Journal of Accounting and Economics 44 (2007) 238–286 www. elsevier. com/locate/jae Corporate disclosures by family ? rms Ashiq Alia,A, Tai-Yuan Chenb, Suresh Radhakrishnana a School of Management, SM41, The University of Texas at Dallas, Richardson, TX 75083-0688, USA b Hong Kong University of Science and Technology, Kowloon, Hong Kong Available online 6 February 2007 Abstract Compared to non-family ? rms, family ? rms face less severe agency problems due to the separation of ownership and management, but more severe agency problems that arise between controlling and non-controlling shareholders.

These characteristics of family ? rms affect their corporate disclosure practices. For S&P 500 ? rms, we show that family ? rms report better quality earnings, are more likely to warn for a given magnitude of bad news, but make fewer disclosures about their corporate governance practices. Consistent with family ? rms making better ? nancial disclosures, we ? nd that family ? rms have larger analyst following, more informative analysts’ forecasts, and smaller bid-ask spreads. r 2007 Elsevier B. V. All rights reserved. JEL classi? cation: G32; M41; M43; M45 Keywords: US family ? ms; Corporate disclosure; Earnings quality; Corporate governance disclosure; Management forecasts 1. Introduction Firms that are managed or controlled by founding families, hereafter, referred to as family ? rms, constitute about one-third of the S&P 500, and operate in a broad array of industries (Anderson and Reeb, 2003a). On average, families own 11% of their ? rms’ cash ? ow rights, representing a signi? cant proportion of the US stock market capitalization, and 18% of their ? rms’ voting rights.

Also, family members serve as top executives or CEO in 63% of family ? ms and serve on the board as directors or chairperson in 99% of the ACorresponding author. Tel. : +1 972 883 6360; fax: +1 972 883 6811. E-mail address: ashiq. [email protected] edu (A. Ali). 0165-4101/$ – see front matter r 2007 Elsevier B. V. All rights reserved. doi:10. 1016/j. jacceco. 2007. 01. 006 ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 239 family ? rms. In their survey of corporate governance literature, Shleifer and Vishny (1997) emphasize the importance of studying the characteristics of such ? rms to better understand the economic ef? iency of different corporate governance mechanisms. As such, several recent papers examine various aspects of US family ? rms. 1 Compared to non-family ? rms, family ? rms in the US face less severe agency problems that arise from the separation of ownership and management (Type I agency problems). However, they are characterized by more severe agency problems that arise between controlling and non-controlling shareholders (Type II agency problems) (Gilson and Gordon, 2003). These characteristics of family ? rms raise interesting issues about their corporate disclosure practices.

In this paper, we examine how these differences in agency problems across family and non-family ? rms in? uence corporate disclosures. We consider the following aspects of corporate disclosures: quality of reported earnings, voluntary disclosure of bad news through management earnings forecasts, and voluntary disclosure of corporate governance practices in regulatory ? lings. 2 We examine whether reported earnings of family ? rms are of better quality than those of non-family ? rms. Family ? rms face less severe Type I agency problems because of their ability to directly monitor the managers (Demsetz and Lehn, 1985).

This enables family ? rms to tie less of management compensation to accounting based performance measures (Chen, 2005), thus their reported numbers are less likely to be manipulated due to managerial opportunism. Moreover, better knowledge of the ? rm’s business activities by family owners (Anderson and Reeb, 2003a) enables them to detect manipulation of reported numbers, thereby keeping this activity in check. Thus, earnings manipulation due to Type I agency problems is likely to occur to a greater extent in non-family ? rms. Family ? rms face more severe Type II agency problems because of families’ signi? ant stock ownership and control over the ? rms’ board of directors. Family ? rms’ boards tend to be less independent and are dominated by family members (Anderson and Reeb, 2003a; Anderson and Reeb, 2004). Type II agency problems may also lead to manipulation of accounting earnings, for example, to hide the adverse effects of related party transactions or to facilitate family members’ entrenchment in management positions. Thus, it is an empirical question whether family ? rms have better or worse earnings quality compared to non-family ? rms. We ? d that compared to non-family ? rms, family ? rms exhibit less positive discretionary accruals, greater ability of earnings components to predict cash ? ows, and larger earnings response coef? cients. These results are consistent with the notion that the difference in agency costs across family and non-family ? rms due to Type I agency problems dominate the difference in agency costs across family and non-family ? rms due to Type II agency problems. We also examine whether, compared to non-family ? rms, family ? rms are more likely to warn for a given magnitude of bad news.

Opportunistic behavior related to both Type I and Type II agency problems may lead to delays in the disclosure of bad news, i. e. , 1 Compared to non-family ? rms, family ? rms in the S&P 500 are more pro? table (Anderson and Reeb, 2003a), have lower cost of debt ? nancing (Anderson et al. , 2003), are less diversi? ed, and have similar level of debt (Anderson and Reeb, 2003b). As in our paper, these studies classify a company as a family ? rm if the founders or descendants continue to hold positions in the top management or on the board, or are among the company’s largest shareholders. Our sample period is 1998–2002, hence our conclusions are applicable to the period prior to Sarbanes Oxley Act, 2002. ARTICLE IN PRESS 240 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 managers of non-family and controlling shareholders of family ? rms face incentives to withhold or delay the release of bad news. Managers of non-family ? rms may withhold bad news to maximize their equity-based compensation or to facilitate entrenchment. Also, controlling shareholders of family ? rms may withhold bad news to reduce scrutiny of their private bene? seeking activities or to facilitate entrenchment in management positions. Thus, again it is an empirical question whether family ? rms are more likely to warn for a given magnitude of bad news than non-family ? rms. We ? nd that family ? rms are more likely to warn for a given magnitude of bad news than non-family ? rms. This result suggests that family ? rms are subject to less opportunistic behavior consistent with the notion that the difference in agency costs across family and non-family ? rms due to Type I agency problems dominate the difference due to Type II agency problems.

Next, we examine whether, compared to non-family ? rms, family ? rms are less likely to make voluntary disclosures about their corporate governance practices. Family ? rms have incentive to reduce the transparency of corporate governance practices to facilitate getting family members on boards without interference from non-family shareholders. Consistent with this argument, we ? nd that family ? rms tend to disclose less information about their corporate governance practices in their proxy statements. Finally, we examine whether better disclosure of ? ancial performance (reported earnings and bad news warning) bene? ts family ? rms in terms of better analyst following, better analysts’ earnings forecasts, and better market liquidity of their stocks. We ? nd that compared to non-family ? rms, family ? rms have larger analyst following, lower dispersion of analysts’ forecasts, smaller forecast errors, less volatile forecast revisions, and smaller bid-ask spreads. To gain additional con? dence that difference in the severity of agency problems across family and non-family ? rms drives our results, we analyze subsamples of family ? ms that are expected to have different severity of agency problems. Villalonga and Amit (2006) provide evidence suggesting that family ? rms with founder CEO have less severe agency problems than those with descendent CEO. They also provide evidence suggesting that family ? rms with dual class shares have more severe agency problems than those without dual class shares. We repeat our analyses after classifying family ? rms accordingly. Our results suggest that family ? rms with founder CEO (rather than those with descendent CEO) are primarily responsible for family ? ms exhibiting better disclosure practices and better disclosure-related economic consequences as compared to non-family ? rms. Our results also suggest that family ? rms without dual class shares (rather than those with dual class shares) are primarily responsible for family ? rms exhibiting better disclosure practices and better disclosure-related economic consequences as compared to non-family ? rms. Overall, our evidence based on the subsamples of family ? rms suggests that the severity of agency problems is a likely factor in the difference in disclosure practices we observe across family and non-family ? ms. Our ? ndings contribute to the literature on corporate disclosures. Healy and Palepu (2001) and Bushman and Smith (2001) note that there is not much evidence in the literature on the effect of agency problems on corporate disclosures. We contribute by showing how the difference in the severity of agency problems across family and non-family ? rms affect the quality of different types of corporate disclosures. A contemporaneous study, Wang (2006), also documents the association between family ? rm membership and earnings quality.

His results are in general consistent with ours. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 241 However, our study differs from his in several ways. We consider a different set of earnings quality measures. Furthermore, we examine the likelihood of management’s warning of a poor earnings report, voluntary disclosure of corporate governance practices, and the effect of differential disclosure practices of family ? rms and non-family ? rms on analyst following, various characteristics of analysts’ forecasts, and bid-ask spreads.

Our ? ndings also contribute to the literature on family ? rms. Given the prevalence of family ? rms in the US and the unique combination of Type I and Type II agency problems associated with them, these ? rms have been the subject of several prior studies (see note 1). Whether the severity of agency problems of US family ? rms is greater or less than that of non-family ? rms is still debatable however (Anderson and Reeb, 2003a). We contribute to this debate by documenting the difference in the quality of disclosures between these two types of ? ms. Our ? nding that family ? rms provide better ? nancial disclosures is consistent with these ? rms being subject to less managerial opportunism due to less severe agency problems. Speci? cally, the difference in agency costs across family and non-family ? rms due to Type I agency problems dominate the difference in agency costs across family and non-family ? rms due to Type II agency problems. The rest of the paper is organized as follows. Section 2 discusses our hypotheses. We describe the agency problems associated with family ? ms and predict their effects on different types of corporate disclosures. Section 3 discusses the sample and Section 4 presents the results. Section 5 concludes the paper. 2. Hypotheses development 2. 1. Family ? rms and agency problems There are two main types of agency problems in public corporations. The ? rst type of agency problem arises from the separation of ownership and management (Type I agency problem). The separation of corporate managers from shareholders may lead to managers not acting in the best interest of the shareholders. The second type of agency problem arises from con? cts between controlling and non-controlling shareholders (Type II agency problem). Controlling shareholders may seek private bene? ts at the expense of noncontrolling shareholders. Below, we discuss how these two types of agency problems differ across family and non-family ? rms. 2. 1. 1. Separation of ownership and management (Type I agency problem) There are several characteristics of family ? rms that reduce the likelihood of managers not acting in the best interest of shareholders. First, families tend to hold undiversi? ed and concentrated equity position in their ? rms.

Thus unlike the free rider problem inherent with small atomistic shareholders, families are likely to have strong incentives to monitor managers (Demsetz and Lehn, 1985). Second, families have good knowledge about their ? rms’ activities, which enables them to provide superior monitoring of managers (Anderson and Reeb, 2003a). Third, families tend to have much longer investment horizons as compared to that of other shareholders. Thus, families help mitigate myopic investment decisions by managers (James, 1999; Kwak, 2003; Stein, 1988, 1989). In summary, compared to non-family ? ms, family ? rms face less severe hidden-action and hidden-information agency problems due to the separation of ownership and management. ARTICLE IN PRESS 242 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 However, certain factors contribute towards mitigating the difference between family and non-family ? rms in the Type I agency problems. Compensating managers based on observable performance measures help align the interest of the managers and stockholders (Demski, 1994; Healy and Palepu, 2001; Lambert, 2001; Bushman and Smith, 2001).

Concern about their reputation in the managerial labor market also contributes towards managers acting in the best interest of the shareholders. In addition, shareholders can bring lawsuits against managers if the managers defraud the shareholders (La Porta et al. , 1998). 2. 1. 2. Con? ict between controlling and non-controlling shareholders (Type II agency problem) In family ? rms, founding family may enjoy substantial control as a result of their concentrated equity holding in their ? rms, their voting rights exceeding their cash ? w rights, and their domination of the board of directors’ membership. This control gives the family power to seek private bene? ts at the expense of other shareholders. Controlling shareholders can seek such private bene? ts by freezing out minority shareholders (Gilson and Gordon, 2003), by engaging in related-party transactions (Anderson and Reeb, 2003a), and through managerial entrenchment (Shleifer and Vishny, 1997). However, certain factors contribute towards mitigating the difference between family and non-family ? rms in the Type II agency problems.

