RELATIONAL INVESTING AND FIRM PERFORMANCE Sanjai Bhagat University of Colorado at Boulder Bernard Black Stanford University Margaret Blair Georgetown University Law School July 2001 We thank seminar participants at Northwestern University and the U.S. Department of Justice for comments on a previous draft of this paper. We gratefully acknowledge financial support of the Alfred P. Sloan Foundation, AIMR, TIAA-CREF, and the Institutional Investor Project of Columbia University. Please address correspondence to any of the co-authors: Sanjai Bhagat, Graduate School of Business, University of Colorado, Boulder, CO 80309-0419. Tel: (303) 492-7821. Email: sanjai.bhagat@colorado.edu Bernard Black, Stanford Law School, Email: bblack@stanford.edu) Margaret Blair, Georgetown University Law School, 600 New Jersey Avenue, NW, Washington, DC 20001;(202) 662-9000 Email: blairm@law.georgetown.edu
Relational Investing And Firm Performance Abstract A substantial academic and popular literature argues that the performance of American corporations might improve if American corporations had long-term outside investors ("relational investors") who held large stakes, actively monitored management performance, and engaged with management in setting corporate policy. Institutional investors could perhaps play this role. We provide the first large-scale test of the hypothesis that relational investing could affect corporate performance. We consider ownership and performance data for more than 1500 large U.S. companies over a 13-year (1983-1995). We document a significant secular increase in large-block shareholding over the of our study, due to increased blockholdings by investment companies, partnerships, investment advisors, and employee benefit plans. However, the overwhelming majority of blockholdings by institutional investors are sold too quickly to qualify as relational investing. Our results provide a mixed answer to the question of whether relational investing affects corporate performance. Our data suggest that there was a in the late 1980s when the presence of a relational investor was associated with higher stock market returns. This cohort of relational investors may have been able to induce corporate restructuring, principally to reduce growth rates while improving profitability. But this pattern was not found in the early 1980s, or repeated in the early 1990s.
Relational Investing And Firm Performance In the last decade, a substantial academic and popular literature has argued that American corporations focus too much on near-term profitability, and that their long-term performance might improve if they had long-term investors ("relational investors") who held large stakes, actively monitored management performance, and engaged with management in helping to set corporate policy. The American prototype is Warren Buffett's holding company, Berkshire Hathaway. The foreign prototype is the lead bank that both lends to a company and owns a large block of its shares, commonly found in Germany and Japan. Large institutional investors, which often hold sizable, relatively illiquid stakes in their portfolio companies, could play this role in the United States. More might do so if legal impediments that currently hinder large blockholdings and institutional activism were removed. (See, e.g., Black, 1990, 1992a, 1992b; Roe, 1994; Coffee, 1991; Jacobs, 1991; Porter, 1992; Symposium, 1998, Twentieth Century Fund, 1992.) At the same time that Americans worry about the absence of strong investors, Europeans worry about their presence. A recent report to the European Commission concludes that "strong blockholders" are a major impediment to good corporate governance of European companies (European Corporate Governance Network, 1997)). However, empirical evidence on the connection between relational investing and corporate performance -- whether positive or negative -- is scarce. Moreover, the concept of relational investing has not been carefully defined -- its proponents have not specified how large a block is large enough to be significant, or how long is "long term," or the nature of the dialogue between the relational investor and corporate management. In this paper, we propose operational definitions of the concept of relational investing, and conduct the first large-scale test of the hypothesis that relational investing can improve the performance of American firms. We collect ownership and performance data on more than 1500 of the largest U.S. companies, over a 13-year (1983-1995). We describe the patterns of long-term, large-block shareholding among large publicly-traded companies. We document a significant secular increase in large-block shareholding over the of our study, with sharp percentage increases in holdings by mutual funds, partnerships, investment advisors, and employee benefit plans. However, most institutional investors, when they purchase large blocks, sell the blocks relatively quickly -- too quickly to be considered relational investors. Our results provide a mixed answer to the question of whether relational investing affects corporate performance.
