Biased forecasts or biased earnings? The role of reported earnings in explaining apparent bias and over/underreaction in analysts’ earnings forecasts

https://doi.org/10.1016/j.jacceco.2003.11.001Get rights and content

Abstract

The extensive literature that investigates whether analysts’ earnings forecasts are biased and/or inefficient has produced conflicting evidence and no definitive answers to either question. This paper shows how two relatively small but statistically influential asymmetries in the tail and the middle of distributions of analysts’ forecast errors can exaggerate or obscure evidence consistent with analyst bias and inefficiency, leading to inconsistent inferences. We identify an empirical link between firms’ recognition of unexpected accruals and the presence of the two asymmetries in distributions of forecast errors that suggests that firm reporting choices play an important role in determining analysts’ forecast errors.

Introduction

Four decades of research have produced an array of empirical evidence and a set of behavioral and incentive-based theories that address two fundamental questions: Are analysts’ forecasts biased? And Do analysts underreact or overreact to information in prior realizations of economic variables? This empirical literature has long offered conflicting conclusions and is not converging to a definitive answer to either question. On the one hand, theories that predict optimism in forecasts are consistent with the persistent statistical finding in the literature of cross-sectional negative (i.e., bad news) mean forecast errors as well as negative intercepts from regressions of forecasts on reported earnings. On the other hand, such theories are inconsistent both with the finding that median forecast errors are most often zero and with the fact that the percentage of apparently pessimistic errors is greater than the percentage of apparently optimistic errors in the cross-section. A similar inconsistency is found in the literature on analyst over/underreaction to prior realizations of economic variables, including prior stock returns, prior earnings changes, and prior analyst forecast errors. Here, again, empirical evidence supports conflicting conclusions that analysts overreact to prior news, underreact to prior news, and both underreact and overreact as a function of the sign of prior economic news. Further reflecting the lack of consensus in the literature, a handful of studies fail to reject unbiasedness and efficiency in analyst forecasts after “correcting” methodological flaws or assuming nonstandard analyst loss functions.1

The accumulation of often inconsistent results concerning analyst rationality and incentives makes it difficult for researchers, practitioners, and policy makers to understand what this literature tells us. This motivates us to reexamine the body of evidence with the goal of identifying the extent to which particular theories for apparent errors in analysts’ forecasts are supported by the data. Such an exercise is both appropriate and necessary at this juncture as it can, among other things, lead to modified theories that will be tested using the new and unique hypotheses they generate.

We extend our analysis beyond a synthesis and summary of the findings in the literature by identifying the role of two relatively small asymmetries in the cross-sectional distributions of analysts’ forecast errors in generating conflicting statistical evidence. We note that the majority of conclusions concerning analyst-forecast rationality in the literature are directly or indirectly drawn from analyses of these distributions. The first asymmetry is a larger number and a greater magnitude of observations that fall in the extreme negative relative to the extreme positive tail of the forecast error distributions (hereafter, the tail asymmetry). The second asymmetry is a higher incidence of small positive relative to small negative forecast errors in cross-sectional distributions (hereafter, the middle asymmetry). The individual and combined impact of these asymmetries on statistical tests leads to three important observations. First, differences in the manner in which researchers implicitly or explicitly weight observations that fall into these asymmetries contribute to inconsistent conclusions concerning analyst bias and inefficiency. Second, a variety of econometric techniques and data adjustments fail to eliminate inconsistencies in inferences across different statistical indicators and conditioning variables. Such techniques include using indicator variables or data partitions in parametric tests, applying nonparametric methods, and performing data truncations and transformations. Third, econometric approaches that choose loss functions that yield consistent inferences—essentially by attenuating the statistical impact of observations that comprise the asymmetries—will not provide definitive answers to the question of whether analysts’ forecasts are biased and inefficient. This is because at this stage in the literature too little is known about analysts’ actual loss functions, and such methods thus leave unresolved the question of why the asymmetries in forecast error distributions are present.