When families engage in private rent seeking, their activities may get revealed to the market and they may incur substantial cost in the form of lower equity value, especially since families have concentrated ownership and tend to hold their ? rms’ equities for long periods. In addition, signi? cant legal protection is accorded to non-controlling shareholders in the United States. La Porta et al. (1998, 2000) show that US is one of the few countries where the legal system gives minority shareholders strong protection against dominant shareholders in the corporate decision making process.

For example, provisions such as ‘‘Proxy by Mail’’ makes it easier for shareholders to cast their votes, ‘‘Cumulative Voting/Proportional Representation’’ gives minority shareholders more power to put their representatives on the board of directors, ‘‘Class action/Derivative lawsuits’’ enables minority shareholders to challenge directors’ decisions in court and force the company to repurchase shares of minority shareholders who object to certain corporate decisions. 2. 2. Family ? rms and corporate disclosures Is the difference in the two types of agency problems across family and non-family ? ms associated with the difference in their corporate disclosure practices? In this study, we examine the following aspects of corporate disclosures: quality of ? nancial statement numbers, speci? cally that of earnings, and the voluntary disclosure of bad news through management earnings forecasts. These features of corporate disclosures have been widely considered in the literature (see e. g. , Francis et al. , 2004; Kasznik and Lev, 1995). In addition, we examine voluntary disclosure of corporate governance practices. 2. 2. 1. Earnings quality As discussed earlier, compared to family ? rms, non-family ? ms have more severe Type I agency problems. To mitigate these problems, non-family ? rms are more likely to compensate their managers based on observable earnings-based performance measures. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 243 Founding families, on the other hand, being more effective monitors of management can reward their managers based on information about managers’ effort obtained through direct monitoring. Also, when family members holding large amount of stocks are managers the problem of separation of ownership and management is limited.

Thus, compared to non-family ? rms, family ? rms are less likely to compensate their managers based on observable earnings-based performance measures. Consistent with the above argument, Chen (2005) provides evidence that earnings-based CEO pay is signi? cantly smaller for family ? rms, both in terms of amount as well as in terms of percentage of total compensation. Since management compensation in family ? rms is less likely to be tied to earnings, family ? rms’ earnings are less likely to be manipulated (Healy and Palepu, 2001; Fields et al. , 2001). Direct monitoring by the families and their better knowledge of the ? ms’ business are additional reasons why managers’ opportunistic behavior is less likely to affect earnings of family ? rms. For example, family members’ knowledge of business conditions and relationship with suppliers and customers will enable them to more effectively detect whether goods have been shipped early to in? ate revenues or unreasonable cuts have been made to certain discretionary spending. The above arguments suggest that because of more severe Type I agency problems, earnings of non-family ? rms are likely to be of lower quality than that of family ? rms. However, certain factors, uch as reputation concerns in the managerial labor market and legal liabilities, help mitigate the difference in Type I agency problems between family and non-family ? rms. While these factors mitigate the difference in Type I agency problems they do not eliminate it. This argument suggests that reported earnings of family ? rms should be of better quality than those of non-family ? rms. Type II agency problems are also likely to have a differential effect on earnings quality across family and non-family ? rms. These agency problems could lead to a greater manipulation of accounting earning by family ? rms.

This manipulation may be done, for example, to hide the adverse effect of a related party transaction or to facilitate family members’ entrenchment in management positions. Moreover, given the high level of in? uence family owners have on their ? rms, if they decide to engage in earnings manipulation they can more easily do so. However, legal liabilities and reduced stock prices that may result from the private bene? t seeking behavior help mitigate the difference in Type II agency problems between family and non-family ? rms. 3 Here again, these factors mitigate the difference in Type II agency problems between family and non-family ? ms but do not eliminate it. Thus, the extent to which family ? rms as compared to nonfamily ? rms are subject to more severe Type II agency problems, the earnings quality of family ? rms will be lower. To summarize, difference in the quality of earnings between family and non-family ? rms would depend on the difference in the severity of their Type I and Type II agency problems. In general, if the difference in Type I agency problems dominates the difference in Type II agency problems, then the total agency problems would be less for family ? rms 3 Adelphia Corporation is an example of family owners very aggressively in? ting the ? rm’s reported earnings to afford Adelphia’s continued access to commercial credit and the capital market, while some of the family members engaged in extensive self-dealing at the expense of other Adelphia stakeholders (SEC Litigation Release No. 17627). However, these activities were discovered and the family owners were subjected to severe penalties, causing loss of most of their wealth (Searcey and Yuan, 2005). ARTICLE IN PRESS 244 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 and their earnings quality would be better, and vice versa. Thus, whether family ? ms’ earnings quality is better or worse than that of non-family ? rms is an empirical question. We summarize these arguments in the following hypotheses:4 H1a. Reported earnings of family ? rms are likely to be of better quality than those of nonfamily ? rms if the difference in their Type I agency problems dominates the difference in their Type II agency problems. H1b. Reported earnings of non-family ? rms are likely to be of better quality than those of family ? rms if the difference in their Type II agency problems dominates the difference in their Type I agency problems. 2. 2. 2.

Management forecasts of earnings Skinner (1994) notes that ? rms may incur legal liabilities and reputation costs if they fail to provide earnings warning prior to an earnings report containing bad news. Consistent with this argument, Skinner (1994) and Kasznik and Lev (1995) show that the likelihood of management earnings forecasts increases with the magnitude of bad news. However, all ? rms are not equally likely to provide earnings warning. Opportunistic behavior related to both Type I and Type II agency problems may lead to managers withholding bad news. Managers of non-family ? ms may withhold bad news to maximize their equity-based compensation or to facilitate entrenchment. Also, controlling shareholders of family ? rms may withhold bad news to reduce scrutiny of their private rent seeking activities or to facilitate entrenchment of family members in management positions. The following hypotheses summarize our expectations. 5 H2a. The association between the likelihood of management earnings forecasts and the magnitude of bad news is stronger for family ? rms than for non-family ? rms if the difference in their Type I agency problems dominates the difference in their Type II agency problems.

H2b. The association between the likelihood of management earnings forecasts and the magnitude of bad news is stronger for non-family ? rms than for family ? rms if the difference in their Type II agency problems dominates the difference in their Type I agency problems. 4 The arguments leading to the hypothesis consider the effect of opportunistic behavior on earnings quality, and not the effect of ef? cient contracting. The ef? cient contracting perspective suggests that ? rms would commit to reporting higher quality earnings to mitigate agency problems (Demski, 1998; Evans and Sridhar, 1996; Fukui, 1996; Arya et al. 1998). Under the ef? cient contracting perspective, if the difference in Type I agency problems between family and non-family ? rms dominates the difference in their Type II agency problems, then the earnings quality of non-family ? rms are predicted to be better than that of family ? rms, and vice versa. However, it is not clear if there exist mechanisms that can enforce ? rms’ commitment to make higher quality disclosures regardless of its content. In fact, Skinner (1993) shows that management opportunism dominates ef? cient contracting in explaining observed accounting choices.

Thus, we propose our hypotheses (H1a and H1b) based on the opportunism perspective. 5 Under the ef? cient contracting perspective, the predictions are opposite of those in hypotheses H2a and H2b: ? rms would commit to provide earnings warnings to mitigate agency problems. However, as before, we propose our hypotheses (H2a and H2b) based on the opportunism perspective. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 245 2. 2. 3. Corporate governance related disclosures Family ? rms’ boards tend to be less independent, with substantial representation by family members (Anderson and Reeb, 2003a, 2004). Family owners prefer to have family members as directors because they tend to be proactive and have a collective desire to maintain unity and preserve their wealth (Business Week, 2003). Maintaining lack of transparency of corporate governance practices may facilitate getting family members on board without much interference from non-family shareholders. The resulting concern that the non-controlling shareholders may have about the lack of transparency in corporate governance practices of family ? rms would be reduced to the extent that these ? rms deliver superior performance (Anderson and Reeb, 2003a).

Regulatory ? lings, such as the proxy statement, contain disclosures on corporate governance practices. However, ? rms have some discretion on disclosures related to certain corporate governance practices such as Voting and Shareholder Meeting Procedures, details of the board committees, and director compensation. Thus, we propose the following hypothesis: H3. Compared to non-family ? rms, family ? rms are less likely to make voluntary disclosures about their corporate governance practices in their regulatory ? lings. 2. 2. 4. Bene? ts of better ? nancial disclosures by family ? ms Disclosure quality has been shown to be related to capital market bene? ts. Welker (1995), Lang and Lundholm (1996), and Healy et al. (1999) show that ? rms with more informative disclosures (measured using analysts’ surveys) have larger analyst following, lower dispersion of analysts’ earnings forecasts, smaller forecast errors, less volatile forecast revisions, and smaller bid-ask spreads. 7 Thus, if earnings quality is better for family ? rms (hypothesis H1a) and family ? rms are more likely to make management forecasts of bad news (hypothesis H2a), we expect greater disclosure-related bene? s will accrue to family ? rms. However, if earnings quality is better for non-family ? rms (hypothesis H1b) and non-family ? rms are more likely to make management forecasts of bad news (hypothesis H2a), we expect greater disclosure-related bene? ts will accrue to non-family ? rms. The following hypotheses summarize our expectations:8 H4a. Compared to non-family ? rms, family ? rms are more likely to have larger analyst following, lower dispersion of analysts’ forecasts, smaller forecast errors, less volatile Anderson and Reeb (2004) report that for the period, 1992–1999, family ? ms had on average 40% inside directors, about half of which were family members. On the other hand, non-family ? rms had only about 22% inside directors. 7 These prior studies argue that more informative disclosure attracts more analysts because information acquisition becomes less costly, which results in superior earnings forecasts and buy-sell recommendations, increasing the demand for analysts’ services. Better disclosure results in lower forecast dispersion because analysts put more weight on public as compared to private information in forming their forecasts.

More informative disclosure improves analyst forecast accuracy. Also, more timely disclosure results in less extreme revisions. Finally, more informative disclosure reduces information asymmetry among market participants, thereby reducing the adverse selection problem and increasing market liquidity. 8 In hypothesis H3, we predict that family ? rms are less likely to make voluntary disclosures about their corporate governance practices. These types of disclosures are not related to ? nancial performance and are therefore unlikely to affect their analyst following and analysts’ earnings forecast roperties, but they may adversely affect their bid-ask spreads. 6 ARTICLE IN PRESS 246 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 forecast revisions, and smaller bid-ask spreads, if these ? rms disclose better quality earnings and are more likely to provide warning for poor earnings. H4b. Compared to family ? rms, non-family ? rms are more likely to have larger analyst following, lower dispersion of analysts’ forecasts, smaller forecast errors, less volatile forecast revisions, and smaller bid-ask spreads, if these ? ms disclose better quality earnings and are more likely to provide warning for poor earnings. 3. Sample We use the Standard and Poor’s 500 ? rms for our analyses, because for the year 2002, BusinessWeek classi? es them into family and non-family ? rms: 177 as family ? rms and the remaining as non-family ? rms. A ? rm is considered a family ? rm if the founder and/or their descendents hold positions in the top management or on the board or are among the companies’ largest shareholders. 9 Considering only S&P 500 ? rms for our analyses has the bene? of making the sample somewhat homogeneous with respect to size. However, there are some disadvantages as well. First, it is likely to reduce the generalizability of our ? ndings. Table 1 reports that family ? rms in our sample operate in a broad array of industries, which should help alleviate to some extent concerns about the generalizability of our results. Second, the small sample reduces power of our tests and may prevent us from detecting certain effects. We address this issue by using 5 years of data, 1998–2002, under the assumption that family ? rm classi? ation is likely to be sticky. That is, we assume that the year 2002 classi? cation applies to the previous 4 years as well. 10 Finally, the test of each of our hypotheses requires data for different sets of variables. For each test, we include in the sample all ? rm-year observations spanning from 1998 to 2002 for which required data are available on Compustat, CRSP, or First Call’s Company Issued Guidance databases. For the test of hypothesis H3, we use the data available from the Standard and Poor’s Transparency and Disclosure database; these data are available for only 2002. 3. 1.

