. fl
fly...
. . ¢ .
Yamfiflmp

W3

1.
tithrs
. ..:

.a. ... é

~ “ma
VA... .

a
. 1.. H a:
unenivn

. 53%

. L. in“

a
5E...
flaw
ta :1

, .

 

2. ;
ini‘I-n‘
ix

- . éfifigmfifiw

 

‘ LIBRARIES
\ MICHIGAN STATE UNIVERSITY
t 1 " ' I EAST LANSING, MICH 48824-1048

; )
“1;;96‘2‘33

\J

This is to certify that the
dissertation entitled

EFFECTS OF INCREASED REPORTING
FREQUENCY ON ACCURACY, DISPERSION AND
CONFIDENCE INTERVALS OF
NONPROFESSIONAL INVESTORS’ EARNINGS
PREDICTIONS

presented by

TERENCE JUDE PITRE

has been accepted towards fulfillment
of the requirements for the

PhD. Degree in Accounting

QD/m 3/ WW

 

 

/ .. Major Piofessor’s Sflnature
«/ / a ~ 0 ~ 0 9/
Date

MSU is an Affirmative Action/Equal Opportunity Institution

---I-o-l-o-v-o-c-a-I-t--o-n-n-a-1-I-n—t--l-t-o-o-o-v-o--t-o--<-I-c-O-I-I-i-C-o-I-a-n-----<-. «—

OC-'--I--a-U--c

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 C2/CIRC/DateDue.p65-p.15

 

EFFECTS OF INCREASED REPORTING FREQUENCY ON
ACCURACY, DISPERSION AND CONFIDENCE INTERVALS OF
NONPROFESSIONAL INVESTORS’ EARNINGS PREDICTIONS

By

TERENCE JUDE PITRE

A DISSERTATION
Submitted to
Michigan State University
In partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Department of Accounting and Information Systems

2004

ABSTRACT

Effects of Increased Reporting Frequency on the Accuracy,
Dispersion and Confidence Intervals of Nonprofessional Investors’ Earnings
Predictions

By

Terence Jude Pitre

Using a between subjects experiment, I analyze how the fi'equency of reporting--
weekly earnings, as opposed to quarterly earnings-affects the accuracy, dispersion and
confidence intervals of earnings predictions by nonprofessional investors. I hypothesize
and find that more frequent reporting results in less accurate predictions and larger
dispersion of predictions for earnings with a strong seasonal pattern. I also hypothesize,
but do not find support, that more frequent reporting significantly increases confidence
interval widths among nonprofessional investors. Joint presentation of weekly and
quarterly earnings reduces the prediction error generated by weekly reporting alone, but
results in an increase in dispersion of predictions when compared to quarterly reporting

alone.

ACKNOWLEDGMENTS

First, I would like to thank God for his strength and wisdom to undertake and
complete this task. I would also like to thank the faculty members of Michigan State
University, in particular Joan Luft, Matthew Anderson, K. Ramesh, Bruce Burns, Richard
Weber, Ed Outslay, Bill McCarthy, Fred Jacobs, Joseph Anthony and Mike Shields.
Finally, I would like to thank my family, friends, classmates, KPMG, AICPA and my
fi'aternity brothers of Kappa Alpha Psi Inc. -Theta Nu Chapter for their support
throughout this process: Betty A. Pitre, Mary L. Williams, Mayor Joseph A. Pitre, Mr. &
Mrs. Albert J. Pitre, Joseph Miles and Family, Frank Anderson and Family, Carole J.
Bartley, Ninas A. Jenkins, Clyde White, Danny Lanier, Christian Mastilak, Alex Woods,
Freddie Coronado, Michael Brown, Anne Farrell, Leslie Campbell, Jan Williams, Angela
Andrews, Nettavia Curry, Ola Smith, Janet Goodwin, Silvia Forero, Liliana Guiterrez,

and Susan Franklin.

iii

TABLE OF CONTENTS

LIST OF TABLES .................................................................................... v
LIST OF FIGURES ................................................................................. vii
1.0 INTRODUCTION ............................................................................... 1
2.0 LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT........................5
Section 2.1 Summary of the Literature Review ........................................................ 6
Section 2.2 Factors that affect confidence intervals .................................................. 8
Section 2.3 Factors That Affect Prediction Accuracy ............................................... 9
Section 2.4 Processing Effects of More Frequent Reporting ................................... 15
Section 2.5 Model of Earnings Behavior ................................................................ 16
Section 2.6 Effects of Increased Frequency on Uncertainty .................................... 16
Section 2.7 Effects of Increased Frequency on Accuracy ....................................... 18
Section 2.8 Effects of Increased Frequency on Dispersions .................................... 20
Section 2.9 Joint Effects of More and Less Frequent Reporting ............................. 21
3.0 RESEARCH DESIGN .......................................................................... 24
Section 3.] Experimental Design and Dependent Variables ................................... 24
Section 3.2 Independent Variables and Data Creation ........................................... 24
Section 3.3 Experiment Materials and Procedures .................................................. 27
Section 3.4 Participants .......................................................................................... 28
4.0 RESULTS: HYPOTHESES 1-3 ............................................................... 29
Section 4.1 Results: Research Questions 1-3: Effects of Joint Presentation ............ 33
5.0 CONCLUSIONS ................................................................................. 37
APPENDIX ........................................................................................... 47
REFERENCES ....................................................................................... 62

iv

LIST OF TABLES

Table l-Descriptive Data Correlations ........................................................................... 39

Table 2-Cell Means and Standard Deviations for Confidence Intervals in Task 1 and 2 .40