Our data suggest that the cohort of relational investors (defined generally as outside shareholders who hold a 10 percent stake for at least 4 years) who held their positions during 1987-90 often targeted firms that had been growing rapidly during the previous 4-year. During the 1987-1990, firms with relational investors outperformed their peers using stock price returns and Tobin's q as performance measures. This is consistent with these having helped their target companies to translate strong growth in the prior (1983-1986) into strong earnings and rising stock prices. But this pattern was not found in the early 1980s, or repeated in the early 1990s. Thus, our data suggest that there may have been a cohort of relational investors who identified a successful investment strategy, or were able to encourage restructuring that improved the performance of their target companies. That strategy could have depended on an active market for hostile takeovers and leveraged restructurings -- a market which flourished during the 1987-1990, was less active in the 1983-1986, and all but disappeared in the first half of the 1990's. Our data do not suggest that relational investing gives firms a sustainable competitive advantage in the current environment of few hostile takeovers and equity prices perhaps making leveraged restructurings unattractive. Another conclusion is that the idea of relational investing must be more carefully specified and clarified in theory. Although our findings are discouraging for a simple-minded theory that large-block shareholders are better monitors, and therefore induce better performance, they leave open the possibility that some kinds of investors might have more effect than others. Ownership of a large block of shares by an officer or director might have a different effect than ownership of a similarly large block by a pension fund or mutual fund. And ownership by an ESOP might have yet a different effect. Quiet, steady ownership may have a different impact on performance than noisy, activist ownership. Although these questions are beyond the scope of this paper, we incorporate their intuition in our work. The remainder of the paper is organized thus. The next section discusses why relational investing might matter. The following section summarizes the extant empirical evidence on large blockholdings and corporate performance. The third section describes the sample and data collection procedure. The fourth section highlights the intertemporal and cross-sectional characteristics of relational investors. Section five includes a discussion of survivorship bias as it pertains to this study. Section six discusses some of the potential problems in measuring the impact of relational investors on firm performance. Sections seven and eight detail the relation between relational investors and stock-market, and accounting measures of performance, respectively. The final section offers some interpretation of our findings and suggests an 4
agenda for future research. I. Why Relational Investing Might Matter? American public corporations have long been characterized by a relative absence of influential shareholders, who hold large blocks of a company's stock for a long of time and actively monitor its performance (sometimes called "relational investors"). The resulting separation of ownership and control has formed the dominant paradigm for understanding our corporate governance system for most of this century (Berle and Means, 1932; see Jensen and Meckling 1976). But the weak shareholder oversight that is the American norm is not inevitable. Internationally, America is unique in the weakness of even the largest shareholders in its major firms. The absence of such investors in the United States, and the presence of strong bank shareholders in Germany and Japan, is perhaps the single defining difference between the capital markets of these three major economies. Moreover, the weakness of American shareholders may reflect political decisions that kept them small and passive, rather than survival of efficient shareholding patterns in a competitive marketplace (Black, 1990; Roe, 1994). The combination of American exceptionalism in having weak shareholders, and the possible political origins of that exceptionalism, raise important policy questions: Would there be economic benefits from relaxing the legal rules that dis courage institutional investors from holding large blocks and intervening actively when management falters? Or has the United States evolved substitute oversight mechanisms that accomplish much the same job that relational investors accomplish elsewhere? If so, adding relational investing to our current corporate governance system wouldn't significantly affect firm performance. If institutions were invited to become relational investors by more favorable legal rules, would they accept the invitation? One potential advantage of a governance system in which more firms have relational investors derives from concerns that managers and shareholders may focus excessively on short-term profitability, with a resulting cost in longterm performance (e.g., Jacobs, 1991; Porter, 1992). The theoretical basis for this concern can be simply stated: If investors have imperfect information about a company's prospects, they may rely on short-term earnings as the best available signal of those prospects. Managers may also overemphasize short-term results, either to please myopic shareholders, or simply to earn this year's bonus (e.g., Shleifer and Vishny, 1990; Stein, 1989, 1996). Alternatively, 5
managers may invest in poor long-term projects, if they believe that shareholders will reward this behavior with higher short-term stock prices (Bebchuk and Stole, 1993). Large shareholders can invest in monitoring, thus reducing the information asymmetry that drives shareholder and manager myopia in these models. Relational investing could also serve as a substitute for, or complement to, the market for corporate control. In the 1980s, hostile takeovers were an important source of monitoring and discipline of corporate managers (e.g, Jensen, 1988; Mikkelson and Partch, 1997). However, hostile takeovers are highly costly, and are feasible only if there is a large gap between a company's value under current management and its potential value if sold or better managed. Moreover, hostile takeovers are now rare, partly because they too are chilled by legal rules that give managers great discretion to block unwanted takeovers (however; see Comment and Schwert, 1995; and Bhagat and Jefferis, 1998). Relational investors potentially could both provide monitoring in normal times (when a firm is not performing