Is all disaggregation good for investors? Evidence from earnings announcements

We use sales as the benchmark because it is disclosed in almost all earnings announcements and is the fundamental driver of profitability. We consider the following earnings components: cost of goods sold; selling, general and administrative expenses; depreciation expenses; research and development expenses; interest expense; non-operating income; tax expense; special items; and discontinued operations.

In untabulated analyses, we also perform an analysis in which we classify the operating income components as homogenous and the nonoperating income, taxes, special items, and discontinued operations components as heterogeneous (as suggested by the classification in Fairfield et al. 1996). We find similar, albeit slightly weaker results, when this simplified classification scheme is used. These results provide further guidance on a practical application of homogenous disaggregation.

Research also shows that disaggregation of management earnings forecasts reduces investor earnings fixation (Elliott et al. 2011) and increases the credibility of the forecast (Hirst et al. 2007; Merkley et al. 2013).

We require firm earnings to occur at least a week prior to their regulatory filing to isolate the market’s reaction to the earnings announcement itself.

Consistent with this assertion, research finds that persistence classifications are useful to investors (Bowen 1981; Lipe 1986; Strong and Walker 1993) as well as analysts (Bradshaw and Sloan 2002; Christensen et al. 2011; Ali et al. 1992). In addition, standard setters have made the classification of earnings items based on their differential persistence a fundamental feature of financial statements. For example, the FASB has required firms to disaggregate items that are unusual, infrequent, or both from other earnings since the 1970s (FASB 1973).

In our sample, 98% of earnings announcements disclose sales. The results are similar to those reported in the tables when we exclude the 2 % of observations in which sales is not disclosed in the earnings announcement.

When identifying the earnings announcement disclosure, we examine both the point-in-time date and corresponding source code to ensure it is the initial earnings announcement disclosure with the most detailed financial statement information.

D’Souza et al. (2010) use legacy databases, Compustat Preliminary History and Unrestated Quarterly, to construct their disclosure ratios for their analyses. In 2013, Compustat consolidated the Prelim History, Unrestated Quarterly, and Point-in-Time databases into one product, Compustat Snapshot. While Compustat ensured the quality and integrity of the data prior to unveiling the Snapshot database, we compared the ratios from the Preliminary History and Unrestated Quarterly databases to the Snapshot database. Results suggest that the ratios are nearly identical across databases. Given this similarity, we rely on the analyses of D’Souza et al. (2010) that verify the accuracy of the data and the procedure in identifying earnings announcement disclosures. However, D’Souza et al. (2010) note that, prior to 2000, the preliminary history database used to identify earnings announcement line items may be prone to identifying Wall Street Journal summary announcements in lieu of the earnings announcement. Because Compustat Snapshot can better identify the source code associated with the information included in the database, we do not think that it suffers from this limitation. That said, we replicate our primary analyses for the sample beginning with the year 2000 and find qualitatively and quantitatively similar results.

Given that interest expense and research and development appear to have a more pronounced difference in persistence compared to sales than the other items considered, we perform robustness tests with these items excluded from the measure of homogeneous versus heterogeneous disaggregation and find similar results.

Interest expense is sometimes considered to be a component of non-operating income and sometimes a component of operating income (such as in the operating section of the statement of cash flows). Fairfield et al. (1996) document that the time-series persistence of interest expense is consistent with the persistence of the other operating components. The results of our classification scheme are consistent with those of Fairfield et al. (1996). Therefore we follow their convention and list it as an operating component in this discussion.

In untabulated analyses, we limit our sample to those with I/B/E/S coverage to include a control for the presence of street or pro forma earnings and to measure earnings news based on analyst expectations. Results are qualitatively similar under this alternative specification. We elect to tabulate the results without restricting for I/B/E/S coverage, due to the substantial increase in sample size. We also performed a test (untabulated) in which we interact homo disagg ratio_i with an indicator variable for the presence of pro forma earnings in the earnings announcement to test if the relation between investor disagreement and homogeneous disaggregation differs based on the inclusion of pro forma earnings in the announcement. We find that the coefficient on the interaction term is insignificant, suggesting that the inclusion of pro forma earnings does not reduce the effect of homogeneous disaggregation on investor disagreement.

Given this distribution of homogenous earnings disaggregation, we also perform each of our primary analyses using an indicator variable set to one if all of the homogenous line items in the subsequent periodic filing are disaggregated in the earnings announcement. Results are qualitatively similar under this alternative specification.

Research has found that bad news announcements are associated with increased investor disagreement, market uncertainty, or both (e.g., Rogers, Skinner, and Van Buskirk 2009), presumably due to their unclear implication for future firm profitability (Hayn 1995).

However, we note that the homo disagg ratio is positively associated (i.e., t-statistic: 2.47) with ∆idvol when we replace the firm fixed effects with industry fixed effects (untabulated).

The estimates of the control variables are not tabulated for brevity.

Performing an over-identifying restrictions test (Larcker and Rusticus 2010) in this specification is not possible, because the number of instruments exactly equals the number of endogenous variables.

We also recalculate the hetero disagg ratio in a similar manner and label this ratio the alt hetero disagg ratio.