Evidence from Archival Data on the Relation Between Security Analysts' Forecast Errors and Prior Forecast Revisions*

Abstract

Abstract. This paper examines the association between analysts' forecast errors at the earnings announcement date and the revisions to those forecasts during the preceding year. The study is an initial effort to use archival data from expert decision makers to test behavioral theories that have support in laboratory environments. Consistent with findings of conservatism in laboratory experiments, we find that analysts systematically underweight new information. This finding is most pronounced when the analysts are consistently revising their estimates downward throughout the year. Résumé. Les auteurs examinent le lien entre les erreurs prévisionnelles des analystes à la date de déclaration des bénéfices et les révisions dont ces prévisions ont fait l'objet au cours de l'exercice précédent. Il s'agit là d'une première tentative d'utilisation des données d'archives des experts en matière décisionnelle dans le but de tester les théories comportementales qui ont été démontrées en laboratoire. Conformément aux observations faites en laboratoire en ce qui a trait à la prudence, les auteurs constatent que les analystes accordent systématiquement trop peu de poids à l'information nouvelle/Cette constatation est très marquée lorsque les analystes révisent continuellement leurs estimations à la baisse au cours de l'année.

Do analysts overreact to extreme good news in earnings?

“…In this sense, analysts underreact to past earnings news. 1 The positive correlation between forecast errors and past news has been corroborated by many other studies using a variety of news measures such as past forecast errors (Mendenhall 1991), past earnings levels (Ali et al 1992), past forecast revisions (Elliott et al 1995), and past stock returns (Abarbanell 1991). 2 Easterwood and Nutt (1999) attempt to provide reconciliation between the results of DeBondt and Thaler (1990) and Abarbanell and Bernard (1992).…”

Loss function assumptions in rational expectations tests on financial analysts’ earnings forecasts

“…Implicitly assuming that financial analysts have a quadratic loss function, prior studies find that financial analysts do not efficiently use information in prior earnings (DeBondt and Thaler, 1990;Abarbanell and Bernard 1992), extreme earnings changes (Easterwood and Nutt, 1999), forecast revisions (Elliott et al, 1995), forecast errors (Mendenhall, 1991), and stock returns (Lys and Sohn, 1990;Ali et al, 1992).…”

“…Mendenhall, 1991;Ali et al, 1992), and past forecast revisions (e.g. Elliott et al, 1995;Amir and Ganzach, 1998) are some information variables studied previously. We adopt the general framework of Runkle (1990, 1998), i.e.…”

“…inefficiency with respect to extreme prior earnings changes under a linear loss function, even though the rational expectations hypothesis is rejected statistically. Elliott et al (1995) report that analysts' forecast errors are positively correlated with their own prior forecast revisions, using sign tests, univariate correlations and OLS regression analysis with control variables. 15 Specifically, they correlate or regress forecast error (Actual-Late Forecast) on forecast revision (Late Forecast-Early Forecast), where Late Forecast is the last consensus IBES forecast before the current annual earnings announcement and Early Forecast is the first consensus annual IBES forecast after the previous annual earnings announcement.…”

Measurement Errors of Expected Returns Proxies and the Implied Cost of Capital

“…This assumption has great intuitive appeal, particularly for studying the measurement errors of ICCs that rely on analyst forecasts of future fundamentals. Because analysts can be slow to incorporate new information (e.g., Lys and Sohn, 1990;Elliot, Philbrick, and Wiedman, 1995;Guay et al, 2011;So, 2013), for example due to an anchoring-and-adjustment heuristic, their forecasts and the resulting ICCs may tend to exhibit persistent (but time-varying) errors. Another possibility is that persistent model misspecification errors gives rise to persistent and time-varying measurement errors.…”