badly enough to warrant a hostile takeover bid), and act as a counterweight to management's incentives to block value-enhancing control changes. At the same time, strong outside shareholders are not an unmitigated blessing. Because they own large stakes, they can overcome the collective action problems that make small shareholders passive, and the information asymmetry that may make small shareholders myopic. But large shareholders can also take advantage of their influence, and the passivity of other shareholders, to extract private benefits from the corporation. For example, a bank that is both a major shareholder and a lender to a company may discourage risk-taking, to protect its position as creditor, or may cause the company to borrow from the bank, when cheaper financing is available elsewhere. Moreover, institutional investors are themselves managed, by agents who face their own agency costs, and may not maximize the value of the institution's stake in a portfolio company (Black, 1992a; Black and Coffee, 1994; Fisch, 1994; Romano, 1993). In light of the risks posed by overly strong shareholders, one of us has previously argued that ownership of moderately large blocks (in the 5-10 percent range) by a half-dozen institutions might produce better governance outcomes than ownership of very large blocks (say 20 percent or more) blocks by one or two major shareholders (Black, 1992a). Hence, any correlation between relational investing and performance could be nonmonotonic: Relational investing might produce benefits up to one ownership level, and costs above that level. Finally, relational investing is only one of a myraid of mechanisms that have evolved to align the 6
interests of managers with that of shareholders: For example, management compensation contracts that emphasize equitysensitive claims; the corporate control market (takeovers, proxy fights); various corporate governance mechanisms such as oversight and monitoring by board members; and finally the discipline of competition in the product market. Thus, from a theoretical perspective, relational investing could be a complement to these monitoring mechanisms and would serve to improve performance. Or, the above monitoring mechanisms, either individually or in comb ination, could be a perfect substitute for relational investing; in this case relational investing would not affect performance. Thus, whether relational investing will improve or degrade corporate performance, or not affect performance strongly one way or another, is uncertain as a theoretical matter, and warrants empirical investigation. II. Prior Empirical Evidence on Relational Investing and Corporate Performance A variety of evidence, some systematic and some anecdotal, has been cited in support of the view that relational investing could improve corporate performance. Some advocates of relational investing draw inferences from descriptions by business historians of the roles that large investors have played in particular companies, such as Pierre DuPont at General Motors, J.P. Morgan and his associates in companies in which they had invested, and, in contemporary times, Warren Buffett at Salomon Brothers (see, for example, Lowenstein, 1991; DeLong, 1991). Kleiman, Nathan and Shulman (1994) report more generally, but still anecdotally, that negotiated large-block investments, some by self-styled "relationship investing" funds, generally predict positive market-adjusted stock price returns, but not when the target obtains the investment as part of a defense to a takeover bid. Direct, quantitative evidence about the impact that large investors have on corporate behavior and performance can be divided into four types: Evidence on the impact of majority shareholdings; evidence on the impact of large blockholdings by corporate insiders; evidence on the impact of large minority-block shareholding by outsiders; and, finally, evidence on the impact of institutional investors. While the third and fourth types are most relevant to the debate over relational investing, most research has focused on the first two categories. We summarize the literature here; for a more detailed survey, see Blair (1994). On majority or control-block holdings: An early study by McEachern (1975) finds weak evidence that 7
firms with a controlling shareholder are more profitable than manager-controlled firms. Salancik and Pfeffer (1980) find that CEO tenure correlates with firm profitability for firms with a controlling shareholder, but not for other firms. Holderness and Sheehan (1988) find that an outsider's purchase of a majority block, without announced plans for a complete takeover, produces a 9.4 percent stock price gain over a 30-day window. However, they find no significant differences in Tobin's q or accounting measures of profitability between majority-owned and diffusely-owned firms. On large blockholdings by corporate insiders: The correlation between inside ownership and profitability remains controverted in the literature. A positive correlation exists up to about 5 percent inside ownership, but there is conflicting evidence on the relationship between performance and insider ownership beyond that point (e.g., Morck, Shleifer and Vishny, 1988, Wruck, 1989, McConnell and Servaes, 1990; Himmelberg, Hubbard and Palia, 1997), and the results are sensitive to whether management ownership is treated as exogenous or endogenous (Palia, 1998). Studies of accounting profitability are variable, but tend to show a positive correlation with managerial ownership (see the survey by Scherer, 1988). Companies with high inside ownership are more likely than manager-controlled companies to agree to a friendly acquisition, and less likely to expand sales at the expense of profits; also, bidders with high inside ownership make fewer conglomerate acquisitions, make better acquisitions generally, and pay lower takeover premiums (see the survey by Black, 1992b). On large minority-block holdings by outsiders: Mikkelson and Ruback (1985) and others find increases in the value of target firms upon the announcement that an investor has taken a large-block position, but most of the positive returns are explained by anticipation of a subsequent takeover of the firm. The gains are reversed for firms that are not subsequently acquired. However, Barclay and Holderness (1992) find a market-adjusted increase in the price of the remaining publicly-traded shares after a transaction in which a large block of shares is acquired at a premium, both for firms that are acquired within one year and for firms that are not acquired, though the increase is smaller for the non-acquired group. Gordon and Pound (1992) study a small sample (18) of "patient capital investments," which they define as transactions "in which an investment partnership purchases a new block of equity and is granted at least one 8
