Refining financial analysts’ forecasts by predicting earnings forecast errors

• Pages: 256-272
• Published date: 02 May 2017
• Date: 02 May 2017
• DOI: https://doi.org/10.1108/IJAIM-06-2016-0065
• Author: Tatiana Fedyk

Rening nancial analysts’

forecasts by predicting earnings

forecast errors

Tatiana Fedyk

Department of Accounting, University of San Francisco, San Francisco,

California, USA

Abstract

Purpose –The purpose of this paper is to examine the way serial correlation in quarterly earnings forecast

errors varies with rm and analyst attributes such as the rm’s industry and the analyst’s experience and

brokerage house afliation. Prior research on nancial analysts’ quarterly earnings forecasts has documented

serial correlation in forecast errors.

Design/methodology/approach –Finding that serial correlation in forecast errors is signicant and

seemingly independent of rm and analyst attributes, the consensus forecast errors are modeled as an

autoregressive process. The model of forecast errors that best ts the data is AR(1), and the obtained

autoregressive coefcients are used to predict consensus forecast errors.

Findings –Modeling the consensus forecast errors as an autoregressive process, the present study predicts

future consensus forecast errors and proposes a series of renements to the consensus.

Originality/value –These renements were not presented in prior literature and can be useful to nancial

analysts and investors.

Keywords Consensus forecast, Financial analysts’ earnings forecasts, Forecast errors,

Predicting forecast errors

Paper type Research paper

1. Introduction

The extensive literature on sell-side analysts’ earnings forecasts reects the importance

those forecasts have acquired in the past two decades. Earnings forecasts play an integral

role in capital markets (Lys and Sohn (1990)), and analysts have incentives to issue accurate

forecasts (Mikhail et al., 1999).

However, sometimes these forecasts fail to fully reect all new information. Moreover,

much of the inefciency in earnings forecasts appears to be persistent. For example,

Eastwood and Nutt (1999) suggest that analysts underreact to negative information but

overreact to positive information, thus on the whole issuing overly optimistic forecasts.

Furthermore, Bulter and Lang (1991) nd that individual analysts tend to issue

systematically optimistic or systematically pessimistic forecasts relative to the consensus.

Their ndings indicate that analysts’ misreactions persist through time. Providing further

evidence of the persistence of analysts’ misreaction, Abarbanell and Bernard (1992) nd

signicant serial correlation between analysts’ quarterly forecast errors, both in by-rm

samples and in pooled cross-section. This effect appears to subside after approximately three

lags, indicating that analyst biases persist for about three quarters. Finally, Bradshaw et al.

(2012) re-examine the conclusion (as proposed by Brown and Hagerman (1987)) that analysts’

forecasts are superior to time-series forecasts and nd instead that analysts’ forecasts do not

present a uniformly better prediction of future earnings than forecasts from time-series

models. For example, they nd that analysts’ forecasts are not consistently better than

The current issue and full text archive of this journal is available on Emerald Insight at:

www.emeraldinsight.com/1834-7649.htm

IJAIM

25,2

256

Received 6 June 2016

Revised 11 September 2016

Accepted 12 September 2016

InternationalJournal of

Accounting& Information

Management

Vol.25 No. 2, 2017

pp.256-272

©Emerald Publishing Limited

1834-7649

DOI 10.1108/IJAIM-06-2016-0065

time-series models for longer horizons, and that time-series model forecasts strongly

dominate analysts’ forecasts for three-year-ahead earnings. A number of recent papers also

re-examine the accuracy and value-relevance of analysts’ earnings and cash ows forecasts

in international settings. For example, Huang and Boateng (2016) study Chinese rms,

Ahmed and Ali (2013) provide evidence from Australia and Liu and O’Farrell (2013) assess

analyst forecast accuracy using an extensive data set from six nations.

From the apparent predictability in nancial analysts’ earnings forecast errors stems the

literature on predicting earnings forecasts, such as Stickel (1990),Abarbanell and Lehavy

(2003) and Konchitchki et al. (2010).Stickel (1990) proposes a model that predicts an

individual analysts’ earnings forecast from the change in mean consensus forecast since that

analyst’s most recent forecast, the deviation of the analyst’s most recent forecast from the

consensus and the cumulative stock return since the date of the analyst’s most recent

forecast. Meanwhile, Abarbanell and Lehavy (2003) posit that the likelihood of small forecast

errors increases with the rm’s stock price sensitivity to earnings news. While both these

studies predict individual analysts’ earnings forecasts, they do not explicitly exploit

predictability in earnings forecast errors. Finally, Konchitchki et al. (2010) consider yearly

earnings forecasts and nd that, although some prior studies suggest that analysts

incorporate information from the contemporaneous year into their forecasts, earnings

forecasts fail to fully incorporate the information available in prior returns over a longer

period. In particular, they posit that stock return in year t⫺1can be used to predict the sign

of the error in the mean of analyst forecasts released in year tfor that scal year.

Literature that specically addresses serial correlation in earnings forecast errors

includes Markov and Tamayo (2006), who provide a potential explanation for the observed

predictability in forecast errors. They suggest that serial correlation in analysts’ quarterly

earnings forecast errors is consistent with a model where analysts face parameter

uncertainty and learn rationally about the parameters over time. Markov and Tamayo (2006)

thus argue that predictability in earnings forecast errors is more consistent with rational

learning than irrationality. These ndings suggest that predictability in earnings forecast

errors is an inherent feature of rational forecasting rather than a passing anomaly,

conrming the persistent nature of the serial correlation.

While these papers study the behavior of nancial analysts as a whole, another sector of

the literature on analyst earnings forecasts considers that different analysts behave

differently and investigates the way earnings forecasts vary with analyst attributes. For

example, Clement (1999) nds that an analyst’s forecast accuracy is positively related to his

or her experience and employer size and negatively associated with the number of stocks

followed by that analyst. On the other hand, Jacob et al. (1999) nd an association between

brokerage house characteristics and forecast accuracy but suggest that forecast accuracy is

not enhanced by experience. Meanwhile, Hong et al. (2000) nd that inexperienced analysts

deviate less from consensus forecasts, are less likely to issue timely forecasts and revise their

forecasts more frequently.

The present paper connects the literature on the predictability of nancial analysts’

forecast errors with the study of differences in analysts’ forecasting ability, by considering

the way an analyst’s experience and brokerage house afliation affect the extent of

autocorrelation in the forecast errors. In particular, I establish that serial correlation in

earnings forecast errors is prevalent across industries and analyst experience levels and

brokerage house afliations, and does not vary systematically with these attributes.

Motivated by this nding, the present paper explicitly uses serial correlation in consensus

forecast errors to predict errors and create a more accurate consensus.

257

Earnings

forecast errors