the total number of hiring decisions. 4.2 Performance Measurement We identify quarter zero as the quarter in which the hiring takes place and then measure performance in four different ways. First, we calculate cumulative excess returns as CERi (t , H ) = t + H −1 s =t ∑ (R i ,s − Rb,s ) As a spot check, we check the value of this index for a handful of plan sponsors for which we obtain direct information on investment restrictions. We find that index values are indeed lower for plan sponsors that have quality and/or quantity restrictions on asst allocations. 18 9 where Ri,s is the return on the mandate type by the investment manager i in quarter s, and Rb,s is the return on the benchmark in quarter s. Second, we calculate the H-quarter buy and hold excess returns for each hiring decision as: t + H −1 s =t BHERi (t , H ) = ∏ (1 + R ) − ∏ (1 + R ) i ,s s =t b ,s t + H −1 Both the cumulative as well as buy and hold returns have their own advantages that are extensively described in the literature on long run returns (see, for example, Fama (1998)). For our purpose, the latter allows us to capture the true investment experience of the plan sponsor over a particular horizon. The former allows us to think of the benchmark return as a passive indexed product that the plan sponsor could presumably have invested in, and the quarter to quarter differential as variation from that benchmark. Another way to capture such variation is to compute information ratios, which we do as below: ER IRi (t , H ) = σ ER where ER is the mean excess return over the appropriate horizon and σ ER is the standard deviation of the excess return. Since information ratios are widely used in the practitioner community, we report them for a subset of our tests. The assessment of the statistical significance is also a non-trivial matter. Plan sponsors hire and investment managers are hired repeatedly in our sample period. This repetition, in combination with overlapping periods in long-horizon returns, introduces cross-sectional and time-series dependencies that render typical standard errors unreliable. We follow Jegadeesh and Karceski (2004) and calculate conservative standard errors based on a calendar time procedure that accounts for cross-correlations, heteroskedasticity and serial correlation. Details of the calculations of standard errors are contained in appendix A. The selection of a benchmark is important and difficult. Plan sponsors typically use mandate-specific benchmarks. As described in section 3.2, our returns data contain benchmark indices for various investment products/mandates. We select one index that we believe is appropriate for the mandate, and for the sake of consistency, try to use as many indices from that 19 organization across investment mandates (see Travers (2004) for a list of commonly used benchmark indices). A list of benchmark indices for each investment mandate is in Table A1. Benchmark adjustments are not risk-adjustments. One alternative is to estimate factors models in the spirit of the mutual fund literature (e.g. Elton, Gruber and Blake (1996) or Carhart (1997). Ideally, we would want to estimate alphas from a factor model before and after hiring for each hiring observation. However, the short time series, in addition to the fact that our returns are quarterly limits our ability to do so. To get around this problem, we follow a calendar time portfolio approach to estimating factor models. This allows us to estimate alphas for each year before and after hiring. The disadvantage is that since we do not obtain alphas for each hiring decision, we cannot examine cross-sectional variation in performance measured by alphas. We calculate separate calendar time portfolio returns for three-years to one-year before and after the hiring decision (in other words, we calculate six separate calendar time portfolios for each asset class). For instance, a hiring decision in December 1998 is included in the threeyear pre-hiring calendar time portfolio from December 1996 to November 1998. We then estimate alphas from factor models with the following specification for each of the calendar time portfolio R p ,t = α p + ∑ β p ,k f k ,t + ε p ,t k =1 K where Rp is the return on the portfolio p, and fk is the kth factor return. The models are estimated separately for domestic equity, fixed income and international equity mandates. For domestic equity mandates, we follow Fama and French (1993) and use the market, size and book-tomarket factors obtained from Ken French’s website. For fixed income portfolios, we use the Lehman Brothers Aggregate Bond Index return, a term spread (computed as the difference between the long-term government bond return and the T-bill return) and a default spread return as the difference between the corporate bond return and the long-term government bond. The default and term spread are obtained from Ibbotson Associates. For international equity mandates, we employ an international version of the three factor model. We obtain the international market return and book-to-market factor from Ken French. The international size factor is computed as the difference between the S&P/Citigroup PMI World index return and the 20 S&P/Citigroup EMI World index return, both of which exclude the United States (see http://www.globalindices.standardandpoors.com). 4.3 Pre- and Post-Hiring Performance Table 3 shows excess returns by all four metrics one, two and three years before and after hiring. Since our returns data end in 2004, two- and three-year returns cannot be calculated for hiring decisions after 2003 and 2002 respectively. To ensure that changing sample sizes do not drive our results, we only report excess returns for a balanced sample in which excess returns can be computed for matched horizons before and after hiring.10 For cumulative excess returns, buy and hold excess returns, and information ratios, we report results for the full sample as well as separately for domestic equity, fixed income and international equity. Standard errors appear in parentheses. Panel A shows results using cumulative excess returns. The pre-hiring excess returns are strikingly positive: the three, two and one-year pre-hiring excess returns are 9.8, 7.4 and 3.5 percent respectively and statistically significant. The same pattern is evident using buy and hold excess returns (Panel B) and information ratios (Panel C). Clearly, and not surprisingly, plan sponsors hire investment managers after a sustained period of superior performance. The magnitude of the pre-hiring excess returns is highest for international equity mandates, followed by domestic equity and then fixed income. In Panel D, factor regressions also show that prehiring alphas are positive, but only for domestic and international equity. Performance subsequent to hiring is statistically flat for the full sample using all methods of measuring excess returns. Cumulative excess returns one, two and three years after hiring are 0.4, 1.1 and 1.8 percent but with standard errors of 0.6, 0.9 and 1.3 percent respectively. The only case in which post-hiring excess returns are positive is for international equity mandates. This effect for international equity mandates appears to be quite robust as all four excess performance measures show positive post hiring performance. 10 For example, the three year pre and post hiring excess returns are based on a matched sample size of 6485 decisions. The sample sizes for the two and one year returns are 7,668 and 8,707 respectively. 21 4.4 Cross-sectional Variation in Post-Hiring Performance The discussion in section 2 suggests that post-hiring returns are likely to be systematically related to incentive structures, size, goals and restrictions of plan sponsors. Capturing these effects without detailed information on the funding ratio, sponsor-specific goals, asset class and security specific restrictions, and composition of the investment committee is almost impossible. However, we can make some headway by grouping plan sponsors into the categories in tables 1 and 2, and by classifying them further by size and the asset allocation index. In essence, these measures are likely to be correlated with the underlying attributes of interest and provide reasonable proxies. Panel A of Table 4 shows cumulative excess returns separately for corporate plans, public plans, endowments, unions and the miscellaneous category (Panel B shows buy and hold excess returns). Pre-hiring returns are consistently positive for all categories. More interesting is the fact that post-hiring returns are statistically indistinguishable from zero for all categories. Despite the fact that corporate plans have incentives to generate superior performance, as a group, they do not appear to do so, at least from their hiring decisions. The economies available to larger plan sponsors might make them better at the selection process than smaller sponsors. We examine the pre and post-hiring performance of small and large plan sponsors using the 25th and 75th percentile of size as cutoffs. Even though pre-hiring excess returns are positive for both small and large plan sponsors, post-hiring returns are positive only for large plan sponsors. The magnitude of the effect is also impressive – the average three year post-hiring excess return for large plan sponsors is 3.2 percent with a standard error of 1.3 percent. The allocation index also appears to be correlated with performance. Again, using the 25th and 75th percentiles as cutoffs, we find that pre-hiring excess returns are large for both low and high index groups. Post-hiring returns, however, are significant only for the high allocation index group, 4.0 percent with a standard error of 1.5 percent. These are useful but simple univariate characterizations of cross-sectional patterns in post-hiring returns. For instance, the correlations between size and plan sponsor categories (and similarly for the allocation index) are important and could provide a richer picture. However, a cross-sectional regression based analysis of post-hiring returns is complicated by the fact that 22 plan sponsors retain consultants to assist in the hiring of investment management firms. Employing a consultant is clearly endogenous to the post-hiring returns and necessitates a procedure that