The effect of legal environment on voluntary disclosure: evidence from management earnings...
I. INTRODUCTION
This paper reports empirical evidence of differences in management earnings forecast disclosure behavior between the United States and Canada. We expect that management forecast disclosure differs between these two countries because of the substantial differences in
Management forecasts affect security prices (Patell 1976; Ajinkya and Gift 1984; Waymire 1984). Therefore, it is important to understand whether and how legal regimes affect forecast disclosure. On one hand, potential litigation costs associated with failure to meet a management forecast mitigate "cheap talk," and thus may enhance the credibility of publicly disclosed forecasts. On the other hand, fear of litigation may alter the conditions under which managers issue forecasts.
As an example of how litigation fears could alter which forecasts are released, Skinner (1994) argues that fear of legal liability relating to Rule 10b-5 suits gives U.S. managers incentive to hasten disclosure of bad news, because (1) early disclosure is a defense against shareholders' claims that management withheld adverse information, and (2) early disclosure shortens the period in which investors can qualify as members of a class action suit. Based on a sample of 93 NASDAQ firms, Skinner (1994) provides evidence that, because of asymmetric incentives to provide timely bad news disclosure, managers are more likely to issue imprecise interim forecasts during periods of large negative earnings surprises than during periods of large positive surprises. Similarly, in a sample of 565 NYSE and AMEX firms, Kasznik and Lev (1995) find that managers are twice as likely to disclose forecasts during periods of large negative earnings news than during periods of large positive earnings news. Policymakers also argue that the threat of litigation reduces a manager's tendency to provide any voluntary disclosure of forward-looking information because of the possibility that earnings may fall short of the forecast (e.g., Breeden 1995). In fact, Congress enacted the Private Securities Litigation Reform Act of 1995 to encourage firms to disclose forward-looking information by reducing legal exposure.
Using a sample of U.S. and Canadian firms from 1993 to 1996, we find that Canadian firms issue more management earnings forecasts than do U.S. firms. Further, although U.S. managers are relatively more likely to issue forecasts during interim periods in which earnings decrease, Canadian managers do not exhibit that tendency. Instead, Canadian managers issue more forecasts when earnings are increasing, and their forecasts are of annual rather than interim earnings. Also consistent with a less litigious environment, Canadian managers issue longer-term and more precise forecasts. These Canadian firms' different disclosure patterns persist after controlling for other conditions that might lead to differential disclosure, including firm size, earnings volatility, information asymmetry, growth, capitalization rates, and membership in high-technology and regulated industries.
Our study contributes to management forecast research by investigating the relation between legal regime and specific management forecast disclosure properties--incidence, timeliness, and precision. Although Skinner's (1994) and Kasznik and Lev's (1995) results are consistent with legal liability leading to more disclosure in bad news periods, it is not clear that the legal environment is driving U.S. firms to issue bad news management forecasts in interim periods. First, extant empirical evidence is mixed on the effectiveness of a bad news management forecast in avoiding litigation. Francis et al. (1994) find that (1) substantial declines in earnings are not sufficient to cause litigation; (2) early disclosure of adverse earnings news is not an effective deterrent to litigation, and failure to provide early disclosure of adverse earnings news does not necessarily lead to litigation; and (3) in a litigation sample of 45 observations, 28 suits were based on a management forecast or other pre-earnings release disclosure rather than on the subsequently released actual earnings announcement. Further, Skinner (1997) finds no evidence that management forecast disclosure increases the probability of lawsuit dismissal. However, he provides some evidence that, after controlling for estimated shareholder damages, timely disclosure reduces settlement amounts. Finally, Johnson et al. (2001) interpret their finding that bad news disclosure frequency increases after the Private Securities Litigation Reform Act of 1995 as inconsistent with the argument that reduced legal exposure leads to proportionally fewer bad news disclosures.
Second, with the exception of Frost (2001) and Johnson et al. (2001), prior research has not examined management forecast disclosures in different legal environments. (2) Skinner (1994, 1997), Francis et al. (1994), and Kasznik and Lev (1995) examine disclosures made in the highly litigious U.S. environment. Our finding that managers are more likely to issue forecasts (both bad and good news) in the less litigious Canadian environment is consistent with Frost (2001) and Johnson et al. (2001), who conclude that managers are more likely to issue voluntary disclosures in less litigious environments. Further, our finding that firms issue proportionally fewer bad news forecasts in the Canadian legal environment (relative to good news forecasts) is consistent with Skinner's (1994) and Kasznik and Lev's (1995) conclusions that legal liability increases U.S. firms' incentives to disclose management earnings forecasts when earnings news is bad. However, our evidence is inconsistent with Johnson et al.'s (2001) evidence that U.S. high-tech firms do not issue proportionally fewer bad news forecasts after passage of the 1995 Private Securities Litigation Reform Act (which was intended to reduce legal liability).
The results of our study are also relevant to accounting policymakers and securities regulators who believe litigation reform can improve financial reporting. The AICPA's Jenkins Committee identified fear of litigation as an obstacle to providing forward-looking information (AICPA 1994). Managers and auditors complained about excessive and frivolous litigation; accordingly, Congress provided additional safe harbors in the Private Securities Litigation Reform Act of 1995 for parties associated with good-faith projections. Evidence on the effect of legal liability on management forecasting can help policymakers understand the costs of forecast disclosure, which is important if they expect cost-reduction strategies to create incentives for forecast disclosure.
In Section II, we develop hypotheses about how differences in legal environment are likely to affect Canadian and U.S. management forecasting. We describe the sample in Section III and the research design in Section IV. We present the primary results in Section V and the results of supplemental tests in Section VI. In Section VII, we discuss the study's limitations and provide suggestions for future research.
II. MANAGEMENT FORECAST DISCLOSURES IN THE U.S. AND CANADA
Evidence suggests that U.S. management forecasting behavior has changed over time. Patell (1976) suggested that, in the 1960s and 1970s, managers selectively released good news. In the early 1980s, Ajinkya and Gift (1984) and Waymire (1984) documented symmetric forecasting. Then, in the late 1980s and early 1990s, litigation against companies and their auditors increased. Accordingly, Skinner (1994) investigated whether legal liability creates an asymmetric incentive favoring timely management forecasting of bad news. Using a sample of U.S. management forecasts from the 1980s, his results are largely consistent with legal liability-based arguments: (1) management issues more forecasts in periods of bad earnings news than in periods of good news (evidence of asymmetric incentives); (2) bad news interim forecasts occur in imprecise form (e.g., range rather than point, open-interval rather than closed-interval); and (3) the bad news disclosures occur in interim periods. Although these findings are clearly consistent with legal-liability arguments, Skinner (1994) does not vary legal regime.
Differences between the U.S. and Canadian legal systems create a natural experiment with two different legal regimes. Clarkson and Simunic (1994) note that, unlike the U.S., (1) courts in Canada generally require unsuccessful plaintiffs to pay the costs of a successful defendant; (2) because plaintiffs have no absolute right to a jury trial in Canada, judges typically hear technical cases and are less likely to award large settlements based on emotional plaintiff appeals; (3) Canadian discovery rules limit whom the plaintiff can examine under oath; (4) Canadian provinces generally do not allow lawyers to work on a contingent-fee basis, and when allowed, the practice is not widespread; and (5) it is considerably more difficult to bring a class action suit in Canada. Clarkson and Simunic (1994) argue that auditors and managers operating in Canada are therefore likely to make decisions with the expectation of lower litigation costs than would be the case in the United States. Although McConomy (1998) notes that the Ontario Class Proceedings Act (effective in 1993) made class action lawsuits more likely, Paskell-Mede (1994, 1999) and Grossman (1996) point out that class action suits remain rare in Canada and continue to face such significant impediments that Canada is still a far less litigious environment than is the U.S.
A particularly important difference between Canadian and U.S. securities legislation is the notion of "fraud on the market" and its relation to secondary security markets (Paskell-Mede 1999). In common law actions, a plaintiff must prove that (1) a misrepresentation exists, (2) the misrepresentation was negligent, and (3) the plaintiff suffered a loss due to reliance on the misrepresentation. In U.S. courts, the "fraud on the market" concept provides a proof that the plaintiff relied on share price that efficiently reflected the misrepresentation. Judges in Canadian courts refuse to apply the "fraud on the market" concept, stating that it has no place in the Canadian legal system. Instead, Canadian law places the burden of proof on the defendant to prove that the plaintiff did not rely on the misrepresentation, but only in the case of initial security offerings. That is, once a plaintiff proves that management made a negligent misrepresentation in the prospectus, the law deems the plaintiff's reliance to have occurred unless the defendant can prove that the plaintiff did not rely on the misrepresentation when purchasing shares (often by proving that the plaintiff was aware of the misrepresentation). However, in the case of secondary trading, the burden of proving reliance remains with the plaintiff. Because plaintiffs in Canadian secondary-markets-related suits must prove individual reliance on a misrepresentation, but cannot use the "fraud on the market" argument, they face a significant barrier to successful litigation. (3)
Hypothesis Development
We test two general hypotheses. First, the concern that legal liability costs curb public disclosure of either favorable or adverse forward-looking information lies at the core of U.S. policymakers' arguments for increased forward-looking information disclosure. In addition, under economic models of symmetric belief adjustment (King et al. 1990), management forecasts increase the firm's exposure to legal liability. Our first hypothesis (stated in the alternative form) is:
H1: Canadian firms, faced with a less-litigious legal environment, engage in more voluntary information disclosure relative to U.S. firms (degree of information disclosure hypothesis).
We define "more voluntary information disclosure" to include: (1) a higher forecast frequency, (2) more precise forecasts, and (3) forecasts over longer horizons. Accordingly, we perform three sets of tests of HI that correspond to these three measures of information disclosure. Hypothesis 1 predicts that (1) Canadian firms are more likely to disclose management earnings forecasts because managers do not fear the legal liability associated with being inaccurate; (2) Canadian managers disclose more precise forecasts even though inaccuracy is easier to detect ex post in more precise forecasts; and (3) Canadian managers disclose forecasts over longer horizons even though the likelihood of inaccuracy increases in forecast horizon.
Our second hypothesis (also stated in the alternative form) relates to prior evidence suggesting that litigation exposure prompts U.S. firms to strategically base the frequency and properties of management forecasts on whether earnings are increasing or decreasing. Specifically, when earnings are decreasing (i.e., in bad news periods), U.S. firms tend to be: (1) more likely to issue a forecast; (2) more likely to issue the forecast in an imprecise form; and (3) more likely to issue the forecast over a short-horizon period. We expect Canadian firms' forecast disclosure properties (frequency, precision, and horizon) to be less associated with earnings decreases:
H2: Canadian firms, faced with a less-litigious legal environment, engage in voluntary information disclosure behaviors that are less related to earnings decreases than do U.S. firms (bad-news-related behavior hypothesis).
