Explaining firms’ earnings announcement stock returns using FactSet and I/B/E/S data feeds

The notion of Street earnings surprise was introduced into the academic literature by Bradshaw and Sloan (2002).

FactSet is a multinational financial data and software company that was founded in 1977 and went public in 1996. I/B/E/S (Institutional Brokers’ Estimates System) was founded by Lynch, Jones & Ryan and Technimetrics and began collecting earnings estimates for US companies in 1976. Barra bought I/B/E/S in 1993, then sold it to Primark in 1995. Thomson Financial (now Thomson Reuters) bought Primark in 2000. We focus on FactSet and I/B/E/S because they are the largest providers of analyst forecast and management guidance online data feeds to US capital markets.

A key part of FactSet’s strategy has been to combine with its own databases the disparate databases of many smaller data vendors that it has acquired. See https://​en.​wikipedia.​org/​wiki/​FactSet.

The FactSet and I/B/E/S analyst forecast data feeds differ in how the data are collected. I/B/E/S data are supplied to I/B/E/S by analysts, while FactSet’s data are primarily gathered manually from analysts’ PDF reports by FactSet employees. This means that the databases are subject to different sources of bias and/or error. I/B/E/S history data constitute at root a voluntary disclosure that for a variety of strategic or other reasons may not exactly reflect the contents of analysts’ PDF reports or full Excel-based financial models. However, the strengths of the I/B/E/S approach are that there is less ambiguity about what analysts are forecasting (since they supply information directly to I/B/E/S in a standardized manner) and that analysts can supply I/B/E/S with better information than they disclose in their PDF reports. In contrast, since FactSet estimates are manually extracted from analysts’ reports, analysts are not able to choose to supply different information in their reports versus their database feeds. Potentially offsetting this advantage is the risk that FactSet employees may misinterpret analysts’ PDFs and/or incorrectly enter the data they contain.

We do not undertake separate analysis on FactSet and I/B/E/S data, because each dataset has been built up over time as FactSet and Thomson Reuters have acquired smaller data providers.

In an untabulated analysis, we include KPI surprises in our annual regressions of earnings announcement returns on analyst forecast and guidance surprises. We find no significant improvement in the overall explanatory power (i.e., adjusted R²) of the model. There are no changes in our top four ranked surprises. The 13 non-KPI surprises with average t-statistics above 1.95 between 2008 and 2016 remain significant. We find 1 KPI, same-store sales for the retail industry, that has an average t-statistic above 1.95 during the 2008–2016 period. Same-store sales data are available for approximately 50–80 firms per quarter.

We require that the analyst consensus period begin no earlier than the first day of the quarter forecasted and no later than the earnings announcement date, and that the earnings announcement date be within 150 days of the quarter-end.

Of the non-KPI item observations that we use in our analysis, 60% are from FactSet and 40% are from I/B/E/S. All KPI items are from FactSet because we did not have access to I/B/E/S KPIs.

Three Items are omitted from appendix 1 because we require item surprises to be available for at least 5% of our sample firms in a given year. The Items that do not meet that threshold in any year are ITEM 5 – “Deferred Revenue – Short Term,” ITEM 7 – “Deferred Revenue – Long Term,” and ITEM14 – “Stock Option Expense.”

We caution against reading figure 1 as suggesting that FactSet adds very little beyond I/B/E/S, or vice versa. This is because we use historical data feeds as of early 2017, and both FactSet and I/B/E/S regularly add to their data feeds forecasts that were available in real time from other vendors’ data feeds but not from their own.

Chuk et al. (2013) hand-collect the actual provision of management forecasts. They find an increase in management forecasts between 1997 and 2001 (implementation of Regulation FD) but a relatively flat or declining trend between 2001 and 2007.

We base our assessment of significance on t-statistics because in panel B the correlation between t-statistics and incremental R² is 0.98.

Guidance firms are those with at least one guidance surprise available at the earnings announcement. High-revenue-growth firms are those whose prior-quarter sales are ≥10% higher than their sales four quarters earlier. Data on sales growth are not available for all firms in our sample. Street-profit firms are those for which the Street earnings announced at the earnings announcement are >0.

Revenue forecasts: Bradshaw et al. (2018), Clark and Elgers (1973), Ertimur et al. (2003), Ertimur et al. (2011), Jegadeesh and Livnat (2006), Jones (2007), Keung (2010), Rees and Sivaramakrishnan (2007), Schreuder and Klaassen (1984), Swanson et al. (1985), and Trueman et al. (2001). Cash flow forecasts: Brown and Christensen (2014), Call et al. (2009, 2013), DeFond and Hung (2003), DeFond et al. (2007), Givoly et al. (2009), McInnis and Collins (2011), Mohanram (2014), and Radhakrishnan and Wu (2014). We identified four papers outside of the top five accounting journals: Brown et al. (2013), Lerman et al. (2007), Pae and Yoon (2011) and Yoon and Pae (2013). We also identified two recent working papers: Calegari and Eames (2016) and Ohlson et al. (2016). Also, Givoly et al. (2019) explore the information content of I/B/E/S’s KPI analyst forecasts, which we do not include in our study.