corrects for this selectivity. We follow Madalla (1983) and estimate the following model. yj = β xj +δ zj +ε j where yj represents three-year post-hiring cumulative excess returns or buy and hold excess returns, xj is a vector of explanatory variable, and zj is a dummy variable for whether a consultant was employed. We use three-year post hiring returns to study post hiring performance since it is less noisy than a shorter horizon return. The selection equation is modeled as z* = γ w j + u j j ⎧ 1, if z * > 0 j where wj is a vector of explanatory variables. The regressions are where z j = ⎨ ⎩0, otherwise jointly estimated via maximum likelihood procedures and standard errors account for clustering where an investment management firm is hired for a mandate in the same style and period by different plan sponsors. The results are shown in Table 5. For the selection equation, we use explanatory variables (wj) that influence the probability that a plan sponsor employs a pension consultant. The logarithm of plan sponsor size captures the idea that larger plan sponsors may have economies in hiring and are thus less likely to employ consultants. We also create two indicator variables, headline risk resistant and headline risk sensitive, that group plan sponsors by their likely sensitivity to adverse publicity. Banks, corporate plans, health and hospital plans, hybrid plans, insurance plans, and nuclear decommissioning trusts are regarded as headline resistant. Because appointments at public plans and unions are often political, we regard them as headline risk sensitive. The remainder, endowments, foundations, annuity funds, trust funds and anonymous plans, are regarded as headline-risk neutral and are captured by the intercept. We also include broad style indicators (corresponding to domestic equity, fixed income, and international mandates) because of mandate specific needs and expertise. To conserve space we show the selection equation from 23 only one model in the table. As expected, plan sponsor size is negatively related to the decision to hire a consultant. Plan sponsors that are more sensitive to headline risk are more likely to employ consultants and those that are resistant to headline risk are less likely to use consultants. The implied increase (decrease) in the probability of employing a consultant from the headline sensitive (resistant) indicator variable is 17 percent (-5 percent). The independent variables (xj) in the return regression measure plan sponsor heterogeneity, scale effects and return reversals. The three-year pre-hiring return determines if return reversal is present in a multivariate setting. We include the logarithm of mandate size because larger mandates may generate smaller post-hiring returns because of diseconomies of scale in investment management (Perold and Salomon (1991)). Plan sponsor heterogeneity is captured in a number of ways. First, we use indicator variables for each type of plan sponsor (and accordingly estimate the regression without an intercept). The indicator variables pick up variation in incentive effects and structure across types of plan sponsors but are nonetheless coarse. Therefore, we also employ interaction effects between these indicators and other determinants of plan sponsors ability, goals and restrictions. We interact plan sponsor size with each of the plan sponsor type indicator variables to pick up scale effects. Similarly, the allocation index used earlier is meant to proxy for the goals and restrictions faced by a plan sponsor. Again, since the effects are likely to be systematically related to the type of plan sponsor, we introduce the allocation index through interaction effects. One could imagine even more complex effects between the type of plan sponsor, size and the allocation index, so we also employ three-way interaction effects. Fixed effects for detailed investment styles allow for intercept shifts in post-hiring returns that are not picked up by the benchmark used to compute excess returns.11 The expected value of employing a consultant, after accounting for the endogenous selection, is ⎡ ⎤ φ ( wi γ ) E ( yi z i = 1) − E ( yi z i = 0) = δ + ρσ ⎢ ⎥ ⎢ Φ ( w j γ ){1 − Φ ( wi γ )} ⎥ ⎣ ⎦ 11 Although the dependent variable is an excess return (say, raw return of a small cap value manager minus the return on a small cap value index), there may still be heterogeneity in investment manager returns within small cap value asset class. For example, one manager might invest in micro-cap securities exclusively, even though it is regarded as small cap. These indicator variables account for such effects. 24 where ρ is the correlation between εj and uj, and Φ is the cdf of a standard normal distribution. Since asset allocation information is only available for a small proportion of the sample, and that too only for large plan sponsors, we estimate an initial model without the allocation index. We use this model to interpret parameter estimates for all variables except asset allocation interactions. A second model, estimated on a subset of data, is used exclusively to conduct inferences regarding asset allocation. The regressions show strong evidence of return reversal at all three horizons. The negative coefficients on the pre-hiring return variable do not imply negative post-hiring returns, just that post-hiring returns are smaller than pre-hiring returns. Mandate size is negatively related to post-hiring returns suggestive of diseconomies of scale in investment management.12 Larger plan sponsors appear to generate superior post-hiring performance, even after accounting for the influence of consultants, consistent with scale economies at the plan sponsor level. Although these effects appear to the largest for corporate and union plans, they are present for all types of plan sponsors. The economic magnitude of some of these effects is quite large. The average impact of a one standard deviation increase in three-year pre-hiring returns (with other variables evaluated at their mean) implies a decrease in three-year post-hiring cumulative excess returns of 4.7 percent. Similarly, one standard deviation increase in plan sponsor size leads to an increase of 3.7 percent in three-year post-hiring returns for corporate plans while one standard deviation increase in mandate size leads to a decrease of 0.8 percent in three-year post-hiring returns. The use of a consultant leads to increase in three-year post-hiring returns between 1.5 and 2.7 percent depending on the specification. Turning now to the second specification that includes asset allocation information, there is a dramatic reduction in sample size. More important, however, is the fact that since asset allocation information is only available for large plan sponsors, a censoring effect is likely to cloud inferences. As a result, we only interpret the coefficients with allocation index 12 We also scale the size of the mandate by the total assets under management in the year prior to the hiring and use the scaled variable in the regression. However, since our asset data are only available from 1996 onwards, the loss in sample size is quite large. Nonetheless, regression results are generally robust to the inclusion of the scaled variable. 25 interactions. Aggressive asset allocation generates superior performance for corporate plans and endowments; these effects are large, 8.1 and 4.5 percent respectively. In contrast, aggressive allocations by public plan sponsors is associated with negative post hiring returns. To the extent that aggressive allocation can be interpreted as originating from fewer restrictions on the investment process, one might interpret our results to suggest that restrictions hinder performance. Overall, the regressions show tremendous heterogeneity in the ability of plan sponsors to time hiring decisions and obtain post-hiring excess returns. The parameter estimates for various types of plan sponsors, their size and asset allocation, and particularly interaction effects between these variables, suggest that this heterogeneity is related to the goals, restrictions and sophistication of plan sponsors. 5. The Termination of Investment Managers 5.1 Sample Description Our firing sample consists of a total of 910 termination decisions. The number of termination decisions captured by our data collection process is substantially smaller than that of hiring decisions for three reasons. First, the data sources that we use (which to our knowledge, are the only available sources) serve a marketing function, designed to inform investment managers that a plan sponsor is searching for an investment manager in a particular asset class / mandate. These sources are not designed to track performance, or to assign blame. As such, the emphasis is on new accounts and revenue. Second, termination decisions are generally viewed with some distaste and there is a natural disinclination to report terminations. Certainly, investment managers have no incentive to report their own terminations. Plan sponsors may choose not to publicize terminations because they may employ the same manager for another mandate, either currently, or in the future. Third, there has been a steady increase in the assets under the administration of plan sponsors over the sample period. Ergo, the number of hiring decisions in the population is likely to be larger than that of firing decisions. Panel A of Table 6 shows the time series distribution of termination decisions. Unlike hiring decisions, the number of firing observations increases over time. This is because our data 26 sources do a better job of capturing such decisions in the later years. Panel B shows the distribution of termination decisions by asset class. As with the hiring decisions, the majority of the firing decisions are for (domestic) equity mandates. We use the textual information in our data sources to categorize the reasons for the termination of the investment manager into three categories. The “organizational” category includes decisions due to personnel turnover, mergers or regulatory actions against the investment management firm. About 13 percent of all termination decisions are identified as such. A termination decision is placed in the “performance” category if the plan sponsor publicly declares it as such. There