seat on the board." Together, Warren Buffett and Corporate Partners Fund account for about half of their sample. They find that "'patient capital' investing has not produced returns that are statistically different from the S&P 500." Bhagat and Jefferis (1994) investigate targeted share repurchases or greenmail transactions where managers agree to repurchase a block of shares at a premium from a single shareholder or group of shareholders. They find that performance of firms that pay greenmail cannot be distinguished from a control group - before or after the repurchase. Fleming (1993) finds that investors who acquired a large equity stake between 1985 and 1989 in a firm that was not subsequently acquired did little to affect the firm's performance. He finds significant positive returns for the target company's shares during the first two months after the the investor's purchase, but significant negative returns over the subsequent two years. Much of Fleming's sample consists of large block acquisitions by corporate "raiders" and arbitrageurs such as Victor Posner and Ivan Boesky. Bethel, Liebeskind, and Opler (1998) examine purchases of large blocks of stock by activist investors during the 1980s. These purchases were followed by abnormal share price appreciation, an increase in asset divestitures, an increase in operating profitability and a decrease in merger and acquisition activity. On the impact of institutional investors: Jennings, Schnatterly, and Seguin (1997) report that higher institutional ownership correlates with lower bid-ask spreads for Nasdaq stocks during 1983-1991, and that a smaller proportion of this spread is attributable to informational asymmetry. Wahal and McConnell (1997) report that firms with high institutional ownership invest more heavily in R&D, consistent with reduced information asymmetry leading to reduced managerial myopia. Denis, Denis and Sarin (1997) report that the presence of an outside blockholder correlates with higher top executive turnover, and with a stronger correlation between turnover and poor firm performance. However, none of these studies explores the impact of institutional ownership on overall firm performance. A number of studies examine the impact of institutional activism on the performance of the targeted firm, and collectively find only limited evidence that activism improves subsequent performance or affects the firm's subsequent actions (see the survey by Black, 1998). 9
In sum, the extant evidence provides modest evidence that large block investments by insiders (management) or by outsiders can increase firm value. There is considerable variance in this finding, however. Most studies discussed above are based on relatively small samples, over relatively short time-s -- perhaps too short for the hypothesized effects of relational investing to show up. Many examine investment by a corporate "raider" -- the antithesis of the model that proponents of relational investing have in mind. Finally, with the exception of Carleton, Nelson, and Weisbach (1997), previous researchers have looked for evidence of performance effects from certain actions that investors or investor groups take (for example, the filing of shareholder resolutions, or activist investors targeting a firm for takeover, or CalPERS or the Council of Institutional Investors targeting of poor performers with negative publicity campaigns). 1 While these studies are helpful in understanding the market s valuation of certain blockholder actions, they may entirely miss the essence of the way relationship investing is supposed to work. Specifically, relational investors are supposed to work constructively with management - most likely, not under media glare or mu ch, if any, public disclosure. Given the above consideration - the only way to determine the impact of relational investors on firm performance is to consider performance over long horizons of several years. III. Sample, Data Collection Procedure, and Definition of relational Investor A. Defining "Relational Investor" The proponents of relational investing have never defined who counts as a "relational investor," beyond the vague requirement that the investor hold a "large" block for a substantial of time, and actively monitor the firm's performance. Nor have they specified how quickly the results of the investor's monitoring should show up in a firm's performance. Thus, an initial question is how to give empirical content to the concept of relational investing. How large a stake is considered "relational"? A lower bound on what percentage stake can be considered large enough to involve a relational investment is set by data availability. Under the securities laws, 1 Carleton, Nelson, and Weisbach (1997) analyze private correspondence between TIAA-CREF and forty large firms they contacted between 1992-1996 on various corporate governance matters. In almost all cases 10
American companies and their shareholders must report 5 percent ownership positions. Below this level, comprehensive ownership information is not available. But a 5 percent shareholder may have little influence, and will often be passive. If the shareholder is dissatisfied with management, it may simply sell its shares, rather than engage with management in an effort to improve future results. Ten percent ownership offers a stricter criterion. A 10 percent shareholder must accept loss of liquidity, both because of the size of its stake, and because the shareholder must forfeit "short-swing" trading profits on a purchase and subsequent sale within a 6-month, under Securities Exchange Act 16(b). Also, during the time of our study, a 10 percent shareholder was required to report ownership on SEC Schedule 13D -- a more complicated form than the Schedule 13G that is available to an institutional investor who remains passive and owns 5-10 percent of a company's shares. Thus, a 10 percent shareholder is more likely to actively monitor, and is less likely (and less able) to simply sell if dissatisfied with management. For these reasons, we use 10 percent ownership as our criterion for when a shareholder owns a large block. We also infer from ownership of a 10 percent block that the shareholder engages in monitoring (we cannot directly observe monitoring). If this block is held for a long enough, we will consider the shareholder to be a "relational investor." However, we also collect data on 5 percent, 15 percent, and 20 percent shareholders, to test the robustness of our results to the