Table 3-Analysis of Effects of Reporting Frequency ...................................................... 41
Table 4-Cel] means and Standard Deviations for ........................................................... 42
Table 5-Analysis of Effects of Reporting Frequency ...................................................... 43
Table 6-Cell Variances for Task 1 and Task 2 ............................................................... 44
Table 7-Analysis of Effects of Frequency on Confidence Intervals ................................ 45
Table 8-Analysis of Effects of Frequency on Absolute Prediction Error" ....................... 46
Table 9-Means and Standard Deviations,First Task ....................................................... 53
Table 10-Analysis of Effects of Frequency on Signed Prediction Error .......................... 53
Table 1 1-Correlations, Post-Experiment Question and Judgment Performance .............. 54

Table lZ-Means and Standard Deviations for Uncertainty in Task 1 and Task 2
Controlling for Direction of Change in Earnings in Prediction Period .................... 55

Table l3-Analysis of Effects of Reporting Frequency on Confidence Intervals When the
Prediction Period Change in Earnings is Positive ................................................... 55

Table l4-Analysis of Effects of Reporting Frequency on Confidence Intervals When the
Prediction Period Change in Earnings is Negative .................................................. 55

Table lS-Means and Standard Deviations of Confidence Intervals in Task 1 and Task 2

Controlling for Random Error Draw ...................................................................... 56
Table 16-Analysis of Effects of Reporting Frequency on Confidence Intervals For

Random Error Draw One ....................................................................................... 56
Table 17-Analysis of Effects of Reporting Frequency on Confidence Intervals For

Random Error Draw Two ...................................................................................... 56
Table 18-Analysis of Effects of Reporting Frequency on Signed Prediction Error ......... 57

Table l9-Analysis of Effects of Reporting Frequency on Signed Prediction Error for
Those Who Received Less Frequent Reporting in the 1“ Task ............................... 57

Table 20-Analysis of Effects of Reporting Frequency on Signed Prediction Error for
Those Who Received More Frequent Reporting in the 1“ Task .............................. 58

Table 21-Means and Standard Deviations for Prediction Error in Task 1 and Task 2
Controlling for Direction of Prediction Period Change in Earnings ........................ 59

Table 22-Analysis of Effects of Reporting Frequency on Prediction Error When the
Prediction Period Change in Earnings is Positive ................................................... 59

Table 23-Analysis of Effects of Reporting Frequency on Prediction Error When the
Prediction Period Change in Earnings is Negative .................................................. 59

Table 24-Means and Standard Deviations for Prediction Error in Task 1 and Task 2
Controlling for Random Error Draw ...................................................................... 60

Table 25-Amlysis of Effects of Reporting Frequency on Prediction Error for Random
Error Draw One ..................................................................................................... 60

Table 26-Analysis of Effects of Reporting Frequency on Prediction Error for Random
Error Draw Two .................................................................................................... 60

Table 27-Means and Standard Deviations for Prediction Error in Task 1 and Task 2 ..... 61

Table 28-Analysis of Effects of Reporting Frequency on Prediction Error for Investors
with Investing Experience ...................................................................................... 61

Table 29-Analysis of Effects of Reporting Frequency on Prediction Error for Investors
with No Investing Experience ................................................................................ 61

vi

LIST OF FIGURES

Figure 1 More Frequent Reporting-Seasonal Data ...................................................... 18
Figure 2 Less Frequent Reporting-Seasonal Data ......................................................... 19

vii

1.0 INTRODUCTION

The optimal frequency of earnings reporting has been much discussed. At one
extreme, some European firms report semiannually and have resisted attempts by
regulators to require quarterly reporting (Commission 2003). At the other extreme, firms
such as Cisco, using in-house technology, have already adopted the concept of real-time
internal reporting, and technologies such as XBRL (eXtensible Business Reporting
Language) are expected to make more frequent external reporting a more realistic
possibility for companies (Watson, McGuire and Cohen, 2000).

More frequent reporting of earnings is often seen as strictly advantageous to
investors, while less frequent reporting is disadvantageous. A former Securities &
Exchange Commission Chairman, Harvey Pitt, argues that “quarterly filings produce an
out-of-date snap shot rather than a real-time window” (Levitt, 2002). Hunton, Wright,
and Wright (2003) find that a sample of 215 financial mangers, analysts and investors
believe that increasing the reporting frequency of earnings would increase the decision
usefulness of financial statements and the quality of earnings.

Little is known about the consequences of more frequent reporting, however. One
major benefit of more frequent reporting is increased timeliness of information provision
to investors.1 Offsetting this benefit, however, are potential negative effects on
subjective judgment. More frequent evaluations made possible by more frequent
reporting of securities returns increase nonprofessional investors’ impressions of the

uncertainty of the returns (Gneezy and Potters 1997), and more frequent evaluations of

 

‘ Other benefits have also been claimed for more frequent reporting. Hunton et al. (2003) argue that more
frequent reporting of earnings would reduce the ability of managers to engage in earnings management and
assist investors to detecting earnings management, because they would have more information to detect
trends, patterns and fluctuations in earnings.

earnings could have similar effects. Moreover, larger data sets resulting from more
frequent reporting could increase individuals’ cognitive load, make pattern recognition
more difficult, and make it more likely that nonprofessionals rely on simple heuristics.
Human information processing theory (Schroder, Driver and Streufert 1967) suggests that
there is an inverted U-Shaped relationship between information load and decision quality.
Larger data sets could increase a number of judgment errors that have been demonstrated
in the psychology literature: failure to detect significant patterns in data series (Klayman
1988), incorrect identification of patterns in a non-random data series (Maines and Hand
1996), or a tendency to see nonexistent patterns in a random series (Andreassen 1987,
1990; Bloomfield et a]. 2001; Lim and O’Connor 1996 and O’Connor et al. 1993).

