By Partha S. Mohanram and Shyam V. Sunder

n October 23, 2000, the Securities and Exchange Commission (SEC) implemented one of its most controversial and far-reaching measures. Regulation Fair Disclosure, or Reg FD, as it is known, required that publicly held companies communicate all material information to all investors at the same time. The SEC’s stated objective for the rule was to eliminate the practice of selective disclosure to favored analysts and institutional shareholders.

Industry groups like the Securities Industry Association (SIA) and the Association for Investment Management Research (AIMR) cried foul. They argued that Reg FD would lead to a deterioration in both the quality and quantity of information reaching investors, and that it would significantly impair analysts’ forecasting ability. The SIA also contended FD would “deter analysts from vigorously competing to glean useful information for their clients and the markets.” The group quoted member surveys that found analysts under the new regime were experiencing considerable difficulty in arriving at reliable earnings forecasts due to a decreased flow of information – even as corporate investor relations professionals claimed to be providing the same amount of information. For their part, the SEC and small investor groups argued that Reg FD should not impact the quality of information available to analysts. Instead, it would merely remove the timing advantage previously enjoyed by analysts and compel these highly paid workers to earn their keep. With reduced information flowing from management, they argued, analysts would have to undertake more private information discovery and analysis in order to maintain accurate coverage of individual companies.

Thus far, Reg FD has inspired a great deal of rhetoric but not much data. So we set out to investigate some of these claims on a quantitative basis. We used the I/B/E/S summary database to create a matched sample of analysts’ quarterly earnings-per-share forecasts in both the pre- and post-FD environments. The pre-FD dataset included forecasts from November 1999 through June 2000, and the post-FD dataset consisted of eight months of forecasts from November 2000 through June 2001. This matched sample design ensured that seasonality and firm-specific characteristics would not impact our results. Then we ran a series of univariate and multivariate tests on the data to answer three broad questions: (1) Has forecasting error increased since October 2000? (2) Are analysts now relying more on alternate sources of information? and (3) Have analysts who were all-stars in the pre-FD world been able to maintain their edge in the new environment?

To a large degree, this was virgin territory. While prior studies have largely focused on assessing the impact of change in quality of information, we aimed to measure the impact of change in access to information. The results are both intriguing and somewhat surprising.

Cloudy Forecasts

The first hypothesis we set out to test was that reduced management disclosure to analysts would result in greater absolute forecast errors in the post-FD world. Table 1 compares the characteristics of analysts’ forecasts before and after Reg FD for our sample. To ensure comparability, we matched our post-FD observations with the corresponding observation for the same firm from the same fiscal quarter for the prior year.

The absolute forecast error is defined as the absolute difference between the actual earnings per share and mean estimate, divided by beginning of period price. That figure, as seen, has increased from a mean of 0.0022 in the pre-FD period to 0.0039 in the post-FD period. This increase is significant for the raw absolute error, which has increased from 4.9 cents in the pre-FD period to around 6.2 cents in the post-FD period. Hence, it appears from this first glance that the ability of analysts to “get it exactly right” has diminished after Reg FD.

In addition, the results show that the range of analysts’ forecasts has also increased significantly in the post-FD period. And when we looked at forecast dispersion, or the standard deviation of analysts’ forecasts at a given point in time, we found that it, too, had increased significantly. In other words, analysts in the post-FD world tend to have a more divergent view of firms’ prospects. When we repeated the analysis for the medians the results are essentially similar, with one significant difference: the absolute forecast error doesn’t change much in the two eras. So for the median analyst, Reg FD has had minimal impact on the ability to get the forecast absolutely correct.

One of the problems with these univariate tests is that Reg FD came into effect around the same time that a downturn started in the US economy. The post-FD observations (November 2000 to June 2001) are likely to be from after the Internet meltdown, while the pre-FD observations (November 1999 to June 2000) are likely to be largely from the height of the Internet boom. Hence, some of the increase in forecast errors is likely to result from the earnings volatility set into motion by the unexpected downturn. To better isolate the effect of Reg FD, we ran multivariate regressions with a pooled sample of information from both the pre-FD and post-FD periods.