definition of "relational investor", and to explore a possible nonmonotonic relationship between ownership stake and the investor's effect on firm performance. How long a holding is "long term"? The constructive engagement with management posited by proponents of relational investing is a multi-year process. Any performance improvements should be expected to emerge only over a several-year. We test for performance effects for firms that had a 10 percent blockholder throughout one of three mostly nonoverlapping, 4-year holding s: 1983-1986, 1987-1990, and 1990-1993.2 In many cases, an investor who held a 10 percent block throughout one of these s also held the stake for a longer. This holding is long enough to permit constructive engagement with management over time, and yet not so long as to compromise data availability, which decreases as we the company eventually adopted the changes proposed by TIAA-CREF. 2 The last two 4-year subs overlap slightly. We only had eleven years worth of block-holding data, 11
lengthen the required holding. We use (mostly) nonoverlapping s because results from overlapping s are likely to be correlated. We also conduct limited robustness tests using shorter, 2-year holding s (1983-1984; 1985-1986; 1987-1988; 1989-1990; 1991-1992) and longer 6 year s (1983-1988 and 1988-1993). Over what will performance effects show up? We measure firm performance over mostly nonoverlapping, 4-year s that (mostly) match our s for measuring long-term block holdings: 1983-1986; 1987-1990; 1991-1995. To allow for the possibility that performance effects will show up after the holding ends, we extend the third performance (1991-1995) for two years after the end of the corresponding (1990-1993) for measuring large-block holdings; and by testing for lead or lag relationships between performance during one performance and the presence of a relational investor during the preceding or subsequent block holding. Which types of investors are "relational"? A relational investor is an outside investor who monitors the firm's performance, and is not himself part of the management team. Thus, we exclude company officers from being considered as relational investors. A relational investing also must have both the ability and incentive to monitor. A typical employee stock ownership plan holds shares for the accounts of a large number of employees, and delegates voting authority to the employees in proportion to their shareholdings. No individual employee has much influence over the company's management, nor much incentive to monitor. Thus, we exclude employee stock ownership plans from consideration as "relational investors." In sum, we define a "relational investor" as a shareholder, other than a company officer or emp loyee stock ownership plan, who holds at least a 10 percent stake, generally for a minimum of four years. B. Sample and Data Collection Procedure The data for this study were assembled by starting with the universe of firms in the Compustat data base, and identifying, for the years 1983 and 1992, the 1,000 non-financial and 100 financial firms with the largest total market capitalization. We used market value of equity and book value of debt to compute total and this seemed the best compromise to make full use of the available data. 12
capitalization, and eliminated foreign-owned companies and subsidiaries of companies whose parent was in our sample already. These criteria produced a list of 1,534 publicly-traded companies, each of which had data in Compustat for at least one year during the 1983-1993. Information on ownership positions in these companies came from CDA/Spectrum, which compiles information from SEC filings into computer-readable form. We considered data on all 13D, 13G, and 14D(1) filings by individual and institutional investors during each of the eleven years from 1983 to 1993. These data were matched to the list of 1,534 companies in our sample to identify all investors who held at least five percent of the equity in any of these companies. The above process led to a list of about 50,000 ownership positions involving about 5,000 different owners. procedures: To ensure the integrity and consistency of our block-ownership data we implemented several 1. Securities laws require that groups of investors who hold and manage their stockholdings as a group file reports with the SEC as a group. In many cases, however, the individual members of the group also filed separately. We aggregated the total ownership positions by all blockholders for every company in every year during 1983-93. If any aggregated position was more than 100 percent we examined the complete history of all reported owners for that firm. We found situations in which, for example, KKR, Kohlberg, Kravis, and Roberts all reported holding, say, 30 percent positions. We eliminated records of the individual filings, and kept only the group filings. 2. We identified every situation in which two or more investors first entered the dataset as blockholders for the same company in the same year, and with the same size holdings, and in which subsequent reports continued to list identical holdings. For example, Jackson Family Trust, Jack Jackson, Jill Jackson, and Jan Jackson Smith all became 6.8 perrcent investors in the same year. Also, nearly every time a Fidelity Fund reported a position in some company, FMR Corporation (Fidelity s parent company) reported a position of the same size. To eliminate double-counting (or triple-counting) blockownership in a company, we eliminated records of all but one position the group position. 3. We identified every situation in which CDA/Spectrum continued to carry information in one year about a position reported by a particular investor in a prior year, even though that investor had filed an updated report. In other words, the source data reported two different positions for the same investor in the same year (in the same company). We retained only the information from the most recent filing. 4. If a single owner reported positions in more than one class of stock for a given company, those positions were aggregated to produce a measure of the share of the total equity capitalization of the firm held by the investor. We eliminated any investors whose aggregated holdings did not total at least five percent of the aggregate equity value of the firm. 5. CDA/Spectrum carries over information from the previous year if there is no new filing in a given year. However, CDA/Spectrum drops 13D filers after five years and 14D(1) filers after two years if they did not make a new filing during that. This meant, for example, Warren Buffet stopped appearing as a significant investor in Berkshire Hathaway after 1987 in the CDA/Spectrum data! We corrected this by identifying all ownership positions that appear to have been dropped for this reason; we filled in missing information by carrying over data from the previous year. Data from prior years are carried over until the end of the sample (1993), or until the investor makes a new filing, or until the firm is dropped from Compustat (because, for example, the firm was taken over), whichever comes first. 13