We present statistical evidence that demonstrates how the two asymmetries in forecast error distributions can indicate analyst optimism, pessimism, or unbiasedness. We also show how observations that comprise the asymmetries can contribute to, as well as obscure, a finding of apparent analyst inefficiency with respect to prior news variables, including prior returns, prior earnings changes, and prior forecast errors. For example, our empirical evidence explains why prior research that relies on parametric statistics always finds evidence of optimistic bias as well as apparent analyst underreaction to prior bad news for all alternative variables chosen to represent prior news. It also explains why evidence of apparent misreaction to good news is not robust across parametric statistics or across prior news variables, and why the degree of misreaction to prior bad news is always greater than the degree of misreaction to prior good news, regardless of the statistical approach adopted or the prior information variable examined.

Finally, while our analysis does not lead to an immediately obvious solution to problems of inferences in the literature, it does reveal a link between the reported earnings typically employed to benchmark forecasts and the presence of the two asymmetries in distributions of forecast errors. Specifically, we find that extreme negative unexpected accruals included in reported earnings go hand in hand with observations in the cross-section that generate the tail asymmetry. We also find that the middle asymmetry in distributions of forecast error is eliminated when the reported earnings component of the earnings surprise is stripped of unexpected accruals. This evidence suggests benefits to refining extant cognitive- and incentive-based theories of analyst forecast bias and inefficiency so that they can account for an endogenous relation between forecast errors and manipulation of earnings reports by firms. The evidence also highlights the importance of future research into the question of whether reported earnings are, in fact, the correct benchmark for assessing analyst bias and inefficiency. This is because common motivations for manipulating earnings can give rise to the appearance of analyst forecast errors of exactly the type that comprise the two asymmetries if unbiased and efficient forecasts are benchmarked against manipulated earnings. Thus, it is possible that some evidence previously deemed to reflect the impact of analysts’ incentives and cognitive tendencies on forecasts is, after all, attributable to the fact that analysts do not have the motivation or ability to completely anticipate earnings management by firms in their forecasts.

This paper's emphasis is on fleshing out salient characteristics of forecast error distributions with an eye toward ultimately explaining how they arise. The analysis highlights the importance of new research that explains the actual properties of forecast error data and cautions against the application of econometric fixes that either fit the data to specific empirical models or fit specific empirical models to the data without strong a priori grounds for doing so. Our findings also represent a step toward understanding what analysts really aim for when they forecast, which is useful for developing more appropriate null hypotheses in tests of analysts’ forecast rationality, and sounder statistical test specifications, as well as the identification of first-order effects that may require control when testing hypotheses that predict analyst forecast errors.

In the next section we describe our data and present evidence of the sensitivity of statistical inferences concerning analyst optimism and pessimism to relatively small numbers of observations that comprise the tail and middle asymmetries. Section 3 extends the analysis to demonstrate the impact of the two forecast error asymmetries on inferences concerning analyst over/underreaction conditional on prior realizations of stock returns and earnings changes, as well as on serial correlation in consecutive-quarter forecast errors. Section 4 presents evidence of a link between biases in reported earnings and the two asymmetries and discusses possible explanations for this link as well as the implications for interpreting evidence from the literature and for the conduct of future research. A summary and conclusions are provided in Section 5.

Section snippets

Data

The empirical evidence in this paper is drawn from a large database of consensus quarterly earnings forecasts provided by Zacks Investment Research. The Zacks earnings forecast database contains approximately 180,000 consensus quarterly forecasts for the period 1985–1998. For each firm quarter we calculate forecast errors as the actual earnings per share (as reported in Zacks) minus the consensus earnings forecast outstanding prior to announcement of quarterly earnings, scaled by the stock

The effect of the two asymmetries on evidence of apparent analyst misreaction to prior stock returns, prior earnings changes, and prior forecast errors

In this section, we demonstrate how observations that comprise the tail and middle asymmetries in forecast error distributions conditional on prior realizations of economic variables contribute to inconsistent inferences concerning the efficiency of analysts’ forecasts. One important message of the ensuing analysis is that the likelihood that a forecast error observation falls into one or the other asymmetry varies by the sign and magnitude of the prior news. This feature of the data links the