Descriptive statistics of characteristics of family and non-family ? rms Panel A of Table 2 provides descriptive statistics on the salient characteristics of family ? rms. We obtain this data from the 2002 proxy statements. On average, family members and/or descendants own 11% of cash ? ow rights and 18% of voting rights in family ? rms. Moreover, 61% of families hold at least 5% cash ? ow rights and 64% of families hold at least 5% voting rights. The differences in cash ? ow rights and voting rights statistics are 9 BusinessWeek adopts this de? nition of family ? rms from Anderson and Reeb (2003a).

In using this de? nition for our analyses, we do not try to exclude ? rms with limited in? uence of founding family. There are several bene? ts of staying with the BusinessWeek classi? cation. First, it is free of any subjective assessment of family in? uence, thus making the results more reliable. Second, to the extent a ? rm classi? ed as family ? rm has only a weak family in? uence, it would introduce a conservative bias in our results. Finally, this de? nition of family ? rm has been used by several recent academic studies on family ? rms (Anderson and Reeb, 2003a, b, 2004; Anderson et al. 2003), thus it makes comparison of our results with these other studies easier. 10 We examine the proxy statements of years 2000 and 2001 and ? nd that ? rms classi? ed as family ? rms in 2002 are family ? rms in years 2000 and 2001 as well. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 Table 1 Number and percent of family and non-family ? rms by two-digit SIC code in the S&P 500, 2002 SIC code Industry description Non-family ? rms (n ? 323 ? rms) 1 12 1 1 0 11 3 1 3 1 6 3 25 4 3 5 6 17 18 15 14 1 4 1 0 1 11 33 1 4 2 7 3 0 1 2 4 2 27 6 7 23 1 1 0 0 Family ? ms (n ? 177 ? rms) Percent of family ? rms in industry (%) 50 33 0 33 100 39 0 75 25 50 40 70 31 33 50 38 14 37 51 12 39 50 0 0 100 67 55 11 50 33 33 42 40 100 75 50 0 71 21 14 30 23 50 83 100 100 247 10 13 14 15 16 20 21 23 24 25 26 27 28 29 30 33 34 35 36 37 38 39 40 42 44 45 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 67 70 72 Metal mining Oil and gas extraction Manufacturing, quarry nonmaterial minerals General building contractors Heavy construction, except buildings Food and kindred products Tobacco products Apparel and other textile products Lumber and wood products Furniture and ? tures Paper and allied products Printing and publishing Chemical and allied products Petroleum and coal products Rubber and miscellaneous plastic products Primary metal industries Fabricated metal products Industrial machinery and equipment Electronic and other electrical equipment Transportation equipment Instruments and related products Miscellaneous manufacturing products Railroad transportation Trucking and warehousing Water transportation Transportation by air Communications Electric, gas, and sanitary services Wholesale trade—durable goods Wholesale trade—nondurable goods Building materials and gardening General merchandise stores Food stores Auto dealers and service stations Apparel and accessory stores Furniture and home furnishings Eating and drinking places Miscellaneous retail Depositing institutions Nondepositing credit institutions Security & commodity brokers Insurance carriers Insurance agents, brokers & service Holding, other investment of? ces Hotels and other lodging places Personal services 1 4 0 1 1 7 0 3 1 1 4 7 11 2 3 3 1 10 19 2 9 1 0 0 1 2 6 4 1 2 1 5 2 2 3 2 0 5 7 1 3 7 1 5 3 1 ARTICLE IN PRESS 248 Table 1 (continued ) SIC code Industry description Non-family ? rms (n ? 323 ? rms) 19 1 1 2 3 0 1 3 Family ? rms (n ? 177 ? ms) Percent of family ? rms in industry (%) 47 0 0 0 40 100 67 0 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 73 75 78 79 80 82 87 99 Business services Auto repair, services, and parking Motion pictures Amusement and recreation services Health services Educational services Engineering and management services Non-classi? cation establishment 17 0 0 0 2 1 2 0 due to the presence of dual class shares; 11% of family ? rms have dual class shares. The difference in voting and cash ? ow rights leads to greater Type II agency problems (Villalonga and Amit, 2006). Thus, we separately examine the disclosure practices of family ? ms with and without dual class shares as a sensitivity check of our conclusions about the relation between the severity of agency problems and disclosure practices. Family member is the CEO in 49% of the family ? rms: the founder is the CEO in 32% and descendant in 17% of family ? rms. The severity of agency problems differs across these two subsamples (Villalonga and Amit, 2006). 11 Thus, we separately examine the disclosure practices of these two subsamples as another sensitivity check of our conclusions about the relation between the severity of agency problems and disclosure practices. Finally, family members can exert their in? uence by holding other important positions. A founding family member or a descendant is a top level manager in 63% of family ? rms, is the chairperson in 67% of family ? ms and sits on the board of directors in 99% of family ? rms. The above characteristics of family ? rms suggest that on average family members exert a non-trivial in? uence on the ? rms that we consider as family ? rms. Panel B of Table 1 reports for family and non-family ? rms certain corporate governance characteristics that have been examined in prior studies (Dechow et al. , 1996; Anderson and Reeb, 2003a). In family ? rms, of? cer and directors own on average 12. 01% of stocks, whereas in non-family ? rms they own only 2. 87% of stocks. This result is consistent with the panel A results that family ? rms almost always have family members in of? er and/or director positions and family members have concentrated ownership in their ? rms. Panel B also shows that the percentage ownership by outside blockholders is 10. 18% in family ? rms and 12. 29% in non-family ? rms. Moreover, 57% of family ? rms and 68% of non-family ? rms have at least one outside blockholder. Outside blockholders are unaf? liated owners holding at least 5% of the ? rm’s outstanding shares. Panel B also shows that in family ? rms 63% of directors are independent, whereas in non-family ? rms 76% are independent. The higher percentage of outside blockholding and independent directors suggest that these two factors would contribute toward non-family ? rms having less severe agency problems.

Finally, the CEO is the chairman of the board in 65% of family ? rms and in 81% of the non-family ? rms, suggesting that this factor contributes Villalonga and Amit (2006) do not examine whether the less severe agency problem when founder is the CEO is because of lower Type I or lower Type II agency problems. 11 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 Table 2 Descriptive statistics of family and non-family ? rms in S&P 500, 2002 Panel A: Ownership and control characteristics of the 177 family ? rms in S&P 500 Percentage of cash ? ow rights controlled by the founding family members or descendents, mean median ? 6. 26%; ? rst quartile ? 2. 1%; third quartile ? 14. 10% Percentage of families holding at least 5% cash ? ow rights Percentage of voting rights controlled by the founding family members or descendents, mean Median ? 9. 40%; ? rst quartile ? 3. 70%; third quartile ? 19. 60% Percentage of families holding at least 5% voting rights Percentage of family ? rms with dual class shares Percentage of family ? rms in which founder is the CEO Percentage of family ? rms in which descendent is the CEO Percentage of family ? rms in which hired executive is the CEO Percentage of family ? rms in which a founding family member or a descendent is a top executive (including CEO) Percentage of family ? ms in which a founding family member or a descendent is the chairperson of the board of directors Percentage of family ? rms in which a founding family member or a descendent is a director (including chairperson) Mean Family ? rms Non-Family ? rms Difference tstatistics Median Family ? rms Non-Family ? rms Difference zstatistics ARTICLE IN PRESS 11% 61% 18% 64% 11% 32% 17% 51% 63% 67% 99% Panel B: Corporate governance characteristics of family and non-family ? rms Of? cers and directors ownership 12. 01% 2. 87% Unaf? liated blockholding 10. 18% 12. 29% Percentage of ? rms with outside 57% 68% blockholders 9. 95*** A3. 01*** A5. 11*** 7. 31% 6. 40% 1. 00 1. 70% 9. 25% 1. 00 10. 01*** A4. 52*** A4. 37*** 249 250

Table 2 (continued ) Mean Family ? rms Non-Family ? rms 76% 81% 323 Difference tstatistics A7. 82*** A3. 55*** Median Family ? rms Non-Family ? rms 75% 1. 00 323 Difference zstatistics A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 A8. 62*** A3. 23*** Percentage of independent directors CEO ? COB No. of observations 63% 65% 177 Mean Family ? rms 64% 1. 00 177 Median ARTICLE IN PRESS Non-Family ? rms Difference tstatistics Family ? rms Non-Family ? rms Difference zstatistics Panel C: Financial characteristics of family and non-family ? rms SIZE 8. 92 9. 01 MB 5. 38 4. 28 0. 06 0. 04 ROA PROA 0. 05 0. 04 No. of observations 177 323 A1. 58 2. 71*** 3. 5*** 4. 76*** 8. 87 3. 55 0. 06 0. 04 177 8. 92 2. 69 0. 04 0. 03 323 A1. 57 2. 71*** 5. 11*** 5. 37*** Variable de? nitions: Of? cers and directors ownership is the equity holdings of all of? cers and directors. Unaf? liated blockholding is the fractional equity stake of unaf? liated owners holding at least ? ve percent of the ? rm’s outstanding shares. Percentage of ? rms with outside blockholders is the percentage of ? rms with at least one unaf? liated owner holding at least ? ve percent of the ? rm’s outstanding shares. Percentage of independent directors is the number of independent directors serving on the board divided by board size. CEO ?

COB is a dummy variable which equals one if CEO is also the chairperson of the board and is zero otherwise. SIZE is the log of market value of equity at the beginning of the ? scal period. MB is a ? rms’ market-to-book ratio de? ned as the market value of equity divided by book value of equity. ROA is earnings before extraordinary item divided by total assets. PROA is the average of prior 5 years’ earnings before extraordinary items divided by the average of prior 5 years’ total assets. . *** indicates signi? cance at the 0. 01 level. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 251 towards non-family ? rms having more severe agency problems.

The above results raise a concern that the difference in disclosure practices that we may observe across family and non-family ? rms could be driven by these other governance factors. To alleviate this concern, we check the sensitivity of our results to controlling the effect of these factors on disclosure practices. Panel C of Table 1 shows that compared to non-family ? rms, family ? rms have better pro? tability. Current period’s return on asset (ROA) and the average return on assets for the prior 5 years (PROA) are signi? cantly greater for family ? rms (t ? 3. 15 and 4. 76, respectively). The market to book ratio (MB) is signi? cantly higher for family ? rms (t ? 2. 71), suggesting that the market views these ? rms to be more pro? able (Villalonga and Amit, 2006). These results are consistent with those reported by Anderson and Reeb (2003a). They attribute these results to less severity in agency problems in family ? rms. They note however that these results are also consistent with the following alternative explanation: Families have superior information about their ? rms’ future prospects and they tend to exit ? rms with poor prospects. In other words, the better performance observed for family ? rms may not be due to less severe agency problems but due to family members continuing only in ? rms with better prospects. Given the survivorship bias due to a ? rm remaining family ? m if performing well and given that better performance is associated with better disclosure (Miller, 2002), there is a potential for spurious correlation between family ? rm membership and disclosure quality. To alleviate this concern, we add measures of ? rm performance, ROA and PROA, as control variables in all our models of disclosure quality. Any association between disclosure quality and the family ? rm indicator variable can then be more reliably attributed to the difference in agency problems across family and non-family ? rms. Of course, to the extent that performance is not adequately controlled for by ROA and PROA variables, our results could be spurious. 4. Results 4. 1.