6. We standardized the names of blockholders, so that if an investor filed as Jackson Family Trust in one year and Jackson Fam. Trst. in another year, one version of the investor name was chosen and applied consistently to all filings by that investor. This would enable our software to tell us with greater accuracy which investors held their positions from one year to the next. The above data-compilation process resulted in 28,614 records; each record has information on name of company, name of blockholder, year of blockholding, and size of blockholding. We then added information to the file from another CDA/Spectrum publication (Insider Holdings), identifying each large-block investor as an "insider" or an "outsider," based on the definition used by CDA/Spectrum. We also added information from Lexis ABI U.S. file, identifying investors by "type" (that is, individual, insurance company, bank, employee pension or benefit plan, investment advisor, broker-dealer, partnership, etc.). When several persons file a joint ownership report, CDA/Spectrum unfortunately lists only the person named first on the report, followed by an "et al". We classifed these joint filings based on the named person. IV. Intertemporal and Cross-sectional Characteristics of Relational Investors Table 1, Panel A, notes the number of blockholders holding at least 5 percent of aggregate equity, summary statistics and distribution of the size of their holdings in our sample of 1,534 largest U.S. firms over the 1983-1993. There is a secular increase in the number of blockholders in our sample firms from 1,457 blockholders in 1983 to 2,402 blockholders in 1993. The mean and median ownership of these blockholders stays approximately the same - 13 percent and 8.5 percent, respectively, - over this 11-year. Table 1, Panel B, notes the number of blockholders of various types for our sample of 1,534 firms over the 1983-1993. The number of blockholders more than doubled during this for each of the following types of investors: employee benefit and pension plans, holding companies, investment advisors, investment companies, and partnerships. The number of blockholders that are broker-dealers, banks, and individuals showed very small increases during this 11-year. Nonetheless, at the end of the, there were still more large-block shareholders who were individual investors than of any other type. In Table 2, Panels A through D note the number of "relational investors" in our sample of firms 14
during all sequential 2-year, 4-year, and 6-year subs of our sample. To qualify as a "relational investor," an investor must hold its position (5 percent of total equity in Panel A, 10 percent in Panel B, 15 percent in Panel C, and 20 percent in Panel D) throughout the sub. Sub-s are identified in the first column. Mean and median fractional ownership of these blockholders, and mean and median number of such blockholders per sample firm are also noted. While our focus is on 10 percent blockholders holding the block for (at least) 4-years, we consider other size and blockholdings for robustness checks. The summary statistics for these definitions of relational investor reveal a secular rise in the total number of relational investors similar to the secular rise in large-block shareholders that we observed in Table 1. (Recall that the latter were identified by their holdings at single points in time -- the end of each calendar year, whereas relational investors, as we define them, must have held for some minimum of time.). We also see stability over time in the distribution of the size of holdings of the relational investors, and even considerable stability in the mean number of relational investors per firm (of firms that have such an investor) over time: About 2 in the two-year s, about 1.7 in the four-year s, and about 1.6 in the six-year s. This suggests that the increase in the total number of relational investors reflects an increase in the number of firms with relational investors, rather than more relational investors taking positions in the same set of firms. This is confirmed by the rise over time in the fraction of sample firms with at least one relational investor in each of the subs. By the end of our sample, more than half the firms in our sample had at least one 5 percent blockholder that held its position for the two years, 1992-93, and more than 13 percent had at least one 20 percent blockholder that held its position throughout this 2- year. These findings run counter to the conventional wisdom that large-block shareholding is rare among large publicly-traded firms. But these tables also make it clear that the way one defines a relational investor in terms of size of block and holding matters when one investigates the prevalence or results of relationship investing in large U.S. corporations. The above definitions of relational investor consider only the size of the block, and the investor's holding. As discussed above, it is possible that different types of investors might have different investment objectives, however, so we might also want to know what type of investor holds each position. 15
We report in Table 3 the frequency of each type of 5 percent (Panel A), 10 percent (Panel B), 15 percent (Panel C), and 20 percent (Panel D) blockholders in our sample over selected, nonoverlapping subs. V. Survivorship Bias This study, like any study that considers long-term financial performance, faces a potential problem with exit from (and entry into) the sample over time. The focus of this study is to understand the impact of relational investing on the performance of the largest U.S. corporations. To this end we constructed our initial sample to include the 1,000 largest non-financial companies and the 100 largest financial companies for 1983. We also wished to consider the performance of these firms over a long time-horizon; specifically, from 1983 to 1995. Over such a long, many of these firms would exit our sample due to bankruptcies, mergers, etc. When such firms exit our sample we would not have stock-market data, accounting data, and/or ownership data for the remainder of the. By the end of our of analysis (1993 for ownership data, 1995 for performance data), we would