Because prior research has shown that U.S. firms' forecast frequency, precision, and horizon are related to earnings decreases, H2 is rejected if either Canadian forecast frequency, precision, and horizon are unrelated to the sign of earnings changes or if these Canadian forecast disclosure properties are related to earnings increases. In either case, Canadian firms do not engage in the kinds of forecasting behaviors that are related to fear of legal liability resulting from bad earnings news.
Country Similarities that Strengthen Internal Validity
The similar business and investment environments in the U.S. and Canada support our inferences that differences across the two countries in the frequency, precision, and horizon of management forecasts are attributable to differences in legal regimes. For example, researchers often group together U.S. and Canadian firms in international tests of value relevance of accounting data, primarily because these firms are similar on so many dimensions and because empirical evidence shows that the association between accounting data and stock prices is relatively similar in the two countries (Ali and Hwang 2000). The U.S. and Canada share important determinants of the value relevance of accounting data--a market-oriented financial system, private sector standard setting, low influence of tax rules on financial accounting measurements, and higher spending on external auditing (Ali and Hwang 2000). Although generally accepted accounting principles differ between Canada and the U.S., Bandyopadhyay et al. (1994) and Barth and Clinch (1996) find that adjustments from Canadian to U.S. GAAP are irrelevant for valuing cross-listed Canadian companies.
Investor sophistication and market efficiency with respect to the interpretation of forecasted data also appear similar in the two countries. McNichols (1989) provides empirical evidence that U.S. investors can assess the ex ante firm-specific bias in seasoned firms' management forecasts. Clarkson et al. (1992) find similar investor sophistication for Canadian firms' management forecasts issued in initial public offerings. With respect to market efficiency, Freund et al. (1997) conclude that the Toronto Stock Exchange (TSE) is informationally efficient. Brown et al. (1991) also document that investors privy to financial analyst forecasts could generate abnormal returns on the TSE, but that the returns are not significantly higher than transactions costs if the trading strategy is not implemented until after the forecasts become public (also see L'Her and Surer [1991] for similar findings). (4) In summary, empirical evidence suggests that revelation of Canadian managers' private information in management forecast form should lead to a relatively complete and immediate price reaction. (5)
Canadian and U.S. exchanges have similar rules on material news disclosure. The TSE's "Policy Statement on Timely Disclosure and Related Guidelines" requires firms to provide investors with equal access to all material information. Material information is "any information relating to the business and affairs of a company that results in or would reasonably be expected to result in a significant change in the market price or value of any of the company's listed securities" (TSE 2000, 1). The TSE requires "immediate disclosure" of "firm evidence of significant increases or decreases in near-term earnings prospects," although "forecasts of earnings and other financial forecasts need not be disclosed" (TSE 2000, 4). Further, the TSE encourages wide dissemination through news services (specifically mentioning Dow Jones). Exceptions to these rules allow delay for proprietary information. Guidelines and rules of law also prohibit insider trading.
Finally, differences in earnings management opportunities may exist between the two countries. However, to create an alternative explanation for our results, (1) differences in earnings management opportunities must exist between the two countries; (2) earnings management must be a successful means of limiting legal liability; and (3) when considering the menu of alternative actions to limit legal liability (e.g., engage in earnings management, issue a voluntary management forecast disclosure of bad news, do not provide a forecast that could turn out to be wrong, etc.), the action chosen must differ between the U.S. and Canada because one alternative performs better on a cost/benefit trade-off. If these conditions all hold, then a finding of, say, fewer management forecasts in the U.S. might indicate that U.S. managers can more easily manage subsequently revealed earnings, that such earnings management fools investors about firm performance, and that earnings management is more beneficial to the firm than voluntary disclosure. We are unaware of any empirical evidence on differential benefits to earnings management between the U.S. and Canada, but we cannot rule out this possibility because controlling for earnings management differences across countries is beyond the scope of our study.
In summary, many similarities between U.S. and Canadian environments strengthen the internal validity of our comparative analysis. Still, differences exist that may affect management forecasts--information asymmetry, analyst coverage, earnings volatility, firm size, membership in high-tech and regulated industries, growth, and capitalization rates. Our empirical tests control for these conditions.
III. SAMPLE
Periods Available for Forecasting
Our sample includes all quarterly and annual periods of U.S. and Canadian firms with available data on Compustat, including beginning-of-period price data, shares, earnings per share, and adjustment factors for the 1993-1996 period. (6) Table 1, Panel A reports that 56,507 firm-years and 152,431 firm-quarters are available. Periods with complete data totaled 115,751. NASDAQ was the most represented U.S. exchange (34.2 percent of all periods in the two-country combined sample), and the TSE was the most represented Canadian exchange (5.0 percent of all periods in the combined sample). In contrast to most prior legal-liability-related research and in order to enhance our results' generalizability, we do not limit our tests to periods with "large" earnings surprises.
Management Earnings Forecasts
We collected management forecasts of both interim and annual earnings from the Dow Jones News Retrieval Service (DJNRS) during the entire 1993-1996 period. (7) The DJNRS contains articles published in the Wall Street Journal and Barron's, as well as unpublished announcements appearing on the Broad Tape. The DJNRS covers the major news wires and newspapers in Canada, especially those specializing in corporate news. In addition, the TSE lists Dow Jones as a preferred means of disclosure. Therefore, we consider DJNRS a reasonable (and unbiased) source of management earnings forecasts. However, if DJNRS provides incomplete Canadian coverage, either because of the DJNRS physical location or because U.S. investors are less interested in smaller foreign firms, then the incomplete coverage would bias against our finding of greater forecast frequency for Canadian firms. We also do not limit the coverage period to only the fourth quarter because Canadian firms are required to file only the first three quarters of interim results (Frost and Kinney 1996).
Table 1, Panel B presents information about the management forecast selection. Our original search identified 2,079 forecasts. We lost 359 forecasts due to changes in Compustat coverage. (8) After deleting forecasts for periods with missing Compustat data (110), forecasts of 1992 and 1997 results (138), and long-range forecasts (89), a total of 1,383 management forecasts remain in the sample: 1,164 issued by U.S. firms and 219 issued by Canadian firms. (9)
The 1,164 U.S. forecasts are made by 687 firms (1.69 per firm). The 219 Canadian management forecasts are made by 64 Canadian firms listed only in Canada and 50 Canadian firms cross-listed on U.S. exchanges (1.92 per firm). Our primary tests do not differentiate locally listed from cross-listed Canadian firms. Pooling the locally listed and cross-listed firms reduces the power of our tests to the extent that cross-listed Canadian firms face some legal exposure in U.S. markets. However, Frost and Pownall (1994) were unable to identify a single instance of formal investigation, administrative proceeding, or court action against a foreign issuer for violation of U.S. disclosure rules. Legal analysis (Paskell-Mede 1999) also raises doubts about the extent to which cross-listed firms are exposed to liability under the U.S. legal system. Frost and Pownall (1994) cite political issues, difficulty in monitoring, and lack of cooperative agreements with other countries as barriers to enforcement of foreign-issuer compliance. Finally, Core (1997) finds that although cross-listing is associated with the magnitude of director and officer liability coverage for Canadian companies that purchase insurance, cross-listing does not explain Canadian firms' decisions to purchase the policy. In summary, cross-listing is unlikely to increase legal exposure to the level faced by U.S. firms. Accordingly, Table 1, Panel B provides descriptive data on the disclosures of the Canadian cross-listed firms vis-a-vis locally listed Canadian and U.S. firms, but our primary tests pool together the cross-listed and locally listed Canadian firms.
IV. RESEARCH DESIGN
In this section, we present our research design for tests of H1 and H2 using the three measures of information disclosure: forecast frequency, forecast precision, and length of forecast horizon.
Forecast Frequency
To test H1 and H2, we estimate the following logistic regression model for a combined sample of U.S. and Canadian companies across all potential forecasting periods (n = 115,751): (10)
(1) [FORECAST.sub.i,t] = [a.sub.0] + [a.sub.1] [DELTA][ESIGN.sub.i,t] + [a.sub.2] [CANADA.sub.i,t] + [a.sub.3] [CANADA.sub.i,t] x [DELTA][ESIGN.sub.i,t] + [a.sub.4] [absolute value of [DELTA][EARNINGS.sub.i,t]] + [a.sub.5] [LSIZE.sub.i,t] + [a.sub.6] [REGULATE.sub.i,t] + [a.sub.7] [HIGHTECH.sub.i,t] + [[epsilon].sub.i,t]
where:
[FORECAST.sub.i,t] = 1 if the firm issued a management earnings forecast during the period, and 0 otherwise;
[DELTA][ESIGN.sub.i,t] = the sign of the earnings change, defined as 1 for [DELTA][EARNINGS.sub.i,t] [greater than or equal to] 0 (i.e., good news) and 0 for [DELTA][EARNINGS.sub.i,t] < 0 (i.e., bad news); where [DELTA][EARNINGS.sub.i,t] = the change in earnings, defined as ([EPS.sub.i,t] - [EPS.sub.i,t-k])/[PRICE.sub.i,t-k], where [EPS.sub.i,t] = earnings (net income) per share for firm i in period t; [EPS.sub.i,t-k] = earnings per share for firm i in period t - 1 for annual and t - 4 for quarterly periods; and [PRICE.sub.i,t-k] = security price for firm i at the end of period t - 1 for annual and t - 4 for quarterly periods (all obtained from Compustat); (11)
[CANADA.sub.i,t] = 1 if the potential forecasting period relates to a Canadian firm (regardless of whether the firm issued a forecast in that period), and 0 otherwise (i.e., 0 when the potential forecasting period relates to a U.S. firm);
[DELTA][EARNINGS.sub.i,t] = the absolute value of [DELTA][EARNINGS.sub.i,t];
LSIZE = the log of market value of equity at the beginning of the period;
REGULATE = 1 if the firm is in a regulated industry, and 0 otherwise; and
HIGHTECH = 1 if the firm is in a high-technology industry, and 0 otherwise.
Figure 1 maps the coefficients in Equation (1) to H1 and H2. Column (1) lists the coefficient sums in earnings increase periods (i.e., good news; [DELTA]ESIGN = 1) for Canadian firms (row 1) and for U.S. firms (row 2). Column (2) provides analogous coefficients for earnings decrease periods (i.e., bad news; [DELTA] ESIGN = 0). The last row indicates the difference between countries in the propensity to issue forecasts in periods of good news (aa + [a.sub.3]) and bad news ([a.sub.2]). Hypothesis 1 predicts that Canadian firms issue more forecasts; thus, we expect both sets of coefficients to be positive ([a.sub.2] + [a.sub.3] > 0, [a.sub.2] > 0).