Accounting conservatism and unexpected accruals

Abarbanell and Lehavy (2003a) argue that an important factor affecting the recognition of accounting accruals is the conservative bent of GAAP. Because conservative accounting principles facilitate the immediate recognition of economic losses but restrict the recognition of economic gains, the maximum amount of possible income-decreasing accruals that a typical firm can recognize in a given accounting period will be larger than the maximum amount of income-increasing accruals (see, e.g., Watts,

Summary and conclusions

In this paper we reexamine the evidence in the literature on analyst-forecast rationality and incentives and assess the extent to which extant theories for analysts’ forecast errors are supported by the accumulated empirical evidence. We identify two relatively small asymmetries in cross-sectional distributions of forecast error observations and demonstrate the important role they play in generating statistical results that lack robustness or lead to conflicting conclusions concerning the

References (47)

  • Abarbanell, J., Lehavy, R., 2003b. An explanation for why prior stock returns and analysts’ earnings forecast revisions...
  • J. Affleck-Graves et al.

    Forecasts of earnings per sharepossible sources of analyst superiority and bias

    Contemporary Accounting Research

    (1990)
  • A. Alford et al.

    A simultaneous equations analysis of forecast accuracy, analyst following, and trading volume

    Journal of Accounting, Auditing & Finance

    (1999)
  • A. Ali et al.

    Analysts’ use of information about permanent and transitory earnings components in forecasting annual EPS

    Accounting Review

    (1992)
  • W. Baber et al.

    The impact of split adjusting and rounding on analysts’ forecast error calculations

    Accounting Horizons

    (2002)
  • Bartov, E., Givoly, D., Hayn, C., 2000. The rewards to meeting or beating earnings expectations. Working paper, New...
  • Basu, S., Markov, S., 2003. Loss function assumptions in rational expectations tests on financial analysts’ earnings...
  • Becker, D., 2001. Analyzing analysts’ remarks” by General Counsel, U.S. Securities & Exchange Commission, Remarks...
  • L. Brown

    A temporal analysis of earnings surprisesprofits versus losses

    Journal of Accounting Research

    (2001)
  • L. Brown et al.

    Univariate time series models of quarterly accounting earnings per sharea proposed model

    Journal of Accounting Research

    (1979)
  • Brown, L., Foster, G., Noreen, E., 1985. Security Analysts’ Multi-year Earnings Forecasts and the Capital Markets....
  • Burgstahler, D., Eames, M., 2002. Management of earnings and analyst forecasts. Working paper, University of...
  • L. Chan et al.

    Momentum strategies

    Journal of Finance

    (1996)
  • Cited by (230)

    • Textual disclosure complexity and analysts’ weighting of information

      2024, Journal of Contemporary Accounting and Economics
    • An international study on the impact of corruption on analysts’ forecasts

      2022, Journal of International Accounting, Auditing and Taxation
    • Stakeholder Value: A Convenient Excuse for Underperforming Managers?

      2023, Journal of Financial and Quantitative Analysis
    View all citing articles on Scopus

    We would like to thank workshop participants at Emory University, the University of North Carolina, Hebrew University, the University of Michigan, New York University, Duke University, the University of Illinois (Champaign-Urbana), Tel Aviv University, the Joint Symposium of the Eleventh Annual Conference of Financial Economics and Accounting and the Seventh Mitsui Life Symposium on Global Financial Markets, the October 2000 NBER (Behavioral Finance) conference, the 2002 Journal of Accounting & Economics Conference, and George Benston, Peggy Bishop, Alon Brav, Robert Bushman, Francesca Cornelli, John Hand, Cam Harvey, David Hsieh, Mark Lang, Thomas Lys, Ed., Maydew, Jim Ohlson, Jay Patel, Christine Petrovits, the reviewer, Brian Rountree, Paul Simko, Ro Verrecchia, Ross Watts (the editor), Peter Wysocki, and Jerry Zimmerman for their helpful comments.

    View full text