be considering only about 70 percent of the 1,100 firms we started out with in 1983. This would lead to two potential problems. First, and more relevant from a policy viewpoint, for the latter of our analysis (roughly, 1990-1995) our results would only be based on a fraction of the largest U.S. companies. If the nature (in terms of industry) of the firms in the economy did not change during 1983-1995, then this would not be a serious limitation. However, a substantial popular and academic literature suggests that during this the economy has shifted dramatically from being dominated by manufacturing firms to being dominated by service and technology firms. Hence, the largest 1,000 non-financials in 1992 are likely to be quite different than the largest 1,000 non-financials in 1983, not just in name but, more important, in the industries they represent. To address this problem, we also included in our sample those firms that were among the largest 1,000 non-financials and 100 largest financials in 1992. (Some of these had been among the largest in 1983 and were already in our sample.) Hence, our sample of 1,534 firms includes the largest 1,000 non-financials and the largest 100 financials in 1983 and 1992. This allows us to study the impact of relational investing on firm performance for the largest U.S. firms for both the earlier (1983) and later (1992) 16
s. The second problem is concerned with the relationship, if any, among survivorship of firms over our of analysis, relative performance of survivors and non-survivors, and the presence (or the lack thereof) of relational investors in such firms. A full analysis of this set of issues is beyond the scope of this paper; we are currently pursuing these issues in another paper. However, we provide some evidence that suggests that our current analysis is not likely to be seriously impacted by this aspect of the survivorship bias problem: Table 4, Panels A and B, summarize the reasons our sample firms exit our sample from Compustat and CRSP, respectively, during 1984-1995. About 400 of the sample firms exit the sample during 1984-1995; more than two-thirds of these firms exit because they were acquired. For the firms that exit our sample because of acquisitions, their stock-market performance measure will not be biased; the last year s return for these firms would include the acquisition premium paid its shareholders. However, we cannot exclude the possibility that accounting measures of performance are different for surviving and nonsurviving firms. For the remaining third of the firms that exit our sample the reasons are bankruptcy, delisting, and exchange for other securities. It is unclear how the exit of these firms would impart a systematic bias in our analysis of the impact of relational investing on firm performance. Furthermore, we find (in the next two sections) that relational investors during the 1987-1990 were perhaps different from relational investors in other s. Table 4, Panels A and B, does not suggest that the years 1987, 1988, 1989, and 1990 are different from the other years in terms of the mix of reasons for exits. VI. Measuring the Impact of Relational Investors on Performance The summary statistics in Table 2 were based on four different block-sizes, and 24 subs (10 overlapping 2-year s, 8 overlapping 4-year s, and 6 overlapping 6-year s), or a total of 96 different definitions of "relational investor." We also developed three different measures of stock price performance, and 13 different accounting measures of performance. Each of these performance measures could be measured for time s that are contemporaneous with the blockholding, for time s that are prior to the blockholding (which might provide insight into characteristics of firms that were 17
targets of investments by large-block investors), and for time s that follow the blockholding (to allow for the possibility that there might be a lag between the blockholder s presence and the effect on performance). With these many variants on the definition of relational investor, and possible ways to measure performance, including varying the amount of lead or lag between blockholder measurement and performance measurement, the number of possible regressions measuring the relationship between corporate performance and the presence of a relational investor becomes unmanageable. To limit this problem, we made several decisions. First, we would consider relational investors defined only over six 2- year non-overlapping subs ('83-'84, '85-'86, '87-'88, '89-'90, '91-'92, and '92-'93), and three 4-year nonoverlapping subs ('83-'86, '87-'90, '90-'93). 3 Second, we would consider performance variables defined over roughly these same 2-year, and 4-year, s. 4 Third, we would report results based on only one measure of stock price performance (the market-adjusted return over the performance s). 5 Fourth, we would use a 2-year lead/lag when using a 2-year measure of performance or of relational investing, and a 4-year lead/lag when using a 4-year measure of performance or of relational investing. 6 For each performance measure in each time considered, we ran a simple linear regression on the sample of firms for which we had the necessary data for that sub-. In each regression, the measure of performance was on the left-hand side, and dummy variables indicating the presence of a 3 Note that the last two 2-year subs, and the last two 4-year subs overlap slightly. Since we only had eleven years worth of block-holding data, this seemed the best compromise to make full use of the data we had. 4 We had performance data through 1995. So for our 2-year blockholding s, the final for measuring performance runs from '93-'95 (three years), whereas the final over which relational investors are defined is '92-'93. For our 4-year blockholding s, the final for measuring performance runs from '91-'95 (five years), while the final over which relational investors are defined runs from '90-'93. 5 We examined the results measured two other ways (cumulative abnormal returns, and standardized abnormal returns, both measured over the relevant performance s), but, as we explain below, we decided that the marketadjusted returns were less likely to be influenced by misspecification problems.. 6 We also considered 6-year s for both relational investing and performance, and examined some regression results involving these longer s. But these results added little insight beyond the results with 2-year and 4-year s, so we do not report them in the main body of the paper. Some regression results considering the 6- year s are contained in Appendix tables A1 through A8. 18
