Regulation Fair Disclosure, analyst following, and analyst forecast dispersion.

SYNOPSIS: This paper presents preliminary evidence of the effect of Regulation Fair Disclosure (FD) on the quantity and quality of firm-specific information released to the market by comparing analyst forecast data from pre-FD to post-FD time periods. By prohibiting selective disclosure of material

information to privileged individuals, the Securities and Exchange Commission intends to provide a level playing field to all investors. However, opponents argue that FD has a negative impact by decreasing the quantity and quality of publicly available information. Consistent with this argument, we document a decrease in analyst following and an increase in forecast dispersion following the passage of FD.

INTRODUCTION

The Securities and Exchange Commission (SEC) passed the Selective Disclosure and Insider Trading Regulation on August 10, 2000. In addition to clarifying two insider-trading issues, this ruling includes a new regulation that prohibits the selective disclosure of material nonpublic information by issuers to privileged individuals. The new regulation, named Regulation Fair Disclosure (FD), states that "when an issuer, or persons acting on its behalf, discloses material nonpublic information to certain enumerated persons (in general, securities market professionals and holders of the issuer's securities who may well trade on the basis of the information), it must make public disclosure of that information" (SEC 2000). The public disclosure must be made "simultaneously" for an intentional selective disclosure and "promptly" for a nonintentional selective disclosure by filing a Form 8-K or through any other medium capable of mass and unbiased distribution (SEC 2000).

In recent times, no other SEC ruling received such widespread attention as FD. Numerous articles in the popular press and surveys of analysts and institutional investors focused on the future implications of this regulation. The SEC argues that the regulation is necessary to provide a level playing field to all investors. Prohibiting selective disclosure forces firms to disclose all information simultaneously to all market participants, an undertaking that opponents contend firms will be reluctant to do in fear of unknowingly releasing proprietary information or facing higher disclosure-related legal liability. Consequently, opponents argue that by attempting to force a level playing field, the regulation will reduce the quantity and quality of information that a firm voluntarily releases to the market.

This study empirically investigates the FD debate by testing the impact of FD on analyst forecast dispersion and analyst following. Both these variables are used in prior literature to proxy for information asymmetry (Lang and Lundholm 1996; Barron and Stuerke 1998). Using conference call data from the pre-FD era, Bowen et al. (2002) find that information availability is negatively correlated with analyst forecast dispersion. Consequently, if FD reduces the quantity and quality of firm-specific information available to all market participants, then we expect lower consensus among analysts and higher forecast dispersion in the post-FD period relative to the pre-FD period. Further, if analysts now have to spend more time researching a firm, then the average number of analysts following a firm will decrease in the post-FD era. Based upon the findings of prior research, reduced analyst following could lead to higher price volatility around earnings announcement dates (Dempsey 1989) and shorter security price lead s on earnings (Ayers and Freeman 2003).

This study uses both univariate and multivariate analysis to investigate the research question. The univariate tests indicate that at the end of every quarter in the post-FD relative to the same quarter in the pre-FD era, analyst following is significantly lower and analyst forecast dispersion is significantly higher. Controlling for confounding factors in a multivariate model provides similar results. These results support the concern of FD opponents regarding the unintended negative impact of the rule on the quantity and quality of corporate disclosures.

REVIEW OF FD AND RELATED RESEARCH

History of FD

Comparing selective disclosure of material information by issuers to insider trading, the SEC argues that new rules and regulations are needed to abolish this practice. The SEC offers three main reasons in support of FD:

1) To elevate investor confidence in the integrity of the capital markets.

2) To address the "potential for corporate management to treat material information as a commodity to be used to gain or maintain favor with particular analysts or investors" (SEC 2000).

3) To take advantage of the technological developments over the past decade that increase the ease with which information can. be disseminated to a wider audience in a relatively short period of time.

These are also the primary reasons cited by supporters of FD in the nearly 6,000 comment letters to the SEC received after the original Proposing Release on December 15 1999; most were from individual investors.

Although selective disclosure has been a controversial issue, the SEC made known its intention to limit selective disclosure in March 1999 (Wall Street Journal 1999).