Nonprofessional investors, who are unlikely to use SOphisticated statistical models to
predict earnings, could be particularly vulnerable to these subjective information-
processing effects. As much as 42% of ownership in the top 1,000 US. companies is
made up of nonprofessional investors (Editorial staff, 2000). As a result, recent research
has focused on nonprofessional investors’ use of earnings reports to make judgments
about future earnings, risk and firm value (Maines and McDaniel 2000; Bartov,
Radhakrishnan and Krinsky 2000; Maines and Hand 1996).

In this study, I experimentally examine how reporting frequency of earnings per share
affects the accuracy, dispersion and confidence intervals of earnings predictions of
nonprofessional investors.

I first test the effects of reporting fiequency when participants are randomly assigned
to either more fiequent (weekly) or less fiequent (quarterly) reporting. The weekly

earnings series were constructed so that the weekly earnings provided no statistical

advantage or disadvantage in estimating quarterly earnings. Consequently this study only
measures judgment effects of more frequent reporting that are not due to differences in
information content. Earnings predictions and confidence intervals were elicited from the
participants. I also investigate the effects on subjective information processing when
weekly and quarterly earnings per Share data are jointly presented to nonprofessional
investors.

The results indicate that more frequent reporting results in less accurate and more
dispersed predictions of quarterly earnings when data are not presented jointly. More
frequent reporting did not affect the confidence intervals of nonprofessional investors.
Joint presentation, however reduces the prediction error generated by more frequent
reporting. Finally, I find that the joint presentation of more and less frequent reporting
results in an increase in dispersion of predictions when compared to the less frequent
reporting condition.

This study contributes to the regulatory debate, both in the United States and abroad,
about the usefulness of more fiequent reporting of earnings. Regulators and researchers
have tended to focus on the benefits of more frequent reporting. This study suggests that
there are potential costs to investors’ use of frequent reporting and these costs should be
incorporated into the debate. This research also contributes to the time series forecasting
literature. Previous research has examined the affects of periodicity,2 trend, seasonality,
noise and instability, but has not examined how the fiequency of reporting can affect

forecasting judgments.3

 

2 Periodicity refers to either, yearly, quarterly or monthly time series data.
3 See Webby and O’Connor (1996) for a review of judgmental and statistical time series literature.

The remainder of this paper is organized as follows: Section 2 contains the literature
review and hypotheses development; Section 3 contains the experimental design;

Section 4 contains the empirical results and section 5 the conclusions.

2.0 LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT

“The annual and quarterly reporting regime is not only on its way to becoming less and
less useful, it is on its way to becoming a dinosaur, an organism that has outlived its
environment.”

—ROBERT K. ELLIOTT, JULY 2000

US Senate Banking Committee hearing on adapting a
1930’s financial reporting model to the 215i century
The relative costs and benefits of increased reporting frequency have long been of
interest both to managers and investors. Although the Securities and Exchange Act of
1934 specifically allowed the SEC to mandate both annual and quarterly financial
statements for exchange-listed corporations, the SEC initially mandated only annual
reports. In 1945, the SEC required quarterly revenue reporting but not net income,
apparently acquiescing to public criticisms that quarterly income information
would be unreliable and potentially misleading due to some businesses’ seasonal nature.
Mandatory quarterly reporting was not officially adopted until 1970 after the 1969 Wheat
Commission. "

In 1995 a speech by SEC Commissioner Wallman called for consideration of more
fi'equent reporting, Stating, “Today, annual and even quarterly reports do not capture and
communicate material developments in sufficient time to meet market informational
needs” (Wallman 1995). In 2003, a proposal for more frequent reporting of earnings in
Europe, motivated by the desire for more transparency, raised concerns that both short-

termism and earnings management would increase as a result of more frequent reporting

(Evans 2003).

 

4 SEC, Securities and Exchange Act of 1934, Release No. 8683, October 15, I969; and SEC, Securities and
Exchange Act of 1934, Release No. 9004, October 28, 1970.

Results in the archival capital markets research literature suggest that nonprofessional
investors are less successful at incorporating the time series properties of earnings into
their forecast than are professional investors’ (Abarbanell and Bernard 1992; Jacob and
Lys 1996). Examining nonprofessional investors’ judgments has the potential to add
convergent validity to the results found in the capital markets literature as well as provide
new insights into nonprofessional investors’ use of time-series data.

More fi'equent reporting can affect the subjective judgment of nonprofessional
investors, and examining some of the potentially negative effects on judgment is the main
purpose of this research. First I summarize and review the literature related to the factors
that affect confidence intervals and the accuracy of predictions. 1 then summarize the
processes by which more frequent reporting can affect the judgment of nonprofessional
investors. Next, I describe the time series model used to generate earnings data in this
study. Finally, using this time series, I motivate my hypotheses about the specific effects
of more frequent reporting on uncertainty, accuracy, and dispersions of predictions.
These hypotheses begin by addressing the effects of nonprofessionals’ using either more
fiequent or less frequent reporting. Because technology will make it possible for
nonprofessional investors to use either form of reporting alone or both together, I then
investigate and report the effects of presenting nonprofessionals with both forms of data.
Section 2.1 Murray gthe Literatye Review

The next 2 sections have described research that is important for understanding the
judgments and decision making with time series data. Specifically the literature review
addressed two major topics: (1) The effects of time series data on confidence intervals

and (2) The effects oftirne series data onprediction accuracy. The dominant findings

from the literature regarding confidence intervals can be summarized as follows (see
following subsections for details):

(a) Individuals do not distinguish between positive and negative autocorrelated
sequences when establishing their confidence intervals.