relational investor served as explanatory variables. 7 Dummy variables were included for relational investors defined in the prior to the performance, for the contemporaneous with the performance, and for the subsequent to the performance, so that up to three dummy variables were included in each regression. For the accounting measures of performance, control variables for firm size and industry were included. This approach was repeated for each of the four size definitions of relational investor (5 percent, 10 percent, 15 percent, and 20 percent), yielding 36 different regressions of stock market performance on relational investors (nine subs, times four different size definitions of relational investor). For the accounting measures of performance, we further narrowed the problem by considering only the three non-overlapping performance s: 83-86, 87-90, and 91-95. Even so, our approach yielded 39 different sets of regressions (13 performance variables times three performance s) for each of four different size definitions of relational investor for a total of 156 regressions on accounting measures of performance. 8 Each of the 192 regressions (156 on accounting variables, and 36 on the stock market performance variable), in turn, included at least two, and sometimes three coefficients of interest (on the dummy variables for relational investor in the prior, the current, and the subsequent ). Thus we generated hundreds of coefficients, any one of which, taken in isolation, could be interpreted as 7 The dummy variables took a value of 1 for a given firm in a given if that firm had at least one relational investor during the relevant, and zero otherwise (that is, if there were no relational investors in that firm, as we defined them). 8 For many of the accounting measures of performance, we also considered variations on all regressions using different industry control variables. See discussion below. 19
telling us something about how relational investing affects corporate performance. 9 We then looked for patterns in the regressions that might provide robust evidence on how relational investing affects performance. This broad range of regressions is appropriate, because our study is not intended to be a test of any particular hypothesis about how relational investors affect corporate performance. It is, rather, intended to provide a comprehensive statistical description of large-block shareholding in the corporate sector, and to conduct an exploratory analysis on the impact of relational investing on firm performance. Our analysis is exploratory rather than a test of a particular hypothesis, because the literature does not contain a precise definition of who a relational investor is, what makes an investor a relational investor, and how such investors add value. Our analysis, hopefully, would motivate others to develop the theoretical underpinnings of this strand of the literature. A. Stock Market Returns VI I. Regression Results: Stock Price Performance Measures Table 5 reports one-year stock price performance summary statistics for the firms in our sample (using stock price data available on the 1996 CRSP tapes). The 1996 CRSP tapes have some return data on 1,336 of the 1,534 sample firms. We measure stock price performance in three ways: (I) Market Adjusted Return (MAR): This technique involves cumulation over the measurement of daily market-adjusted returns (MAR t ) for the entire sample: MAR t = sample return on day t (R t ) minus the return on the S&P 500 index (RM t ) for day t, without an adjustment for β. (ii) Cumulative AbnormalRreturn (CAR): This technique also treats the entire sample as a single portfolio, but with an adjustment for β. We estimate daily abnormal returns over the measurement (AR t ) for the entire sample based on the market model: AR t = R t - α - β*rm t. The market model parameters α and β are estimated during the year preceding the measurement, using the S&P 500 index as the market index. Under the null hypothesis of no abnormal performance and 9 Of course, when considering so many regressions and their coefficients, statistically significant coefficients could be obtained due to pure random chance. 20
stationarity of the returns-generating process over time, the CAR for the sample should be zero. (iii) Standardized AbnormalReturn (SAR): Cumulation over the measurement of daily standardized abnormal returns for each firm (SAR i,t ) (as in Dodd and Warner (1983 )), where the market model parameters α, β,and the standard deviations of the sample firms' abnormal returns, σ, are estimated during the year preceding the measurement, using the S&P 500 index as the market index. This technique controls for heteroscedasticity in the abnormal returns across firms. Under the null hypothesis of no abnormal performance and stationarity of the returns-generating process over time, the firm SARs should be distributed unit normal (mean = 0, standard deviation = 1), and the portfolio should have SAR = 0, assuming independence across the n sample firms, standard deviation = 1/n 0.5. Many of the single year MAR, CAR, and SAR returns reported in Table 5 are large. However, there is no apparent sign pattern for these returns. This suggests either that the net-of-market returns to the firms in our sample are not independent of each other, or that long-horizon stock-performance measures are misspecified. 10 Kothari and Warner (1997) argue that the misspecification problem in tests of long-horizon stock-performance is less severe for MAR than for CAR and SAR. For this reason, and because we also observe that the standard deviations are higher for both CAR and SAR measures relative to MAR, we rely on the MAR measure of stock performance for our regression analysis. 11 10 See Kothari & Warner (1997) and Barber & Lyon (1997), for discussions of the misspecification problem. 11 The large single-year portfolio returns are not an artifact of our choice of market index. We also computed MAR, CAR, and SAR series using the CRSP equally weighted index as the market index, instead of the S&P 500 index. The entries in individual years were different, but the combination of no clear overall trend with large single-year and multiyear returns persisted. The sensitivity of our portfolio returns to the choice of market index is further evidence that long-horizon tests for stock price returns are badly specified. Barber & Lyon (1997) find that the misspecification of longhorizon returns can be corrected by matching sample firms to control firms that are similar in size and book-to-market ratio. This correction was not possible for our study because our sample is essentially the universe of large U.S. public firms; a control sample does not exist. 21