Brian Lane, the SEC corporate finance director, explained the purpose of FD as "a way to level the playing field" (Bloomberg News 1999). A strong advocate of leveling the playing field is former SEC Chairman Arthur Levitt, who noted on several occasions that certain companies warned selected individuals, predominantly analysts, before making the information public. When the information was revealed, often several hours later, some companies had stock price movements as large as 25 percent (Bloomberg News 1999). For example, shares of General Motors rose 3.2 percent in January 1999 after the automaker told analysts at an invitation-only meeting about plans to raise production of certain vehicles. In February 1999, shares of Lehman Brothers surged 6.8 percent after Chairman Richard Fuld told analysts at a lunch meeting of a positive earnings surprise (Bloomberg News 1999). On June 9, 2000, the stock of Electronic Data Systems plummeted at the start of trading after an earnings warning privately given to some an alysts the night before (Sugawara 2000). These and other instances prompted the SEC to take action to reduce and/or eliminate selective disclosure and protect the integrity of the securities markets.

Why Firms Prefer to Selectively Disclose Information

From the SEC's initial concerns in March 1999 through the eventual adoption of FD in August 2000 and continuing today, there has been an ongoing debate about how this law will impact corporate information released to the security markets. Though opponents do not refute the SEC's claim that FD provides a level playing field for all investors, they argue that the quantity and quality of information will be lower in the post-FD time period. There are several reasons why firms prefer to release information privately to a few analysts instead of releasing it publicly to the entire investment community.

* It is safer from a legal liability point of view as the information may be improperly interpreted or the actual results may differ from the forecasted results.

* It is safer from a competitive point of view, since firms can protect proprietary information and avoid answering probing analyst questions without being exposed to larger audiences.

* Firms can require analysts to use the information solely to improve their earnings per share (EPS) forecasts and not disclose specific details to the investment community.

* Research finds that firms privately provide information to analysts in return for allowing firms to manage analysts' EPS forecasts and consequently beat them when actual EPS are announced (Richardson et al. 2001). (1)

Has FD been successful in eliminating selective disclosure or did the opponents correctly observe that although a noble idea it lacks practicality? The answer depends on whom you ask. (2) A November 2000 survey conducted by Financial Executives International (FEI) after F]) passed in August 2000 revealed that over 97 percent of the respondents said that they either currently or will shortly open their conference calls to the general public. Eighty-five percent said that their companies intend to continue participating in broker sponsored and one-on-one analyst meetings, subject to compliance with FD. A July 2001 National Investor Relations Institute (NIRI) provides similar results.

On the other hand, in a March 2001 Association for Investment Management and Research survey, 57 percent of its members said that the quantity of substantive information released by companies they research decreased while only 14 percent said that it increased. Similar percentages (56 percent and 15 percent) are reported for the quality of information. Eighty-one percent said that companies could use FD as a pretense for minimizing communication and 71 percent said that the reduced information flow caused by FD contributes to increased market volatility. A May 2001 survey of 30 buy-side and sell-side analysts by the Securities Industry Association shows similar results. Contrary to the results provided by the FEI and NIRI surveys, all of the sell-side analysts interviewed report fewer one-on-one discussions with management (Securities Industry Association 2001).

Review of Related Academic Research

Several academic researchers examine the debate and report conflicting results that relate to the research method used and/or the time period studied. Using data ending in December 2000, Heflin et al. (2003) find that prior to earnings announcements in the post-FD period there is no deterioration in the information environment. The study reports a lower stock return volatility around earnings announcements and no change in analyst forecast bias, accuracy or dispersion. In a related study, Heflin et al. (2001) find no significant net increase in stock return volatility around all post-FD earnings information release dates. On the other hand, Straser (2002) finds that despite an increase in the quantity of available public information, the quality of that information, as measured by information asymmetry levels, declined in the post-FD era. One limitation of all these studies is that their post-FD periods are very short-two months in the case of Heflin et al. (2001, 2003) and three months for Straser (2002). Co nsequently, their results may not be representative of the entire post-FD period.

Mohanram and Sunder (2002) find lower forecast accuracy and higher forecast dispersion in a longer post-FD period. The study claims that financial analysts performed more independent analysis in the post-FD era by providing evidence of a significant increase in the idiosyncratic component of information. Using a different methodology that controls for the total amount of earnings information in the pre- and the post-FD era, Zitzewitz (2002) argues that the Mohanram and Sunder (2002) findings result from an increased arrival rate of bad earnings information as the economy entered a recession. According to Zitzewitz (2002), FE) succeeded in reducing selective disclosure without reducing the total amount of information disclosed by firms. However, these results are strong only for the fourth quarter of 2000 and have since reversed. Finally, Shane et al. (2002) find significantly reduced stock market reactions to earnings announcements after FD. Contrary to Mohanram and Sunder (2002), accuracy of analyst forecast s made at the end of the quarter after FE) is similar to pre-FD accuracy, even though forecasts made at the beginning of the quarter are less accurate after FD.