The last column of Figure 1 provides coefficients associated with differences between good and bad news periods (i.e., "sign-related" forecasting behavior) in Canada (row 1) and in the U.S. (row 2). The last row in the column shows that the coefficient [a.sub.3] measures the difference between countries in sign-related forecasting behavior. If legal-liability-created asymmetric forecast disclosure incentives in the U.S. lead to more forecasts in bad news periods, then we expect [a.sub.1] < 0. Hypothesis 2 predicts [a.sub.3] > 0, indicating that Canadian managers are less likely to skew forecast disclosure toward bad news periods. Given our focus on the differences associated with different legal regimes, we base all hypothesis tests on the difference between countries, not the conditions within a country.
The remaining variables in Equation (1) control for other potential nonlegal-liability-related differences between the two countries that could affect management forecasting behavior. Firms with more volatile earnings face greater risk of inaccurate forecasts, with the associated legal and reputational costs. We control for volatility using the absolute value of the current-period earnings change, [absolute value of [DELTA]EARNINGS], because this captures current-period volatility and does not introduce the survivorship bias associated with an historical measure. We expect [a.sub.4] < 0 because managers are less willing to forecast when earnings are volatile. Prior research finds that larger firms issue more management forecasts (Cox 1985; Waymire 1985), so we control for the log of market value of equity at the beginning of the period, LSIZE, and expect [a.sub.5] > 0. Regulatory bodies require regulated firms to produce large amounts of information, and thus, management earnings forecasts are likely less beneficial (Kasznik and Lev 1995). Using Kasznik and Lev's (1995) classification scheme, REGULATE equals 1 when a firm belongs to Telephone (4812-4813), TV (4833), Cable (4841), Communications (4811-4899), Gas (4922-4924), Electricity (4931), Water (4941), or Financial (6021-6023, 6035-6036, 6141, 6311, 6321, 6331) industries, and 0 otherwise. We expect [a.sub.6] < 0 if regulated firms issue fewer forecasts. High-tech firms are riskier, and the greater value of private information may lead less-informed investors to demand more voluntary, publicly released management forecasts to level the informational playing field. On the other hand, high-tech firms may be less likely to issue a management forecast to avoid conveying proprietary information. Accordingly, we cannot predict the sign of [a.sub.7]. Following Kasznik and Lev (1995), HIGHTECH equals 1 if a firm belongs to Pharmaceuticals (Compustat SIC codes 2833-2836), R&D Services (8731-8734), Programming (7371-7379), Computers (3570-3577), or Electronics (3600-3674) industries, and 0 otherwise.
Forecast Precision
King et al. (1990) argue that managers are concerned about potential litigation if a forecast turns out to be inaccurate. Less precise forecasts (e.g., a range, minimum, maximum, or general impression) are less likely to turn out to be inaccurate than are point forecasts. Accordingly, researchers have argued that, when faced with perceived higher expected litigation costs, managers will issue less precise forecasts. Empirical evidence in U.S. markets is consistent with the assertion that forecasts are less precise when the firm is performing poorly (Skinner 1994; Baginski and Hassell 1997; Bamber and Cheon 1998).
Although Canadian firm managers are subject to adverse reputation effects of inaccuracy, the Canadian legal system exacts lower legal penalties for inaccuracy than does the U.S. system. Canadian managers are therefore likely to issue more precise management forecasts (H1) and to make forecast precision choices that are less likely to depend on whether the firm is performing poorly during the period (H2). To test these hypotheses, we estimate the following ordered logistic regression model for a pooled sample of all forecasts issued by U.S. and Canadian firms (n = 1,383):
(2) [PRECISE.sub.i] = [b.sub.o] + [b.sub.1] [DELTA][ESIGN.sub.i] + [b.sub.2] [CANADA.sub.1] + [b.sub.3] [CANADA.sub.i] x [DELTA][ESIGN.sub.i] + [b.sub.4] [LSIZE.sub.i] + [b.sub.5] [REGULATE.sub.i] + [b.sub.6] [HIGHTECH.sub.i] + [b.sub.7] [FHORIZON.sub.i] + [v.sub.i].
We measure management forecast precision using an ordinal coding scheme that assigns the highest value to the most precise forecasts. PRECISE equals 3, 2, 1, and 0 for point, closed-interval, open-interval, and general impression forecasts, respectively. Hypothesis 1 predicts that Canadian firms will issue more precise forecasts because the legal penalties for inaccuracy are smaller. For earnings decreases, this suggests that [b.sub.2] > 0, and for earnings increases, it suggests that [b.sub.2] + [b.sub.3] > 0. If fear of legal liability leads U.S. firms to issue less precise forecasts when the firm is performing poorly, then [b.sub.1] > 0. Hypothesis 2 predicts that Canadian forecast precision is less skewed toward poor performance than is U.S. forecast precision ([b.sub.3] < 0). (12)
Based on Baginski and Hassell's (1997) conclusion that larger firms garner lower benefits of precise forecasting, we expect a negative coefficient on firm size ([b.sub.4] < 0). AS in Baginski and Hassell (1997) and Bamber and Cheon (1998), we also control for forecast horizon (FHORIZON), the number of calendar days between the forecast and period-end. The longer the forecast horizon, the greater management's uncertainty about earnings, which should lead to less precise forecasts. Thus, we expect [b.sub.7] < 0. We continue to expect [b.sub.5] < 0 (less precise forecasts for regulated firms), and we make no prediction for [b.sub.6].
Length of Forecast Horizon
Longer forecast horizons provide investors with information on a timelier basis. However, longer-horizon forecasts have a greater chance of being ex post inaccurate and, as a result, increase the firm's legal exposure. Canadian firms are more likely to enjoy the benefits of providing investors with more timely information because Canadian firms face lower legal exposure when issuing longer-horizon forecasts. Thus, H1 predicts that Canadian firms issue longer-horizon forecasts for both good and bad news.
Skinner (1994) finds that, faced with substantial legal exposure, U.S. firms' longer-horizon annual forecasts reflect primarily good news, but their shorter-horizon interim forecasts reflect primarily bad news. Hypothesis 2 predicts that Canadian firms, facing relatively less legal exposure, are less likely to restrict bad news forecasts to short horizons.
To test these predictions, we estimate the following OLS regression models for the pooled sample of U.S. and Canadian management forecasts (n = 1,204):
(3) Model 1: [FHORIZON.sub.i] = [c.sub.o] + [c.sub.1] [NEWS.sub.i] + [c.sub.2] [CANADA.sub.i] + [c.sub.3] [CANADA.sub.i] x [NEWS.sub.i] + [c.sub.4] [LSIZE.sub.i] + [c.sub.5] [REGULATE.sub.i] + [c.sub.6] [HIGHTECH.sub.i] + [[mu].sub.i].
(4) Model 2: [ALTFHORIZON.sub.i] = [c.sub.o] + [c.sub.1] [NEWS.sub.i] + [c.sub.2] [CANADA.sub.i] + [c.sub.3] [CANADA.sub.i] x [NEWS.sub.i] + [c.sub.4] [LSIZE.sub.i] + [c.sub.5] [REGULATE.sub.i] + [c.sub.6] [HIGHTECH.sub.i] + [[eta].sub.i].
As in Baginski and Hassell (1997) and Skinner (1997), we operationalize forecast horizon (FHORIZON) as the number of calendar days until period-end, regardless of whether the period is an interim or annual forecasting period. However, viewing an interim forecast as part of an annual period suggests an alternative measure of horizon (ALTFHORIZON), the number of days until fiscal year-end. For example, a first-quarter forecast 15 days before quarter-end is 288 days before year-end. Accordingly, we estimate the model for each of the two forecast horizon definitions.
When investigating the relation between legal liability and forecast horizon, Skinner (1994) defines good and bad news in terms of the change in contemporaneous expectations. Accordingly, we define a dichotomous variable, NEWS, to proxy for the sign of the news management forecasts convey relative to contemporaneous expectations. Good news disclosures include management forecast point estimates above expected earnings, range estimates where the midpoint exceeds expected earnings, and all minimum estimates. Expected earnings equals the median I/B/E/S analyst forecast when available. Otherwise, expected earnings equals the random-walk EPS for annual forecasts and seasonal random-walk for quarterly forecasts. Bad news disclosures include point estimates below expected earnings, range estimates where the midpoint is less than expected earnings, and all maximum estimates. NEWS equals 1 for good news forecasts and 0 for bad news forecasts. We discard 32 neutral news forecasts (point estimates equaling expected earnings and range estimates where the midpoint equals expected earnings) and all general impression forecasts. (13)
Hypothesis 1 predicts that Canadian firms will be more likely to engage in riskier longer-range forecasting for both bad ([c.sub.2] > 0) and good ([c.sub.2] + [c.sub.3] > 0) news. Given Skinner's (1994) finding that U.S. firms tend to voluntarily release good news in longer-horizon annual forecasts and bad news in shorter-horizon interim forecasts, we expect [c.sub.1] > 0, indicating that U.S. good news forecasts are of a longer horizon. Hypothesis 2 predicts that Canadian firms will have less tendency to choose longer forecast horizons for good news relative to U.S. firms ([c.sub.3] < 0). For the control variables, we expect [c.sub.4] > 0 (longer forecast horizons for larger firms), [c.sub.5] < 0 (shorter forecast horizons for regulated firms), and we make no prediction for the coefficient [c.sub.6] on high-tech firms.
V. RESULTS
Variable Distributions and Other Descriptive Statistics
Table 2, Panel A describes variable distributions for our sample of 115,751 forecasting periods used to estimate Equation (1) (forecast frequency tests). Mean and median [DELTA]EARNINGS are slightly higher than 1 percent and 0.1 percent of share price, respectively. Mean and median [absolute value of [DELTA]EARNINGS] are approximately 8.5 percent and 1.1 percent of price, respectively. Mean earnings change is more positive for Canadian firms (2.15 percent relative to 1.13 percent for U.S. firms), but median earnings change is almost identical (0.16 percent relative to 0.14 percent for U.S. firms). A similar relationship holds for absolute earnings change. (14)
Median firm size (per observation) is nearly $87 million, which is much smaller than in Kasznik and Lev (1995), who report median firm sizes of $381 and $436 million for earnings increase and decrease periods, respectively. We do not require I/B/E/S quarterly analyst forecasts for our primary tests, so our sample includes many more small and lightly followed firms. Also, in contrast to Kasznik and Lev (1995), we include only earnings forecasts, not sales forecasts, and we sample from the full year instead of only the fourth quarter where management forecast frequency is the greatest. Not surprisingly then, Panel A shows that forecast frequency is only 1.09 percent for U.S. firms and 2.41 percent for Canadian firms, compared to Kasznik and Lev's (1995) rates of between 9 percent and 21 percent.