We included a critical independent variable called SURPRISE, which controls for macroeconomic effects such as the large decline in earnings that firms experienced around the time Reg FD came into effect. Then we ran the regressions both with and without SURPRISE, in order to isolate the importance of controlling for the earnings surprise. Our variable of interest is POSTFD, which indicates whether an observation was before or after FD. In the first specification that excludes SURPRISE, POSTFD has a significant and positive coefficient, confirming the univariate results from earlier. However, when we include SURPRISE, the coefficient on POSTFD is insignificant. That indicates that the large unexpected negative surprises in the post-FD period may have been responsible for the increase in absolute forecast error. So while the performance of analysts may have declined in the post-FD period, one cannot attribute the decline to the passage of the regulation after one controls for the changing macroeconomic environment.

Common Knowledge vs. Private Knowledge?

Analysts surveyed by the AIMR regard conversations with management as the most important source of information for forming opinions about a company. In the post-FD environment, analysts’ ready access to management has potentially been curtailed. This may reduce the precision of the shared or common knowledge that analysts have at their disposal. In this environment, one would assume that analysts would have to increase the effort expended on their own private information gathering and analysis. So we set out to test two hypotheses. First, that the level of common knowledge reflected in analyst forecasts in the post-FD period would be lower than the level in the pre-FD period. Second, that the level of idiosyncratic knowledge reflected in analyst forecasts in the post-FD period is higher than the level in the pre-FD period.

These are abstract concepts. But researchers Orie Barron, Oliver Kim, Steve Lim, and Douglas Stevens in 1998 developed a model for inferring the information environment that analysts operate in by studying the distribution of forecasts. Information here does not necessarily refer only to managerial disclosures but also to any other relevant information or insights that analysts may uncover on their own. Under the so-called BKLS model, there are a few important metrics. The precision of public information measures the extent to which analysts rely on common or public information while coming up with their forecasts. The precision of private information measures the extent to which analysts rely on private or idiosyncratic information while doing so. The sum of the two is the precision of total information.

Again, we used information from the I/B/E/S summary database to calculate the precision measures. Table 2 presents univariate tests that compare the precision of information before and after Reg FD. Panel A presents the means of the variables. The results indicate that while the precision of public information (PUB) has declined after the passage of Reg FD, the precision of private information (PRIV) has increased. As a result, the precision of total information (TOT) is essentially unchanged. When we repeated the test for medians, the results are similar. Thus, at the univariate level, the results indicate that analysts appear to be compensating for the reduced precision of common information by acquiring more idiosyncratic knowledge on their own.

To confirm the univariate results, we ran regressions with our precision measures as dependent variables. As before, we included SURPRISE as a dependent variable, as the unexpected downturn may also have had an impact on the precision of information. We also included controls for earnings variability, sales growth, firm size, and price-to-book ratio – all factors that may affect the precision of information.

In the regression for the precision of public information (PUB), POSTFD has a significant negative coefficient, consistent with our univariate results. Hence, there does appear to have been a decline in the precision of public information, potentially because of reduced disclosures from management to the analyst community. However, in the regression for the private information (PRIV), POSTFD has a positive coefficient, indicating that there has been a corresponding increase in the precision of analysts’ idiosyncratic information. POSTFD has an insignificant coefficient in the regression for total information (TOT), showing that analysts may have essentially compensated for the reduced disclosure through private information gathering and analysis.

Among the control variables, SURPRISE has a significant negative coefficient in all three regressions, indicating that macroeconomic factors reduced the precision of all information and made forecasting very difficult. In addition, we found there was lower precision of information for firms with volatile earnings and rapid growth, and greater information for larger firms.

Taken together, these results provide strong support for some of the SEC’s conjectures regarding the efficacy of Reg FD and refute the SIA’s viewpoint that analysts incentives to gather information will be dampened in the new environment.

