Professional Documents
Culture Documents
Acknowledgements
List of Tables/Figures
Table of Contents
1. Introduction
2. Prior Literature
Sinha et al (1997) criticised O’Brien (1987) for falsely concluding in his study
that there were “no differences in analysts’ earnings forecast accuracy” as
there were no “control of differences in the recency of forecasts issued.”
Moreover, Lawrence D Browne (2001) also asserted that “analyst ratings are
important to those that believe past accuracy portends future forecast
accuracy.” An analyst’s prior ability (measured via prior forecast accuracy) to
forecast a specific firm’s earnings is a good indicator for the accuracy of her
earnings forecasts for that firm.
Hypothesis
The variables of interest here are DV = Analyst forecast accuracy & IV = Firm-
specific experience.
Forecast accuracy (DV) can be measured by applying Clement’s (1999)
‘Proportional Mean Absolute Forecast Error’ (PMAFE) formula
𝐏𝐌𝐀𝐅𝐄𝐢𝐣𝐭=(𝐀𝐅𝐄𝐢𝐣𝐭− 𝐀𝐅𝐄𝐣𝐭̅)/𝐀𝐅𝐄𝐣𝐭̅[= 𝑫𝑨𝑭𝑬𝒊𝒋𝒕 / 𝑨𝑭𝑬𝒋𝒕̅] . This measure
accounts for time-series and cross sectional differences where the individual
analyst’s forecast error that year minus the mean forecast error of all analysts
following the specific company that year scaled by the mean of forecast errors
of all analysts following the firm in the particular year.
Independent variables used will be derived from past studies such as Clement
(1999).
NOTATION:
𝑭𝑬𝑿𝑷𝒊𝒋 = Measures the analyst's firm-specific experience. i.e. the number of
years an analyst ‘i’ produced earnings forecasts for their financial firm ‘j’.
Dependent variables used will also come from the Clement (1999) case study.
NOTATION:
𝐀𝐅𝐄𝐢𝐣𝐭 = absolute difference between forecasted EPS by analyst ‘i’ and the
actual EPS of firm ‘j’ in year ’t’
𝐀𝐅𝐄𝐣𝐭̅ = Represents the absolute forecast error of analysts as a whole for
financial firms j, at time t.
𝑫𝑨𝑭𝑬𝒊𝒋𝒕 = 𝐀𝐅𝐄𝐢𝐣𝐭 - 𝐀𝐅𝐄𝐣𝐭̅
PMAFE is a good measure for several reasons, one of which, is that it controls
firm-year effects. Clement (1999) demonstrated how there were greater
deviations in 𝑫𝑨𝑭𝑬𝒊𝒋𝒕 amongst firms with higher EPS figures, which therefore
creates bias. Heteroskedasticity suggests variability is dependent on the
constant and to an extent, changes in the original independent variable which
is an issue. According to Clement (1999) greater unpredictability occur in
forecast errors for larger firms. So 𝐴𝐹𝐸𝑗𝑡̅ reduces heteroskedasticity by
reducing 𝐷𝐴𝐹𝐸𝑖𝑗𝑡 or by deflating AFEijt - AFEjt by AFEjt. Nonetheless, this
highlights how a smaller/negative PMAFE figure would suggest a better
performance (higher than average) in terms of forecast accuracy, vice-versa.
This is because accuracy and forecast error are conversely related.
Univariate Analysis……
This analysis will be used to establish and describe data regarding the firm-
specific experience and its effect on forecast accuracy. This should allow us to
determine whether a simple relationship exists between the two variables. For
consistency with the PMAFE measure, independent variables ie. other types of
experience such as industry are to be firm-year mean adjusted. The sample of
analysts will be split in to a firm-specific experienced group and a less firm-
specific experience group where average forecast accuracy will be measured.
There isn’t a strict method in classifying an analyst as experienced or
inexperienced thus the sample can be split into analysts with 0-5 years firm-
specific experience and those with more than 5 years (based on even division
of firm-specific experience years). It is also important that there are equal
sample numbers in both groups before performing the t-test in order to
determine any significance in value between the groups.
Multivariate Analysis…