U.S. firms report a slightly higher proportion of earnings increases (58.4 percent) during the time period sampled than do Canadian firms (55.4 percent). The U.S. also has a higher percentage of both high-tech and regulated firms. Canadian management earnings forecasts appear in the most precise point form 46.1 percent of the time, nearly twice as often as those of U.S. firms.
Table 2, Panel B presents data for the sample of 1,236 management forecasts possessing all data necessary to estimate Equations (2) through (4) (precision and horizon tests). Panel B's sample equals Panel A's sample of 1,383 management forecasts minus 154 general impression forecasts plus 7 forecasts for which an analyst forecast was available on I/B/E/S when data for constructing a random walk forecast (i.e., prior period EPS) were not available on Compustat. Comparison of Panel B (periods for which the firm issues a forecast) to Panel A (all periods regardless of whether the firm issues a forecast) reveals several expected differences. Firms issuing forecasts are on average much larger; absolute change in earnings during the period ([absolute value of [DELTA]EARNINGS]) is much lower; and regulated firms rarely issue forecasts. Panel B also shows that mean (median) forecast horizon (FHORIZON) is 188.1 (215.0) calendar days for Canadian firms but only 70.5 (30.0) days for U.S. firms. U.S. firms are much more apt to issue interim forecasts (59.6 percent of U.S. forecasts) than are Canadian firms (7.6 percent of Canadian forecasts). Even measuring forecast horizon relative to the fiscal year-end, mean (median) ALTFHORIZON is 205.5 (228.0) for Canadian firms but only 142.5 (139.0) days for U.S. firms. Therefore, Canadian firms' annual forecasts occur earlier in the fiscal year than U.S. firms' interim forecasts. In addition, 73.6 percent of the firms are covered by at least one I/B/E/S analyst. The coverage is slightly greater for Canadian firms that issue management forecasts (77.6 percent relative to 72.9 percent for U.S. firms that issue management forecasts). Finally, 62.2 percent of the forecasts are good news. The distributions of NEWS for the U.S. and Canada are fairly similar.
Primary Results: Forecast Frequency
Table 3 presents the Equation (1) logistic regression tests of H1 and H2. Coefficient [a.sub.2] is significantly positive (p = 0.006), so Canadian firms are more likely to issue forecast disclosures during bad news periods relative to U.S. firms. The sum [a.sub.2] + [a.sub.3] is also significantly positive (p < 0.001), revealing that Canadian firms are more likely to issue forecast disclosures during good news periods relative to U.S. firms. These results support H1's prediction that lower legal liability in Canada leads to more forecast disclosure for both good and bad news periods. (15) These results are consistent with the higher univariate forecast rates for Canadian firms (2.41 percent) relative to U.S. firms (1.09 percent) reported in Table 2, Panel A.
With respect to H2, the results are also consistent with our expectations. The tendency to skew disclosure toward bad news periods relative to good news periods differs between the U.S. and Canada. U.S. firm behavior is as expected: coefficient [a.sub.l] is significantly negative (p = 0.009), indicating more forecast disclosure in bad news periods relative to good news periods. Coefficient [a.sub.3], which measures the difference between Canadian and U.S. sign-related behavior, is significantly positive (p < 0.001), indicating that Canadian firms are less apt to skew disclosure toward bad news periods than are U.S. firms. (16) In fact, Table 3 reveals that Canadian firms skew disclosure toward good news. The sum [a.sub.l] + [a.sub.3] measures the extent to which Canadian forecast frequency is related to the sign of earnings changes (i.e., the sum [a.sub.l] + [a.sub.3] is the Canadian equivalent to coefficient [a.sub.1] for U.S. firms). This sum is positive and significant (p < 0.001; results not tabulated), indicating that Canadian forecasts occur more often in good news periods. The univariate forecast rates reported in Table 2, Panel A are consistent with this result. (17)
These differences between Canadian and U.S. forecasting behavior arise even after controlling for the absolute value of the earnings surprise ([a.sub.4] is significantly negative as expected, p = 0.007), firm size (as is significantly positive as expected, p < 0.001), regulated industry membership ([a.sub.6] is significantly negative as expected, p < 0.001), and high-tech industry membership (a.sub.7] is insignificantly different from 0). (18)
Primary Results: Forecast Precision
Table 4 presents results of estimating Equation (2). As H1 predicts, the significantly positive [b.sub.2] (p < 0.001) indicates that Canadian firms issue more precise forecasts in bad news periods than do U.S. firms, and the significantly positive sum [b.sub.2] + [b.sub.3] (p < 0.001) indicates that Canadian firms also issue more precise forecasts in good news periods than do U.S. firms. Coefficient [b.sub.1] on [DELTA]ESIGN is significantly positive (p = 0.028), indicating that U.S. firms on average issue less precise forecasts when earnings are declining. However, the data do not support H2's prediction that the sign of earnings change has less effect on forecast precision in Canada ([b.sub.3] is insignificant, p = 0.301). These results arise after controlling for the expected negative effects of long forecast horizons on forecast precision (negative [b.sub.7]; p < 0.001), high-tech firms that issue less precise forecasts (negative [b.sub.6], p = 0.090), and firm size and regulated firms (insignificant [b.sub.4] and [b.sub.5], respectively). In summary, the results reported in Table 4 indicate that Canadian firms issue more precise forecasts than do U.S. firms.
Primary Results: Forecast Horizon
Table 5 presents OLS regression estimates of Equations (3) and (4), designated Model 1 and Model 2, respectively. Hypothesis 1 predicts that Canadian firms issue longer-horizon forecasts. Consistent with H1, the significantly positive [c.sub.2] (p < 0.001 in both models), indicates that Canadian firms issue longer-horizon bad news forecasts than do U.S. firms, and the significantly positive sum [c.sub.2] + [c.sub.3] (p < 0.001 in both models), indicates that Canadian firms also issue longer-horizon good news forecasts than do U.S. firms. (19) U.S. firms forecast good news over longer horizons relative to bad news (significantly positive [c.sub.1], p < 0.05). Further, the data also support the H2 prediction that the sign of forecast news has less effect on Canadian firms' forecast horizon. As expected, Canadian firms are less apt to choose longer forecast horizons for good news relative to bad news (significantly negative [c.sub.3], p = 0.031).
VI. SENSITIVITY CHECKS
Effects of the Private Securities Litigation Reform Act on U.S. Forecasting Behavior
On December 22, 1995, Congress voted to override a presidential veto, thereby establishing the Private Securities Litigation Reform Act of 1995. One intent of the Act was to encourage firms to disclose forward-looking information by reducing legal penalties. If management forecasting behavior changed after the Act, then including post-Act U.S. forecasts in our sample biases the results in favor of the null hypothesis of no differences between U.S. and Canadian forecasting behavior. We reject the null hypothesis in each of our tests except for H2 in Table 4, where we find no difference between Canada and the U.S. in the association of the sign of the earnings change and forecast precision. Accordingly, we ran a model analogous to Equation (2) on U.S. firms only, replacing the Canadian regime indicator variable (CANADA) with a 1,0 indicator variable for whether the forecast was issued after the Act in 1996. We find that, in 1996, U.S. forecasting firms still issue less precise forecasts in periods of earnings decreases (p = 0.022; results not tabulated). However, running Equation (2) without 1996 U.S. forecasts does not change our inference of no significant difference between Canada and the U.S. in the association of the sign of the earnings change and forecast precision.
Information Asymmetry
The benefits of voluntary management forecast disclosure increase when private information exists (King et al. 1990). Fowler et al. (1979) and Jorion and Schwartz (1986) document thin-trading for many TSE firms, a symptom of information asymmetry. (20) If Canadian managers wish to reduce such asymmetry to decrease the firm's cost of capital (King et al. 1990), then they will be more likely to issue management earnings forecasts, and the forecasts will on average be longer-horizon. Although Canadian managers may also wish to issue more precise forecasts when faced with greater information asymmetry, their forecasts will likely be less precise because information asymmetry is strongly linked to uncertainty, and Baginski and Hassell (1997) document that management forecast precision is inversely related to uncertainty.
We replicated our entire study for the subset of firm-periods for which more than one analyst follows the firm at the beginning of the period. We use the standard deviation of these analyst forecasts as a proxy for predisclosure information asymmetry (e.g., Atiase and Bamber 1994). Specifically, we divide the standard deviation by beginning of the period price, rank the deflated variable, and include it as an additional control variable. Greater information asymmetry leads to greater forecasting benefits (i.e., removal of information asymmetry). Thus, our proxy should be associated with more management forecasts, longer forecast horizons, and due to its link to uncertainty, with less precise forecasts. The results (not tabulated) are consistent with these expectations, but our conclusions concerning H1 and H2 remain unchanged. (21)
VII. SUMMARY AND CONCLUSIONS
Policymakers regularly call for U.S. corporations to increase disclosure of forward-looking information (e.g., AICPA 1994; Breeden 1995). In this paper, we show that the frequency and characteristics of voluntary earnings forecast disclosures differ across two countries with different legal regimes. Canadian managers, faced with a less litigious environment than U.S. managers, disclose more earnings forecasts (in periods of both good and bad news), and their forecasts are more precise and of longer range. In addition, using a larger and more representative sample of firms, we confirm the findings in prior research that U.S. forecasts occur more often, are less precise, and are of a shorter horizon when earnings news is bad. In contrast, we document that Canadian forecasts are less likely to exhibit these bad-news-related tendencies. In fact, Canadian firms issue more forecasts in earnings increase relative to earnings decrease periods.
Our evidence that lower legal liability is consistent with more voluntary management forecast disclosure is consistent with evidence in Frost (2001) and Johnson et al. (2001). Our finding that Canadian firms issue proportionally more good news forecasts in the lower legal liability regime is consistent with Skinner's (1994) and Kasznik and Lev's (1995) conclusions that legal exposure leads to asymmetric incentives favoring proportionally more disclosure of bad news. However, the finding is inconsistent with Johnson et al. (2001), who fail to find a proportional decrease in bad news forecasts (or proportional increase in good news forecasts) after passage of the Private Securities Litigation Reform Act that was intended to reduce legal liability. This pattern of evidence suggests that either (1) the difference in legal exposure between the U.S. and Canada is greater than the change in legal exposure in the U.S. brought about by the Private Securities Litigation Reform Act, or that (2) post-Act behavior of firms in high-tech industries differs substantially from that of firms in other industries.