Tomorrow’s All-Stars

Analysts can distinguish themselves through superior performance in terms of their ability to make earnings forecasts, set price targets, and pick stocks. Such superiority does not go unnoticed, as highly rated analysts receive lucrative remuneration packages.

Now, empirical research has shown that superior analysts are more likely to be well connected with the companies they follow. In the post-FD setting, such linkage advantages are likely to be less important. This may level the playing field for all analysts and lead to a convergence in performance among analysts. So we set out to check the following hypothesis: analysts who had superior performance in the pre-FD period will be unable to maintain their superiority in the post-FD period.

Using a sample of I/B/E/S forecasts from the detailed database from 1997 to November of 1999 (just prior to the beginning of our matched pre-FD and post-FD sample), we classified analysts into four quartiles based on their average absolute forecast error in this testing period. The top quartile consists of analysts with the lowest absolute forecast errors. We then compared the performance of analysts in these quartiles in our pre-FD and post-FD period. The results are presented in Panel A of Table 3.

In the pre-FD period, the superior analysts plainly maintained their superiority. The data show a gradual increase in mean absolute forecast errors from 0.0019 for the top analysts to 0.0022 for the 3rd quartile, to 0.0025 for the 2nd quartile, to 0.0030 for the bottom analysts. The difference between the top and bottom quartiles is highly significant, and this result holds for means as well as medians.

But when we compared the four groups in the post-FD period, we found a dramatic change. The four groups are virtually indistinguishable in the post-FD period! The mean absolute forecast error is identical for the first three groups at 0.0038 and insignificantly greater for the formerly bottom quartile at 0.0042. Hence, there appears to have been a convergence among the four groups, and the analysts who used to be superior in the pre-FD setting were unable to maintain their superiority in the post FD world.

What accounts for this? We set out to examine the characteristics of analysts in these four groups by looking at three characteristics. We identified analysts as belonging to a “large” brokerage if their firm was ranked among the top 10 investment banks in 1999. Second, we identified analyst as affiliated analysts if they worked for the same firm that was the lead underwriter in the company’s IPO.

"While prior studies have largely focused on assessing the impact of change in quality of information, we aimed to measure the impact of change in access to information."

Finally, we used Institutional Investor’s 1999 rankings of all-star analysts to see how all star analysts performed in the post-FD setting. The results are presented in Panel B of Table 3.

It turns out that analysts who were top analysts in the pre-FD environment were significantly more likely to belong to work for brokerage houses than bottom analysts (43.6% vs. 32.3%). Further, top analysts were more likely to be affiliated analysts than bottom analysts (4.9% vs. 2.9%), and more likely to be classified as all-star analysts (34.9% vs. 24.2%). These results are interesting because such analysts have been alleged to have much greater access to firms than other analysts. And the fact that top analysts were unable to maintain their advantages in the post-FD setting may reflect the diminished value of such linkages after the passage of the regulation.

A Level Playing Field

Reg FD had two important goals at its core – to level the informational playing field in the capital markets, and to cut the umbilical cord that appeared to exist between managements of firms and analysts following the firms. And all without causing undue dislocation and disruption in the markets. Has it met those goals?

While absolute forecast error has increased after the passage of Reg FD, the impact of the regulation itself is insignificant after one controls for macroeconomic factors such as the downturn that started in late 2000. Further, while the precision of public information does appear to have declined after Reg FD, this has been compensated for by the increased precision of private information, reflecting the greater efforts of analysts in private information gathering and analysis. Finally, the fact that well-connected all-star analysts were unable to maintain their superior performance in the post-FD period suggests that all types of analysts are now having to prove their worth through their wits and not through their connections to executives.

Partha S. Mohanram is an assistant professor of accounting at NYU Stern.

Shyam V. Sunder, Stern Ph.D. 2002, is an assistant professor of accounting at the Kellogg School of Management at Northwestern University