Earnings quality We assess the quality of earnings in the following four ways: the level of discretionary accruals in earnings, the ability of earnings components to predict future cash ? ows, the persistence of earnings, and the association of earnings with contemporaneous stock returns. 4. 1. 1. Discretionary accruals We estimate the following models to examine the relation between discretionary accruals and whether a ? rm is a family or a non-family ? rm. The models are similar to that used by Ashbaugh et al. (2003). To their model, we add the family ? rm membership variable and control variables used in other studies (e. g. War? eld et al. , 1995). ABSPADCA ? a ? b1 FAMILYFIRM ? b2 L1ACCRUAL ? b3 SIZE ? b4 MA ? b5 FINANCING ? b6 LITIGATION ? b7 LEVERAGE ? b8 MB ? b9 LOSS ? b10 CFO ? b11 INSTITUTION ? 12 VAR ? b13 BETA X ri INDUSTRY i ? error, ? 1? ? b14 ROA ? b15 PROA ? ARTICLE IN PRESS 252 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 PADCA ?? a ? b1 FAMILYFIRM ? b2 L1ACCRUAL ? b3 SIZE ? b4 MA ? b5 FINANCING ? b6 LITIGATION ? b7 LEVERAGE ? b8 MB ? b9 LOSS ? b10 CFO ? b11 INSTITUTION ? b12 VAR ? b13 BETA ? b14 ROA ? b15 PROA X ? ri INDUSTRY i ? error. ?2? The dependent variable in Eq. (1) is ABSPADCA, which stands for absolute value of performance adjusted discretionary accruals. The dependent variable in Eq. (2), PADCA, is the same as ABSPADCA, except that it is not transformed to absolute value. We follow Kothari et al. 2005) for measuring PADCA. We ? rst estimate the modi? ed Jones model cross-sectionally using all ? rm-year observations in the same two-digit SIC code. Discretionary accruals from this model are then differenced with discretionary accruals of a ? rm with the same two-digit SIC code and with the closest return on assets in the current year. Other variables are de? ned as follows. FAMILYFIRM is a dummy variable which equals one for family ? rms, and zero otherwise; L1ACCRUAL is last year’s total current accruals and equals net income before extraordinary items plus depreciation and amortization minus operating cash ? ow scaled by beginning of year total assets.

This variable captures the reversal of accruals over time. SIZE is the log of a ? rm’s market capitalization. MA is 1 if the ? rm has engaged in a merger and acquisition, and 0 otherwise. FINANCING is 1 if MA is not equal to 1 and number of outstanding shares increased by at least 10%, or long-term debts increased at least 20%, or the ? rm ? rst appears on the CRSP monthly returns database during the ? scal year, and 0 otherwise. LITIGATION is 1 if the ? rm operates in a high-litigation industry, and 0 otherwise; LEVERAGE is the ratio of total debt to total assets at the beginning of the ? scal period; MB is market-to-book ratio; LOSS is 1 if the ? rm reports a net loss for the ? cal period, and 0 otherwise; CFO is cash ? ow from operations scaled by beginning of year total assets; INSTITUTION is the percentage of stocks held by institutional investors; VAR is the standard deviation of quarterly earnings for the period 1997–2002; BETA is systematic risk; ROA is current year’s return on assets; PROA is prior 5 years’ return on assets; and INDUSTRYi is a dummy variable for industry membership. We use the 12 industry groups in Fama and French (1997). 12 Panel A of Table 3 reports the descriptive statistics of the variables in Eqs. (1) and (2). ABSPADCA is signi? cantly greater for family ? rms (t ? 1. 96, z ? 2. 25). However, PADCA is signi? cantly maller for family ? rms (t ? A4. 31, z ? A4. 44). Many of the control variables are signi? cantly different across family and non-family ? rms. Thus, it is important to control for these variables to draw proper conclusion about the relation between discretionary accruals and family ? rm membership. 12 For estimating discretionary accruals and the predictability of cash ? ows (Section 4. 1. 2), we use industry groups based on the two-digit SIC codes. Prior studies have done the same. We use all the ? rms in the COMPUSTAT for which the required data is available in the estimation process and therefore have a reasonable number of ? rms in each industry group.

However, when estimating the difference in disclosure practices across family and non-family ? rms, our sample is limited to S&P500 ? rms. Using two-digit SIC codes for de? ning industry results in too few ? rms in some of the industry groups. Thus, we use the Fama-French industry de? nition for these tests. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 253 Panel B of Table 3 presents the regression estimates of Eqs. (1) and (2). 13 The coef? cient on FAMILYFIRM is insigni? cant in the ABSPADCA model (A0. 64, t ? A0. 05) and is negative and signi? cant in the PADCA model (A1. 36, t ? A1. 98). 14 The coef? cients on the control variables, whenever signi? ant, are consistent with the signs predicted by prior studies (Ashbaugh et al. , 2003; War? eld et al. , 1995). 15 Overall, the results suggest that discretionary accruals are more negative for family ? rms as compared to non-family ? rms. Assuming that on average mangers have incentives to increase income, this result seems consistent with less opportunistic behavior in family ? rms. 16 The greater negative accruals by family ? rms may also be motivated by the desire to minimize tax or reduce political costs. If these factors drive the result then it would lead to lower quality of earnings in terms of value relevance. However, in the subsequent tests if we ? nd that family ? ms’ earnings are better at predicting future cash ? ows, have higher persistence, and have higher association with contemporaneous returns, then it would suggest that less severe opportunistic behavior is primarily responsible for the more negative discretionary accruals in family ? rms. 4. 1. 2. Predictability of cash ? ows Following Dechow et al. (1998), Barth et al. (2001), and Cohen (2004), we assess the quality of reported earnings by examining the ability of its components to predict future cash ? ows. Speci? cally, we use the residuals obtained from the regression of future cash ? ow from operations on prior period’s earnings components.

We estimate the following equation: CF Oit? 1 ? a0 ? a1 CF Oit ? a2 DARit ? a3 DINV it ? a4 DAPit ? a5 DEPRit ? a6 OTHERit ? eit? 1 , ?3? 13 For all model estimations in the paper, we use the Huber-White procedure. Also, throughout the paper, our conclusions about the effect of family ? rm membership are robust to outlier deletions as well as the use of binary transformation of control variables. Finally for all the models in the paper, we carry out year-by-year estimations. We ? nd that the year-by-year coef? cients in most cases have signs consistent with that reported for the pooled regression. In the few cases where????? signs are??????????????? the p not consistent, the coef? cients are never statistically signi? cant. We P also compute Z-statistic ? ?1= N ? N tj = kj =kjA2 , where t is t-statistic, k is the degrees of freedom for year j, j? 1 and N is the number of years. This statistic controls for the effect of time-series correlation of the variables. Our conclusions remain unaffected. 14 If family ? rms have younger assets, these ? rms may report higher depreciation and/or amortization expenses. To the extent that the property, plant, and equipment variable in the Jones model does not completely control for this effect, family ? rms’ discretionary accruals estimates would be more negative.

To alleviate this concern, we repeat our analysis after adding depreciation and amortization expenses to total accruals. The results lead to the same conclusion. 15 We ran all the analysis in the paper with the corporate governance variables as additional controls. Speci? cally, we include the following variables: the percentage of stocks held by outside blockholders (or an indicator variable for a large outside blockholder), percentage of directors who are independent, and an indicator variable on whether the CEO is also the chairman of the Board of Directors. Given that we have only three years (2000–2002) of data for these variables readily available to us, e estimate our models for the three years. We ? nd that our conclusions with respect to the FAMILYFIRM variable remain unchanged. Moreover, none of these additional control variables are signi? cant in any of the models. Given that we do not have data for these additional variables for all the ? ve sample years, we do not report these results in the text. 16 A more compelling test for opportunistic behavior would be to identify situations (? rm-years) where there would be income increasing or income decreasing incentives and then examine separately for each category the difference in discretionary accruals across family and non-family ? rms. 254 Table 3 Family ? ms and discretionary accruals, 1998–2002 Mean Family ? rms Panel A: Descriptive statistics ABSPADCA (%) PADCA (%) L1ACCRUAL SIZE MA FINANCING LITIGATION LEVERAGE MB LOSS CFO INSTITUTION VAR BETA ROA PROA No. of Observations Variables Non-family ? rms Difference t-stat. Median Family ? rms Non-family ? rms Difference z-stat. A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 7. 55 A3. 11 A7. 14 8. 89 0. 25 0. 16 0. 38 0. 23 5. 61 0. 13 0. 15 0. 61 0. 41 1. 07 0. 06 0. 05 593 Predicted sign 6. 99 A1. 12 A4. 32 8. 94 0. 20 0. 24 0. 23 0. 29 4. 31 0. 14 0. 13 0. 67 0. 54 0. 84 0. 04 0. 04 1009 1. 96** A4. 31*** A1. 97** A1. 51 2. 01** A3. 65*** 5. 72*** A7. 30*** 5. 2*** A0. 57 4. 44*** A3. 58*** A5. 11*** 7. 51*** 4. 24*** 4. 76*** 5. 55 A2. 57 A1. 97 8. 86 0. 00 0. 00 0. 00 0. 20 3. 71 0. 00 0. 14 0. 62 0. 27 0. 96 0. 06 0. 04 593 4. 68 A1. 07 A1. 21 8. 91 0. 00 0. 00 0. 00 0. 29 2. 85 0. 00 0. 12 0. 69 0. 40 0. 80 0. 04 0. 03 1009 2. 25** A4. 44*** A1. 59 A1. 01 2. 09** A3. 62*** 5. 85*** A8. 01*** 6. 63*** A0. 57 4. 49*** A3. 71*** A7. 29*** 8. 22*** 5. 01*** 5. 37*** ARTICLE IN PRESS Dependant var. ? ABSPADCA Coeff. t-stat. Predicted sign Dependant var. ? PADCA Coeff. t-stat. Panel B: Regression estimates Intercept FAMILYFIRM L1ACCRUAL SIZE MA FINANCING LITIGATION ? ? A A + + + 5. 41 A0. 64 A1. 02 A0. 1 0. 46 0. 92 A1. 69 2. 23** A0. 05 A1. 33 A0. 03 0. 78 1. 68* A1. 41 ? ? A A + + + 2. 62 A1. 36 A2. 35 A0. 01 1. 11 0. 35 1. 00 0. 75 A1. 98** A1. 43 0. 11 1. 12 0. 08 0. 52 LEVERAGE MB LOSS CFO INSTITUTION VAR BETA ROA PROA Adjusted R2 (%) No. of observations A + + A A + + ? ? A1. 97 0. 27 0. 82 7. 78 A1. 15 0. 47 0. 38 A7. 22 A0. 05 A1. 20 4. 72*** 3. 01*** 1. 49 A0. 21 3. 38*** 1. 45 A1. 54 A0. 02 12. 81 1602 A + A A A ? ? ? ? A1. 37 A0. 03 A1. 28 A27. 75 A1. 01 0. 19 A0. 37 8. 42 3. 60 5. 62 A0. 52 A0. 55 A1. 31 A8. 21*** A0. 35 1. 25 A0. 62 1. 56 0. 32 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 1602 Variable de? itions: PADCA is the performance-matched modi? ed-Jones model discretionary accruals. ABSPADCA is the absolute value of PADCA; FAMILYFIRM is a dummy variable which equals one for family ? rms, and zero otherwise; L1ACCRUAL is last year’s total current accruals scaled by beginning of year total assets. SIZE is the log of a ? rm’s market capitalization. MA is 1 if the ? rm has engaged in a merger and acquisition, and 0 otherwise. FINANCING is 1 if MA is not equal to 1 and number of outstanding shares increased by at least 10%, or long-term debts increased at least 20%, or the ? rm ? rst appears on the CRSP monthly returns database during the ? scal year, and 0 otherwise.