Although we are unaware of any documented biases in our data sources that could confound our analysis, sampling of periods available for disclosure and actual forecast disclosures is limited by the completeness of sources such as Compustat and DJNRS. Although our sample is from the relatively recent 1993-1996 period, our results may not be generalizable to the present. U.S. voluntary disclosure policy is constantly evolving, and managers may now have more evidence on whether the Private Securities Litigation Reform Act of 1995 actually reduces disclosure-related legal exposure.
Our evidence suggests that legal regime is associated with management forecasting behavior. Future research could assess whether lower legal liability undermines the credibility of Canadian management forecasts. To date, research on the relation between legal liability and forecast credibility focuses on the passage of the Private Securities Litigation Reform Act of 1995, and its results are mixed. Johnson et al. (2001) find no change in management forecast accuracy and bias post-Act; however, Ali and Kallapur's (2001) results suggest that shareholders reacted negatively to the Act's restrictions on their ability to bring suit. Substantial difference in legal regimes across countries might provide a powerful treatment effect for detecting differences in forecast credibility and shareholders' perceptions of how their wealth is affected by protections under different legal regimes.
FIGURE 1
Mapping Equation (1) Coefficients into Hypothesis Tests (a)
Good News Period
[DELTA]ESIGN = 1
Canadian
(CANADA = 1) [a.sub.0] + [a.sub.1] + [a.sub.2] + [a.sub.3]
U.S.
(CANADA = 0) [a.sub.0] + [a.sub.1]
Difference (H1 test during good
between countries news period):
(row 1 - row 2) [a.sub.2] + [a.sub.3] > 0
Difference across
Sign of
Bad News Period Earnings Change
[DELTA]ESIGN = 0 (column 1 - column 2)
Canadian
(CANADA = 1) [a.sub.0] + [a.sub.2] [a.sub.1] + [a.sub.3]
U.S.
(CANADA = 0) [a.sub.0] [a.sub.1]
Difference (H1 test during bad Difference between
between countries news period): countries in
(row 1 - row 2) [a.sub.2] > 0 sign-related
behavior (H2):
[a.sub.3] > 0
(a) Equation (1): [FORECAST.sub.i,t] = [a.sub.0] + [a.sub.1]
[DELTA][ESIGN.sub.i,t] + [a.sub.2] [CANADA.sub.i,t] + [a.sub.3]
[CANADA.sub.i,t] x [DELTA][ESIGN.sub.i,t] + [a.sub.4]
[absolute value of [DELTA][EARNINGS.sub.i,t]] + [a.sub.5]
[LSIZE.sub.i,t] + [a.sub.6] [REGULATE.sub.i,t] + [a.sub.7]
[HIGHTECH.sub.i,t] + [[epsilon].sub.i,t]
To map equations (2), (3), and (4) into the hypotheses, replace the
"a" coefficients with "b," "c," and "c" coefficients, respectively,
and reverse the sign predictions on b3 and c3. Then refer to the
following models:
Equation (2): [PRECISE.sub.i] = [b.sub.0] + [b.sub.2]
[DELTA][ESIGN.sub.i] + [b.sub.2] [CANADA.sub.i] + [b.sub.3]
[CANADA.sub.i] x [DELTA][ESIGN.sub.i] + [b.sub.4] [LSIZE.sub.i]
+ [b.sub.5] [REGULATE.sub.i] + [b.sub.6] [HIGHTECH.sub.i] +
[b.sub.7] [FHORIZON.sub.i] + [v.sub.i]
Equation (3): [FHORIZON.sub.i] = [c.sub.0] + [c.sub.1] [NEWS.sub.i]
+ [c.sub.2] [CANADA.sub.i] + [c.sub.3] [CANADA.sub.i] x [NEWS.sub.i]
+ [b.sub.4] [LSIZE.sub.i] + [b.sub.5] [REGULATE.sub.i] + [b.sub.6]
[HIGHTECH.sub.i] + [[mu].sub.i]
Equation (4): ALTFHORIZON = [c.sub.0] + [c.sub.1] [NEWS.sub.i] +
[c.sub.2] [CANADA.sub.i] + [c.sub.3] [CANADA.sub.i] x [NEWS.sub.i]
+ [c.sub.4] [LSIZE.sub.i] + [c.sub.5] [REGULATE.sub.i] + [c.sub.6]
[HIGHTECH.sub.i] + [[eta].sub.i]
[DELTA]ESIGN = 1 for [DELTA]EARNINGS [greater than or equal to] 0
(i.e., good news), 0 for [DELTA]EARNINGS < 0 (i.e., bad news), where
[DELTA][EARNINGS.sub.i,t] = [[EPS.sub.i,t]
- [EPS.sub.i,t-k])]/[PRICE.sub.i,t-k];
[EPS.sub.i,t] = earnings per share for firm i in period t;
[EPS.sub.i,t-k] = earnings per share for firm i in period t-k;
[PRICE.sub.i,t-k] = security price for firm i in period t-k; and
k = 1 and 4 for annual and quarterly observations, respectively.
FORECAST = 1 if the firm issued a management forecast during the
period, and 0 otherwise;
PRECISE = 3 (2, 1, 0) for point (range, open-interval, general
impression) forecasts;
CANADA = 1 if the potential forecasting period relates to a Canadian
firm, and 0 if it relates to a U.S. firm;
NEWS = 1 if the forecast is good news, and 0 if the forecast is bad
news. Good news disclosures include management forecast point
estimates above expected earnings, range estimates where the midpoint
exceeds expected earnings, and all minimum estimates. Expected earnings
equals the median I/B/E/S analyst forecast when available. Otherwise,
expected earnings equals the random-walk EPS for annual forecasts and
seasonal random-walk for quarterly forecasts. Bad news disclosures
include point estimates below expected earnings, range estimates where
the midpoint is less than expected earnings, and all maximum estimates;
FHORIZON = the number of calendar days between the date of forecast and
the end of the period being forecast;
ALTFHORIZON = the number of calendar days between the date of forecast
and the end of the firm's fiscal year;
LSIZE = natural log of the market value of equity at the beginning of
the period;
HIGHTECH = 1 if a firm belongs to Pharmaceuticals (Compustat SIC codes
2833-2836), R&D Services (8731-8734), Programming (7371-7379),
Computers (3570-3577), or Electronics (3600-3674) industries, and 0
otherwise; and
REGULATE = 1 if a firm belongs to Telephone (4812-4813), TV (4833),
Cable (4841), Communications (4811-4899), Gas (4922-4924), Electricity
(4931), Water (4941), or Financial (6021-6023, 6035-6036, 6141, 6311,
6321, 6331) industries, and 0 otherwise.
TABLE 1
Sample Selection Criteria and Sample Characteristics
Panel A: Selection of 1993-1996 Firm-Quarters and Firm-Years
U.S. Canadian
Firms Firms Total
Firm-years available
on Compustat 51,434 5,073 56,507
Missing data (24,895) (2,102) (26,997)
Surviving firm-years 26,539 2,971 29,510
Firm-quarters available
on Compustat 132,737 19,694 152,431
Missing data (52,600) (13,590) (66,190)
Surviving firm-quarters 80,137 6,104 86,241
Total firm-years and 106,676 9,075 115,751
firm-quarters sample
Distributed across
Following Exchanges Firms % of total
NASDAQ 39,591 34.2
NYSE 34,412 29.7
OTC 25,583 22.2
American Stock Exchange 8,737 7.5
Toronto Stock Exchange 5,757 5.0
Regional exchanges 893 0.8
Vancouver Stock Exchange 378 0.3
Nonlisted Canadian 253 0.2
Montreal Exchange 147 0.1
Total firm-years and
firm-quarters sample 115,751 100.0
Panel B: Selection of 1993-1996 Management Earnings Forecasts
Issued Issued
by by
U.S. Canadian
Firms Firms Total
Earnings forecasts
identified by keyword
search of Dow Jones
News Retrieval Service 1,688 391 2,079
Lost due to changes in
Compustat coverage (281) (78) (359)
Listed on Compustat 1,407 313 1,720
Missing Compustat data (74) (36) (110)
Complete data on
Compustat 1,333 277 1,610
Forecasts of 1992 and
1997 results (112) (26) (138)
Long-range (> 365 days)
forecasts (57) (32) (89)
Final management earnings
forecast sample 1,164 219 1,383
Distributed across % of
Following Exchanges Forecasts Total
NYSE 649 46.9
NASDAQ 376 27.2
Toronto Stock Exchange 206 14.9
Other U.S. 139 10.1
Other Canadian 13 0.9
Final management earnings
forecast sample 1,383 100.0
Forecasts
Distributed across Firms Firms (%) (%)
U.S. firms that issued:
one forecast 398 (57.9) 398 (34.2)
two forecasts 181 (26.4) 362 (31.1)
three forecasts 64 (9.3) 192 (16.5)
four or more forecasts 44 (6.4) 212 (18.2)
687 (100.0) 1,164 (100.0)
Canadian locally listed
firms that issued:
one forecast 38 (59.4) 38 (34.9)
two forecasts 16 (25.0) 32 (29.3)
three forecasts 5 (7.8) 15 (13.8)
four or more forecasts 5 (7.8) 24 (22.0)
64 (100.0) 109 (100.0)
Canadian cross-listed
firms that issued:
one forecast 20 (40.0) 20 (18.2)
two forecasts 16 (32.0) 32 (29.1)
three forecasts 8 (16.0) 24 (21.8)
four or more forecasts 6 (12.0) 34 (30.9)
50 (100.0) 110 (100.0)
TABLE 2
Variable Distributions for 1993-1996 Sample of 115,751 Potential