(b) Confidence intervals tend to be particularly narrow for flat series but close to
that of a statistical model for both upward and downward series.

(c) Confidence intervals increase as the level of seasonality increases and are not
found to be affected by the levels of noise in the series.

The dominant findings from the literature regarding prediction accuracy can be
summarized as follows:

(a) Individuals are able to distinguish between positively autocorrelated
sequences and random sequences, but sometimes see patterns in random data
when they do not exist. Furthermore, individuals are not able to distinguish
between negatively autocorrelated sequences and random sequences.

(b) Both trend and seasonality affect the ability of individuals to make accurate
predictions.

(c) Prediction accuracy is enhanced when judgmental forecasts are combined
with other judgmental forecasts or other quantitatively derived forecasts, extra
causal information is used, a directive search strategy is employed and when
decision aids are used to generate forecasts. Prediction accuracy is best when

the forecast period is short.

((1) Prediction accuracy is worsened by the presence of discontinuities in the data
series and also by the presence of incentives to bias the forecast (i.e.,
professional relationships with the firm being analyzed.)
(e) Individual forecasts are typically dampened slightly. That is, individuals tend
to underforecast upward-trend series and overforecast down-trend series.
Section 2. 2 Factors that cat con dence intervals

My research theorizes that more frequent reporting will affect the uncertainty
(confidence intervals) of predictions generated fi'om time series data. In this section, I
review the key literature related to factors that affect confidence intervals.

Eggleton (1976, 1982) investigated and found that the width of the confidence
intervals was affected by the nature of the statistical variation in the data. Results
indicated that individuals did not distinguish between positively autorcorrelated,
negatively autorcorrelated and random sequences when establishing confidence intervals.
That is, individuals ignored the nature of the underlying sequence when setting their
credible intervals.

Lawrence and Makridakis (1989) investigated the effects of the type of trend
(upwards, flat or downwards), randomness and presentation on forecast accuracy and
confidence intervals. They found that confidence intervals were particularly narrowed
(overconfident) for the flat series, but were closer to confidence intervals generated by
the statistical model for upward and downward trends.

Similarly, O’Connor and Lawrence (1989) used 33 real-life time series and found that
confidence intervals surrounding predictions were excessively narrow, indicating

overconfidence, but that these intervals increased over time with feedback. Because

previous studies (Lichtenstein and F ischohoff, 1978) found that task difficulty in
forecasting affected the width of confidence intervals, O’Connor and Lawrence (1989)
also examined the effects of task difficulty (as defined by forecast error).5 Specifically,
they found that for medium to high difficulty tasks, confidence intervals were too narrow,
indicating overconfidence. However, confidence intervals in the low task difficulty group
were larger and significantly different from the medium and high difficulty task. 6

In a subsequent study, O’Connor and Lawrence (1992) found a positive relationship
between the level of seasonality and confidence intervals suggesting that as the level of
seasonality increases, the size of the confidence interval will also increase. Surprisingly,
the authors did not find support for an influence of noise on the width of confidence
intervals. They also found that as the level oftrend7 increased, the width of the
confidence interval increased.
Section 2. 3 Factors That Aflect Prediction Accuracy

My research also theorizes that more frequent reporting will affect the accuracy of
predictions generated by nonprofessional investors. In this section I review key
literature related to both the ability of individuals to make accurate predictions using time
series data and the factors (i.e., seasonality and trend) that have been demonstrated to
affect the accuracy of predictions.

Eggleton (1976, 1982) argued that the accuracy of individuals’ estimates of future

observations and their confidence intervals are a function of their ability to (1) correctly

 

’ Lichtenstein and Fischohoff, (1978) found that when the forecasting task is easy, underconfidence
abounds, and when forecasting is difficult, overconfidence is prevalent.

5 Analysis of the histograms of the magnitudes of the differences between the judgmental and statistical
methods reveals that the judgmental confidence intervals for the low difficulty task was different from that
of other groups, but the result did not occur for the high group.

7 Trend was defined as the slope of the regression line fitting some interval of the deseasonalized data.

assess the nature of the underlying relationship in the data, (2) cognitively represent the
characteristics of that relationship and (3) generate bias-free predictions from that
cognitive representation. Based on ANOVA results and a post experiment questionnaire,
Eggleton found that individuals were able to distinguish between positively
autocorrelated (trend) sequences and random sequences, but were unable to distinguish
between negatively autocorrelated (alternating) sequences and random sequences.
Individuals’ prediction performance can be summarized as follows: (A) for random
sequences, predictions were accurate close to the mean of the sequence; (B) for positive
autocorrelated (trend) sequences, they were below the statistical estimate, but above the
mean of the sequence; and (C) for negatively autocorrelated (alternating) sequences, they
were above the statistical estimate and close to the mean of the sequence. Eggleton
concluded that the differences between individuals’ predictions and that of the statistical
model were due to the nature of the error and the characteristics of the time series. He
also found that participants were more likely to read a systematic pattern into random
series when the variability in the series increased.

Research from accounting literature examined the ability of individuals (investors) to
recognize patterns. Maines and Hand (1996) examined whether individuals’ earnings
forecasts correctly reflect the positive autocorrelation in seasonal quarterly earnings and
the negative fourth-order moving average term documented by Brown and Rozeff (1979).
Their results indicated that participants’ forecasts reflected the autoregressive and fourth-
order moving average, but underweighted them Additionally, the authors found that

participants incorrectly identified an autoregressive component in data from a seasonal

10

random walk series, illustrating that people sometimes see patterns where they do not
exist.