are 292 termination decisions (32 percent) that fall into this category. There are 102 termination decisions in the “reallocation” category, which take place because the plan sponsor has decided to reduce or eliminate its position in the asset class that the investment manager is responsible for. For 394 termination decisions (43 percent of the total), we do not have enough information to place the termination decision in any of these categories. There are two important caveats associated with the termination reasons described above. First, the reasons are self-identified by the plan sponsor. For example, if a plan sponsor publicly states that an investment management firm is being terminated for performance reasons, we take the sponsor’s word for it. It is entirely possible that the underperformance allegation is disingenuous. Second, the above categories are not homogenous. For instance, terminations caused by personnel turnover (i.e., the departure of a key portfolio manager or founder) and those caused by regulatory action against the firm are both classified as organizational, even though the causal mechanisms are quite different. Similarly, elements of current or future underperformance can easily creep into non-performance categories. An acquisition of one investment management by another can take place after underperformance. Alternatively, a plan sponsor may terminate an investment manager after the departure of key personnel because it believes that the departure will cause underperformance in the future. Thus, we interpret termination reasons with caution. Panel D shows the distribution of firing decisions across types of plan sponsors. Almost half of the terminations originate from public plan sponsors. Panel E shows the mean and median size of plan sponsors for the subsample of termination decisions. The average size of the 27 plan sponsor is $7.9 billion, about $2 billion larger than that for hiring decisions. This is most likely because our data sources do a better job of covering the firing decisions of larger plan sponsors. The mean (and median) mandate size for terminated managers is substantially larger than that for hired managers, $145 million ($50 million) compared to $78 million ($24 million). 5.2 Pre- and Post-Firing Performance Table 7 shows average pre- and post-firing returns using all four performance measures and associated standard errors. As before, we ensure that pre- and post-firing returns are comparable by requiring a balanced pre- and post-event sample. This shrinks sample sizes; the three (two, one) year returns are based on 300 (510, 810) termination decisions. Panels A and B show cumulative and buy and hold excess returns respectively, Panel C shows information ratios and Panel D shows calendar time alphas from factor models. Two main results stand out. First, pre-firing returns are generally statistically indistinguishable from zero. For instance, the three (one) year pre-firing cumulative excess return is 0.6 percent (-1.0 percent) with a standard error of 1.3 percent (0.7 percent). Second, three years after the firing decision, post-firing returns are positive. Again, focusing on cumulative excess returns, the three-year post-firing return is 4.3 percent with a standard error of 1.3 percent. It is not surprising that the standard errors associated with pre-firing returns are large. Ex ante, we expect negative pre-firing excess returns for investment managers terminated due to poor performance. But it is not obvious that pre-firing returns should be negative for other categories. A termination due to a reallocation decision by the plan sponsor may occur because the investment style is out of favor, or because the plan sponsor desires an asset allocation shift to accommodate accelerated or decelerated retirement payments. In the former case, if a style is out of favor, this should also be reflected in the benchmark return, so that pre-firing excess returns should be zero. In the latter, asset-liability mixes should be unrelated to performance, implying that pre-firing returns should be zero. Finally, it is difficult to predict pre-firing excess returns for terminations due to organizational reasons. On the one hand, it is possible that organizational restructuring (a merger or personnel turnover) takes place after poor performance. 28 One the other hand, it is also possible that such restructuring takes place because of superior performance (e.g., a star portfolio manager is hired by another firm or decides to start his/her own firm). We break up the returns of terminated investment managers by the reason for the firing and present the results in Table 8. Panel A presents results with cumulative excess returns and Panel B shows results for buy and hold returns. The largest negative returns are for performance-related termination. As mentioned earlier, it is possible that plan sponsors are sometimes disingenuous in alleging underperformance and consequently terminating investment managers. Therefore, we also separately tabulate returns of investment managers that have strictly negative pre-firing returns over each return horizon. Interestingly, post-firing returns are positive for all these subsamples. Following our analysis with post-hiring returns, we break up pre- and post-firing returns by plan sponsor type, size and the asset allocation index. Pre-firing excess returns are only negative for endowments. Over longer horizons (three years), pre-firing returns are actually positive for corporate plans, public plans and unions. But, with the exception of unions, postfiring returns are positive for all remaining types of plan sponsors. There is not much dispersion in pre-firing or post-firing returns across small and large plan sponsors, or by the magnitude of the asset allocation index. These univariate cuts of the data are certainly suggestive of more complex interactions. Although any attempt at understanding variation in post-firing returns is hindered by small sample sizes, we estimate selectivity-corrected cross-sectional regressions for post-firing returns, analogous to those for hiring returns. However, we make two modifications to the specifications. First, we add three indicator variables that capture the reason for the firing. Second, we do not use asset allocation information because its inclusion drastically reduces the sample size. The results of these regressions are reported in Table 9. The pre-firing return continues to be negative and is statistically significant in all three return regressions. In general, the coefficients on other variables such as plan sponsor size and mandate size are similar to the hiring regressions but lack statistical significance, probably 29 because of small sample sizes. The most important result from the plan sponsor type indicators and interactions with size is that large endowments earn significantly positive post-firing returns. As a whole, our data appear to indicate that plan sponsors show limited timing ability in terminating investment managers. In some cases, they fire investment managers after a sustained period of underperformance, but subsequent to that, returns rebound. The extent to which such (mis) timing damages the performance of the plan sponsor depends on their true economic loss, which in turn, depends on the performance of investment managers hired to replace terminated managers. Although it is tempting to simply compare post-hiring returns with post-firing returns, we refrain from doing so because firing and hiring decisions are coordinated using complicated mechanisms. Therefore, we proceed to an analysis of such “round-trips” below. 6. Round-trip Termination and Selection of Investment Managers The best way to illustrate the complexity of a round-trip termination and selection decision is by way of examples. Example 1 In the first quarter of 2000, the St. Louis Employees Retirement System terminated 1838 Investment Advisors for its core long-term fixed income portfolio, reportedly because of poor performance. It then hired Reams Asset Management to handle this $45 million portfolio. Watson Wyatt Investment Consulting assisted in the search. Example 2 In the first quarter of 2002, the Arapahoe County Employees Retirement System hired Barclays Global Investors to manage $15 million in passive global large-cap equity, Artisan Partners for a $10 million active international all-cap equity mandate, Brazos for $9 million in active domestic micro-cap equity and Royce for $5 million in active domestic small-cap equity. The Barclays’ hiring was funded by reallocating $15 million from a $44 million active domestic large cap growth equities portfolio managed by Fayez Sarofin. Artisan’s allocation came from terminating a $10 million active international all-cap equities portfolio managed by Brinson Partners. Brazos and Royce were funded by terminating a $14 million active domestic madcap growth equities portfolio managed by Denver Investment Advisors. The first example is a straightforward round-trip firing and hiring decision in which the mandate size and type is the same, and the reason for the decision clearly delineated. The second contains two round-trip observations: (i) Denver Investment Advisors is terminated and replaced by Brazos and Royce. The mandates for the hired investment managers are different from the 30 terminated investment manager and the allocation of the $14 million portfolio is not even. (ii) Brinson Partners is terminated and replaced by Artisan Partners in the same mandate. Note that the Barclays Global Investors hiring does not create a round-trip observation since it is not the result of a termination but an allocation adjustment for an ongoing investment manager. In both of the above examples, the fired and hired investment managers are clearly identified. Some plan sponsors can replace terminated investment managers relatively quickly, particularly if the terminated manager was “on watch” prior to the termination. If terminations are abrupt or unplanned, however, plan sponsors often “park” assets in index funds until the search process is completed. This makes the identification of round-trip decisions more difficult, since the hired manager (an index provider) is only temporarily managing the assets. 