LITIGATION is 1 if the ? rm operates in a high-litigation industry (SIC codes of 2833–2836, 3570–3577, 3600–3674, 5200–5961, and 7370), and 0 otherwise. LEVERAGE is the ratio of total debt to total assets at the beginning of the ? scal period. MB is a ? rms’ market-to-book ratio. LOSS is 1 if it reports a net loss in the ? scal period, and 0 otherwise. CFO is cash ? ow from operations scaled by beginning of year total assets. INSTITUTION is the percentage of stocks held by institutional investors. VAR is the standard deviation of quarterly earnings for the period 1997–2002; BETA is systematic risk. ROA is earnings before extraordinary item divided by total assets.

PROA is the average of prior 5 years’ earnings before extraordinary items divided by the average of prior 5 years’ total assets. The regression model includes dummy variables for industry membership. We use the Fama-French de? nition of industry. For brevity, we do not report the industry dummy coef? cients. The predicted signs on the control variables are based on prior studies. The t-statistics are corrected using the Huber-White procedure. *** indicates signi? cance at the 0. 01 level, ** indicates signi? cance at 0. 05 level, and * indicates signi? cance at the 0. 10 level. ARTICLE IN PRESS 255 ARTICLE IN PRESS 256 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 where CFOit is the cash ? w from operations for ? rm i in year t minus the accrual portion of extraordinary items and discontinued operations; DARit is change in accounts receivable; DINVit is change in inventory; DAPit is change in accounts payable and accrued liabilities; DEPRit is depreciation and amortization expense; and OTHERit is net of all other accruals, calculated as (EARNA(CFO+DAR+DINVADAPADEPR)), where EARN is income before extraordinary items and discontinued operations. We estimate Eq. (3) for each ? scal year separately from 1998 to 2002 for each two-digit SIC industry code with at least 20 observations, and use the estimated coef? cients to calculate ? rm-speci? c residuals.

The empirical measure of reporting quality is the absolute value of these residuals: RES ? |eit+1|. These residuals re? ect the magnitude of future operating cash ? ows unrelated to current disaggregated earnings. Lower absolute values of the residuals indicate higher quality ? nancial reporting. 17 To examine the association between earnings quality and family ? rm membership, we estimate the following equation. The control variables in this model are from Cohen (2004). They capture the various costs and bene? ts associated with disclosing high-quality ? nancial information. QUALITY ? a ? b1 FAMILYFIRM ? b2 OWNER ? b3 CAPITAL ? b4 HERFINDEX ? b5 SALESGROW ? b6 MARGIN ? b7 LEVERAGE ? b8 OC ? b9 SEGMENT ? b10 SIZE ? 11 ROA X ri INDUSTRY i ? error, ? b12 PROA ? ?4? where the dependent variable, QUALITY, is a binary variable which equals 1 if RES is less than the median value of RES. FAMILYFIRM is a binary variable which equals 1 if the ? rm is a family ? rm, and 0 otherwise. OWNER is the natural log of the number of shareholders of a ? rm minus the natural log of median number of shareholders for the same two-digit SIC code; CAPITAL is net plant, property and equipment divided by total assets; HERFINDEX is the Her? ndahl Index, calculated as the sum of squares of market shares of the ? rms in the industry (two-digit SIC code); SALESGROW is current year’s rowth in sales; MARGIN is gross margin percentage; LEVERAGE is long term debt plus debt in current liabilities divided by total assets; OC is operating cycle (in days) and is calculated as [(ARt+ARtA1)/ 2C(SALES/360)]+[(INVt+INVtA1)/2C(COGS/360)] where AR is the ? rm’s accounts receivable, INV is inventory, and COGS is cost of goods sold; SEGMENT is the number of two-digit SIC industry codes the ? rm operates in; SIZE is natural logarithm of market capitalization at the end of the ? scal year; ROA is current year’s return on assets; PROA is prior 5 years’ return on assets; and INDUSTRY is a dummy variable for industry membership. We use the 12 industry groups in Fama and French (1997). 7 The mean (median) number of observations per industry to estimate Eq. (3) for industries represented by family ? rms is 602 (264) and for industries represented by non-family ? rms is 657 (281). The average number of observations per industry is large and is of similar order of magnitude across the two groups. Thus, it is unlikely that our results are driven by the downward bias in the RES variable for one of the group, because of an over ? t of Eq. (3) due to too few observations in some of the industries represented by that group. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 257 Cohen (2004) provides the following arguments for the explanatory variables in Eq. (2).

OWNER and LEVERAGE capture the higher demand for ? rm-speci? c information. CAPITAL, HERFINDEX, SALESGROW and MARGIN capture proprietary costs of disclosures. OC captures the predictability of future cash ? ows resulting from the length of operating cycle. SEGMENT captures the effect of the complexity of the ? rm’s operating environment on information quality. SIZE captures the difference in ? rms’ information environment along with other aspects. Finally, ROA and PROA control for the effect of pro? tability. Table 4, panel A provides the descriptive statistics of the variables in Eq. (4). QUALITY is not signi? cantly different across family and non-family ? ms in the univariate tests. However, several of the control variables are signi? cantly different across family and nonfamily ? rms. Thus, it is important to control for these variables to draw proper conclusions on the relation between earnings quality and family ? rm membership. The results of estimating Eq. (4) are presented in panel B of Table 4. The coef? cient on the FAMILYFIRM is positive and signi? cant (0. 21, w2 ? 3. 81). The coef? cients on the control variables, when signi? cant, have the signs as predicted by prior studies (Cohen, 2004); the only exception is SALESGROW. 18 Overall, the results suggest that compared to non-family ? rms, family ? ms’ earnings components are signi? cantly better at predicting future cash ? ows. 4. 1. 3. Earnings persistence Another commonly used measure of earnings quality is its persistence. We measure earnings persistence for a ? rm by estimating the following time-series model for the period 1995–2002: DEPS t ? l0 ? l1 DEPS tA1 ? error, (5) where DEPSt is the change of earnings before extraordinary items divided by the number of outstanding shares. The slope coef? cient, l1, represents the persistence of earnings, referred to as PERSISTENCE. To examine whether PERSISTENCE varies across family and non-family ? rms, we estimate the following model. PERSISTENCE ? a ? 1 FAMILYFIRM ? b2 SIZE ? b3 ROA X ri INDUSTRY i ? error. ? b4 PROA ? ?6? The control variables in this model are based on Lev (1983). He shows that earnings persistence is associated with ? rm size and various industry characteristics, such as, type of products, degree of competition, and operating leverage. We use industry membership indicator variable to capture the industry characteristics. We use 12 industry groups as in Fama and French (1997). ROA and PROA, de? ned earlier, control for the effect of pro? tability on PERSISTENCE. Panel A of Table 5 reports descriptive statistics of the variables in Eq. (6). The mean value of PERSISTENCE is signi? antly greater for family ? rms (t ? 1. 97). Panel B of Table 5 reports the regression estimate of Eq. (6). The coef? cient on 18 The coef? cient on FAMILYFIRM remains signi? cant when one control variable at a time is excluded from Eq. (4). This result suggests that the insigni? cant difference in QUALITY across family and non-family ? rms in the univariate tests is due to a correlated omitted variable bias associated with a combination of variables. ARTICLE IN PRESS 258 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 Table 4 Family ? rms and predictability of future cash ? ows, 1998–2002 Mean Family ? rm Non-family ? rm Difference t-stat.

Median Family ? rm Non-family ? rm Difference z-stat. Panel A: Descriptive statistics QUALITY OWNER CAPITAL HERFINDEX SALESGROW (%) MARGIN (%) LEVERAGE OC SEGMENT SIZE ROA PROA No. of observations 0. 52 2. 38 0. 30 0. 06 11. 03 41. 22 0. 23 142. 62 5. 07 8. 83 0. 06 0. 05 671 0. 49 3. 30 0. 36 0. 06 7. 77 37. 96 0. 29 231. 77 5. 92 9. 01 0. 04 0. 04 1165 0. 97 A9. 58*** A5. 63*** 0. 65 3. 30*** 3. 37*** A7. 30*** A2. 77*** A6. 14*** A1. 55 4. 24*** 4. 76*** 1. 00 2. 27 0. 24 0. 04 11. 04 37. 91 0. 20 105. 70 5. 00 8. 93 0. 06 0. 04 671 0. 00 3. 35 0. 30 0. 04 6. 71 35. 27 0. 29 106. 78 6. 00 8. 86 0. 04 0. 03 1165 0. 97 A11. 78*** A6. 22*** 1. 5* 5. 73*** 2. 86** A8. 01*** A0. 81 A7. 31*** 1. 25 5. 01*** 5. 37*** Dependent var. ? QUALITY Variables Panel B: Logistic model estimates Intercept FAMILYFIRM OWNER CAPITAL HERFINDEX SALESGROW MARGIN LEVERAGE OC SEGMENT SIZE ROA PROA Likelihood ratio (p-value) No. of observations Predicted sign Coeff. Wald w2 Marginal probability ? ? + + ? A ? + A ? + ? ? A0. 73 0. 21 A0. 02 1. 34 2. 46 0. 46 A1. 11 0. 71 0. 01 0. 03 0. 14 A4. 44 A4. 05 1. 96 3. 81** 0. 33 22. 17*** 6. 21** 3. 08* 12. 41*** 3. 33** 0. 81 2. 21 6. 11** 28. 66*** 7. 91*** 358. 64 0. 00 1836 A 0. 12 0. 00 0. 28 0. 51 0. 10 A0. 23 0. 15 0. 00 0. 01 0. 03 A0. 92 0. 16 Variable de? itions: QUALITY is a binary variable which equals 1 if RES is less than the median value of RES, where RES is the absolute value of the residual obtained from a regression of future cash ? ow from operation on prior period’s earnings components (see Eq. (1)); FAMILYFIRM is a dummy variable which equals one for family ? rms, and zero otherwise; OWNER is the natural log of the number of shareholders of a ? rm minus the natural log of median number of shareholders for the same two-digit SIC code; CAPITAL is net plant, property and equipment divided by total assets; HERFINDEX is the Her? ndahl Index, calculated as the sum of squares of market shares of the ? ms in the industry (two-digit SIC code); SALESGROW is current year’s growth in sales; MARGIN is gross margin percentage; LEVERAGE is long term debt plus debt in current liabilities divided by total assets; OC is operating cycle (in days) and is calculated as [(ARt+ARtA1)/2C(SALES/360)]+[(INVt+INVtA1)/2C(COGS/360)], where AR is accounts receivable, INV is inventory , and COGS is cost of goods sold; SEGMENT is the number of two-digit SIC industry codes the ? rm operates in; SIZE is natural logarithm of market capitalization at the end of the ? scal year. ROA is earnings before extraordinary item divided by total assets. PROA is the average of prior 5 years’ earnings before extraordinary items divided by the average of prior 5 years’ total assets. The regression model includes dummy variables for industry membership. We use the Fama-French de? nition of industry. For brevity, we do not report the industry dummy coef? cients. The predicted signs on the control variables are based on prior studies.