Research has found that factors such as seasonality and trends also affect prediction
accuracy. Lawrence, Edmundson and O’Connor (1985) documented the ability of
individuals to generate accurate forecasts using time series data that contained a seasonal
component. In a subsequent study Lawrence, Edmundson and O’Connor (1986) found
that combining judgmental forecasts with either other judgmental forecasts or with
quantitatively derived forecasts can improve the accuracy of the forecast. They
documented the greatest benefit of combining forecasts arises when the forecast period is
Short and the task is easier.8

Lawrence and Makridakis (1989) investigated the effects of the type of trend
(upwards, flat or downwards), randomness and presentation on forecast accuracy. They
found that the deviations between participants’ subjective forecasts and those generated
by a regression model were influenced by the type of trend in time series data and that
estimates were not influenced by noise (consistent with Mosteller et al. 1981). Similar to
the results found in Eggleton (1976; 1982), individual forecasts tended to be dampened
slightly. That is, individuals tended to underforecast upward-trend series and
overforecast down-trend series.

Increasing the frequency of reporting has been shown to affect the saliency of the
patterns or trends. Using a model of judgmental extrapolation based on exponential

smoothing, Andreassen and Kraus (1990) presented support for the hypothesis that

 

' In order to segment the ease of forecasting, the time series were stratified into three groups according to
Mean Absolute Percentage Error (MAPE): those 25% with the highest MAPE were classified as more
difficult, those 25% with the lowest average MAPE were considered less difficult and the remaining 50%
that fell in between.

11

judgmental extrapolation forecasts vary as a function of salience of changes. They
demonstrated that focusing participants’ attention on price gha_nges as opposed to price
l_e_ve_ls_ led to more accurate assessments of the trend.9 They also demonstrated that
changing the statistical properties (i.e., sample size) in the stock prices series so that the
changes were more likely to be salient affected judgments.lo They find that strong signals
relative to noise levels affected the ability of participants to detect trends. This result was
in direct contrast to the results of Mosteller et al. (1981) and Lawrence and Makridakis
(1989). One major difference and possible explanation was that participants in the
Andreassen and Kraus study were only given the latest value in the series and the change
from the previous value, while participants in the other two studies were given graphs of
series. I

An implicit argument in the previous studies is that people expect past trends to
continue. Using 38,000 forecasts of stock price data, De Bondt (1993), showed that non-j
experts expected continuation of past trends in prices. That is, they were optimistic in
bull markets and pessimistic in bear markets. Interestingly, participants hedged their
forecast. If they expected positive changes, participants’ subjective probability
distribution of future prices was left skewed, while the probability distribution was right

skewed for negative changes.

 

9 Participants were required to choose one stock for 120 trials after the current price and the price change
from the previous trial. Determination of the trend can be inferred from the pattern of trading. To measure
the pattern of trading, authors computed a correlation between the price on a given trial and the number of
shares the subject held at the end of the trial. The correlations were converted to z scores and referred to as
a tracking measure. Detection of trend would result in participants holding more shares when average
prices changes were positive and the price is relatively high and holding less shares when the average share
changes have been negative and the price is relatively low.

'° .lncreasing sample size (holding the mean and variance constant) should increase the salience of the
changes leading to positive shifts in the tracking measure.

12

Prior research has also studied how factors other than trend, saliency, and seasonality
affect forecast accuracy. Lim and O’Connor (1996) found that people could distinguish
between low causal cues and high causal cues. Low causal cues were those with low
vhlidity and high causal cues where those with high validity. They also found that
participants could incorporate the extra-model causal information to make up for what the
statistical time-series model lacked. In this study, extra-model causal information
consisted of information beyond that of a statistical forecast. The authors found that
people could use the extra-model information to generate forecasts that were superior to
statistical models.

Whitecottton (1996) examined the effects of experience and use of a decision aid on
forecast accuracy. The probability predictionll for a sample of 16 firms was used as a
decision aid. Undergraduates, MBA students and professional analysts were used to
proxy for three different levels of experience. She found that both experience and use of
a decision aid had a positive effect on the accuracy as measured by the mean probability
score. Interestingly, there was no significant difference in accuracy between MBA
students and financial analysts. This was consistent with the contention of Yates,
McDaniel and Brown (1991) that MBA students may be considered “semi-experts.”

The type of search strategy an individual utilizes can affect the prediction accuracy as
well. Hunton and McEwen (1997)12 utilized an experiment to investigate the search

strategy of financial analysts. They found that analysts generated more accurate forecasts

 

” The statistical probability that future earnings would be more than expected based on a sample of 385
firms including the sample firms used in the task. Participants were informed that the decision aid would
be correct 75% of the time.

'2 It is important to note that the subjects in the Hunton and McEwen experiment had qualitative
information in addition to the time series data. However, there findings are still likely to apply to pure
time-series forecasting.

13

when they used a directive search strategy and less accurate forecast when they used a
sequential forecast. Additionally, the authors examined if certain motivational incentives
can influence analysts’ forecast accuracy and lead to more optimistic estimates of
earnings. Analysts were assigned to either a control group (no relationship with firm for
which forecast is being made), firm following group (the analyst will be following the
company in the future) or firm underwriting group (the brokerage firm will be
underwriting a stock issue for the company. They found that analysts in both the firm
following group and the firm underwriting group had higher forecast error than those in
the control group.

