Analyst reputation and the issuance of disaggregated earnings forecasts to I/B/E/S

In the limit, reputable analysts may not need I/B/E/S at all. That is, the sample of analysts who provide earnings forecasts to I/B/E/S may exclude many of the more reputable analysts. While we cannot discern the reason for an analyst’s exclusion from I/B/E/S, as an empirical matter, we observe significant variation in our proxy for analyst reputation, which suggests that at least some reputable analysts stay on I/B/E/S.

To substantiate this claim, we randomly sampled 50 instances in I/B/E/S where an analyst forecasted earnings but not revenue on a given date. We were able to locate the corresponding research reports on the Investext or Reuters databases in 22 cases and found explicit numerical revenue forecasts in 20 (91%) of the research reports. Both reports not stating an explicit revenue forecast stated that the entire revision of aggregate earnings was due to operating expenses, thereby implying that the existing revenue forecast was unchanged. Thus, all analysts in this random sample have generated revenue and expenses to arrive at total earnings but chose to supply I/B/E/S only the aggregate earnings forecasts.

Full-information forecasting, commonly taught in valuation courses, involves forecasting the full set of factors that drive operating income. The first step in full-information forecasting is projecting revenues (Penman 2004; Stickney et al. 2004).

An alternative disaggregation of earnings is cash flows and accruals (Call et al. 2009). While important in its own right, we do not use this disaggregation because analysts tend to supply cash flow forecasts to I/B/E/S when earnings are uninformative (DeFond and Hung 2003). This implies that cash flow forecasts are viewed as substitutes for earnings forecasts from a valuation standpoint. In contrast, revenue forecasts complement investors’ understanding of earnings.

Williams (1995) reports that, prior to our sample period, contributors to First Call (now owned by I/B/E/S’s parent company, Thomson Financial) paid a substantial fee to be included in their database. We have corresponded with Thomson employees as well as sell-side analysts to understand the current arrangement between I/B/E/S and contributing analysts. Thomson Financial explicitly stated that there is “no monetary compensation between contributing analysts and Thomson Financial.” We then obtained “Thomson Financial Contributor Qualifications” and “Contributor Approval Policy for Estimates and Research,” two documents that summarize the conditions that must be met to have analyst data reside as part of Thomson data platforms, including I/B/E/S. We noted no reference to analyst or brokerage payments as a criteria for analyst data inclusion with Thomson Financial. Rather, we noted only minimum competency and legitimacy requirements.

Regarding why an analyst would provide information to I/B/E/S and the suite of Thomson Financial products without any monetary payment, a Thomson Financial representative said, “Via Thomson Financial products, analysts get vast exposure on The Street.” Other exposure benefits include ratings by investor services and evaluation by potential employers. For example, StarMine, an investor services firm, uses earnings forecast data from I/B/E/S to rate analysts. See http://​www.​forbes.​com/​2007/​04/​29/​starmine-data-security-pf_​07topanalysts-cz_​0501starmethodol​ogy.​html for information about the methodology StarMine use for its analyst rankings. Brokerages often advertise their analysts’ StarMine rankings through press releases and even use StarMine’s analyst ratings service to measure the performance of their analysts. (See http://​db.​riskwaters.​com/​public/​showPage.​html?​page=​422822 for an article about Merrill Lynch and UBS’s decisions to rely on StarMine analyst rankings service. See http://​www2.​stifel.​com/​docs/​pdf/​pressreleases/​2007/​ForbesStarMine.​pdf and http://​www.​ubs.​com/​1/​e/​investors/​annual_​reporting2005/​handbook/​0024.​html for examples of press releases.).

In our analysis, we include only those analysts with distinct identifiers in I/B/E/S. Analysts choosing to anonymize their forecasts in I/B/E/S ex ante will lower the power of our tests and could potentially bias our results.

Note that nothing prevents analyst B or C from mimicking analyst A after analyst A has disclosed her forecast. However, the investing community will not view the mimicking strategy as valuable. Academic research and anecdotal evidence suggests the value of analysts stems not only from the quality of their outputs but also from the speed with which they communicate relevant information to their clients relative to other analysts (Cooper et al. 2001; Lowengard 2006; Guttman 2008).

This requirement allows us to focus on relatively active analysts and to calculate walk-down bias in analyst earnings forecasts, an important control variable in both the disaggregation determinant model and the career outcome model (Ke and Yu 2006) estimated later in the study.