The w2s are corrected using the Huber-White procedure. *** indicates signi? cance at the 0. 01 level, ** indicates signi? cance at 0. 05 level, and * indicates signi? cance at the 0. 10 level. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 Table 5 Family ? rms and earnings persistence, 1998–2002 Mean Family ? rms Panel A: Descriptive statistics PERSISTENCE A0. 21 SIZE 8. 89 ROA 0. 06 PROA 0. 05 No. of observations Variables 173 Non-family ? rms Difference t-stat. Median Family ? rms Non-family ? rms Difference z-stat. 259 A0. 32 8. 94 0. 04 0. 04 314 1. 97** A1. 51 3. 71*** 4. 21*** A0. 27 8. 87 0. 06 0. 04 173 A0. 33 8. 91 0. 04 0. 03 314 1. 31 A1. 3 4. 84*** 4. 67*** Dependent var. ? PERSISTENCE Predicted sign Coeff. t-stat. Panel B: Regression estimates Intercept FAMILYFIRM SIZE ROA PROA Adjusted R2 (%) No. of observations ? ? + ? ? A0. 38 0. 09 0. 02 A0. 71 0. 25 3. 32 487 A2. 06** 1. 52 1. 33 A1. 97** 0. 35 Variable de? nitions: FAMILYFIRM is a dummy variable which equals one for family ? rms, and zero otherwise; PERSISTENCE is the slope coef? cient, l1, from the following time-series model: DEPSt ? l0+l1DEPStA1+error; DEPSt is the change of earnings before extraordinary items divided by the number of outstanding shares. The model is estimated from 1995 to 2002 to yield ? rm-speci? l1; SIZE is the log of market value of equity at the beginning of the ? scal period. ROA is earnings before extraordinary item divided by total assets. PROA is the average of prior 5 years’ earnings before extraordinary items divided by the average of prior 5 years’ total assets. The regression model includes dummy variables for industry membership. We use the Fama-French de? nition of industry. For brevity, we do not report the industry dummy coef? cients. The predicted signs on the control variables are based on prior studies. The t-statistics are corrected using the Huber-White procedure. *** indicates signi? cance at the 0. 01 level, ** indicates signi? cance at 0. 05 level, and * indicates signi? ance at the 0. 10 level. FAMILYFIRM is positive but not signi? cant (0. 09, t ? 1. 52). Even the coef? cient on SIZE is not signi? cant. Given that prior studies ? nd a signi? cantly positive coef? cient on SIZE, the insigni? cant results we obtain could be due to the lack of power of our test. 4. 1. 4. Earnings response coef? cient In the above sections, we measure earnings quality in terms of the predictability of only next period’s cash ? ows or persistence with respect to only next period’s earnings. Earnings response coef? cient (ERC) captures the ability of earnings to predict future cash ? ows and the persistence of earnings more comprehensively.

To test the difference between ERCs of family and non-family ? rms, we estimate the following ARTICLE IN PRESS 260 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 equation: RETURN ? a ? b1 EARNINGS ? b2 FAMILYFIRM ? b3 EARNINGS A FAMILYFIRM ? b4 EARNINGS A VAR ? b5 EARNINGS A LEVERAGE ? b6 EARNINGS A MB ? b7 EARNINGS A SIZE ? b8 EARNINGS A BETA ? b9 EARNINGS A ROA X b11 EARNINGS ? b10 EARNINGS A PROA ? A INDUSTRY i ? error, ? 7? where RETURN is the cumulative abnormal return for the 12-month period ending 3 months after the ? scal year end; FAMILYFIRM is a dummy variable which equals one for family ? rms, and zero otherwise; EARNINGS is the annual change in earnings per share de? ted by the price at the beginning of the return accumulation period; VAR is the standard deviation of quarterly earnings for the period 1997–2002;19 LEVERAGE is the ratio of total debt to total assets at the beginning of the ? scal period; MB is market-tobook ratio at the beginning of the ? scal period; SIZE is the log of market value of equity at the beginning of the ? scal period; BETA is systematic risk; ROA is current year’s return on assets; PROA is prior 5 years’ return on assets; and INDUSTRY is a dummy variable for industry membership. We use 12 industry groups as in Fama and French (1997). We predict that b3 will be positive, indicating that the ERC of family ? rms is greater than that of non-family ? rms. Other interaction variables in Eq. 7) control for previously identi? ed determinants of ERCs (see, e. g. , Collins and Kothari, 1989; Kothari, 2001). The descriptive statistics of the variables in Eq. (7) are presented in panel A of Table 6. All of the determinants of ERC are signi? cantly different across family and non-family ? rms. Thus, it is important to control for these variables. The regression results are presented in panel B of Table 6. The ERC of family ? rms is signi? cantly higher than that of non-family ? rms both with and without the control variables. For the full model, the coef? cient on the interaction term, EARNINGS A FAMILYFIRM is 1. 25 (t ? 4. 37). The coef? ients on the control variables, when signi? cant, have the predicted signs, except for the coef? cient on EARNINGS A BETA. The results in this table are consistent with that in Tables 3–5, suggesting that as compared to non-family ? rms, family ? rms’ earnings are of higher quality, thereby providing support to hypothesis H1a. 4. 2. Management forecasts of earnings We examine the likelihood of management issuing quarterly earnings forecasts across family and non-family ? rms. For this purpose, we use the data on quarterly earnings guidance obtained from Thompson First Call’s, Company Issued Guidance (CIG) ? le. We use a model similar to that in Kasznik and Lev (1995).

MGMT_FORECAST ? a ? b1 CHEPS ? b2 FAMILYFIRM ? b3 CHEPS A FAMILYFIRM ? b4 SIZE ? b5 BM ? b6 HIGHTECH ? b7 REGULATION ? b8 ROA ? b9 PROA ? error, ? 8? We use the standard deviation of the prior sixteen quarters’ earnings to measure VAR and ? nd qualitatively similar results. 19 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 Table 6 Family ? rms and earnings response coef? cients, 1998–2002 Mean Family ? rms Panel A: Descriptive statistics RETURN 0. 00 EARNING A0. 01 VAR 0. 41 LEVERAGE 0. 23 MB 5. 04 8. 95 SIZE BETA 1. 07 ROA 0. 06 PROA 0. 05 No. of observations 852 Non-family ? rms t-stat. Median Family ? rms Non-family ? ms z-stat. ARTICLE IN PRESS 0. 00 A0. 01 0. 54 0. 29 3. 90 9. 02 0. 84 0. 04 0. 04 1450 0. 45 0. 37 A5. 11*** A12. 11*** 5. 82*** A1. 98** 7. 51*** 4. 24*** 4. 76*** A0. 05 0. 00 0. 27 0. 23 3. 61 8. 92 0. 96 0. 06 0. 04 852 Dependent var. ? RETURN Coeff. t-stat. A0. 05 0. 00 0. 40 0. 28 2. 69 8. 91 0. 80 0. 04 0. 03 1450 A0. 12 0. 58 A9. 29*** A12. 63*** 6. 63*** A2. 03** 7. 22*** 5. 01*** 5. 37*** Dependent var. ? RETURN Variables Panel B: Regression estimates Intercept EARNINGS FAMILYFIRM EARNINGS A FAMILYFIRM Predicted sign Coeff. t-stat. Dependent var. ? RETURN Coeff. t-stat. ? + ? ? A0. 19 0. 79 A3. 39*** 15. 51*** A0. 20 0. 74 A0. 01 0. 92 A4. 8*** 13. 92*** A0. 62 3. 44*** A0. 19 0. 78 A0. 01 1. 25 A7. 23*** 0. 82 A0. 06 4. 37*** 261 262 Table 6 (continued ) Dependent var. ? RETURN Variables EARNINGS A VAR EARNINGS A LEVERAGE EARNINGS A MB EARNINGS A SIZE EARNINGS A BETA EARNINGS A ROA EARNINGS A PROA Adjusted R2 (%) No. of observations Predicted sign A A + + A ? ? 20. 81 2302 21. 01 2302 Coeff. t-stat. Dependent var. ? RETURN Coeff. t-stat. Dependent var. ? RETURN Coeff. A0. 11 0. 62 0. 19 A0. 14 0. 12 2. 09 A6. 42 t-stat. A1. 39 0. 83 2. 86*** A0. 76 1. 41 1. 42 A1. 51 24. 79 2302 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 ARTICLE IN PRESS Variable de? itions: RETURN is the cumulative abnormal return for the 12–month period ending three months after the ? scal year end; FAMILYFIRM is a dummy variable which equals one for family ? rms, and zero otherwise; EARNINGS is the annual change in earnings per share de? ated by the price at the beginning of the return accumulation period; VAR is the standard deviation of quarterly earnings for the period 1997–2002; LEVERAGE is the ratio of total debt to total assets at the beginning of the ? scal period; MB is market-to-book ratio at the beginning of the ? scal period; SIZE is the log of market value of equity at the beginning of the ? scal period; BETA is systematic risk. ROA is earnings before extraordinary item divided by total assets.

PROA is the average of prior 5 years’ earnings before extraordinary items divided by the average of prior 5 years’ total assets. The full regression model includes interaction of EARNINGS with dummy variables for industry membership (see Eq. (7)). We use the Fama-French de? nition of industry. For brevity, we do not report the coef? cients on the industry dummy interaction variables. The predicted signs on the control variables are based on prior studies. The t-statistics are corrected using the Huber-White procedure. *** indicates signi? cance at the 0. 01 level, ** indicates signi? cance at 0. 05 level, and * indicates signi? cance at the 0. 0 level. ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 263 where MGMT_FORECAST is an indicator variable that takes the value of one if the managers make an earnings forecast of quarterly earnings, and zero otherwise, CHEPS is the change in earnings per share from that of the same quarter in the previous ? scal year, de? ated by stock price at the beginning of the quarter; SIZE is the natural log of market capitalization at the beginning of the ? scal quarter; BM is the natural log of the book-tomarket ratio at the beginning of the quarter; HIGHTECH is an indicator variable that takes a value of one if the ? m operates in any of the following industries: Drugs, Computers, Electronics, Programming, R&D services, and is zero otherwise; REGULATION is an indicator variable that takes on a value of one if the ? rm operates in any of the following industries: Telephone, TV, Cable, Communications, Gas, Electricity, Water, and is zero otherwise; ROA is current year’s return on assets; PROA is prior ? ve years’ return on assets. Kasznik and Lev (1995) estimate their model (Eq. (8) without the FAMILYFIRM and CHEPS A FAMILYFIRM variables) separately for good news (positive CHEPS) and bad news (negative CHEPS) ? rms. They obtain a signi? cantly negative coef? ient on CHEPS for bad news ? rms. Their result suggests that the likelihood of management earnings forecasts increases with the magnitude of bad news. Moreover, they do not ? nd a signi? cant coef? cient on CHEPS for good news ? rms. We predict that the strength of the relation between the likelihood of management forecast and the magnitude of bad news would differ across family and non-family ? rms. Thus, we expect that the coef? cient b3 will be either negative (hypothesis H2a) or positive (hypothesis H2b) when Eq. (8) is estimated using observations with CHEPSo0. The other variables in Eq. (8) are control variables, similar to that used in Kasznik and Lev (1995).

SIZE is found to be positively related to the likelihood of management forecasts, probably because of economies of scale (Lang and Lundholm, 1993). BM is included to control for risk as well as growth. HIGHTECH is expected to have a positive coef? cient, re? ecting exposure to larger risk of shareholder lawsuits due to larger price ? uctuations. Finally, REGULATION is expected to have a negative coef? cient, re? ecting a smaller demand for management forecasts because of regulated ? rms’ practice of providing considerable amount of information to the regulatory body and therefore indirectly to the investors. Finally, ROA and PROA control for the effect of pro? tability on the likelihood of management forecast.

The descriptive statistics of the variables in Eq. (8) are presented in panel A of Table 7. The likelihood of family ? rms making management forecasts is greater than that for nonfamily ? rms both when CHEPS40 and CHEPSo0. However, most of the control variables have signi? cantly different values across family and non-family ? rms. Thus, to draw proper conclusions, it is important to control for these variables. The results from estimating Eq. (8) are presented in panel B of Table 7. We ? rst estimate the models without the FAMILYFIRM variables and obtain results similar to that in Kasznik and Lev (1995). Coef? cient on CHEPS is insigni? cant for the good news case and is negative and signi? ant for the bad news case, A1. 97 (p-valueo0. 01). For bad news ? rms, the results of the full model show that the coef? cient on CHEPS A FAMILYFIRM is negative and signi? cant, A2. 71 (p-valueo0. 05). The coef? cients on the control variables, when signi? cant, are consistent with the predictions in prior studies (Kasznik and Lev, 1995). Overall, the results suggest that the association between the likelihood of management forecast of earnings and the magnitude of bad news is stronger for family ? rms as compared to non-family ? rms, consistent with hypothesis H2a. ARTICLE IN PRESS 264 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 4. 3.