Finally, increasing frequency of reporting will result in a larger number of breaks or
discontinuities or changes in direction within a given time period. O’Connor, Remus and
Griggs (1993) analyzed subjective predictions when using time series data that contained
five discontinuities and two levels of randomness. They find that people performed
significantly worse than statistical models in the presence of discontinuities. They also
found that people responded to random fluctuations in data as if they were valid signals.
It should be noted the participants were informed that the nature of the study was to
investigate people’s ability to recognize changes in time series data. This may have
caused participants to be hypervigilant towards the changes.

The above review has identified factors tint have been documented to affect the
accuracy of predictions and confidence intervals surrounding those predictions. Some of
the factors listed above that affect predictions will be controlled for in my study, while
some will be varied independently of reporting fiequency, and still some will explain the

expected effects of more frequent reporting. The following sections will more

14

specifically discuss and theorize how more frequent reporting will affect the accuracy of
predictions and confidence intervals surrounding those predictions.
Section 2. 4 Processing E fleets 01 More Frequent Reporting

"I Analysis of the time series behavior of earnings fer US. firms suggests that at least
some seasonality is present in nearly every major industry (Palepu, Bernard and Healy,
1996). Seasonality is a pattern that could be more difficult for nonprofessional investors
to detect, estimate, and use appropriately when more fi'equent reporting increases the
volume of data available to them. I therefore use earnings data with a strong seasonal
pattern, generated from the Foster model, which is commonly taught in financial
statement analysis classes as a relatively simple and effective earnings-prediction model
that captures these seasonal effects.13

EIQII=QH+6+,IQH-QI-5]

Where: Q, = Quarterly earnings in current period.
Q” = Quarterly earnings one quarter prior
,4 = Quarterly earnings 4 quarters prior (1 )
Q,-5 = Quarterly earnings 5 quarters prior
6 =the long run trend in year-to-year quarterly earnings increases.
a = year to year change in quarterly earnings experienced most recently

[Qt-l “ QM]-

Maines and Hand (1996) Show that nonprofessionals can subjectively estimate and
use seasonal and autoregressive components in quarterly earnings series. I therefore
expect that nonprofessionals will be able to use quarterly earnings data based on the
Foster model to predict future quarterly earnings, although it may be more difficult for

them to do so when they have more frequently reported earnings.

 

'3 While the Brown and Rozeff model is believed to more accurately model earnings behavior, it requires
iterative statistical techniques for parameter estimates that non-professional investors are not likely to be
able to reproduce without the aid of sophisticated statistical software. On the other hand, the Foster model,
which is a special case of the Brown and Rozeff model, can be easily estimated with a simple linear
regression, which is more likely to be approximated by subjective judgment.

15

Section 2.5 Model 01 Earnings Behavior

More frequent reporting will result in larger data set for a given time period, which I
expect will make pattern identification more difficult and lead to less accurate
predictions. Larger data sets can increase the cognitive load imposed on
nonprofessionals, thus increasing the use of simple heuristics such as representativeness
and recency. More frequent reporting also affects perceived volatility of earnings. More
fi'equent reporting can have the same effect as more fi'equent reviews of a portfolio,
which have been linked to perceived higher risk because people judge risk by the
fi'equency of negative returns (Gneezy and Potters, 1997). More frequently reported
earnings are likely to have more reversals (switches between positive and negative
earnings changes) within a given time period and so are likely to lmve more fiequent
negative changes which may have the same effect on risk perceptions as fi'equency of
negative returns. Because nonprofessional investors have been shown to erroneously use
the fiequency of past earnings reversals as an indication of the likelihood of firture
reversals (Bloomfield and Hales 2002) a large number of reversals could lead to a
perception of higher uncertainty.

Section 2.6 Eflects QfIncretfid F regyency on Uncertain_r_v

Previous experimental studies have documented that investors’ risk judgments
increase with the variability of earnings (Maines and McDaniel 2000; Lipe 1998). More
frequent earnings reporting can increase subjective uncertainty or perceived variability of
earnings in nonprofessional investors, because they will not aggregate data temporally to
the relevant time horizon and will use short-term fluctuations in earnings as a basis for

judging uncertainty at a longer time horizon. If, as is oflen the case, nonprofessionals are

16

buy-and-hold investors rather than frequent traders, their time horizons are relatively
long, but if more frequent reporting is available to them, they may focus on shorter-term
fluctuations, like investors who review their retirement portfolio quarterly or annually
although their time horizon is twenty years (Benartzi and Thaler 1995). In the
experiment, I ask individuals for predictions and confidence intervals for quarterly
earnings, regardless of whether they have weekly or quarterly earnings reports. The
relevant time horizon in this case is quarterly, and the volatility of quarterly earnings is
more relevant to individuals’ judgment than the volatility of weekly earnings. It is
likely, however, that at least some individuals will not spontaneously aggregate weekly
data into quarterly data, and will use weekly fluctuations to judge the uncertainty of
quarterly earnings.

Using asset-return data, Gneezy and Potters (1997) demonstrate that more frequent
evaluation of investments increases their perceived risk because people judge risk by the
frequency of negative returns. (More frequent negative returns can be Observed over a
given period if people examine monthly returns, for example, rather than annual returns.)
Similarly, Thaler, Tversky, Kahneman and Schwartz (1997) demonstrated that as the
frequency of evaluation increases, investors chose to invest less in risky assets. Like
returns, more frequently reported earnings will include more fluctuations (switches
between negative and positive changes), and thus more negative earnings changes. The
frequency of such changes could lead nonprofessional investors to perceive more
fi'equently reported earnings as more volatile, leading to higher uncertainty about the

future earnings. More fiequent reporting is likely to have both frequent negative changes

17

and more fluctuations in the data, which will lead nonprofessional investors to state wider
confidence intervals around their quarterly earnings predictions.
Hypothesis 1: Nonprofessional investors who receive more frequent
earnings reports will exhibit wider confidence intervals

around their quarterly earnings predictions than those who
receive Iessfi'equent earnings reports.