We require revenue forecasts in both the first and second halves of the year to be consistent with our sample requirement that each analyst issue an earnings forecast in both the first and second halves of the year. Both revenue and earnings forecasts are necessary for the earnings disaggregation into revenue and expense.

The Carter-Manaster rankings are based on the hierarchy of the listing of underwriters in the prospectus of the security offering. Prestigious underwriters are listed higher in the underwriting section of the prospectus. For our sample period, we use the updated Carter-Manaster rankings constructed by Loughran and Ritter (2004).

As in Chen and Jiang (2005), we restrict the analysts for whom we calculate this ability measure to those whose first forecasts for any firm appeared on or after January 1, 1985, in order to ensure that the measure is calculated consistently and accurately across analysts.

While sample selection criteria (3) above partially controls for variation in demand, it is unlikely to be a sufficient control. This motivates us to include firm characteristics in the model. As an alternative, we estimate Eq. 1 with firm fixed effects. Untabulated results reveal that our inferences are unchanged.

We model disaggregation as an analyst-firm level decision. However, (i) analysts may make the decision to disaggregate (to ultimately reap exposure benefits) across the portfolio of firms they cover or (ii) the decision to issue disaggregated forecasts to I/B/E/S may be a brokerage-level decision. We explore these possibilities in unreported analyses. To address (i), we include a control variable in model (1) that measures the percentage of other firms in the analyst’s portfolio that contains disaggregated forecasts. We find that this percentage is the best predictor of whether an analyst disaggregates for a given firm, but inferences regarding the role of reputation in the disaggregation decision and the sign and significance level of most of our firm-level control variables remain unchanged. Alternatively, we aggregate the dependent and independent variables to the analyst level and repeat the analysis. Inferences are identical to those reported in column B of Table 4. To address (ii) we estimate model (1) after excluding analysts who work at brokerages that appear to have a policy of issuing or not issuing revenue forecasts to I/B/E/S (i.e., brokerages where in a given year none or more than 95% of the unique analyst-firm combinations contain at least one revenue forecast). Our inferences are unchanged in this specification.

Additionally, using a time trend and squared time trend as control variables instead of year fixed effects yields inferences identical to those reported in Table 4. This time trend and the trend in year fixed effects indicates the other model variables do not explain the increase in disaggregation over our sample period. Untabulated estimations of model (1) in 2 year increments indicate the increase is also not explained by a temporal change in model coefficients. Untabulated univariate results also reveal no obvious temporal trend in the independent variables. Thus, our study cannot fully explain the increase in disaggregation over time. However, the objective of our study is not to explain the trend in disaggregation; it is to examine which analysts are more likely to take advantage of the opportunity to provide disaggregated earnings forecasts.

We use earnings guidance from First Call because revenue guidance is not available in machine readable form. Note that our empirical models already contain many of the firm-level determinants of management guidance (Lansford and Tucker 2008).

Reported results are robust when we include trend and trend squared in the specification instead of year fixed effects (β₄ = 0.4556, p < 0.001). Additionally, using a continuous measure of disaggregation representing the proportion of disaggregation in an analyst’s portfolio instead of the dichotomous A_DIS does not alter the inferences drawn (β₄ = 0.3344, p < 0.001). Finally, the dependent variable assumes that analysts that disappear from I/B/E/S in the subsequent year experience negative career outcomes. This assumption may not be valid. Re-estimating model (2a) after eliminating 1,839 analyst-year observations where the analyst is not identifiable on I/B/E/S in the subsequent year does not change the inference on our variables of interest. The coefficients on A_DIS is 0.5797 (p < 0.001). In contrast, the coefficients on accuracy and walk-down bias are not statistically significant at conventional levels under this alternative specification.

Forecast revision dates in I/B/E/S represent the dates the forecasts are entered into the I/B/E/S system and therefore the dates the market receives the signal. If there is (1) a lag between when an analyst publishes her report and when the forecast is entered into the I/B/E/S system and (2) the market reacts to the published report, we are less likely to observe any market reaction on the date that I/B/E/S posts the data.

The high percentage of disaggregated earnings forecasts is likely the result of two factors. First, we are assessing revenue forecasts for only the set of less reputable analysts, who have more incentives to disaggregate than reputable analysts. Second, if the frequency of forecasting is another mechanism by which disaggregating analysts seek exposure, we should observe a high concentration of disaggregated forecasts among this set of less reputable analysts.