Forecasting Periods (Panel A) and 1993-1996 Sample of 1,236
Management Earnings Forecasts (Panel B)
Panel A: Variable Distributions for Sample of 115,571 Potential
Forecasting Periods and Frequencies and Types of Forecasts Released
during those Periods
Continuous Standard
Variables n Mean Deviation
[DELTA]EARNINGS 115,751 0.0121 0.2258
U.S. 106,676 0.0113 0.2237
Canadian 9,075 0.0215 0.2494
[absolute value of
[DELTA]EARNINGS] 115,751 0.0850 0.2096
U.S. 106,676 0.0838 0.2077
Canadian 9,075 0.0991 0.2298
SIZE (millions) 115,751 984.3 4315.3
U.S. 106,676 1001.5 4459.5
Canadian 9,075 782.3 1926.4
Continuous Lower Upper
Variables Quartile Median Quartile
[DELTA]EARNINGS -0.0083 0.0014 0.0124
U.S. -0.0082 0.0014 0.0120
Canadian -0.0083 0.0016 0.0182
[absolute value of
[DELTA]EARNINGS] 0.0027 0.0105 0.0462
U.S. 0.0027 0.0130 0.0452
Canadian 0.0033 0.0130 0.0589
SIZE (millions) 22.5 86.7 396.1
U.S. 22.2 85.1 386.3
Canadian 29.2 107.3 535.4
Classification Variables Yes (%) No (%)
High-tech firm (HIGHTECH = 1) 20,058 (17.3) 95,693 (82.7)
U.S. 18,913 (17.7) 87,763 (82.3)
Canadian 1,145 (12.6) 7,930 (87.4)
Regulated firm (REGULATE = 1) 15,116 (13.1) 100,635 (86.9)
U.S. 14,541 (13.6) 92,135 (86.4)
Canadian 575 (6.3) 8,500 (93.7)
Good news period (Earnings
increase; [DELTA]ESIGN = 1) 67,378 (58.2) 48,373 (41.8)
U.S. 62,348 (58.4) 44,328 (41.6)
Canadian 5,030 (55.4) 4,045 (44.6)
Classification Variables Total (%)
High-tech firm (HIGHTECH = 1) 115,751 (100.0)
U.S. 106,676 (100.0)
Canadian 9,075 (100.0)
Regulated firm (REGULATE = 1) 115,751 (100.0)
U.S. 106,676 (100.0)
Canadian 9,075 (100.0)
Good news period (Earnings
increase; [DELTA]ESIGN = 1) 115,751 (100.0)
U.S. 106,676 (100.0)
Canadian 9,075 (100.0)
Forecast Rates in Potential Forecast Forecasts Periods Rate
Periods
Canadian firms 219 9,075 2.41%
Good news periods ([DELTA]EARNINGS 150 5,030 2.98%
[greater than or equal to] 0)
Bad news periods
([DELTA]EARNINGS < 0) 69 4,045 1.71%
U.S. firms 1,164 106,676 1.09%
Good news periods 685 62,348 1.10%
Bad news periods 479 44,328 1.08%
Management Forecast U.S. Firm Canadian Total (%)
Type Firm (%)
Point 282 (24.2) 101 (46.1) 383 (27.7)
Range 292 (25.1) 32 (14.6) 324 (23.4)
Minimum 309 (26.5) 48 (21.9) 357 (25.8)
Maximum 151 (13.0) 14 (6.4) 165 (12.0)
General impression 130 (11.2) 24 (11.0) 154 (11.1)
Total 1,164 (100.0) 219 (100.0) 1,383 (100.0)
Panel B: Variable Distributions for 1993-1996 Sample of 1,236
Management Earnings Forecasts for which Forecast News Can Be
Calculated (General Impression Forecasts Excluded)
Continuous Standard
Variables n Mean Deviation
FHORIZON 1,236 89.2 107.4
U.S. 1,040 70.5 98.1
Canadian 196 188.1 100.3
ALTFHORIZON 1,236 152.5 111.6
U.S. 1,040 142.5 112.9
Canadian 196 205.5 87.6
SIZE (millions) 1,236 1,789.5 4,523.0
U.S. 1,040 1,893.3 4,862.4
Canadian 196 1,238.5 1,795.9
[absolute value of
[DELTA]EARNINGS] 1,236 0.0343 0.1013
U.S. 1,040 0.0347 0.1079
Canadian 196 0.0324 0.0536
Continuous Lower Upper
Variables Quartile Median Quartile
FHORIZON 6.0 49.0 188.0
U.S. 2.0 30.0 105.5
Canadian 145.0 215.0 246.0
ALTFHORIZON 63.0 160.0 245.0
U.S. 54.0 139.0 239.5
Canadian 197.0 228.0 258.0
SIZE (millions) 74.9 300.7 1397.8
U.S. 66.6 280.4 1374.9
Canadian 138.8 541.9 1526.7
[absolute value of
[DELTA]EARNINGS] 0.0023 0.0080 0.0247
U.S. 0.0019 0.0072 0.0221
Canadian 0.0038 0.0113 0.0361
Classification
Variables Yes (%) No (%) Total (%)
High-tech firm
(HIGHTECH = 1) 211 (17.1) 1,025 (82.9) 1,236 (100.0)
U.S. 193 (18.6) 847 (81.4) 1,040 (100.0)
Canadian 18 (9.2) 178 (90.8) 196 (100.0)
Regulated firm
(REGULATE = 1) 45 (3.6) 1,191 (96.4) 1,236 (100.0)
U.S. 39 (3.8) 1,001 (96.2) 1,040 (100.0)
Canadian 6 (3.1) 190 (96.9) 196 (100.0)
Interim forecast 635 (51.4) 601 (48.6) 1,236 (100.0)
U.S. 620 (59.6) 420 (40.4) 1,040 (100.0)
Canadian 15 (7.7) 181 (92.3) 196 (100.0)
Analyst following 910 (73.6) 326 (26.4) 1,236 (100.0)
U.S. 758 (72.9) 282 (27.1) 1,040 (100.0)
Canadian 152 (77.6) 44 (22.4) 196 (100.0)
Forecast News Total (%) U.S. (%) Canadian (%)
Good news
(NEWS = 1) 769 (62.2) 650 (62.5) 119 (60.7)
Neutral news
(discarded) 32 (2.6) 25 (2.4) 7 (3.6)
Bad news
(NEWS = 0) 435 (35.2) 365 (35.1) 70 (35.7)
Total 1,236 (100.0) 1,040 (100.0) 196 (100.0)
FHORIZON = the number of calendar days between the forecast and the
end of the period being forecast.
ALTFHORIZON = the number of calendar days between the forecast and
the end of the firm's fiscal year;
[DELTA][EARNINGS].sub.i,t] = [[EPS.sub.i,t] - [EPS.sub.i,t-k]]/
[PRICE.sub.i,t-k];
[EPS.sub.i,t] = earnings per share for firm i in period t;
[EPS.sub.i,t-k] = earnings per share for firm i in period t - k;
[PRICE.sub.i,t-k] = security price for firm i in period t - k; and
k = 1 and 4 for annual and quarterly observations, respectively.
[DELTA]ESIGN = 1 for [DELTA]EARNINGS [greater than or equal to] 0
(i.e., good news), 0 for [DELTA]EARNINGS < 0 (i.e., bad news);
[absolute value of [DELTA][EARNINGS.sub.i,t]] = absolute value of
[DELTA][EARNINGS.sub.i,t];
SIZE = market value of equity at the beginning of the period;
NEWS = 1 if the forecast is good news, and 0 if the forecast is bad
news. Good news disclosures include management forecast point
estimates above expected earnings, range estimates where the
midpoint exceeds expected earnings, and all minimum estimates.
Expected earnings equals the median I/B/E/S analyst forecast when
available. Otherwise, expected earnings equals the random-walk EPS
for annual forecasts and seasonal random-walk for quarterly
forecasts. Bad news disclosures include point estimates below
expected earnings, range estimates where the midpoint is less than
expected earnings, and all maximum estimates;
HIGHTECH = 1 if a firm belongs to Pharmaceuticals (Compustat SIC
codes 2833-2836), R&D Services (8731-8734), Programming
(7371-7379), Computers (3570-3577), or Electronics (3600-3674)
industries, and 0 otherwise; and REGULATE = 1 if a firm belongs to
Telephone (4812-4813), TV (4833), Cable (4841), Communications
(4811-4899), Gas (4922-4924), Electricity (4931), Water (4941), or
Financial (6021-6023, 6035-6036, 6141, 6311, 6321, 6331)
industries, and 0 otherwise.
TABLE 3
Management Earnings Forecast Frequency in the U.S. and Canada
[FORECAST.sub.i,t] = [a.sub.0] + [a.sub.1] [DELTA][ESIGN.sub.i,t]
+ [a.sub.2] [CANADA.sub.i,t] + [a.sub.3] [CANADA.sub.i,t] x
[DELTA][ESIGN.sub.i,t] + [a.sub.4] [absolute value of
[DELTA][EARNINGS.sub.i,t]]
+ [a.sub.5] [LSIZE.sub.i,t] + [a.sub.6] [REGULATE.sub.i,t]
+ [a.sub.7] [HIGHTECH.sub.i,t] + [[epsilon].sub.i,t]
Independent Variable (Coefficient) Expected Sign (Hypothesis)
Intercept ([a.sub.0]) None predicted
[DELTA]ESIGN ([a.sub.1]) -
CANADA ([a.sub.2]) + (H1 for bad news)
CANADA x [DELTA]ESIGN ([a.sub.3]) + (H2)
[absolute value of
[DELTA][EARNINGS.sub.i,t]]
([a.sub.4]) -
LSIZE ([a.sub.5]) +
REGULATE ([a.sub.6]) -
HIGHTECH ([a.sub.7]) None predicted
Coefficients [a.sub.2] + [a.sub.3] + (H1 for good news)
n
Chi-square
p-value
Independent Variable (Coefficient) Coefficient Estimate
(p-value (a))
Intercept ([a.sub.0]) -5.65 (0.001)
[DELTA]ESIGN ([a.sub.1]) -0.14 (0.009)
CANADA ([a.sub.2]) 0.33 (0.006)
CANADA x [DELTA]ESIGN ([a.sub.3]) 0.59 (0.001)
[absolute value of
[DELTA][EARNINGS.sub.i,t]]
([a.sub.4]) -0.52 (0.007)
LSIZE ([a.sub.5]) 0.27 (0.001)
REGULATE ([a.sub.6]) -1.51 (0.001)
HIGHTECH ([a.sub.7]) -0.10 (0.149)
Coefficients [a.sub.2] + [a.sub.3] 0.92 (0.001)
n 115,751
Chi-square 764.8
p-value (0.001)
(a) p-values are one-tailed for signed predictions.
[absolute value of [DELTA][EARNINGS.sub.i,t]] = absolute value of
[DELTA][EARNINGS.sub.i,t], where [DELTA][EARNINGS.sub.i,t]
= [[EPS.sub.i,t] - [EPS.sub.i,t-k]]/[PRICE.sub.i,t-k];
[EPS.sub.i,t] = earnings per share for firm i in period t;
[EPS.sub.i,t-k] = earnings per share for firm i in period t - k;
[PRICE.sub.i,t-k] = security price for firm i in period t - k; and
k = 1 and 4 for annual and quarterly observations, respectively.