Two papers examine the effects of the regulation on information asymmetry as proxied for by conference calls. Contrary to Straser (2002), Bushee et al. (2002) fail to find evidence of a negative impact on information quality, proxied by information provided during conference calls. In addition, the study finds increased stock price volatility in the post-FE) period, a finding that contradicts the results given in Heflin et al. (2001b). The increased volatility is attributed to broadened access to information and investors using that information to trade in real-time. Sunder (2002) examines the bid-ask spreads of firms that previously used open conference calls vs. those that previously used restricted conference calls. The finding that pre-FD firms that restricted access to conference calls exhibited higher bid-ask spreads is consistent with selective disclosure being associated with higher information asymmetry. This difference between the two sets of firms is no longer present in the post-FD period.

The preceding discussion of conflicting research findings indicates that the effects of FD remain unclear. Our research provides more evidence by examining the effect of FD on analyst following and analyst forecast dispersion. Ours is the first study to investigate the effect of the new regulation on analyst following. Moreover, although some earlier studies examine forecast dispersion, we measure the dispersion of annual EPS forecasts to provide a longer horizon test of the effect of FD on information asymmetry. Finally, in addition to extending the post-FD period to 11 months and the pre-FD period to 5 years, we control for confounding factors by examining the effects of the new law on these two variables in a multivariate context.

RESEARCH METHOD

Lower quantity and quality of corporate financial information should affect all users of such information. This study examines financial analysts, a subset of such users considered sophisticated in their use of financial information and included by the SEC in its definition of "enumerated persons." We use data ending in September 2001 to investigate the impact of the regulation on analyst following and analyst forecast dispersion. (3)

Analyst following is frequently used to proxy for the informativeness of a firm's information environment (Walther 1997; Ayers and Freeman 2003). Similarly, Lang and Lundholm (1996) find that analyst following is positively associated with the informativeness of a firm's disclosure policy, measured by ratings given in Financial Analyst Federation (FAF) reports. Further, accessibility to management and the firm's willingness to respond to analysts in a timely fashion has the most influence on the FAF ratings. This result corroborates Lees' (1981) survey finding that analysts consider interviews with company executives as the most important input to their earnings forecasts.

By prohibiting selective disclosure, FD eliminates what analysts consider a vital input into their earnings forecast models. This makes analysts' original research requiring higher information collection and processing costs even more important. Clement (1999) finds that the complexity of the analysts' task, as measured by the number of firms and industries followed, is inversely related to forecast accuracy. The need for accurate forecasts and original research along with limited resources could also lead to analysts following fewer firms in the post-FD period.

Forecast dispersion is used as a proxy for information asymmetry (Lang and Lundholm 1996; Barron and Stuerke 1998). High dispersion implies low consensus among analyst forecasts, a sign of high information asymmetry. If FD causes firms to reduce the quantity and quality of available information, then analysts will be forced to rely more on their own research and herding behavior should decline. Ceterus pari bus, spending more time on individually researching firms and reaching varied conclusions should lead to higher forecast dispersion in the post-FD period. (4) This argument is consistent with the Barron et al. (1998) finding that dispersion captures the extent to which private information differs across analysts. We test the above implications with univariate and multivariate methods.

Univariate Tests

Using analyst forecast data provided by First Call from 1995-2001, the univariate tests compare the end of quarter means of annual EPS forecast dispersion and analyst following in the pre- and post-FD time periods. (5) Including only firms with December 31 fiscal year-ends provides an equal forecast horizon for all firms in a given quarter. Fourth quarter of 1995 (Q4/1995) through the third quarter of 1996 (Q3/1996) is denoted as period -5, and so on, such that period 0 is from Q4/2000 to Q3/2001. Periods -5 through-i comprise the pre-FD time period and period 0 is the post-FD time period. Annual BPS forecasts used in a given quarter are for the year in which that quarter ends. For instance, the Q4/2000 forecasts are for the fiscal year ending December 31, 2000, while Q112001 forecasts are for the fiscal year ending December 31,2001. The file containing summary analyst forecast data provides the necessary data. (6) We measure analyst following by the total number of most recent EPS estimates at the end of ea ch quarter. Each firm's analyst forecast dispersion at the end of each quarter is defined as the standard deviation of the most recent annual BPS forecasts divided by the firm's stock price at the beginning of the period.

The analysis includes only firms with no missing observations for all periods ending in a given quarter resulting in the largest possible sample. For example, firm A is included for Q1 if it has no missing observations for all Q1s throughout the time period. This procedure results in identical sample sizes for the same quarter across all periods and different sample sizes across different quarters. In addition, firms included in the final sample must be followed by at least two analysts in a given month to compute forecast dispersion, and must have at least one new forecast during a quarter. This latter criterion addresses the problem of stale forecasts and ensures that the analysis includes only firms with an active analyst following.