6.1 Sample Construction and Description Because of the complexity of the process described above, we cannot mechanically associate hiring and firing decisions, and must therefore build a sample using manual procedures. We start with our sample of firing decisions. For each firing decision, we match hiring decisions by the same plan sponsor up to one quarter after the firing date.13 This produces 2,206 candidate firing-hiring decisions, which contain duplications. The duplications occur because of two reasons. First, a hiring decision can be associated with more than one firing decision and viceversa. Second, some round-trip decisions can involve the use of transition managers that hold parked assets while a new manager search is conducted. For each candidate observation, we then search for articles detailing the decisions in the following trade journals: Pensions and Investments (P&I), Investment Management Weekly (IMG), Money Management Letter (MML), and Dow Jones Money Management Alert (DJMMA). We mark each round-trip with an ID that allows us to track these decisions and eliminate duplications. This process identifies 729 round-trip firing-hiring decisions. We then match these round-trip decisions with our returns database, keeping only decisions for which we have some returns. As before, this eliminates decisions involving investments in hedge funds, 13 We restrict our search for matching hiring decisions to one quarter after the firing to limit the amount manual data collection required. 31 venture capital funds and private equity. Our final sample consists of 660 round-trip firinghiring decisions between 1996 and 2003. On average, each round-trip decision is associated with the firing and hiring of 1.1 investment managers, with a maximum of 11 investment managers hired or 7 investment managers fired in a particular decision. The average mandate size for firing is $116 million while the average mandate size for hiring is $102 million. 6.2 Round-trip Performance If more than one firm is fired (or hired), we compute the excess return for that round-trip observation as the average across the fired (or hired) firms. In example 2 described above, preand post-firing returns for Denver International Advisors would be compared to the average of the pre- and post-hiring returns of Brazos and Royce. Both hired and fired firms are required to have returns over a particular evaluation horizon. Panel A (B) of Table 10 shows average pre- and post-event cumulative (buy and hold) excess returns for fired and hired firms separately for the entire sample. Consistent with earlier results, the pre-firing returns of fired firms are negative, and the three-year post-firing return is positive and statistically significant. Also, excess returns are large and positive prior to hiring, and statistically insignificant after hiring. This is reassuring because it suggests that our roundtrip sample is similar to that of the earlier (larger) hiring and firing samples. In addition to hired and fired firm’s returns, we also report return differences (hired firm’s excess returns minus fired firm’s excess returns) with corresponding standard errors.14 Prior to the firing/hiring decision, the return differences are large, positive and statistically significant. The three-year (one year) cumulative excess return difference prior to the firing/hiring is 9.2 percent (4.7 percent) with a standard error of 1.2 percent (0.95 percent). After the hiring/firing decision, the performance of the fired firms’ exceeds that of the newly hired firms’ over all three horizons but with larger standard errors; the three year cumulative excess return difference is –2.2 percent but with a standard error of 2.2 percent. 14 Note that return differences can be directly computed from the fired and hired firm’s returns but the standard errors of these return differences cannot. 32 7. Robustness Our tests involve numerous decisions regarding empirical design and performance measurement. We perform a battery of tests to see if our results are sensitive to these decisions. Although all the tests are too numerous to report, we discuss the important ones below. 7.1 Date Identification Our data identify hiring and firing dates with some imprecision, but the imprecision is likely to be in one direction – there can only be a delay in reporting the decision date. If the delay is more than one quarter for hiring decisions, then our post-hiring returns are biased downward. Similarly, if the delay is greater than one quarter, the post-firing returns are upward biased but the pre-firing returns should not be affected. To determine if these effects are important in our data, we shift our hiring/firing date back by one quarter and retabulate our results. The results (not reported), suggest that on average, it does not appear that event time misalignment significantly influences our inferences. 7.2 Backfilling and Survivorship Issues The returns data are self-reported by investment management firms. Given that a successful track history of returns is critical for hiring, it is possible that some investment management firms “amend” prior year’s returns in updating return information. We ensure that this is not the case – Mercer informs us that investment managers provide each quarter’s return soon after the end of the quarter and are not permitted to update prior returns. Another potential concern is one of survivorship bias, particularly since post-firing returns (over a three-year interval) are positive and statistically significant. We perform three checks to determine if survivorship bias influences our results. First, we compute attrition rates of investment managers and ensure that return histories disappear over time. Simple tabulations of return histories show an attrition rate of approximately 4 percent per year in our sample (by comparison, Carhart et. al. (2002) report an average annual attrition rate of 3.6 percent for mutual funds). Second, we calculate the number of instances where pre-firing returns are 33 available but post-firing returns are not. To do this, we eliminate firing decisions where postfiring returns would be unavailable because our returns data end in 2004 and tabulate sample sizes. We find that the loss in data is trivial (10 observations for a one-year horizon), suggesting that post-firing returns do not disappear from the sample because the pre-firing returns are negative. Third, we re-examine the portion of our firing database for which we have no returns (either pre- or post-firing). The vast majority of firing decisions for which we have no returns are where the mandate is unknown or in an asset class not covered by our returns database (private equity, venture capital, real estate investments etc.). 7.3 Benchmarking, Return Computation and Changes in Risk Our excess returns could be sensitive to our choice of benchmarks. We repeat all our analysis by randomly choosing a benchmark from the feasible set of benchmarks provided by Mercer, not including our originally chosen benchmark. These alternative benchmarks do not significantly influence our results for hiring, firing and round-trip decisions. If an investment manager has more than one product in an investment mandate, we average the returns across products in our analysis. To determine if this averaging procedure influences our results, we redo our analysis using only data that are uniquely matched between the investment mandate and product returns. The post-hiring excess returns one, two and threeyears after hiring are 0.9, 1.6, and 2.2 percent respectively while the post-firing returns are 1.8, 3.5, and 7.6 percent respectively. Investment manager termination could be correlated with changes in portfolio risk before and after termination. For example, Brown, Harlow and Starks (1996), Chevalier and Ellison (1999) and Busse (2001) show that underperforming managers increase portfolio risk in an attempt to generate superior returns. Gallo and Lockwood (1999) show correlated changes in investment style. Such behavior may be prevalent in institutional investment managements firms as well but is more likely to influence our termination sample. Our specification of our calendar time factor model regressions allows us to test if these pre- and post-event betas are different from each other. However, we mostly fail to reject the null hypothesis of constant beta. For instance, for domestic equity mandates, we find that the pre-hiring beta on the market portfolio is 34 between 0.98 and 0.99 while post-hiring beta on the market portfolio is around 1.03 – the difference is statistically insignificant. The differences in pre-firing and post-firing betas are a bit larger. The beta on the market is 1.07 for one-year prior to firing and 1.01 for one-year postfiring and this difference is only marginally significant. The differences in pre-firing and postfiring betas for longer horizons of two- to three-years are positive but statistically insignificant. We suspect two reasons for the small changes in beta. First, most investment management firms have a large stable of clients. Losing one or two clients is unlikely to dramatically influence risk-taking incentives. Second, plan sponsor monitoring of tracking error (Del Guercio and Tkac, 2002) is likely to reduce incentives to change risk profiles dramatically. 8. Discussion We do not know the true performance of plan sponsors – that would require detailed knowledge of asset allocation and the returns to each mandate with investment managers (not just composite returns). But our results are still informative about the raison d’être of plan sponsors. To summarize, we find that plan sponsors hire investment managers after superior performance but on average, post-hiring excess returns are zero. There is tremendous variation in post-hiring performance that is related to attributes of the plan sponsors. Plan sponsors fire investment managers (mostly) after sub-standard performance but post-firing excess returns are frequently positive and sometimes statistically significant. How does one interpret this evidence? One way to think about this is in terms of opportunity costs and frictions. For hiring decisions that are necessitated by the termination of an existing investment manager (due to performance, organizational or reallocation reasons), the opportunity costs of hiring can be identified as the returns that the fired manager would have delivered relative to what the hired manager actually delivers. Our round-trip results suggest that these opportunity costs are positive. If one adds transition costs discussed in the introduction (say, 1.0 to 2.0 percent) to these opportunity costs, the overall costs of firing and hiring investment managers rise further.15 15 Subtracting a constant from the mean return obviously does not change the standard errors and will “make” the excess returns statistically significant. 