[DELTA]ESIGN = 1 for [DELTA]EARNINGS [greater than or equal to] 0
(i.e., good news), 0 for [DELTA]EARNINGS < 0 (i.e., bad news);
FORECAST = 1 if the firm issued a management earnings forecast
during the period, and 0 otherwise;
CANADA = 1 if the potential forecasting period relates to a
Canadian firm, and 0 if it relates to a U.S. firm;
LSIZE = natural log of the market value of equity at the beginning
of the period;
HIGHTECH = 1 if a firm belongs to Pharmaceuticals (Compustat SIC
codes 2833-2836), R&D Services (8731-8734), Programming
(7371-7379), Computers (3570-3577), or Electronics (3600-3674)
industries, and 0 otherwise; and
REGULATE = 1 when a firm belongs to Telephone (4812-4813), TV
(4833), Cable (4841), Communications (4811-4899), Gas (4922-4924),
Electricity (4931), Water (4941), or Financial (6021-6023,
6035-6036, 6141, 6311, 6321, 6331) industries, and 0 otherwise.
TABLE 4
Determinants of Management Forecast Precision in the U.S. and
Canada
Model: [PRECISE.sub.i] = [b.sub.0] + [b.sub.1] [DELTA][ESIGN.sub.i]
+ [b.sub.2] [CANADA.sub.i] + [b.sub.3] [CANADA.sub.i] x
[DELTA][ESIGN.sub.i] + [b.sub.4] [LSIZE.sub.i + [b.sub.5]
[REGULATE.sub.i] + [b.sub.6] [HIGHTECH.sub.i] + [b.sub.7]
[FHORIZON.sub.i] + [[upsilon].sub.i]
Independent Variable (Coefficient) Expected Sign (Hypothesis)
[DELTA]ESIGN ([b.sub.1]) +
CANADA ([b.sub.2]) + (H1 for bad news)
CANADA x [DELTA]ESIGN ([b.sub.3]) - (H2)
LSIZE ([b.sub.4]) -
REGULATE ([b.sub.5]) -
HIGHTECH ([b.sub.6]) None predicted
FHORIZON ([b.sub.7]) -
Coefficients [b.sub.2] + [b.sub.3] + (H1 for good news)
n
Model Chi-square
probability
Independent Variable (Coefficient) Coefficient Estimate
(p-value (a))
[DELTA]ESIGN ([b.sub.1]) 0.211 (0.028)
CANADA ([b.sub.2]) 1.060 (0.001)
CANADA x [DELTA]ESIGN ([b.sub.3]) -0.141 (0.301)
LSIZE ([b.sub.4]) -0.007 (0.393)
REGULATE ([b.sub.5]) -0.247 (0.155)
HIGHTECH ([b.sub.6]) -0.178 (0.090)
FHORIZON ([b.sub.7]) -0.003 (0.001)
Coefficients [b.sub.2] + [b.sub.3] 0.919 (0.001)
n 1,383
Model Chi-square 67.13
probability (0.001)
(a) p-values are one-tailed for signed predictions.
PRECISE = 3 (2, 1, 0) for point (range, open-interval, general
impression) forecasts;
[DELTA][EARNINGS.sub.i,t] = [[EPS.sub.i,t] - [EPS.sub.i,t-k]]/
[PRICE.sub.i,t-k];
[EPS.sub.i,t] = earnings per share for firm i in period t;
[EPS.sub.i,t-k] = earnings per share for firm i in period t - k;
[PRICE.sub.i,t-k] = security price for firm i in period t - k; and
k = 1 and 4 for annual and quarterly observations, respectively.
[DELTA]ESIGN = 1 for [DELTA]EARNINGS [greater than or equal to] 0
(i.e., good news), 0 for [DELTA]EARNINGS < 0 (i.e., bad news);
CANADA = 1 if a Canadian firm issued the forecast, and 0 if a U.S.
firm issued the forecast;
LSIZE = natural log of the market value of equity at the beginning
of the period;
HIGHTECH = 1 if a firm belongs to Pharmaceuticals (Compustat SIC
codes 2833-2836), R&D Services (8731-8734), Programming
(7371-7379), Computers (3570-3577), or Electronics (3600-3674)
industries, and 0 otherwise;
REGULATE = 1 if a firm belongs to Telephone (4812-4813), TV (4833),
Cable (4841), Communications (4811-4899), Gas (4922-4924),
Electricity (4931), Water (4941), or Financial (6021-6023,
6035-6036, 6141, 6311, 6321, 6331) industries, and 0 otherwise; and
FHORIZON = the number of calendar days between the forecast and the
end of the period being forecast.
TABLE 5
Determinants of Management Forecast Horizons in the U.S. and Canada
Model 1: [FHORIZON.sub.i] = [c.sub.0] + [c.sub.1] [NEWS.sub.i]
+ [c.sub.2] [CANADA.sub.i] + [c.sub.3] [CANADA.sub.i] x [NEWS.sub.i]
+ [c.sub.4] [LSIZE.sub.i] + [c.sub.5] [REGULATE.sub.i] + [c.sub.6]
[HIGHTECH.sub.i] + [[mu].sub.i]
Model 2: [ALTFHORIZON.sub.i] = [c.sub.0] + [c.sub.1] [NEWS.sub.i]
+ [c.sub.2] [CANADA.sub.i] + [c.sub.3] [CANADA.sub.i] x [NEWS.sub.i]
+ [c.sub.4] [LSIZE.sub.i] + [c.sub.5] [REGULATE.sub.i] + [c.sub.6]
[HIGHTECH.sub.i] + [[eta].sub.i]
Independent Expected Sign
Variable (Coefficient) (Hypothesis)
Intercept ([c.sub.0]) None predicted
NEWS ([c.sub.1]) +
CANADA ([c.sub.2]) + (H1 for bad news)
CANADA x NEWS ([c.sub.3]) - (H2)
LSIZE ([c.sub.4]) +
REGULATE ([c.sub.5]) -
HIGHTECH ([c.sub.6]) None predicted
Coefficients [c.sub.2] + [c.sub.3] + (H1 for good news)
n
Adjusted [R.sup.2]
Model F
Independent Estimated Coefficients (p-values (a))
Variable (Coefficient) Model 1 Model 2
Intercept ([c.sub.0]) 31.69 (0.001) 110.76 (0.001)
NEWS ([c.sub.1]) 30.61 (0.001) 11.67 (0.050)
CANADA ([c.sub.2]) 141.82 (0.001) 80.78 (0.001)
CANADA x NEWS ([c.sub.3]) -35.18 (0.013) -33.34 (0.031)
LSIZE ([c.sub.4]) 3.78 (0.004) 4.86 (0.002)
REGULATE ([c.sub.5]) 12.35 (0.800) 20.43 (0.891)
HIGHTECH ([c.sub.6]) -15.47 (0.038) -18.91 (0.012)
Coefficients [c.sub.2]
+ [c.sub.3] 106.64 (0.001) 47.44 (0.001)
n 1,204 1,204
Adjusted [R.sup.2] 0.1931 0.0556
Model F 48.97 12.80
(a) p-values are one-tailed for signed predictions.
FHORIZON = the number of calendar days between the forecast and the
end of the period being forecast;
ALTFHORIZON = the number of calendar days between the forecast and
the end of the firm's fiscal year;
NEWS = 1 if the forecast is good news, and 0 if the forecast is bad
news. Good news disclosures include management forecast point
estimates above expected earnings, range estimates where the
mid-point exceeds expected earnings, and all minimum estimates.
Expected earnings equals the median I/B/E/S analyst forecast when
available. Otherwise, expected earnings equals the random-walk EPS
for annual forecasts and seasonal random-walk for quarterly
forecasts. Bad news disclosures include point estimates below
expected earnings, range estimates where the midpoint is less than
expected earnings, and all maximum estimates;
CANADA = 1 if a Canadian firm issued the forecast, and 0 if a U.S.
firm issued the forecast;
LSIZE = natural log of the market value of equity at the beginning
of the period;
HIGHTECH = 1 if a firm belongs to Pharmaceuticals (Compustat SIC
codes 2833-2836), R&D Services (8731-8734), Programming
(7371-7379), Computers (3570-3577), or Electronics (3600-3674)
industries, and 0 otherwise; and
REGULATE = 1 if a firm belongs to Telephone (4812-4813), TV (4833),
Cable (4841), Communications (4811-4899), Gas (4922-4924),
Electricity (4931), Water (4941), or Financial (6021-6023,
6035-6036, 6141, 6311, 6321, 6331) industries, and 0 otherwise.
REFERENCES
Ajinkya, B., and M. Gift. 1984. Corporate managers' earnings forecasts and symmetrical adjustments of market expectations. Journal of Accounting Research 22 (Autumn): 425-444.
Ali, A., and L. Hwang. 2000. Country-specific factors related to financial reporting and the value relevance of accounting data. Journal of Accounting Research 38 (Spring): 1-21.
--, and S. Kallapur. 2001. Securities price consequences of the Private Securities Litigation Reform Act of 1995 and related events. The Accounting Review 76 (July): 431-460.
American Institute of Certified Public Accountants (AICPA), Special Committee on Financial Reporting. 1994. Improving Business Reporting--A Customer Focus. New York, NY: AICPA.
Atiase, R., and L. S. Bamber. 1994. Trading volume reactions to annual accounting earnings announcements: The incremental role of predisclosure information asymmetry. Journal of Accounting and Economics 17 (May): 309-329.
Baginski, S., J. Hassell, and G. Waymire. 1994. Some evidence on the news content of preliminary earnings estimates. The Accounting Review 69 (January): 265-273.
--, and --. 1997. Determinants of management forecast precision. The Accounting Review (April): 303-312.
Bamber, L. S., and Y. Cheon. 1998. Discretionary management earnings forecast disclosures: Antecedents and outcomes associated with forecast venue and forecast specificity choices. Journal of Accounting Research 36 (Autumn): 167-190.
Bandyopadhyay, S., J. D. Hanna, and G. Richardson. 1994. Capital market effects of U.S.-Canada GAAP differences. Journal of Accounting Research 32 (Autumn): 262-277.
Barth, M., and G. Clinch. 1996. International accounting differences and their relation to share prices: Evidence of U.K., Australian, and Canadian firms. Contemporary Accounting Research (Spring): 135-170.
Breeden, R. 1995. Testimony before the Securities Subcommittee of the Senate Committee on Banking, Housing, and Urban Affairs. April 6, 1995. Washington, D.C.: Government Printing Office.
Brown, L., G. Richardson, and C. Trzcinka. 1991. Strong-form efficiency on the Toronto Stock Exchange: An examination of analyst price forecasts. Contemporary Accounting Research 7: 323-346.
Clarkson, P., A. Dontoh, G. Richardson, and S. Sefcik. 1992. The voluntary inclusion of earnings forecasts in IPO prospectuses. Contemporary Accounting Research (Spring): 601-626.
--, and D. Simunic. 1994. The association between audit quality, retained ownership, and firm-specific risk in U.S. vs. Canadian IPO markets. Journal of Accounting and Economics 17 (January): 207-228.
Core, J. 1997. On the corporate demand for directors' and officers' insurance. Journal of Risk and Insurance 64 (March): 63-87.