Multivariate Tests

The univariate analysis assumes that, except for FD, other factors that potentially influence the test variables remain constant over the studied time period. This assumption can be relaxed by a multiple regression that includes relevant control variables. The multivariate analysis redefines the periods to span from Q1 through Q3 of the same fiscal year since several control variables are computed using annual data from the prior fiscal year. As reported later, since there was little variation in analyst following across different quarters, we use average analyst following over the three quarters of each period in the multivariate analysis. On the other hand, we estimate separate quarterly regressions for forecast dispersion. Applying the above-mentioned criteria to forecast data from periods -2 through 0 in the multivariate analysis results in a sample larger than the one used in univariate analysis and provides a meaningful comparison between the pre-FD and the post-FD era. We obtain financial and stock pri ce data from

Compustat and stock return data from CRSP. We delete all observations with incomplete data for the control variables and adjust all per share data for stock splits and dividends.

We use the following multivariate model to test the effect of FD on analyst following: (7)

[FOL.sub.i,t] = [alpha] + [[beta].sub.1][FD.sub.t] + [[beta].sub.2][FOL.sub.i,t-2] + [[beta].sub.3][TA.sub.i,t-1] + [[beta].sub.4][EPS.sub.i,t-1] + [[beta].sub.6][EPSCH.sub.i,t-1] + [[beta].sub.6][SALCH.sub.i,t-1] + [[beta].sub.7][AR.sub.i,t] + [[beta].sub.8][EPSVOL.sub.i,t-1] + [[beta].sub.9][RETVAR.sub.i,t] + [[beta].sub.10][SIC.sub.i,t] + [[epsilon].sub.i,t] (1)

where:

[FOL.sub.i,t] = mean of analyst following for firm i over the first three quarters of period t (t = -2, -1, 0), where analyst following equals the total number of most recent annual EPS estimates at the end of each quarter;

[FD.sub.t] = 1 in period t=0 (2001, post-FD) and 0(1999-2000, pre-FD) otherwise;

[FOL.sub.i,t-2] = analyst following for firm i in period t-2;

[TA.sub.i,t-1] = the natural logarithm of total assets for firm i in period t-1;

[EPS.sub.i,t-1] = annual EPS before discontinued operations and extraordinary items for firm i in period t-1;

[EPSCH.sub.i,t-1] = change in annual EPS for firm i in period t-1 divided by t-2 beginning stock price;

[SALCH.sub.i,t-1] = change in annual net sales revenue for firm i in period t-1 divided by t-2 net sales;

[AR.sub.i,t] = firm i's compounded return over a one-year period ending in the third quarter of period t less the compounded return for the value weighted market index for the same period;

[EPSVOL.sub.i,t-1] = standard deviation of annual EPS (using EPS from t-1 through t-4) for firm i divided by the stock price at the beginning of t-4;

[RETVAR.sub.i,t] = standard deviation of daily firm i returns for a one-year period ending in the third quarter of each period;

[SIC.sub.i,t] = 1 if firm i belongs to a regulated industry (SIC 4900-4999,6000-6411, or 6500-6999) and 0 otherwise; and

[[epsilon].sub.i,t] = normally distributed error term.

A negative [[beta].sub.1] is consistent with FD having a negative impact on analyst following. Following Walther and Willis (1999), we include [FOL.sub.i,t-2] to control for correlated omitted firm characteristics. An alternative is to estimate the dependent variable in changes rather than levels. However, such a procedure is not used because it constrains the coefficient of [FOL.sub.i,t-2] to equal 1.We include the remaining variables as control variables following prior research that finds these variables to affect analyst following.

[TA.sub.i,t-1] proxies for the richness of the firm's information environment (Bhushan 1989). The richer the environment, the larger the number of analysts following a given firm. Prior research also finds firm performance to be positively correlated with analyst following. Following Walther and Willis (1999), we include [EPS.sub.i,t-1], [EPSCH.sub.i,t-1], [SALCH.sub.i,t-1], and [AR.sub.i,t] as proxies for firm performance. The first three capture prior firm performance while [AR.sub.i,t] measures current performance. Alford and Berger (1999) find that analysts prefer to follow firms for which earnings are easier to forecast. Consequently, we include [EPSVOL.sub.i,t] and expect a negative association with analyst following. We also predict a negative relationship between [RETVAR.sub.i,t] and analyst following. Bhushan (1989) argues that analyst following is positively correlated with stock return variability, since analysts' private information gathering is more valuable for firms with high stock return volat ility. However, Bhushan's (1989) regression model does not control for the effect of abnormal returns. Finally, we include [SIC.sub.i,t] it to control for the effect of regulatory status. O'Brien and Bhushan (1990) argue that analysts prefer regulated industries because of the supplementary information provided as a result of regulatory oversight.