35 If the costs associated with hiring and firing investment managers are important, then at the margin, they should play a role in retention decisions. Typically, an investment management firm is hired for a given term, but then can be “rehired” for a subsequent term. If replacement costs are relevant, then the pre-rehiring performance that justifies retention should be lower than for brand new hiring. To determine if that is the case, we create a sample of retentions. We examine a random sample of 350 plan sponsors in Nelson’s Directory of Plan Sponsors (2005). Nelson’s reports the name of investment managers with mandates from each plan sponsor as of 2004, the year that investment manager was originally hired, and the investment mandate. We manually record this information for investment management firms that are in our returns database, where the mandate amount is recorded and where the original hiring year is before 2000. We then assume that a retention decision is made every three years. For example, if XYZ Asset Management was originally hired by ABC Plan Sponsor in 1996, we assume a retention decision is made in 1999 and 2002. In total, our sample consists of 1,867 retention decisions. We then compute pre-retention returns in the same manner as before and compare them to prehiring returns for the same plan sponsors. We find that the average one year (three year) cumulative excess return for retentions is 2.4 percent (6.1 percent), compared with 4.9 percent (14.7 percent) for hiring decisions by the same plan sponsors. This suggests that, in making retention decisions, plan sponsors incorporate the costs associated with hiring and firing. For hiring decisions necessitated by the growth of sponsor assets, the identification of opportunity costs is more difficult. One way to approach the problem is to consider the role of persistence in investment manager returns. If there is little or no persistence in the performance of investment managers in general, then on average, hiring decisions should produce zero excess returns. This implies that plan sponsors achieve their objectives, since on average, hired investment managers deliver benchmark returns. A full scale analysis of persistence is beyond the scope of our paper. However, Christopherson et al. (1998) and Busse et al. (2006) undertake such an analysis for institutional investment managers and find evidence of persistence amongst winners for up to one year, and in some cases, longer. Their persistence results indicate that plan sponsors could generate excess returns by appropriately timing hiring decisions, and suggest that the opportunity cost of zero post-hiring returns is positive. 36 It could be that the costs documented and discussed above have compensating benefits that we are unable to measure. From an efficiency perspective, terminating investment managers could be critical to maintaining discipline among incumbents and maintaining a competitive marketplace. It is also possible that the agency relationships described by Lakonishok et al. (1992) create such high barriers to change so as to make it impossible to eliminate the costs. Moreover, if one views hiring and firing as analogous to fund flows, then our results are consistent with the equilibrium model performance and flows of Berk and Green (2004). 9. Conclusion In this paper, we examine the selection and termination of investment managers by plan sponsors. To do so, we build a dataset that comprises hiring and firing decisions by 3,500 plan sponsors over a 10-year period from 1994 to 2003. We find that plan sponsors hire investment managers after these managers earn significant excess returns. On average, post-hiring returns are statistically indistinguishable from zero. However, we note that there is considerable heterogeneity in post-hiring returns appears to be related to sponsors attributes. In contrast, plan sponsors terminate investment managers after poor performance but the performance of these investment managers appears to rebound after firing. We also examine a set of round-trip firing and hiring decisions and find that the post-firing returns of fired investment managers are generally larger than the post-hiring returns of hired investment managers. Given the magnitude of the return differences, and the transactions costs associated with transitioning portfolios from fired investment managers (legacy portfolios) to hired investment managers (target portfolios), our results suggest that the termination and selection of investment managers is an endeavor fraught with risk. 37 References Allen, Franklin, 2001, Do financial institutions matter? Journal of Finance, 56, 11651176. Barber, Brad, and Terrance Odean, 2000, Trading is hazardous to your wealth: The common stock investment performance of individual investors, Journal of Finance, 773-806. Barber, Brad, Terrance Odean, and Lu Zheng, 2003, Out of sight, out of mind: The effects of Expenses on mutual fund flows, forthcoming, Journal of Business. Berk, Jonathan, and Richard C. Green, 2004, Mutual fund flows and performance in rational markets, Journal of Political Economy 112, 1269-1295. Brown, Stephen, and William Goetzmann, 1995, Performance persistence, Journal of Finance, 50, 679-698. Brown, Keith, W.V. Harlow, and Laura Starks, 1996, Of tournaments and temptations: A analysis of managerial incentives in the mutual fund industry, Journal of Finance, 51, 85-110. Bollen, Brian, 2004, Lost in transition? Financial News, April 19, 2004. Bollen, Nicholas, and Jeffrey Busse, 2005, Short-term persistence in mutual fund performance, Review of Financial Studies, 18, 569-597. Brull, Steven, 2006, Rich plan, poor plan, Institutional Investor, 40, 30-35. Busse, Jeffrey, 2001, Another look at mutual fund tournaments, Journal of Financial and Quantitative Analysis 36, 53-73. Busse, Jeffrey, Amit Goyal, and Sunil Wahal, 2006, Performance Persistence in Institutional Investment Management, working paper Emory University. Carhart, Mark, 1997, On persistence in mutual fund performance, Journal of Finance, 52, 57-82. Carhart, Mark M., Jennifer N. Carpenter, Anthony W. Lynch, and David K. Musto, 2002, Mutual fund survivorship, Review of Financial Studies 5, 1439-1463. Chevalier, Judith, and Glen Ellison, 1999, Career concerns of mutual fund managers, Quarterly Journal of Economics, 114, 389-432. 38 Christopherson, Jon A, Wayne Ferson, and Debra Glassman, 1998, Conditioning manager alphas on economic information: Another look at the persistence of performance, Review of Financial Studies 11, 111-142. Coggin, T, Frank Fabozzi, and Shafiqur Rahman, 1993, The investment performance of US equity pension fund managers: An empirical investigation, Journal of Finance, 48, 1039-1055. Cornell, Bradford, and Richard Roll, 2004, A delegated agent asset-pricing model, forthcoming, Financial Analysts Journal. Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997, Measuring mutual fund performance with characteristic-based benchmarks, Journal of Finance 52, 1035-1058. Del Guercio, Diane, and Paula Tkac, 2002, The determinants of the flow of funds of managed portfolios: Mutual funds versus pension funds, Journal of Financial and Quantitative Analysis 37, 523-557. Elton, Edwin, M. Gruber, S. Das, and M. Hlavka, 1992, Efficiency with costly information: A reinterpretation of the evidence for managed portfolios, Review of Financial Studies 6, 1-22. Elton, Edwin, M. Gruber, and Christopher Blake, 1996, The persistence of risk-adjusted mutual fund performance, Journal of Business. Fama, Eugene, and Kenneth French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3-56. Fama, Eugene, 1998, Market efficiency, long-term returns and behavioral finance, Journal of Financial Economics 49, 283-306. Fabozzi, Frank, 1997, Pension fund investment management, Frank J. Fabozzi Associates, New Hope, Pennsylvania. Gallo, J.G., and Larry Lockwood, 1997, Benefits of proper style classification of equity portfolio managers, Journal of Portfolio Management 23, 47-55. Goetzmann, William, and Roger Ibbotson, 1994, Do winners repeat? Patterns in mutual fund performance, Journal of Portfolio Management, 20, 9-18. Grinblatt, Mark, and Sheridan Titman, 1992, The persistence of mutual fund performance, Journal of Finance 47, 1977-1984. 