Voluntary disclosure of corporate governance practices To examine whether family ? rms are less likely to make voluntary disclosures related to corporate governance practices (hypothesis H3), we use the Transparency and Disclosure (T&D) database. 20 It provides transparency and disclosure scores collected by Standard and Poor’s for the S&P 500 ? rms. The scores are computed using the company’s annual report and regulatory ? lings, such as the 10-K and proxy statements. The scores are available for 98 questions organized in 12 groups (Patel and Dallas, 2002). For each question that is answered in the af? rmative, the company receives a score of one, and receives a score of zero otherwise. In general, an af? mative answer to a question indicates the presence of a disclosure item. These questions are listed in Appendix A. In panel A of Table 8, we consider those groups of questions that are related to shareholder rights and corporate governance structure and practices. The score for each group indicates the average number of questions answered in the af? rmative within that group. For two of these groups, Information on Auditors (#8) and Board Structure and Composition (#9), almost all ? rms have an af? rmative answer, probably because there is no discretion available, i. e. , information pertaining to these aspects are mandatory. For the remaining groups, ? rms seem to have some discretion.

For four of these groups, Concentration of Ownership (#2), Voting and Shareholder Meeting Procedures (#3), Role of Board (#10), and Director Training and Compensation (#11), the scores for family ? rms are signi? cantly different than that for non-family ? rms, with t-statistics of 4. 51, A4. 42, A4. 69 and A2. 61, respectively. To better understand the reasons for these differences, we list the scores of all the questions in each of these four groups (panel B of Table 8). The category, Concentration of Ownership, have higher scores for family ? rms than non-family ? rms. However, this may simply re? ect that these questions are more relevant for family ? rms, and so these companies are more likely to respond. Thus, family ? rms end up getting a higher score than non-family ? ms in this category. In other words, this result does not indicate greater voluntary disclosure of Concentration of Ownership by family ? rms. 21 For the other three groups related to corporate governance practices, the disclosure scores are signi? cantly less for family ? rms than for non-family ? rms. For the group Voting and Shareholder Meeting Procedures, the questions for which family ? rms provide signi? cantly less disclosure are: how shareholders convene an extraordinary general meeting (t ? A1. 86), how shareholders nominate directors to board (t ? A2. 76) and does the annual report refer to or publish the corporate governance charter (t ? A3. 49).

For the group Role of the Board, the questions for which family ? rms provide signi? cantly less disclosure are: is there a list of board committees (t ? A1. 86), is there a nomination committee (t ? A3. 31), disclosure of names on nomination committee (t ? A3. 40), other 20 Khanna et al. (2004) use this database to examine differences in disclosure practices of companies across countries. 21 It is possible that the response to questions in some of the other categories may also be affected by whether the particular issue is relevant for the ? rm or not. For example, the group Related Party Structure and Transaction is more relevant for family ? rms and less so for non-family ? rms.

In Panel A of Table 8, we ? nd that the score is not signi? cantly different across the family and non-family ? rms. The insigni? cant difference could be due to the offsetting effect of family ? rms’ unwillingness to voluntarily disclose information about these transactions. It is dif? cult to control for this type of problem in our analyses of the T&D data. Our results should therefore be interpreted with caution. Table 7 Family ? rms and voluntary management forecasts, 1998–2002 Mean Family ? rms Panel A: Descriptive statistics I. Good news ? rms (CHEPS40) MGMT_FORECAST CHEPS SIZE BM HIGHTECH REGULATION ROA PROA No. of observations II. Bad news ? ms (CHEPSo0) MGMT_FORECAST CHEPS SIZE BM HIGHTECH REGULATION ROA PROA No. of observations Non-family ? rms Difference t-stat. Median Family ? rms Non-family ? rms Difference z-stat. A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 0. 14 0. 01 8. 96 0. 35 0. 26 0. 08 0. 05 0. 07 1855 0. 18 A0. 02 8. 98 0. 35 0. 28 0. 07 0. 01 0. 05 1458 0. 12 0. 02 9. 05 0. 42 0. 12 0. 09 0. 03 0. 04 2961 0. 16 A0. 02 9. 01 0. 47 0. 11 0. 11 0. 00 0. 03 2728 1. 75* A3. 97*** A2. 11** A7. 51*** 12. 16*** A1. 65* 3. 99*** 4. 82*** 0. 00 0. 00 8. 88 0. 26 0. 00 0. 00 0. 06 0. 05 1855 0. 00 0. 00 8. 89 0. 36 0. 00 0. 00 0. 04 0. 03 2961 0. 00 0. 00 8. 81 0. 40 0. 0 0. 00 0. 00 0. 03 2728 1. 74* A5. 93*** A2. 11** A11. 21*** 12. 86*** A1. 62 4. 21*** 5. 21*** ARTICLE IN PRESS 1. 65* A0. 78 A0. 65 A10. 63*** 14. 60*** A4. 57*** 3. 12*** 4. 28*** 0. 00 0. 00 8. 88 0. 26 0. 00 0. 00 0. 01 0. 05 1458 1. 02 4. 21*** 0. 39 A12. 91*** 14. 26*** A4. 56*** 3. 96*** 4. 99*** Dependent Var. ? MGMT_FORECAST Variables Predicted sign Coeff. w2 Dependent Var. ? MGMT_FORECAST Coeff. w2 Marginal probability Panel B: Logistic model estimates I. Good news ? rms (CHEPS40) Intercept CHEPS FAMILYFIRM CHEPS A FAMILYFIRM ? ? ? ? A2. 63 A0. 26 55. 19*** 0. 14 A2. 67 A1. 07 0. 13 1. 91 55. 64*** 0. 84 1. 54 1. 59 A 0. 24 0. 04 0. 4 265 266 Table 7 (continued ) Dependent Var. ? MGMT_FORECAST Variables Predicted sign Coeff. w2 Dependent Var. ? MGMT_FORECAST Coeff. w2 Marginal probability A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 SIZE BM HIGHTECH REGULATION ROA PROA Likelihood ratio (p-value) No. of observations II. Bad news ? rms (CHEPSo0) Intercept CHEPS FAMILYFIRM CHEPS A FAMILYFIRM SIZE BM HIGHTECH REGULATION ROA PROA Likelihood ratio (p-value) No. of observations + + + A ? ? 0. 08 0. 08 0. 17 0. 07 A0. 34 3. 34 11. 36 0. 05 4816 7. 53*** 0. 14 2. 93* 0. 08 0. 31 0. 79 0. 11 0. 06 0. 15 0. 05 A0. 33 3. 30 6. 86*** 0. 24 1. 96 0. 11 0. 27 0. 8 15. 38 0. 03 4816 0. 03 0. 01 0. 03 0. 01 A0. 07 0. 13 ARTICLE IN PRESS ? A ? ? + + + A ? ? A2. 88 A1. 97 54. 98*** 6. 42*** 0. 12 0. 19 0. 19 0. 08 A0. 06 0. 92 23. 15 0. 00 4186 7. 72*** 1. 46 2. 02 0. 21 0. 02 0. 19 A2. 90 A0. 96 0. 01 A2. 71 0. 11 0. 16 0. 17 0. 07 A0. 05 0. 90 54. 33*** 1. 21 0. 02 3. 78** 7. 31*** 1. 27 2. 00 0. 26 0. 01 0. 17 28. 41 0. 00 4186 A A0. 18 0. 00 A0. 51 0. 03 0. 03 0. 03 0. 01 A0. 01 0. 04 Variable de? nitions: MGMT_FORECAST is an indicator variable which equals one if the managers make an earnings forecast of quarterly earnings, and zero otherwise, FAMILYFIRM is a dummy variable which equals one for family ? ms, and zero otherwise; CHEPS is the change in earnings per share from that of the same quarter in the previous ? scal year, de? ated by stock price at the beginning of the quarter; SIZE is the natural log of market capitalization at the beginning of the ? scal quarter; BM is the natural log of the book-to-market ratio, computed using the book value of equity at the beginning of the quarter divided by the market capitalization at the beginning of the quarter; HIGHTECH is an indicator variable that takes on a value of one if the ? rm operates in any of the following industries: Drugs, Computers, Electronics, Programming, R&D services, and is zero otherwise; REGULATION is an indicator variable that takes on a value of one if the ? m operates in any of the following industries: Telephone, TV, Cable, Communications, Gas, Electricity, Water, and is zero otherwise. ROA is earnings before extraordinary item divided by total assets. PROA is the average of prior 5 years’ earnings before extraordinary items divided by the average of prior 5 years’ total assets. Table 8 Family ? rms and Standards & Poor’s Transparency and Disclosure data, 2002 (T&D group#) T & D group name Number of questions Mean of number of questions answered (Mean of percentage of questions answered) A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 All ? rms (N ? 451) Family ? rms (N ? 161) Non-family ? rms (N ? 290) Difference t-statistics

Panel A: Transparency and disclosure groups related to corporate governance practices (1) Transparency of Ownership 11 8. 02 (73%) 7. 96 8 2. 48 (31%) 2. 90 (2) Concentration of Ownership (3) Voting and Shareholder Meeting 9 3. 68 (41%) 3. 34 Procedures 4 1. 03 (26%) 1. 07 (7) Related Party Structure and Transaction (8) Information on Auditors 4 4. 00 (100%) 4. 00 (9) Board Structure and 8 7. 93 (99%) 7. 93 Composition (10) Role of the Board 12 9. 19 (77%) 8. 81 (11) Director Training and 6 3. 10 (52%) 3. 02 Compensation (12) Executive Compensation and 9 7. 25 (81%) 7. 17 Evaluation T & D group Question (72%) (36%) (37%) (27%) (100%) (99%) (73%) (50%) (80%) 8. 06 (72%) 2. 24 (28%) 3. 87 (43%) 1. 01 (25%) 4. 0 (100%) 7. 94 (99%) 9. 39 (78%) 3. 15 (53%) 7. 30 (81%) A0. 77 4. 51** A4. 42*** 0. 69 . A0. 41 A4. 69*** A2. 61*** A1. 60 ARTICLE IN PRESS Fraction of ? rms that answer the question All ? rms (N ? 451) Family ? rms (N ? 161) Non-family ? rms (N ? 290) Difference t-statistics Panel B: Details of T&D groups with signi? cantly different response (2)Concentration of Top 1 shareholder disclosed? Ownership Top 3 shareholders disclosed? Top 5 shareholders disclosed? Top 10 shareholders disclosed? Shareholders owning more than 3% is disclosed? across family and non-family ? rms 0. 82 0. 89 0. 40 0. 50 0. 09 0. 15 0. 02 0. 03 0. 06 0. 12 0. 78 0. 34 0. 05 0. 1 0. 03 3. 01*** 2. 96*** 3. 57*** 1. 12 3. 64*** 267 268 Table 8 (continued ) T & D group Question Fraction of ? rms that answer the question All ? rms (N ? 451) Family ? rms (N ? 161) 0. 77 0. 54 0. 02 0. 96 0. 01 0. 95 0. 10 0. 66 0. 06 0. 46 Non-family ? rms (N ? 290) 0. 69 0. 41 0. 03 0. 97 0. 04 0. 97 0. 16 0. 79 0. 09 0. 61 Difference t-statistics A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 1. 57 2. 45** 0. 92 A0. 95 A1. 31 A0. 91 A1. 86* A2. 76*** A0. 82 A3. 49*** Shareholders owning more than 5% is disclosed? Shareholders owning more than 10% is disclosed? Does the company disclose percentage of cross-ownership? 3) Voting and Shareholder Meeting Procedures Is there a calendar of important shareholder dates? Review of shareholder meetings (could be minutes)? Describe procedure for proposals at shareholder meetings? How shareholders convene an extraordinary general meeting? How shareholders nominate directors to board? Describe the process of putting inquiry to board? Does the annual report refer to or publish Corporate Governance Charter? Does the annual report refer to or publish Code of Best Practice? Are the Articles of Association or Charter Articles of Incorporation published? Details about role of the board of directors at the company? 0. 72 0. 45 0. 2 0. 97 0. 03 0. 97 0. 13 0. 74 0. 08 0. 56 ARTICLE IN PRESS 0. 09 0. 20 0. 08 0. 15 0. 09 0. 23 A0. 47 A1. 51 (10) Role of the Board 0. 95 0. 93 0. 96 A1. 25 Is there disclosed a list of matters reserved for the board? Is there a list of board committees? Review last board meeting (could be minutes)? Is there an audit committee? Disclosure of names on audit committee? Is there a remuneration/compensation committee? Names on remuneration/compensation committee)? Is there a nomination committee? Disclosure of names on nomination committee? Other internal audit functions besides audit committee? Is there a strategy/investment/? nance committee? 11) Director Training and Compensation Disclose whether they provide director training? Disclose the number of shares in the company held by directors? Discuss decision-making process of directors’ pay? Are speci? cs of directors’ salaries disclosed (numbers)? Form of directors’ salaries disclosed (cash, shares, etc. )? Speci? cs disclosed on performancerelated pay for directors? 0. 13 0. 99 0. 02 1. 00 1. 00 0. 99 0. 99 0. 83 0. 81 0. 84 0. 50 0. 00 0. 98 0. 09 0. 97 0. 98 0. 07 0. 11 0. 99 0. 02 1. 00 1. 00 0. 99 0. 99 0. 74 0. 72 0. 89 0. 42 0. 00 0. 98 0. 06 0. 94 0. 97 0. 06 0. 14 1. 00 0. 02 1. 00 1. 00 1. 00 0. 99 0. 87 0. 87 0. 96 0. 55 0. 00 0. 98 0. 11 0. 98 0. 99 0. 08 A0. 98 A1. 86* A0. 04 . . A.

Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 A1. 31 A0. 38 A3. 31*** A3. 40*** A2. 38** A2. 35** A0. 75 A0. 34 A1. 71* A1. 91* A1. 12 A0. 95 ARTICLE IN PRESS In panel A, for T&D Group 1, 8. 02 (73%) represents the mean across all ? rms of the number (percentage) of 11 questions to which they provide an answer. Appendix A lists all the S&P transparency and disclosure practice questions. The difference column provides the t-statistic of the difference across family ? rms and non-family ? rms. *** indicates signi? cance at the 0. 01 level, ** indicates signi? cance at 0. 05 level, and * indicates signi? cance at the 0. 10 level. 69 ARTICLE IN PRESS 270 A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 internal audit function besides audit committee (t ? A2. 38), and is there a strategy/ investment/? nance committee (t ? A2. 35). For the group director training and compensation, the questions for which family ? rms provide signi? cantly less disclosure are: discuss decision-making process of directors’ pay (t ? A1. 71) and are speci? cs of directors’ salaries disclosed (t ? A1. 91). Overall, the results in Table 8 suggest that family ? rms provide less disclosure about their corporate governance practices. This evidence supports hypothesis H3. 4. 4.

Analyst following, analysts’ forecast properties and bid-ask spreads Since we ? nd that family ? rms disclose higher quality earnings and are more likely to provide warning for bad news, we test for hypothesis H4a and not H4b. We investigate how family and non-family ? rms differ on analyst coverage, dispersion in analysts’ forecasts, analyst forecast accuracy, volatility in forecast revisions, and bid-ask spread. For this examination, we adopt the models used in Lang and Lundholm (1996) and Healy et al. (1999). 4. 4. 1. Analyst following We estimate the following equation. COVERAGE ? a0 ? a1 FAMILYFIRM ? a2 SIZE ? a3 STDROE ? a4 CORR ? a5 INVPRICE ? a6 RETVAR ? a7 RD ? a8 EFFORT ? 9 BROKER ? a10 ROA ? a11 PROA ? error ? 9? The dependent variable analyst coverage, COVERAGE, is de? ned as the 12-month average of the number of analysts who issued annual earnings forecasts in IBES. Our main independent variable, family ? rm membership, is denoted by FAMILYFIRM. Following Lang and Lundholm (1996), we include the following control variables. SIZE, de? ned as the natural logarithm of market value of equity at the beginning of the ? scal year, is predicted to have a positive coef? cient. Bhushan (1989) argues that larger ? rms are more widely held with more potential transaction business for analysts’ brokerage houses. STDROE, de? ed as the standard deviation of return-on-equity during the preceding 10-year period, is predicted to have a positive coef? cient. Bhushan (1989) explains that expected trading bene? ts based on private information is higher for a ? rm with higher return variability because it increases the conditional expected returns. CORR, de? ned as the Pearson correlation between ROE and annual stock return in the preceding 10-year period, is predicted to have a positive coef? cient. Bhushan (1989) argues that it is easier for analysts to predict future stock price for ? rms with higher return-earnings correlations. We include the following additional control variables beyond those included in Lang and Lundholm (1996). INVPRICE, de? ed as the inverse of stock price at the beginning of the year, is predicted to have a positive coef? cient. Brennan and Hughes (1991) argue that inverse of stock price proxies for the rate of the brokerage commission and the higher the brokerage commission the greater will be analysts’ incentive to follow the ? rm. RETVAR, de? ned as daily stock return variance estimated over the last 200 days prior to end of the year, is predicted to have a positive coef? cient. RETVAR is an additional measure for return variability and hence the reason for the prediction is the same as that discussed ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 271 above for STDROE. RD, de? ed as the annual research and development expense divided by total assets at the beginning of the ? scal year, is predicted to have a positive coef? cient. Barth et al. (2001) argue that intangible assets typically are not recognized, making ? nancial statements less informative and providing greater incentive for analysts to follow ? rms with greater research and development expenses. EFFORT is de? ned as the negative of the average number of ? rms followed by the ? rm’s analysts in a particular year divided by the number of analysts covering the ? rm in that year. This variable captures the notion that if a particular ? rm requires more effort to cover it, then the ? rm’s analysts will cover fewer ? ms (Barth et al. , 2001). BROKER is de? ned as the average number of analysts employed by the brokerage houses that employ the ? rm’s analysts. Larger brokerage houses have greater resources and can therefore follow more ? rms. The inclusion of BROKER in the model controls for cross-sectional difference in EFFORT that is related to the size of the brokerage houses, thereby making the EFFORT variable more effective (Barth et al. , 2001). Finally, ROA and PROA, de? ned earlier, control for the effect of pro? tability on analyst coverage. 4. 4. 2. Forecast dispersion, forecast accuracy, and revision volatility To investigate how family and non-family ? ms differ in terms of dispersion in analysts’ earnings forecasts, forecast accuracy, and volatility in forecast revisions, we use the following equations. The control variables are primarily from Lang and Lundholm (1996): DISP ? a0 ? a1 FAMILYFIRM ? a2 SIZE ? a3 STDROE ? a4 CORR ? a5 ACHEPS ? a6 RD ? a7 ROA ? a8 PROA ? error, FERROR ? a0 ? a1 FAMILYFIRM ? a2 SIZE ? a3 STDROE ? a4 CORR ? a5 ACHEPS ? a6 RD ? a7 ROA ? a8 PROA ? error, REVISION ? a0 ? a1 FAMILYFIRM ? a2 SIZE ? a3 STDROE ? a4 CORR ? a5 ACHEPS ? a6 RD ? a7 ROA ? a8 PROA ? error, ? 12? In Eq. (10), the dependent variable, DISP, is dispersion in individual analyst earnings forecasts, de? ed as 12-month average of the standard deviation of analysts’ forecasts. In Eq. (11), the dependent variable, FERROR, is the absolute value of 12-month average of analyst forecast error de? ned as actual earnings minus the median analyst forecast. For both DISP and FERROR, we compute a simple average across the twelve months corresponding to the ? rm’s ? scal year. We also de? ate both the variables by beginning of ? scal year stock price. In Eq. (12), the dependent variable, REVISION, is volatility in forecast revisions, de? ned as the standard deviation of monthly forecast revisions over the ? scal year, de? ated by the beginning of ? scal year price, where forecast revision is de? ed as current month median forecast minus previous month median forecast. Eqs. (10)–(12) include SIZE, STDROE, CORR, and RD as control variables. As discussed before, these variables represent factors that affect analysts’ incentives to collect information and are therefore likely to affect the properties of their forecasts. In these models, we also control for ACHEPS, de? ned as the absolute value of annual change in earnings per share de? ated by the beginning of ? scal year price. It controls for the fact that dispersion in analysts’ earnings forecasts, forecast errors, and volatility in forecast revisions are likely to increase ? 11? ?10? ARTICLE IN PRESS 272 A.

Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 with the magnitude of the forthcoming earnings information. Finally, ROA and PROA, de? ned earlier, control for the effect of pro? tability on the three forecast properties. 4. 4. 3. Bid-ask spread Finally, to examine the difference in bid-ask spread between family and non-family ? rms, we use the following equation: SPREAD ? a0 ? a1 FAMILYFIRM ? a2 SIZE ? a3 LTURNOVER ? a4 LPRICE ? a5 ROA ? a6 PROA ? error. ?13? Eq. (13) is similar to that in Healy et al. (1999). SPREAD is de? ned as the annual average of the daily closing bid-ask spread as a percentage of daily closing price.

SIZE and LTURNOVER, de? ned as the natural logarithm of the annual median value of daily trading volume divided by total shares outstanding, are included to control for the possibility that bid-ask spreads are narrower for larger ? rms or for ? rms whose shares are traded more often. LPRICE, de? ned as the natural logarithm of the beginning of year stock price, is included because ? xed order costs are spread across more dollars in stocks that have a higher price and consequently the percentage spread is lower for these stocks (Stoll, 1978). Finally, ROA and PROA, de? ned earlier, control for the effect of pro? tability on bid-ask spreads. 4. 4. 4.

Results Panel A of Table 9 presents descriptive statistics for all the dependent and independent variables in Eqs. (9)–(13) and panel B presents the regression estimates of these models. The coef? cient on FAMILYFIRM is positive and signi? cant for the analyst coverage model (0. 94, t ? 3. 46), suggesting that family ? rms enjoy greater analyst coverage than non-family ? rms. The coef? cient on FAMILYFIRM is negative and signi? cant for the forecast dispersion model (A0. 08, t ? A4. 24), suggesting that for family ? rms there is less disagreement on earnings forecasts among analysts. The coef? cient on FAMILYFIRM is negative and signi? cant for the forecast error model (A0. 12, t ? A2. 7), suggesting that for family ? rms analysts’ forecasts tend to be more accurate. The coef? cient on FAMILYFIRM is negative and signi? cant for the volatility of forecast revision model (A0. 06, t ? A3. 01), suggesting that forecast revisions for family ? rms are less extreme. The coef? cient on FAMILYFIRM is negative and signi? cant for the bid-ask spread model (A0. 66, t ? A3. 17), suggesting that family ? rms enjoy greater liquidity. The control variables in all models, when signi? cant, have the predicted signs, except in two cases. The coef? cients on CORR and RD have the opposite signs in the forecast dispersion, forecast error and forecast revision models.

Overall, the results in Table 9 are consistent with hypothesis H4a, suggesting that family ? rms enjoy larger analyst following, better analysts’ forecast properties and greater liquidity, probably due to better quality of their reported earnings and because of their reputation of disclosing bad news through management forecasts. 4. 5. Family ? rm subsamples To gain additional con? dence that difference in the severity of agency problems across family and non-family ? rms are responsible for our results, we analyze subsamples of ARTICLE IN PRESS A. Ali et al. / Journal of Accounting and Economics 44 (2007) 238–286 273 family ? rms that are expected to have difference in the severity of agency