Section 2. 7 Eflects 01 Increased Frequengy on Accuragy

More fiequent reporting can also affect the accuracy of nonprofessional investors’
earnings predictions because of the increase in noise associated with more temporally
disaggregated data. This can easily be illustrated visually. Consider, for example, the

identical seasonal data presented in both weekly and quarterly format (Figure 1 and 2
respectively).

Figure 1 More Frguent Refining-Seasonal Data

fir

Earnings Weekly Earnings

 

 

 

 

 

 

’

1 10 19 28 37 46 55 64 73 82 91 100109118127136145154
Time Period-Weeks i

 

 

The weekly fluctuations in the data in Figure 1 can obscure the 4th quarter earnings
increase that appears more clearly in Figure 2. If the seasonal (or other) pattern is

obscured by weekly fluctuations, less accurate predictions can be expected.

Figure 2 Less Frequent Reportinfleasonal Data

 

 

 

 

 

 

 

 

 

 

Earnings Quarterly Earnings

52.50 T”..-— W...” , - ___---.__ _____~.__--

52.00 b ew/

51.50 W

$1.00

50.50 L

50.00 . 4 e . . n—— . . r r . 5
1 2 3 4 5 6 7 8 9 10 11 12

Time Period - Quarter

 

 

 

More frequent reporting (i.e., Figure 1) could also lead to large cognitive demands,
resulting in the use of heuristics and predictable biases. Research suggests that
information overload can occur with as little as seven plus or minus two items (Miller
1956). When nonprofessional investors process large quantities of data, they are not
likely to focus consistently on the most relevant subset of the data (Bouvvman 1982).
Faced with high levels of information load, nonprofessional investors are likely to resort
to heuristics to aid in the decision making process. One possibility is that they will
reduce their information load by using only the most recent data as a basis for their
predictions. Recency effects (Tversky and Kahneman 1974) represent an application of
the representativeness heuristic, because individuals believe the most recent data best
represents current characteristics of the earnings series. Hunton and McEwen (1997) find
that recency can partially explain less accurate earnings forecasts by analysts.

The effects of recency will be different in each reporting frequency condition,
however. Consider, for example, a scenario in which nonprofessional investors are given
three years of reported earnings. Assume that nonprofessional investors can only

cognitively process 12 data points. In the case of quarterly reporting, 12 data points

19

provide three years of data, which gives nonprofessional investors an opportunity to
identify and use the seasonal preperties of earnings. In the weekly reporting condition, in
contrast, the last 12 data points are insufficient as a basis for estimating seasonal
properties of earnings. Consequently, recency effects will likely lead to less accurate
predictions in the more frequent reporting condition than in the less frequent reporting
condition.

Hypothesis 2: Nonprofessional investors ’ absolute prediction error will
be greater in the more frequent reporting condition than in
the less frequent reporting condition.

Section 2.8 Ewtiects QfIncrecged F regyency on Dispersions

Dispersion of expectations about earnings is important in practice because it can
lead to increased trading volume. The error predicted in H2 includes both dispersion and
bias; H3 focuses on dispersion alone.

Earnings predictions are likely to be dispersed even when individuals’ subjective
predictions are based on the same data, because ability to manage increased cognitive
load and strategies for doing so are likely to differ across individual nonprofessional
investors. Individual differences are expected to have larger effects with more frequent
reporting. When reporting is less fi'equent, fewer reports are generated over a given
period of time. Most nonprofessional investors can then use a large enough data set (e. g.
12 quarterly reports over three years) to identify important patterns in the data, resulting
in low dispersion of predictions. Three years of weekly data, however, include 156

earnings reports. Many nonprofessional investors will not use all of this data, and they

20

will differ with respect to how much of it they use and what conclusions they draw from
it.

People often tend to see trends or streaks in truly random data series (Andreassen,
1987, 1990; Bloomfield et al., 2001; Lim and O’Connor, 1996 and O’Connor et al.,
1993). Nonprofessional investors have also been demonstrated to have difficulty in
identifying patterns in nonrandom data series. For example, Maines and Hand (1996) find
that nonprofessional investors incorrectly detected autoregressive components in
quarterly seasonal random walk data. The larger number of data points and more
frequent changes of direction in the more frequent reporting condition offer opportunities
for individuals to see a greater variety of nonexistent patterns, as well as obscuring actual
patterns in the data. Because of individual differences in cognitive processing, more
dispersion in predictions is likely to result in the more fiequent reporting condition.

Hypothesis 3: Variance of predictions will be larger in the more

frequent reporting condition than in the less fiequent
reporting condition.

Section 2. 9 Joint Eflects at More and Less Frequent Remrting

Hypotheses 1-3 predict that if nonprofessional investors examine only weekly
earnings reports, their predictions will be less accurate, more uncertain and more
dispersed than if they examine only quarterly reports. The question remains whether or
not these judgment effects will persist when both weekly and quarterly reports are
presented jointly. Online earnings reporting could offer investors their choice of
reporting intervals (so that investors could choose to see only quarterly or only weekly

earnings), or it could always present quarterly and weekly earnings together. In order to

21

investigate the effects of these alternatives, I examine how joint presentation of weekly
and quarterly reporting affects investor judgments.

Some prior research suggests reasons for believing that predictions based on joint
presentation of weekly and quarterly earnings reports will be similar to predictions based
on quarterly reports only. Other research, however, suggests reasons for believing that
predictions based on joint presentation will be similar to predictions based on weekly
reports only. It is not clear ex ante which of these sets of reasons will dominate in
individuals’ earnings-prediction behavior. I therefore summarize arguments on both
sides and state research questions rather than directional hypotheses about the effects of
joint presentation.