The following model investigates the effect of FD on analyst forecast dispersion. To control for the forecast horizon, we conduct the investigation at the end of each of the three quarters: (8)

[DISP.sub.i,q(t)] = [alpha] + [[beta].sub.1][FD.sub.t] + [[beta].sub.2][TA.sub.i,t-1] + [[beta].sub.3][LOSS.sub.i,t-1] + [[beta].sub.4][EPSVOL.sub.i,t-1] + [[beta].sub.5][RETVAR.sub.i,q(t)] + [[epsilon].sub.i,q(t)] (2)

where variables not previously defined are:

[DISP.sub.i,q(t)] = standard deviation of the most recent annual EPS forecasts for firm i at quarter-end q of period t (t = -2, -1, 0) divided by the beginning of period stock price;

[LOSS.sub.i,t-1] = 1 if firm i reported a negative annual EPS in period t-1, and 0 otherwise;

[RETVAR.sub.i,q(t)] = standard deviation of daily firm i returns for a one-year period ending in the period t quarter q; and

[[epsilon].sub.i,q(t)] = normally distributed error term.

A positive [[beta].sub.1] indicates an increase in analyst forecast dispersion as a result of FD. Once again, [TA.sun.i,t-1] proxies for the richness of firm's information environment. Lang and Lundholm (1996) find that firms with more informative disclosures exhibit lower dispersion. Dispersion is also found to be positively associated with loss firms (Brown 2001; Ciccone 2001). Consequently, we expect a positive [[beta].sub.3]. We also expect positive [[beta].sub.4] and [[beta].sub.5]. Ceteris paribus, high past earnings volatility should make predicting future earnings difficult and thus lead to lower consensus among analysts and higher forecast dispersion. Similarly, high firm-specific market uncertainty measured by stock return variability should also lead to high dispersion. Lobo and Tung (1998) find a positive association between stock price variability around earnings announcements and forecast dispersion.

FINDINGS

Figures 1 and 2 provide the univariate results. Figure 1 indicates a decrease in the average number of analysts in the post-FD period. Univariate tests (not reported) find this decrease to be statistically significant. In the pre-FD period, we observe a statistically significant increase in analyst following for most quarters reflecting the rise in the analyst community during the last decade. The decrease in analyst following is consistent with the decrease in the quantity and quality of information made available by firms, thereby making analysts' task more complex and time-consuming.

Figure 2 plots the mean analyst forecast dispersion on a quarterly basis for the six periods. Once again, univariate tests (not reported) find the increase in mean forecast dispersion in period 0 (post-FD) to be statistically significant. As expected, the mean forecast dispersion is highest in the first quarter and generally gets lower as the forecast horizon decreases. Results are qualitatively similar, albeit lower in magnitude, for median forecast dispersion (not reported).

Table 1 provides the multivariate results for analyst following. Most of the variables are statistically significant and in the hypothesized direction. As a whole, the model has significant explanatory power with an F-statistic of 1937.65 and an adjusted [R.sup.2] of 84 percent. Including [FOL.sub.i,t-2] produces approximately half of the explanatory power. However, we observe qualitatively similar results for the remaining variables in its absence. The negative coefficient on FD is significant, consistent with fewer analysts following a firm in the post-FD time period. This result supports the univariate findings and is consistent with the arguments given in Mohanram and Sunder (2002).

Table 2 presents the quarterly regression tests of the effect of FD on forecast dispersian. The last row of the panel provides results using a combined sample of all firm-quarter observations. The model has significant explanatory power in all quarters as evidenced by high F-statistics and adjusted [R.sup.2]s ranging from 0.1377 to 0.2393. The coefficient on FD, [[beta].sub.1], is positive and significant (p-values less than 1 percent) in all quarters. Except for [TA.sub.i,t-1], all other control variables are consistently significant in the hypothesized direction. Once again, these results are consistent with the univariate findings reported earlier.

The above evidence indicates that FD has a negative impact on the financial analyst community. Whether this will also adversely affect a firm's market valuation is a question to be answered by future research as more data become available to enable more reliable estimations. Because the results presented in this paper capture the early reaction to FD, they may disappear as firms and analysts become more accustomed to the new rule. Finally, the implication that FD has a negative impact on the quantity and quality of information is subject to analyst following and analyst forecast dispersion being valid proxies for these constructs.