Cox, C. 1985. Further evidence on the representativeness of management earnings forecasts. The Accounting Review 60 (October): 692-701.
Francis, J., D. Philbrick, and K. Schipper. 1994. Shareholder litigation and corporate disclosures. Journal of Accounting Research 32 (Autumn): 137-104.
Fowler, D., C. H. Rorke, and V. Jog. 1979. Heteroscedasticity, [R.sup.2] and thin trading on the Toronto Stock Exchange. Journal of Finance 34: 1201-1210.
Freund, W., M. Larrain, and M. Pagano. 1997. Market efficiency before and after the introduction of electronic trading at the Toronto Stock Exchange. Review of Financial Economics 6: 29-56.
Frost, C., and G. Pownall. 1994. Accounting disclosure practices in the United States and the United Kingdom. Journal of Accounting Research 32 (Spring): 75-102.
--, and W. Kinney, Jr. 1996. Disclosure choices of foreign registrants in the United States. Journal of Accounting Research 34 (Spring): 67-84.
--. 2001. Characteristics and information value of corporate disclosures of forward-looking information in global equity markets. Working paper, Dartmouth College.
Grossman, B. 1996. Class rules. CA Magazine 129 (March): 31-33, 43.
Grundfest, J., and M. Perino. 1997. Security litigation reform: The first years' experience. Working paper, Stanford University.
Guenther, D., and A. Rosman. 1994. Differences between COMPUSTAT and CRSP SIC codes and related effects on research. Journal of Accounting and Economics 18 (July): 115-128.
Johnson, M., R. Kasznik, and K. Nelson. 2001. The impact of securities litigation reform on the disclosure of forward-looking information by high technology firms. Journal of Accounting Research 39 (September): 297-327.
Jorion, P., and E. Schwartz. 1986. Integration vs. segmentation in the Canadian stock market. Journal of Finance 41 (July): 603-614.
Kasznik, R., and B. Lev. 1995. To warn or not to warn: Management disclosures in the face of an earnings surprise. The Accounting Review 70 (January): 113-134.
King, R., G. Pownall, and G. Waymire. 1990. Expectations adjustment via timely management forecasts: Review, synthesis, and suggestions for future research. Journal of Accounting Literature 9:113-144.
Levine, D., and A Pritchard. 1998. The Securities Litigation Uniform Standards Act of 1998: The sun sets on California's Blue Sky Laws. Business Lawyer 54 (November): 1- 54.
L'Her, J., and J. Suret. 1991. The reaction of Canadian securities to revisions of earnings forecasts. Contemporary Accounting Research (Spring): 378-406.
McConomy, B. 1998. Bias and accuracy of management earnings forecasts: An evaluation of the impact of auditing. Contemporary Accounting Research 15 (Summer): 167-195.
McNichols, M. 1989. Evidence of informational asymmetries from management earnings forecasts and stock returns. The Accounting Review 64 (January): 1-27.
Paskell-Mede, M. 1994. What liability crisis? CA Magazine 127 (May): 42-43.
--. 1999. Cross-border lawsuits. CA Magazine 132 (April): 37-38.
Patell, J. 1976. Corporate forecasts of earnings per share and stock price behavior: Empirical tests. Journal of Accounting Research (Autumn): 246-276.
Skinner, D. 1994. Why firms voluntarily disclose bad news. Journal of Accounting Research 32 (Spring): 38-60.
--. 1997. Earnings disclosure and stockholder lawsuits. Journal of Accounting and Economics 23 (November): 249-282.
Toronto Stock Exchange (TSE). 2000. The Toronto Stock Exchange's Policy Statement on Timely Disclosure. Toronto, Ontario.
Waymire, G. 1984. Additional evidence on the information content of management earnings forecasts. Journal of Accounting Research 22 (Autumn): 703-718.
--. 1985. Earnings volatility and voluntary management forecast disclosure. Journal of Accounting Research 23 (Spring): 268-295.
We thank workshop participants at the University of Georgia, Washington University, and Indiana University for comments on earlier versions of this paper. We also benefited from discussions with Jamie Pratt, Jerry Salamon, Jim Wahlen, Grace Pownall, Jim Seida, and discussants' comments by Paquita Davis-Friday and Lenny Softer in conjunction with our presentations of this paper at the Second Globalization Conference (2000), jointly sponsored by the American Accounting Association and the British Accounting Association, and the 2000 American Accounting Association Annual Meeting, respectively. Professor Baginski thanks the Center for International Business Education Research (CIBER) at Indiana University for support of the study. We also gratefully acknowledge the contribution of I/B/E/S International Inc. for providing earnings per share forecast data. These data have been provided as part of a broad academic program to encourage earnings expectations research.
Submitted August 2000
Accepted June 2001
(1) Technically, management's failure to disclose material bad news that it should have released to correct misleading or inaccurate prior disclosures triggers legal liability. In practice, however, the mere existence of undisclosed bad news increases the firm's exposure to shareholder lawsuits (see Skinner [1994] and Bamber and Cheon [1998] for related arguments).
(2) Johnson et al. (2001) capitalize on an expected change in legal liability created by passage of the Private Securities Litigation Reform Act of 1995 in the U.S. to test whether management forecasting behavior changed. They examine 523 firms in three high-technology industries and find that managers issue more good news and bad news management earnings and sales forecasts after the Act. However, confining the analysis to high-technology finns limits the results' generalizability. The extent that the Act reduced legal exposure was likely not clear to most managers and their legal counsels. Although the rate of post-Act fraud suit dismissals increased from 40 to 60 percent (Levine and Pritchard 1998), 40 percent of the suits continued. Further, Grundfest and Perino (1997) document a shift of post-Act filings from federal to state courts, presumably to avoid the defendant's additional protections under the Federal Act.
Frost (2001) provides comparative descriptive evidence on disclosure of forward-looking information in the U.S. and several European countries for firms in the manufacturing sector. Samples of 80 firms in each country reveal that managers issue more forecasts and more precise forecasts in the non-U.S. countries where legal and regulatory environments are less stringent.
(3) Empirical evidence suggests that even the stronger Canadian laws for primary offerings do not introduce significant fear of litigation. Clarkson and Simunic (1994) note that approximately 50 percent of Canadian initial public offerings include a management forecast in the offering prospectus (a virtually nonexistent practice in U.S. initial public offerings), and these management forecasts are on average optimistic (Clarkson et al. 1992).
(4) Our data requirements result in a Canadian management forecast sample that is almost entirely composed of large firms listed on the TSE, approximately three-quarters of which have analyst following. The market for shares of large firms with analyst following is more likely to be informationally efficient. Further, if managers release forecasts to move stock prices, a lack of market efficiency would likely result in fewer Canadian management forecasts relative to the U.S. (a bias against the null hypothesis that we reject in tests described later). Also, a bad news management forecast would be a more beneficial legal liability-limiting strategy in Canada because the company could reap the benefits of public disclosure without the associated negative stock price reaction (a bias in favor of the null hypothesis that Canadian management forecasts exhibit the same bad news bias as do U.S. management forecasts--a null that we reject in tests described later).
(5) In an untabulated regression of percentage change in analyst forecasts on percentage change in expectations conveyed by management forecasts, the slope coefficient is significantly positive (p < 0.001). Significant differences do not exist (1) between the slope coefficient for Canadian and U.S. firms, and (2) for the Canadian firms, between locally listed firms and firms cross-listed in the U.S. These results suggest that analysts update their expectations in response to management forecasts.
(6) Like most research, our sample is limited to firms covered by Compustat, which covers only about half of the firms reporting to the SEC in the middle of our sample period. (In the 1994 manual, Compustat reports coverage of approximately 6,800 U.S. companies; Guenther and Rosman [1994] report that 13,424 firms filed with the SEC in 1994.) Although we do not have data on the number of Canadian firms in 1994, the most recent TSE web site lists 1,407 traded firms. The 1998 Compustat tape includes 674 TSE firms, nearly half of the TSE population, which according to its web site, represents 95 percent of market capitalization in Canada. Thus, Compustat Canadian firm coverage spans an important segment of the Canadian market.
(7) Our DJNRS search keywords are: "expects earnings," "expects net," "expects income," "expects losses," "expects profits," "expects results," and three similar lists with first words "forecasts," "predicts," and "sees."
(8) A major source of data loss is a combination of CUSIP and name changes that we are unable to track. Conversations with Compustat personnel have not enabled us to recover these observations. Because the rate of management earnings forecast observation loss is higher for Canadian firms (78/391 = 20 percent) than U.S. firms (281/1,688 = 16.6 percent), and because we find greater forecast frequency for Canadian firms, the loss of observations does not affect our inferences for H1.
(9) Canadian forecasts represent 18.8 percent of the original sample (391 of 2,079). However, 36 percent of the discarded long-range forecasts (32 of 89) are Canadian, indicating that Canadian firms provide more long-range (> 365 days) forecasts.
(10) We also ran OLS regressions, and the results are qualitatively identical. We code single news stories containing both annual and interim earnings forecasts as two separate observations.
(11) Kasznik and Lev (1995) employ alternative quarterly expectation models (analyst forecasts and seasonal random-walk) and document no difference in results. To maximize sample size, results tabulated for Equation (1) are based on the random-walk (seasonal random-walk) expectations to compute annual (quarterly) earnings surprise. Replication with a random-walk for quarterly earnings does not affect our conclusions. Analyst forecast coverage is significantly greater for our U.S. firms (mean analyst coverage of 1.84) than for our Canadian firms (mean analyst coverage of 1.29, Wilcoxon rank sum test p-value < 0.001 on the difference). We were able to obtain I/B/E/S median financial analyst forecasts for 40 percent of the sample. Replication substituting I/B/E/S analyst forecasts when available and limiting the analysis to firms covered by analysts does not affect our inferences.
Stephen P. Baginski John M. Hassell Michael D. Kimbrough Indiana University
In addition, make sure to read these articles:
- Traveling Overseas? Get the Skinny on Currency Exchange
- Laid Off: How to Make Losing Your Job a Winning Venture
- eBay Q3 Earnings and eBay Express
- Canadian firms help produce Mexican economic development conference.
- ADVISORY/News Conference Today on Delta Water Quality, Levee Break.
- News update.
- Alaska/Canada trade values $330M in '04
- Ness Energy: Increased Revenues and Assets, Management Forecast: 'Trend to Continue This...
- Iraq - Turkish & Canadian Firms Recommended For Field Development Contracts.
- CRITICAL ISSUES & VALUABLE SOLUTIONS RISK MANAGEMENT FORECAST: 2001.
- Delta launches Spanish version of website.
- AENS to use Delta-T Corp. ethanol technology.