39 Grinblatt, Mark, Sheridan Titman, and Russ Wermers, 1995 Momentum investing strategies, portfolio performance and herding: A study of mutual fund behavior, American Economic Review, 85, 1088-1105. Gruber, Martin, 1996, Another puzzle: The growth in actively managed mutual funds, Journal of Finance, 51, 783-810. Hendricks, D, J. Patel, R. Zeckhauser, 1993, Hot hands in mutual funds: Short-run persistence of performance, 1974-1988, Journal of Finance 48, 93-130. Heisler, Jeffrey, Christopher R. Knittel, John J. Neumann and Scott Stewart, 2004, Why do Institutional Plan Sponsors Fire Their Investment Managers? Working paper, Boston University. Investment Company Institute, 2004, Trends in mutual fund investing, www.ici.org/stats/index.html Ippolito, Richard, 1989, Efficiency with costly information: A study of mutual fund performance, 1965-1984, Quarterly Journal of Economics 104, 1-23. Jegadeesh, Narasimhan, and Jason Karceski, 2004, Long-run performance evaluation: Correlation and heteroskedasticity-consistent tests, working paper, Emory University. Jensen, Michael, 1968, The performance of mutual funds in the period 1945-1964, Journal of Finance 48, 389-416. Lakonishok, Josef, Andrei Shleifer, and Robert Vishny, 1992, The structure and performance of the money management industry, Brookings Papers: Microeconomics , 339-391. Logue, Dennis E., and Jack S. Rader, 1998, Managing pension plans: A comprehensive guide to improving plan performance, Harvard Business School Press, Boston, MA. Madalla, G.S., 1983, Limited dependent variables and qualitative variables in econometrics, Econometric Society Monographs, Number 3, Cambridge University Press, Cambridge. Money Market Directory of Pension Funds and Their Investment Advisors, 2003. Charlottesville, VA. Nelson’s Directory of Investment Managers, 2002, Nelson Publications, Port Chester, NY. 40 Newey, Whitney, and Kenneth West, 1987, A simple, positive definite, heteroskedasticity, and autocorrelation consistent covariance matrix, Econometrica 55, 703-708. Odean, Terrance, 1998, Are investors reluctant to realize their losses? Journal of Finance 53, 1775-1798. Odean, Terrance, 1999, Do investors trade too much? American Economic Review, 89, 1279-1298. Perold, Andre, and Robert Salomon, Jr, 1991, The right amount of assets under management, Financial Analysts Journal 47, 31-39. Proszek, Stan, 2002, Transition management: Simple – but not easy, Benefits and Pensions Monitor 12, October 2002. Sirri, Erik, and Peter Tufano, 1998, Costly search and mutual fund flows, Journal of Finance, 53, 1589-1622. Travers, Frank J., 2004, Investment manager analysis: A comprehensive guide to portfolio selection, monitoring and optimization, John Wiley & Sons. Wermers, Russ, 2000, Mutual fund performance: An empirical decomposition into stockpicking talent, style, transactions costs, and expenses, Journal of Finance 55, 1655-1695. Werner, Bob, 2001, The true costs and benefits of portfolio transition management, www.russell.com/AU/press_room/Press_Releases/PR20011004_AU_p.asp. Zheng, Lu, 1999, Is money smart? A study of mutual fund investors’ fund selection ability, Journal of Finance, 54, 901-933. 41 Table 1 Distribution of Hiring Decisions by Plan Sponsors Each plan sponsor is categorized into one of the categories listed below. Public plans include state, county and city plans. Unions include single and multi-employer unions and Taft-Hartley plans. The “miscellaneous” category includes anonymous corporate plans, insurance and trusts. The size of the mandate in the hiring decision and the size of the plan sponsor are in millions of dollars as of year of the hiring decision. Panel B shows the distribution of hiring decisions classified by whether the plan sponsor employed a consultant in the decision. Panel C shows the distribution of hiring decisions in each calendar year. Number of Hirings Plan Sponsor Size ($M) Mean Median N Mandate Size ($M) Mean Median N Panel A: Distribution by type of plan sponsor Corporate Public Endowments & Foundations Unions Miscellaneous All 1,978 3,735 1,739 919 843 9,214 1,684 10,955 503 1,132 12,235 6,276 309 1,300 175 250 1,000 460 1,666 3,644 1,422 791 482 8,005 41 120 23 32 156 78 20 40 11 18 40 24 1,709 3,524 1,540 855 530 8,158 Panel B: Use of consultant Consultant employed No Consultant employed 5,907 3,307 5,063 8,910 450 500 5,482 2,523 75 84 25 22 5,426 2,732 Panel C: Time series distribution of hiring decisions 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Number of Hirings 939 907 1,016 968 971 1,056 1,100 758 752 747 Percent 10.1 9.8 11.0 10.5 10.5 11.5 11.9 8.2 8.1 8.1 42 Table 2 Asset Allocations and Hiring Mandates Panel A shows the average allocation (in percent) across the asset classes reported for various types of plan sponsors. Alternative assets include buyout funds, venture capital and hedge funds. Other assets include balanced, GICs, cash, real estate, timber, oil & gas. The number of observations across asset classes are not equal because of data collection procedures; as a result, the sum of allocations is not equal to 100 percent. The panel also shows the percentage of total assets that are indexed and managed internally. Again, the number of observations in each cell vary accordingly to data availability. The allocation index is the average of the allocation to equity (both domestic and international), alternative assets, non-indexed assets and externally-managed assets. For plan sponsors without data on indexation and externally managed assets, the average is computed from the equity and alternative asset allocation. Panel B shows the number of hiring decision across asset classes. Asset Class Domestic Fixed Income International Alternative Others Assets Indexed Internally Allocation Equity Equity Assets Managed Index Panel A: Average asset allocation by plan sponsor type Corporate Public Endowments Unions Misc. 47.5 45.6 49.3 41.6 52.3 26.4 34.7 26.3 38.4 31.7 10.7 11.0 7.5 2.1 5.2 13.0 2.8 8.6 11.1 8.0 8.8 7.0 6.1 11.8 8.0 9.4 21.5 NA 7.7 12.5 3.3 13.1 NA 0.0 18.5 0.63 0.49 0.34 0.44 0.45 Panel B: Number of hiring decisions in asset class by type of plan sponsor Corporate Public Endowments Unions Misc. 1,098 1,746 1,022 566 423 372 809 338 203 246 353 763 247 30 128 NA NA NA NA NA 130 367 110 102 32 - 43 Table 3 Excess Returns of Investment Managers Before and After Hiring Panel A presents average cumulative excess returns computed by summing quarterly excess returns (raw minus benchmark return). Panel B shows excess buy and hold returns computed by subtracting compounded benchmark returns from raw returns. Information on benchmarks is provided in Table A1. Heteroskedasticity, serial and cross-correlation consistent standard errors standard errors are calculated using the procedure described in Jegadeesh and Karceski (2004). Panel C shows information ratios calculated by scaling the average excess return by its standard deviation. Panel D shows estimates of alphas from calendar time regressions factor regressions with standard errors in parentheses. For domestic equity mandates, we use the Fama and French (1993) three-factor model with market, size and book-to-market factors. For fixed income mandates, we employ a three factor model with the Lehman Brothers Aggregate Bond Index return, a term spread (the difference between the long-term government bond return and the T-bill return), and a default spread (the difference between the corporate bond return and the long-term government bond return). For international equity mandates, we use international versions of the domestic equity three factor models. Pre-Hiring Period (years) Hiring Post-Hiring Period (years) -3 to 0 -2 to 0 -1 to 0 Qtr 0 0 to 1 0 to 2 0 to 3 Panel A: Cumulative Excess Returns Full Sample 9.88 7.44 3.55 0.40 0.45 1.18 1.88 (1.79) (1.68) (1.04) (0.25) (0.67) (0.96) (1.30) Domestic Equity 12.06 9.58 4.56 0.52 -0.08 0.14 0.76 (2.72) (2.71) (1.61) (0.34) (0.95) (1.47) (2.12) International Equity 15.59 11.26 5.54 0.62 3.26 6.94 9.83 (3.65) (2.69) (1.39) (0.50) (1.27) (1.77) (2.41) Fixed Income 3.84 2.50 1.11 0.18 0.37 0.69 0.83 (0.27) (0.27) (0.25) (0.11) (0.24) (0.47) (0.64) Panel B: Buy and Hold Excess Returns Full Sample 14.50 9.33 4.06 0.40 0.56 1.04 1.24 (2.75) (2.27) (1.32) (0.25) (0.79) (1.06) (1.58) Domestic Equity 18.93 12.65 5.36 0.52 0.07 -0.20 -0.40 (4.25) (3.85) (2.09) (0.34) (1.14) (1.75) (2.84) International Equity 20.86 12.88 6.02 0.62 3.53 7.80 11.73 (4.78) (3.09) (1.58) (0.50) (1.41) (2.14) (3.44) Fixed Income 4.93 2.90 1.20 0.18 0.40 0.82 1.04 (0.37) (0.33) (0.27) (0.11) (0.26) (0.53) (0.84) Panel C: Information Ratios Full Sample 3.63 2.68 1.61 0.43 0.76 0.95 Domestic Equity 2.99 2.40 1.41 -0.07 0.13 0.19 International Equity 4.11 3.29 2.05 1.41 2.40 3.17 Fixed Income 5.69 3.80 2.36 1.45 1.88 1.99 Panel D: Calendar-Time Alphas from Factor Regressions Domestic Equity 1.10 1.10 1.10 -0.15 -0.16 -0.11 (0.25) (0.29) (0.35) (0.16) (0.16) (0.18) International Equity 1.45 1.49 1.27 0.77 0.62 0.60 (0.44) (0.51) (0.56) (0.33) (0.33) (0.31) Fixed Income -0.26 -0.15 -0.02 -0.02 -0.02 -0.03 (0.19) (0.19) (0.23) (0.38) (0.37) (0.37) 44 Table 4 Distribution of Excess Returns of Investment Managers Before and After Hiring Cumulative excess returns are computed by summing quarterly excess returns over one, two and three year intervals. Excess buy and hold returns are computed by subtracting compounded benchmark returns from raw returns over the same interval. The allocation index is the average of the allocation to equity (both domestic and international), alternative assets, non-indexed assets and externally-managed assets. For plan sponsors without data on indexation and externally managed assets, the average is computed from the equity and alternative asset allocation. The 25th and 75th percentiles are used as cutoffs to place plan sponsors in small, medium (not shown) and large categories. Pre-Hiring Period Post-Hiring Period -3 to 0 -2 to 0 -1 to 0 0 to 1 0 to 2 0 to 3 Panel A: Cumulative Excess Returns Corporate Plans 9.77 7.31 3.60 0.42 1.28 1.89 (1.73) (1.61) (0.90) (0.77) (1.06) (1.66) Public Plans 10.34 7.33 3.40 0.90 1.70 2.48 (2.08) (1.72) (1.12) (0.73) (0.99) (1.26) Endowments 10.47 8.29 4.05 -0.02 0.54 1.56 (2.06) (2.07) (1.13) (0.72) (1.13) (1.54) Unions 7.14 6.06 2.61 -0.23 0.73 1.25 (1.27) (1.60) (0.84) (0.63) (1.29) (2.01) Miscellaneous 9.71 7.92 4.08 0.20 0.35 0.61 (2.38) (2.35) (1.43) (0.62) (0.95) (1.20) Sponsor Size: small 8.38 6.40 3.06 -0.06 0.09 -0.31 (1.37) (1.46) (0.93) (0.64) (1.10) (1.75) Sponsor Size: large 10.81 7.71 3.62 0.62 2.15 3.27 (2.15) (1.88) (1.21) (0.70) (1.09) (1.34) Allocation Index: low 10.48 7.69 3.54 0.22 0.89 1.69 (1.69) (1.66) (1.00) (0.63) (1.00) (1.50) Allocation Index: high 11.76 9.16 4.10 1.36 2.66 4.02 (2.07) (2.09) (1.26) (0.95) (1.20) (1.51) Panel B: Excess Buy and Hold Returns Corporate Plans 14.48 9.20 4.12 0.49 0.91 0.75 (2.51) (1.90) (1.10) (0.89) (1.26) (2.36) Public Plans 14.86 8.95 3.81 1.06 1.76 2.10 (3.16) (2.34) (1.40) (0.84) (1.03) (1.16) Endowments 15.83 10.78 4.73 0.09 0.35 1.09 (3.52) (3.01) (1.51) (0.84) (1.24) (2.14) Unions 10.70 7.68 3.05 -0.29 0.25 -0.05 (1.53) (1.92) (1.03) (0.75) (1.54) (2.83) Miscellaneous 14.02 10.02 4.77 0.35 0.36 0.40 (3.65) (3.26) (1.85) (0.75) (1.04) (1.23) Sponsor Size: small 12.12 7.86 3.54 -0.01 -0.58 -2.13 (1.97) (1.80) (1.20) (0.79) (1.35) (2.70) Sponsor Size: large 15.28 9.28 4.09 0.78 2.52 2.98 (3.01) (2.38) (1.54) (0.78) (1.34) (1.18) Allocation Index: low 15.49 9.65 4.10 0.32 0.73 1.24 (2.51) (2.22) (1.27) (0.73) (1.01) (1.88) Allocation Index: high 17.48 11.50 4.65 1.52 2.80 3.85 (3.28) (2.88) (1.57) (1.10) (1.30) (1.54) 45 Table 5 Maximum-likelihood Post-Hiring Selectivity-Corrected