CONCLUSION

FD stirred more debate regarding its ability to deter selective disclosure without stiffing corporate information flow than any other regulation passed by the SEC in recent times. Using analyst forecast data provided by First Call, we present preliminary findings of FD's impact on the quantity and quality of firm-specific information released to the market, as proxied by analyst following and analyst forecast dispersion. We document a decrease in analyst following and an increase in analyst forecast dispersion, findings which support the "chilling" of information argument given by FD opponents (Opdyke 2000).

Future research in this area should focus on evaluating the longer-term effects of FD. Issues to be examined include:

* Extending the preliminary results presented in this paper concerning the effect of FD on financial analyst forecast properties.

* Empirically examining the effect of the rule on the quantity and quality of voluntary disclosures disseminated by individual firms compared to the past levels.

* Investigating the effect of FD on market valuation and volatility.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

TABLE 1

Relationship between Analyst Following and FD

(n = 3,690)


[FOL.sub.i,t] = [alpha] + [[beta].sub.1][FD.sub.t] +
[[beta].sub.2][FOL.sub.i,t-2] + [[beta].sub.3][TA.sub.i,t-1]
+ [[beta].sub.4][EPS.sub.i,t-1] + [[beta].sub.5]
[EPSCH.sub.i,t-1] + [[beta].sub.6][SALCH.sub.i,t-1] +
[[beta].sub.7][AR.sub.i,t] + [[beta].sub.8][EPSVOL.sub.i,t-1]
+ [[beta].sub.9][RETVAR.sub.i,t] + [[beta].sub.10][SIC.sub.i,t]
+ [[epsilon].sub.i,t] (a)

                    Expected
                      Sign     Coefficient

Intercept              ?         -0.7917 *
[FD.sub.i,t]           -         -0.7875 *
[FOL.sub.i,t-2]        +          0.9132 *
[TA.sub.i,t-1]         +          0.2227 *
[EPS.sub.i,t-1]        +          0.0938 *
[EPSCH.sub.i,t-1]      +          0.5701 +
[SALCH.sub.i,t-1]      +          0.7779 *
[AR.sub.i,t]           +          0.3297 *
[EPSVOL.sub.i,t-1]     -          0.3051
[RETVAR.sub.i,t]       -         -5.6362 +
[SIC.sub.i,t]          +          0.2769 *
Adjusted [R.sup.2]   0.8400
F-statistic         1937.65 *

* Statistically different from zero at significance levels of 1 percent
or less.

+ Statistically different from zero at 5 percent significance level.

(a)Except for the intercept, all the other p-values are based on
one-tailed t-tests

[FOL.sub.i,t] = mean of analyst following for firm i over the first three quarters of period t (t = -2, -1, 0), where analyst following equals the total number of most recent annual EPS estimates at the end of each quarter;

[FD.sub.t] = 1 in period t=0 (2001, post-FD) and 0 (1999-2000, pre-FD) otherwise;

[FOL.sub.i,t-2] = analyst following for firm i in period t-2;

[TA.sub.i,t-1] = the natural logarithm of total assets for firm i in period t-l;

[EPS.sub.i,t-1] = annual EPS before discontinued operations and extraordinary items for firm i in period t-1;

[EPSCH.sub.i,t-1] = change in annual EPS for firm i in period t-1 divided by t-2 beginning stock price;

[SALCH.sub.i,t-1] = change in annual net sales revenue for firm i in period t-1 divided by t-2 net sales;

[AR.sub.i,t-1] = firm i's compounded return over a one-year period ending in the third quarter of period t less the compounded return for the value weighted market index for the same period;

[EPSVOL.sub.i,t-1] = standard deviation of annual EPS (using EPS from t-1 through t-4) for firm i divided by the stock price at the beginning of t-4;

[RETVAR.sub.i,t] = standard deviation of daily firm i returns for a one-year period ending in the third quarter of each period;

[SIC.sub.i,t] = 1 if firm i belongs to a regulated industry (SIC 4900-4999, 6000-6411, or 6500-6999) and 0 otherwise; and

[[epsilon].sub.i,t] = normally distributed error term.

TABLE 2

Relationship between Analyst Forecast Dispersion and FD

[DISP.sub.i,q(t)] = [alpha] + [[beta].sub.1][FD.sub.t] +
[[beta].sub.2][TA.sub.i,t-1] + [[beta].sub.3][LOSS.sub.i,t-1] +
[[beta].sub.4][EPSVOL.sub.i,t-1] + [[beta].sub.5][RETVAR.sub.i,q(t)] +
[[epsilon].sub.i,q(t)] (a)

                 n    [alpha]  [[beta].sub.1]  [[beta].sub.2]

Expected Sign            ?           +               -

Quarter 1       3983   0.0002     0.0011 *        -0.0001
Quarter 2       4045   0.0004     0.0008 *        -0.0001
Quarter 3       3913  -0.0015     0.0009 *         0.0001

All Quarters   11941  -0.0003     0.0009 *         0.0001

               [[beta].sub.3]  [[beta].sub.4]  [[beta].sub.5]

Expected Sign        +               +               +

Quarter 1         0.0082 *        0.0451 *        0.0822 *
Quarter 2         0.0074 *        0.0233 *        0.1037 *
Quarter 3         0.0063 *        0.0147 *        0.1027 *

All Quarters      0.0074 *        0.0261 *        0.0983 *

               Adjusted [R.sup.2]  F-Statistic

Expected Sign

Quarter 1            0.2393          251.48 *
Quarter 2            0.1640          159.66 *
Quarter 3            0.1377          125.95 *

All Quarters         0.1745          505.90 *

* Statistically different from zero at significance levels of 1 percent
or less.

(a)Except for the intercept, all the other p-values are based on
one-tailed t-tests.

[DISP.sub.i,q(t)] = standard deviation of the most recent annual EPS forecasts for firm i at quarter-end q of period t (t = -2, -1, 0) divided by the beginning of period stock price;

[FD.sub.t] = 1 in period t=0 (2001, post-FD) and 0 (1999-2000, pre-FD) otherwise;

[TA.sub.i,t-1] = the natural logarithm of total assets for firm i in period t-1;

[LOSS.sub.i,t-1] = 1 if firm i reported a negative annual EPS in period t-1 and 0 otherwise;

[EPSVOL.sub.i,t-1] = standard deviation of annual EPS (using EPS from t-1 through t-4) for firm i divided by the stock price at the beginning of t-4;

[RETVAR.sub.i,q(t)] = standard deviation of daily firm i returns for a one-year period ending in the period t quarter q;

[[epsilon].sub.i,q(t)] = normally distributed error term.

Submitted: August 2001

Accepted: September 2002

(1.) As stated earlier, abolishing the management practice of using information as a commodity is one of the three main objectives of FD.

(2.) The surveying organization's web site is the source for the surveys mentioned in this study.

(3.) To be consistent across all years, the cutoff date used for Q3 is September 10 instead of September 30 because of the tragic events of September 11, 2001.

(4.) According to Zacks Investment Research, Chicago, the median difference between the high and the low analyst forecast estimates for stocks in the Standard & Poor's 500 stock index grew 40 percent in 2000-01 to 35 cents from 25 cents in the earlier years (Elstein 2001).

(5.) Comparison of monthly means yields identical results.

(6.) First Call uses the following procedure to deal with the problem of stale forecasts. It uses only the most recent forecast issued by an individual analyst to compute the various summary measures such as the mean, median, and standard deviation at any point in time. Further, First Call discards forecasts by analysts who subsequently stop following a firm. We manually reconciled the detail file with the summary file for a small sample of firms and therefore have confidence in using the summary file.

(7.) In estimating Equation (1), we eliminate 216 extreme observations associated with studentized residuals [greater than or equal to] \2\ or Cook's D statistic [greater than or equal to] 1. Methods of addressing the outlier problem, the studentized residual equals the residual divided by its standard error while Cook's D measures the change to the estimates that results from deleting each observation. Qualitative results are not sensitive to this elimination, even though the adjusted [R.sup.2] is lower.

(8.) In estimating Equation (2), we eliminate 233 monthly observations associated with studentized residuals [greater than or equal to] \2\ or Cook's D statistic [greater than or equal to] 1. Once again, this elimination has no qualitative effect on the results.

REFERENCES

Alford, A., and P. Berger. 1999. A simultaneous equations analysis of forecast accuracy, analyst following, and trading volume. Journal of Accounting, Auditing and Finance 14 (Summer): 219-240.

Ayers, B., and R. Freeman. 2003. Evidence that analyst following and institutional ownership accelerate the pricing of future earnings. Review of Accounting Studies 8: 47-67.

Barron, O., and P. Stuerke. 1998. Dispersion in analysts' earnings forecasts as a measure of uncertainty. Journal of Accounting, Auditing and Finance 13 (Summer): 243-268.