Return Regressions The return regression is: y j = β x j + δ z j + ε j where yj represents three year post-hiring returns, xj is a vector of explanatory variables, and zj is a dummy variable for whether a consultant was employed. The explanatory variables ⎧ 1, if z * > 0 * j , and wj is a are computed as in earlier tables. The selection equation is z j = γ w j + u j where z j = ⎨ ⎩0, otherwise vector of explanatory variables. The headline resistant indicator is equal to one for banks, corporate plans, health and hospital plans, hybrid plans, insurance plans, and nuclear decommissioning trusts, zero otherwise. The headline sensitive indicator is equal to one for public plans, single and multi-employer union plans, zero otherwise. Headline neutral plans, which include endowments, foundations, annuity funds, trust funds and anonymous plans, are captured by the intercept. The results for only one selection equation are shown. Standard errors (in parentheses) account for clustering in observations where the investment manager is hired for a mandate in the same style and period by different plan sponsors. The expected value of a consultant is computed as E ( yi zi = 1) − E ( yi z i = 0) . Constant Log (Plan Sponsor Size) Headline Resistant Indicator Headline Sensitive Indicator Broad Style Indicator Detailed Style Indicators Pre-Hiring Return Log (Mandate Size) Corporate Plan Indicator Public Plan Indicator Endowment Indicator Union Indicator Other Plan Indicator Corporate*Plan Sponsor Size Public *Plan Sponsor Size Endowment*Plan Sponsor Size Union*Plan Sponsor Size Other*Plan Sponsor Size Corporate*Allocation Index Public*Allocation Index Endowment*Allocation Index Union*Allocation Index Other*Allocation Index Corporate*Size*Index Public*Size*Index Endowment*Size*Index Union*Size*Index Other*Size*Index Expected value of consultant Number of Observations Selection Eqn. 0.91 (0.15) -0.05 (0.01) -0.13 (0.05) 0.39 (0.05) Yes No - Cumulative Excess Return No No Yes Yes -0.3 (0.01) -0.3 (0.02) -0.7 (0.23) -0.4 (0.18) -5.7 (3.5) -3.6 (7.8) -2.8 (3.8) 27.0 (4.3) -1.6 (3.7) 8.4 (10.3) -8.2 (4.5) 8.6 (10.7) -2.3 (4.3) -3.9 (30.3) 1.5 (0.4) 3.6 (1.3) 0.8 (0.2) -1.2 (0.5) 0.7 (0.3) 1.7 (1.5) 1.9 (0.5) 1.9 (1.8) 0.8 (0.4) 1.9 (4.7) 35.2 (12.2) -16.1 (7.8) 50.7 (28.0) 10.7 (29.8) 41.2 (71.2) -5.5 (2.1) 2.9 (1.0) -8.1 (4.0) -0.8 (4.6) -4.7 (10.4) 1.57 5,045 46 Buy and Hold Returns No No Yes Yes -0.1 (0.01) -0.12 (0.01) -1.2 (0.3) -0.5 (0.4) -10.1 (4.0) -11.6 (10.4) -1.5 (4.1) 41.9 (5.9) -2.7 (4.2) 5.8 (13.1) -9.3 (5.2) 18.1 (14.2) 2.4 (5.2) 29.4 (39.1) 2.6 (0.4) 6.3 (1.7) 0.9 (0.2) -1.91 (0.7) 1.4 (0.5) 3.3 (1.9) 2.5 (0.6) 2.5 (2.4) 0.5 (0.6) -1.5 (6.1) 57.9 (16.0) -23.0 (10.4) 105.7 (35.2) 6.2 (39.5) 6.4 (92.0) -9.2 (2.7) 3.8 (1.4) -15.8 (5.0) 0.7 (6.1) -0.4 (13.5) 2.7 5,045 3.8 2,986 1.33 2,986 Table 6 Descriptive Statistics for Firing Decisions Panel A shows the distribution of firing decisions for each calendar year. Panel B shows the distribution of firing decisions by the investment style of the mandate for which the firing took place. Style definitions are the same as those in Table 2. Panel C shows the distribution of firing decisions by the stated reason for the firing. The “organizational” category includes firing of investment managers because of personnel turnover, regulatory action and acquisitions. The “reallocation” category refers to firings because the plan sponsor has decided to move away from the asset allocation / investment style offered by the investment manager. Panel D shows the distribution of firing decisions for each plan sponsor category. Number of Firings Percent Panel A: Distribution of firing decisions by calendar year 1996 1997 1998 1999 2000 2001 2002 2003 21 45 20 54 138 102 228 302 Panel B: Distribution of firing decisions by investment style 2.3 4.9 2.2 5.9 15.2 11.2 25.0 33.2 65.1 22.6 8.5 3.7 All Domestic Equity 593 Fixed Income 206 International 77 Others 34 Panel C: Distribution of firing decisions by stated reason Organizational 122 13.4 Performance 292 32.1 Reallocation 102 12.0 Not Available 394 43.3 Panel D: Distribution of firing decisions by type of plan sponsor Corporate 94 10.3 Public 450 49.5 Endowments & Foundations 83 9.1 Unions 69 7.6 Miscellaneous 213 23.4 Panel E: Descriptive statistics of plan and mandate size Plan Size Mandate Size Mean 7,988 145 Median 735 50 N 712 751 47 Table 7 Excess Returns of Investment Managers Before and After Firing Panel A presents average cumulative excess returns computed by summing quarterly excess returns. Panel B shows excess buy and hold returns computed by subtracting compounding benchmark returns from raw returns. Information on benchmarks is provided in Table A1. Heteroskedasticity, serial and cross-correlation consistent standard errors standard errors are calculated using the procedure described in Jegadeesh and Karceski (2004). Panel C shows information ratios calculated as the average excess return scaled by the standard deviation of the excess return. Panel D shows estimates of alphas from calendar time regressions factor regressions with standard errors in parentheses. For domestic equity mandates, we use the Fama and French (1993) three-factor model with market, size and book-to-market factors. For fixed income mandates, we employ a three factor model with the Lehman Brothers Aggregate Bond Index return, a term spread computed as the difference between the long-term government bond return and the T-bill return, and a default spread computed as the difference between the corporate bond return and the long-term government bond return. For international equity mandates, we use an international version of the domestic equity three factor model. Pre-Hiring Period Hiring Post-Hiring Period -3 to 0 -2 to 0 -1 to 0 Qtr 0 0 to 1 0 to 2 0 to 3 Panel A: Cumulative Excess Returns Full Sample 0.61 -1.62 -1.00 0.32 0.77 1.91 4.34 (1.34) (1.50) (0.75) (0.22) (0.75) (1.37) (1.33) Domestic Equity 0.00 -2.84 -1.58 0.29 0.52 1.89 4.87 (2.20) (1.83) (0.85) (0.30) (1.03) (1.80) (2.33) International Equity 8.63 4.14 2.25 0.99 1.29 2.94 7.32 (0.68) (2.07) (1.72) (0.42) (1.37) (3.49) (2.47) Fixed Income -0.45 -1.17 -0.89 0.13 0.86 0.82 0.12 (0.47) (1.39) (0.69) (0.19) (0.64) (0.98) (0.81) Panel B: Excess Buy and Hold Returns Full Sample -0.17 -1.77 -1.08 0.32 0.76 1.32 3.90 (1.98) (1.27) (0.78) (0.22) (0.81) (1.40) (1.27) Domestic Equity -1.52 -3.11 -1.72 0.29 0.50 1.04 4.06 (3.24) (1.38) (0.88) (0.30) (1.13) (1.99) (2.63) International Equity 11.32 4.98 2.44 0.99 1.13 1.69 7.48 (1.11) (2.37) (1.80) (0.42) (1.39) (3.43) (2.65) Fixed Income -0.50 -1.23 -0.89 0.13 0.93 0.93 0.20 (0.52) (1.61) (0.71) (0.19) (0.68) (1.12) (0.97) Panel C: Information Ratios Full Sample 0.07 -0.15 -0.13 0.69 1.19 1.69 Domestic Equity -0.58 -0.29 -0.29 0.08 0.67 1.58 International Equity 2.38 1.65 0.99 0.47 1.64 3.03 Fixed Income 0.83 -0.59 -0.14 2.28 2.02 0.80 Panel D: Calendar-Time Alphas from Factor Regressions Domestic Equity -0.01 -0.48 -0.72 0.47 0.35 0.42 (0.24) (0.17) (0.19) (0.46) (0.38) (0.39) International Equity -0.54 -0.36 -0.02 0.23 0.23 0.11 (0.36) (0.27) (0.22) (0.39) (0.44) (0.21) Fixed Income 0.43 0.05 -0.62 0.92 0.10 0.11 (0.27) (0.30) (0.63) (0.23) (0.23) (0.21) 48 Table 8 Distribution of Excess Returns of Investment Managers Before and After Firing The allocation index is the average of the allocation to equity (both domestic and international), alternative assets, non-indexed assets and externally-managed assets. The 25th and 75th percentiles are used as cutoffs to place plan sponsors in small and large categories. The “organizational” category includes firing of investment managers because of personnel turnover, regulatory action and acquisitions. The “reallocation” category refers to firings because the plan sponsor has decided to move away from the asset allocation / investment style offered by the investment manager. Firing decisions labeled “performance” specifically identify poor performance as the stated reason for termination. Pre-Hiring Period Post-Hiring Period -3 to 0 -2 to 0 -1 to 0 0 to 1 0 to 2 0 to 3 Panel A: Cumulative Excess Returns Organizational Reason 5.41 -0.21 0.75 2.40 5.33 7.82 (4.09) (2.94) (1.12) (0.50) (1.01) (3.06) Performance Reason -2.79 -4.57 -3.01 0.35 1.54 5.18 (2.10) (1.97) (1.23) (1.20) (2.22) (2.02) Reallocation Reason -1.58 -2.04 -0.15 0.68 1.35 4.77 (0.81) (0.97) (0.60) (1.56) (1.69) (1.66) Not Available Reason 3.11 0.35 -0.28 0.64 1.52 2.76 (1.80) (1.45) (0.87) (0.79) (1.04) (1.37) [-3, 0] Negative Return [-2, 0] Negative Return [-1, 0] Negative Return Corporate Plans Public Plans Endowments Unions Other Plans Sponsor Size (small) Sponsor Size (large) Allocation Index (small) Allocation Index (large) -9.32 (1.69) -8.40 (1.52) -5.48 (1.75) 9.29 (3.84) 3.10 (1.27) -7.36 (3.10) 5.49 (1.46) -2.09 (1.14) -1.31 (1.54) 12.54 (1.52) -1.73 (1.68) 1.59 (4.49) -10.69 (2.73) -11.54 (3.06) -9.73 (3.35) 3.35 (2.75) -2.40 (3.24) -5.15 (1.60) 4.10 (1.18) -1.75 (1.57) -1.19 (1.44) -1.12 (5.46) -3.48 (1.15) -0.76 (2.66) 49 -4.23 (0.89) -4.78 (0.61) -5.96 (0.51) -1.42 (1.81) -0.81 (1.35) -1.51 (1.01) 0.91 (0.96) -1.64 (0.65) -1.65 (0.67) -0.73 (1.69) -0.85 (1.01) -1.28 (1.09) 0.87 (1.54) 0.72 (1.43) 0.59 (1.27) -0.64 (2.38) 1.00 (0.51) 0.46 (2.67) 0.48 (0.89) 1.15 (1.27) 0.70 (1.31) 0.43 (0.67) 1.27 (1.21) 0.57 (0.95) 3.06 (3.07) 2.48 (2.69) 2.44 (2.17) 4.48 (1.82) 0.66 (1.15) 7.12 (2.92) -0.22 (1.32) 2.06 (2.20) 2.58 (2.44) -1.20 (0.48) 2.22 (1.42) 2.98 (2.01) 6.17 (2.80) 5.93 (2.13) 6.25 (2.54) 7.53 (2.86) 2.03 (0.62) 14.83 (4.09) -0.72 (2.51) 4.93 (2.16) 5.83 (2.63) -1.81 (1.32) 4.32 (2.78) 5.24 (4.30) Organizational Reason Performance Reason Reallocation Reason Not Available Reason [-3, 0] Negative Return [-2, 0] Negative Return [-1, 0] Negative Return Corporate Plans Public Plans Endowments Unions Other Plans Sponsor Size (small) Sponsor Size (large) Allocation Index (small) Allocation Index (large) Panel B: Excess Buy and Hold Returns 7.50 1.22 0.79 2.43 (5.88) (3.03) (1.13) (0.68) -4.90 -4.60 -3.03 0.17 (2.84) (1.37) (1.17) (1.24) -2.46 -2.67 -0.30 0.91 (1.05) (0.98) (0.64) (1.84) 2.78 -0.11 -0.41 0.69 (2.42) (1.51) (1.00) (0.83) -13.92 (2.23) -12.62 (1.77) -9.88 (2.66) 12.35 (6.04) 2.92 (1.71) -7.84 (5.04) 7.27 (1.61) -4.32 (1.72) -3.03 (2.41) 13.92 (1.64) -2.70 (2.14) 1.05 (5.94) -11.85 (1.96) -11.46 (1.31) -9.78 (1.87) 4.03 (2.80) -2.00 (2.82) -5.53 (1.62) 4.88 (1.26) -2.93 (1.91) -2.20 (1.79) -1.42 (5.39) -3.52 (0.87) -0.68 (2.76) -4.74 (0.86) -4.97 (0.57) -6.00 (0.48) -1.28 (1.70) -0.80 (1.43) -2.01 (0.88) 1.00 (0.85) -1.86 (0.66) -1.81 (0.66) -1.06 (1.91) -0.79 (1.00) -1.42 (1.10) 0.91 (1.61) 0.85 (1.49) 0.52 (1.44) -0.82 (2.65) 1.05 (0.50) -0.02 (2.72) 0.54 (0.94) 1.26 (1.55) 0.71 (1.55) 0.45 (0.87) 1.20 (1.22) 0.83 (1.14) 5.08 (1.43) 0.76 (2.26) 0.64 (1.71) 1.04 (1.23) 2.24 (2.97) 1.62 (3.01) 1.72 (2.22) 4.66 (1.59) 0.18 (1.05) 6.84 (3.04) 0.28 (1.16) 0.88 (2.68) 1.38 (2.96) -2.15 (0.43) 1.59 (1.43) 3.00 (1.77) 8.00 (3.71) 4.61 (1.96) 5.17 (2.09) 1.92 (1.84) 5.74 (2.75) 5.18 (2.49) 5.12 (2.48) 9.29 (1.78) 1.85 (0.60) 13.24 (3.61) -0.85 (2.66) 4.10 (2.57) 5.04 (3.10) -3.50 (0.78) 3.11 (3.85) 5.67 (3.37) 50 Table 9 Maximum-likelihood Post-Firing Selectivity-Corrected Return Regressions The return regression is: y j = β x j + δ z j + ε j where yj represents three year post-firing returns, xj is a vector of explanatory variables, and zj is a dummy variable for whether a consultant was employed. The explanatory variables are computed as in earlier tables. The “organizational” category includes firing of investment managers because of personnel turnover, regulatory action and acquisitions. The “reallocation” category refers to firings because the plan sponsor has decided to move away from the asset allocation / investment style offered by the investment manager. Firing decisions labeled “performance” specifically identify poor performance as the stated reason for termination. ⎧ 1, if z * > 0 j The selection equation is z * = γ w j + u j where z j = ⎨ , and wj is a vector of explanatory variables. The j ⎩0, otherwise headline resistant indicator is equal to one for banks, corporate plans, health and hospital plans, hybrid plans, insurance plans, and nuclear decommissioning trusts, zero otherwise. The headline sensitive indicator is equal to one for public plans, single and multi-employer union plans, zero otherwise. Headline neutral plans, which include endowments, foundations, annuity funds, trust funds and anonymous plans, are captured by the intercept. The results for only one selection equation are shown. Standard errors (in parentheses) account for clustering in observations where the investment manager is hired for a mandate in the same style and period by different plan sponsors. The expected value of a consultant is computed as E ( yi zi = 1) − E ( yi zi = 0) . Selection Eqn. 0.50 (0.35) -0.01 (0.03) -0.25 (0.20) 0.54 (0.18) Yes No Cumulative Excess Return No No Yes Yes -0.3 (0.06) -0.2 (0.05) -0.3 (0.7) -0.5 (1.3) 11.7 (9.3) 44.7 (18.7) 4.3 (9.6) 42.1 (15.7) 15.6 (8.1) -30.3 (18.9) 3.7 (9.5) -1.2 (15.3) 9.2 (8.5) 0.0 (2.0) -0.9 (0.9) 8.3 (2.7) -1.1 (2.0) 3.8 (3.4) 3.9 (3.2) -0.5 (2.3) -0.2 (2.2) 0.4 (2.7) -0.2 (2.7) 2.0 270 -2.9 174 Buy and Hold Returns No No Yes Yes -0.2 (0.05) -0.2 (0.03) -0.4 (0.9) -0.6 (1.3) 10.5 (10.9) 8.4 (14.9) 1.2 (11.1) 5.4 (12.5) 12.1 (9.5) -30.1 (18.3) 1.4 (11.5) -1.0 (15.3) 6.5 (9.9) 0.4 (1.9) -0.8 (0.9) 7.1 (2.6) -0.1 (1.9) 4.9 (4.0) 4.9 (3.1) 0.2 (2.7) 0.7 (2.1) 1.8 (3.2) 0.9 (2.6) 2.7 270 3.8 174 Constant Log (Plan Sponsor Size) Headline Resistant Indicator Headline Sensitive Indicator Broad Style Indicator Detailed Style Indicators Pre-Hiring Return Log (Mandate Size) Corporate Plan Indicator Public Plan Indicator Endowment Indicator Union Indicator Other Plan Indicator Corporate*Plan Sponsor Size Public *Plan Sponsor Size Endowment*Plan Sponsor Size Union*Plan Sponsor Size Other*Plan Sponsor Size Organizational Indicator Performance Indicator Reallocation Indicator Expected value of consultant Number of Observations 51 Table 10 Buy and Hold Returns of Investment Managers Before and After Firing and Hiring Panel A presents average cumulative excess returns computed by summing quarterly excess returns over given intervals. Panel B shows excess buy and hold returns computed by compounding raw returns over appropriate intervals and subtracting benchmark returns compounded over the same interval. Information on benchmarks is provided in appendix A. Heteroskedasticity and serial correlation consistent standard errors (HSC_V) standard errors are calculated using the procedure described in Jegadeesh and Karceski (2004) and appear in parentheses. Pre-Event Period Event Post-Event Period (years) Quarter (years) -3 to 0 -2 to 0 -1 to 0 Qtr 0 0 to 1 0 to 2 0 to 3 Panel A: Cumulative Excess Returns Fired Firms 1.42 -0.76 -0.25 0.47 0.92 2.29 4.40 (1.19) (1.31) (0.64) (0.19) (0.74) (1.56) (1.99) Hired Firms Return Diff. (Hired-Fired) N Fired Firms Hired Firms Return Diff. (Hired-Fired) 10.70 (1.79) 9.28 (1.25) 175 8.59 (1.76) 9.36 (1.78) 284 4.52 (1.47) 4.77 (0.95) 409 0.40 (0.25) -0.07 (0.27) 440 0.93 (0.64) 0.01 (0.98) 409 1.91 (0.97) -0.38 (1.90) 284 1.91 (1.43) 1.75 (1.18) -0.16 (2.06) 2.21 (1.55) -2.20 (2.23) 175 3.80 (2.15) 1.50 (1.69) -2.30 (2.77) Panel B: Buy and Hold Excess Returns 0.82 -0.67 0.25 0.47 1.01 (1.86) (1.07) (0.6) (0.19) (0.80) 16.07 (3.65) 15.25 (3.02) 10.09 (2.54) 10.75 (2.12) 5.03 (1.94) 5.58 (1.35) 0.40 (0.25) -0.07 (0.27) 0.99 (0.76) -0.01 (1.15) 52 Table A1 Investment Mandates and Indices Investment Mandate Description Domestic Equity Largecap Largecapcore Largecapgrowth Largecapindex Largecapvalue Midcap Midcapcore Midcapgrowth Midcapindex Midcapvalue Smallcap Smallcapcore Smallcapgrowth Smallcapindex Smallcapmicro Smallcapvalue Smid Smidcapcore Smidcapgrowth Smidcapvalue Equitygrowth Equityvalue Equitycombined International equity Emergmkteq Europeincuk Europeincuksm Globaleq Intleq Intleqsmall Intlpassive Pacbasinincj Fixed income Convertibles Fixed1-3yrs Fixedcore Fixedcoreinvest Large-cap equity Large-cap – between growth & value Large-cap – growth Large-cap – indexed Large-cap – value Mid-cap equity Mid-cap – between growth and value Mid-cap – growth Mid-cap – indexed Mid-cap – value Small-cap equity Small-cap – between growth and value Small-cap – growth Small-cap – indexed Small-cap – value Small-cap equity Small to mid-cap equity Small to mid-cap – between growth and value Small to mid-cap – growth Small to mid-cap – indexed All equity - growth All equity – value All equity Index S&P 500 S&P 500 S&P 500/BARRA Growth S&P 500 S&P 500/BARRA Value S&P Midcap 400 S&P Midcap 400 S&P/BARRA Mid Cap Growth S&P Midcap 400 S&P/BARRA Mid Cap Value S&P Small Cap 600 S&P Small Cap 600 S&P/BARRA Small Cap Growth S&P Small Cap 600 S&P Small Cap 600 S&P/BARRA Small Cap Value Russell 2500 Russell 2500 Russell 2500 Growth Russell 2500 Value Russell 3000 Growth Russell 3000 Value Russell 3000 Emerging market equity Europe incl. U.K. Europe incl. U.K. – small-cap Global equity (incl. U.S.) International equity International equity – small-cap International equity – passive Pacific basin incl. Japan Convertibles Duration between 1 and 3 years Inv. and non-inv. grade, duration 3-7 years Inv. grade, duration 3-7 years 53 MSCI Emerging Mkts Free MSCI Europe 15 MSCI Europe S/C MSCI World Free MSCI EAFE Free MSCI EAFE S/C MSCI EAFE MSCI AC Pacific Free Merrill Lynch Inv Grade Convertible Merrill Lynch Govt/Corp 1-3 Years Lehman Aggregate Lehman Aggregate Fixedcoreopportun Fixedhighyield Shortterm Fixedintermed Fixedlongdura Mortgageb Fixedcombined Emergmktdebt Globalfixhedg Globalfixunhedg Intlfixedhedg Intlfixedunhedg Others Realestate Realestateselect Reits Taa Balanced Non-inv. grade, duration 3-7 years High yield securities Duration between 1 and 2.4 years Duration between 2 and 4.6 years Duration greater than 6 years Mortgage-backed securities All fixed income Emerging market debt Global fixed income - hedged Global fixed income - unhedged International fixed income – hedged International fixed income - unhedged Real estate Real estate select Reits Tactical asset allocation Balanced Lehman Aggregate Lehman High Yield Composite Citigroup 3-Month T-Bill Lehman Int. Aggregate Lehman Long Govt/Credit Lehman Mortgages Lehman Aggregate JP Morgan ELMI+ Lehman Global Aggregate (Hedged) Lehman Global Aggregate (Unhedged) Citigroup Non-US WGBI (Hedged) Citigroup Non-US WGBI (Unhedged) NCREIF Property NCREIF Property NAREIT S&P 500 S&P 500 54 Appendix A: Standard Error Calculation The sample comprises N hiring/firing decisions of investment managers by plan sponsors (“events”). We wish to test whether the managers exhibit excess return performance from the event date through a H-quarter holding period. We define the H-quarter excess return (ER) for investment manager i that starts at the beginning of the event quarter t as either the cumulative excess return or buy and hold excess return. ⎧ t + H −1 ∑ (Ri,s − Rb,s ) ⎪ ⎪ ERi (t , H ) = ⎨t + H −1 s =t or ⎪ ∏ (1 + Ri ,s − Rb,s ) − 1 ⎪ s =t ⎩ where Ri,s is the return on the mandate type by the investment manager i in quarter s, and Rb,s is the return on the benchmark b in quarter s. Define: 1 N ER sample ( H ) = ∑ ERi (t , H ) N i =1 Let Nt equal the number of events in the sample in qurter t, and let N be the total number T of events in the sample. Therefore N = ∑t =1 N t . Define the average abnormal return for each event quarter t across all events in that quarter (we refer to this group of events as a quarterly cohort) as: ⎧ 1 Nt ER (t , H ) , if N t > 0 ⎪ ER (t , H ) = ⎨ N t ∑ i i =1 ⎪0 otherwise ⎩ Let ER ( H ) be a T × 1 column vector where the tth element equals ER(t , H ) . ER( H ) is the average long-run excess return of each quarterly cohort. Define w as a T × 1 column vector of weights where the tth element is the ratio of the number of events that occur in quarter t divided by N. Specifically, w(t ) = N t N . Note that the sample average excess return is equal to the quarterly weight vector w times the average excess return of each quarterly cohort: ER sample ( H ) = w' ER( H ) The variance of ER sample ( H ) is given by: σ 2 ER sample ( H ) = w' Vw where V is the T × T variance covariance matrix of ER( H ) . Our estimator for V allows for heteroskedasticity as well as serial correlation and is denoted as HSC. The stth element of HSC_V is ⎧ (H − l ) ⎪ ER ( s, H ) ER (t , H ), if l = s − t < H hscst = ⎨ l ⎪ 0 otherwise ⎩ This estimator uses Newey and West (1987) weighting scheme that ensures that HSC_V is positive definite. ( ) 55