-----, O. Kim, S. Lim, and D. Stevens. 1998. Using analysts' forecasts to measure properties of analysts' information environment. The Accounting Review 73 (October): 421-433.

Bhushan, R. 1989. Firm characteristics and analyst following. Journal of Accounting and Economics 11: 255-274.

Bloomberg News. 1999. SEC may bar selective disclosures by firms. Los Angeles Times (March 17).

Bowen, R., A. Davis, and D. Matsumoto. 2002. Do conference calls affect analysts' forecasts? The Accounting Review 77 (April): 285-316.

Brown, L. 2001. A temporal analysis of earnings surprises: Profits vs. losses. Journal of Accounting Research 39 (September): 221-241.

Bushee, B., D. Matsumoto, and G. Miller. 2002. Managerial and investor responses to disclosure regulation: The case of Reg FD and conference calls. Working paper, University of Pennsylvania.

Ciccone, S. 2001. Analyst forecast properties, financial distress, and business risk. Working paper, University of New Hampshire.

Clement, M. 1999. Analyst forecast accuracy: Do ability, resources, and portfolio complexity matter? Journal of Accounting and Economics 27 (June): 285-303.

Dempsey, S. 1989. Pre-disclosure information search incentives, analyst following, and earnings announcement price response. The Accounting Review 64 (October): 748-757.

Elstein, A. 2001. Earnings estimates vary more widely. Wall Street Journal (August 2).

Heflin, F., K. Subramanyam, and Y. Zhang. 2001. Stock return volatility before and after Regulation FD. Working paper, Purdue University.

-----, -----, and -----. 2003. Regulation FD and the financial information environment: Early evidence. The Accounting Review 78 (January): 1-37.

Lang, M., and R. Lundholm. 1996. Corporate disclosure policy and analyst behavior. The Accounting Review 71 (October): 467-492.

Lees, F. 1981. Public Disclosure of Corporate Earnings Forecasts. New York, NY: The Conference Board.

Lobo, G., and S. Tung. 1998. Financial analysts' earnings forecast dispersion and intraday stock price variability around quarterly earnings announcements. Working paper, Syracuse University.

Mohanram, P., and S. Sunder. 2002. Has Regulation Fair Disclosure affected financial analysts' ability to forecast earnings? Working paper, New York University.

O'Brien, P., and R. Bhushan. 1990. Analyst following and institutional ownership. Journal of Accounting Research 28 (Supplement): 55-82.

Opdyke, J. D. 2000. The Big Chill: Street feels effect of the new "Fair Disclosure" rule. Wall Street Journal (October 23).

Richardson, S., S. Teoh, and P. Wysocki. 2001. The walkdown to beatable analyst forecasts: The roles of equity issuance and insider trading incentives. Working paper, Massachusetts Institute of Technology.

Securities and Exchange Commission (SEC). 2000. Selective Disclosure and Insider Trading. Release 33-7881. Washington, D.C.: SEC.

Securities Industry Association. 2001. Costs and Benefits of Regulation Fair Disclosure. New York, NY: Securities Industry Association.

Shane, P., N. Soderstrom, and S. Yoon. 2002. The effects of Reg. FD on earnings management, analysts' earnings forecasts errors, and the market's earnings surprise: A preliminary analysis. Working paper, University of Colorado at Boulder.

Straser, V. 2002. Regulation Fair Disclosure and information asymmetry. Working paper, University of Notre Dame.

Sunder, S. 2002. Investor access to conference call disclosures: Impact of Regulation Fair Disclosure on information asymmetry. Working paper, New York University.

Sugawara, S. 2000. SEC to bar selective disclosures by firms. The Washington Post (August 4). Wall Street Journal. 1999. SEC study probes "selective disclosure" to stock analysts. (March 17).

Walther, B. 1997. Investor sophistication and market earnings expectations. Journal of Accounting Research 35 (Autumn): 157-179.

-----, and R. Willis. 1999. Are earnings surprises costly? Working paper, Northwestern University and Duke University.

Zitzewitz, E. 2002. Regulation Fair Disclosure and the private information of analysts. Working paper, Stanford University.

We thank Robert Freeman and the two anonymous referees for their invaluable suggestions. We also thank First Call Corporation, a Thomson Financial company, for providing the analyst forecast data. Professor Irani gratefully acknowledges the financial support provided by the Virginia Paul Dee Professorship.

Corresponding author: Afshad J. Irani

Email: afshad.irani@unh.edu

Afshad J. Irani is an Assistant Professor at the University of New Hampshire, and Irene Karamanou is a Lecturer at the University of Cyprus.