Nonprofessional investors’ judgments in the joint reporting condition could resemble
those of the less fiequent reporting condition. Because of the large cognitive load
presented by more frequently reported data, nonprofessional investors could voluntarily
choose to ignore the larger data set and focus on the smaller data» set when it appears to
provide sufficient information for the earnings prediction task. In this case,
nonprofessional investors in the joint reporting condition are likely to make judgments
very similar to those in the less frequent condition.

On the other hand, judgments in the joint reporting condition could resemble those in
the more frequent reporting condition. If nonprofessional investors believe that more
frequent reporting is more informative and usefirl (Hunton et al. 2003), they will focus on
it more strongly when making their judgments. Another possibility is that

nonprofessional investors will have difficulty ignoring the more frequent earnings reports

22

even if they focus primarily on the less frequent earnings reports (i.e., the curse of

knowledge, Camerer 1989; Fischhoff 1977).

These arguments lead to the following set of research questions:

R1:

How will the provision of both more and less frequent earnings reports
aflect the confidence intervals around nonprofessional investors ’
quarterly earnings prediction, compared to the provision of either more or
less frequent reports?

How will the provision of both more and less frequent earnings reports
aflect the accuracy of quarterly earnings prediction of nonprofessional
investors, compared to the provision of either more or less frequent
reports?

How will the provision of both more and Iessfi‘equent earnings reports
affect the dispersion around quarterly earnings prediction of
nonprofessional investors, compared to the provision of either more or
less frequent reports?

23

3.0 RESEARCH DESIGN
Section 3.1 Experimental Design and Dependent Variables

The research design is a 2 (reporting frequency: quarterly or weekly) x 2 (earnings
series created with two different random error draws) x 2 (direction of change: increases
or decreases between actual earnings in the last reported period and expected earnings as
predicted by the Foster model in the forecasted period) x 2 (tasks: in the first task,
participants predicted quarterly earnings based on either quarterly or weekly data and in
the second task, participants made decisions using both quarterly and weekly data). See
Section 3.2 for an explanation of the random draw and direction variables. The first three
variables are manipulated between subjects and the last variable is manipulated within
subjects.

The dependent variables examined were 1) uncertainty, as measured by participants’
confidence intervals, 2) accuracy of predictions of next quarter’s earnings, measured by
the absolute difference between the prediction and earnings forecasted by the model, and
3) dispersion of predictions, measured by the variance of participants’ predictions.

Hypotheses 1-3 were tested by comparing judgment performance in the first task
across frequency conditions. Research questions 1-3 were tested by comparing judgment
performance in both tasks.

Section 3.2 Independent Variables and Data Creation

 

Participants’ task was to predict quarterly earnings per share based on historical data.
In the more frequent reporting condition, participants received weekly historical earnings
per share data. In the less frequent reporting condition, participants received quarterly

historical earnings per share data.

24

The Foster model was used to create earnings data with strong fourth quarter
seasonality. Initial quarterly earnings were those of Compaq Computer Company for the
year 1993, which a widely used textbook employs as an example of earnings that can be
predicted with the Foster model (Palepu, Bernard & Healy, 1996). The expected earnings
in the subsequent quarter were generated using Compaq’s values for the phi (p)
parameter of .49 and the delta (6) parameter of .09. In order to create weekly data the
first 5 quarters generated fi'om the initial process were selected. Each quarter was divided
into 13 equal weeks and the original Foster model was adapted to create weekly data,
resulting in the following equation:

EIWII= Wu4+6+¢qu.,- Wit-5] (2)
Where: W, = Weekly earnings in current week.
Wu.) = The earnings in week i one quarter prior
W“ = The earnings in week i 4 quarters prior
Wu.5= The earnings in week i 5 quarters prior
6.. = Year-to-year weekly earnings increases.
ow = A fraction of the year to year increase in weekly earnings
experienced most recently[W .1- 1 - WM]
The Foster model adjusted for weekly calculations was then used to generate weekly
earnings per share. A small amount of random error N(0, .01) was added to each weekly
data point. The final weekly instrument consisted of 156 weeks presented in 13 week
quarters. Quarterly data and instruments were created by summing each 13 weeks of data
and presenting only quarterly totals.

The first earnings series was created as described above. The second was created by

using a different 156-week period from data generated by the Foster model” with a

different direction of change between the last reported quarter’s earnings and the earnings

predicted for the next quarter by the Foster model. In one series (forecast-period

 

" The parameters were identical in both data series.

25

increase), the next quarter’s expected earnings were higher than the last reported
quarter’s earnings, while in the other series (forecasted period decrease), they were lower.
Two versions of the increase series and two versions of the decrease series were created
by adding two different sets of random error draws from the same N (0, .01) distribution
to the weekly expected earnings generated fi‘om the weekly Foster model. The direction
of prediction and the random error draw were manipulated to make sure that results were
not unique to a single data series.

In the first and second task, individuals received earnings series that differed with
respect to both direction and random error draw. For example, if they received an
increase series with random error draw 1 in the first task, they received a decrease series
with random error draw 2 in the second task. If they had received identical or very
similar data sets in both tasks, demand effects might have driven them to make identical
judgments in the two tasks.

The weekly data used in this experiment were constructed to ensure that real
informtion content was constant across weekly and quarterly conditions and any
difference between experimental conditions was due to differences in subjective
processing. If actual weekly data had been less informative than quarterly data, this
would add to the negative effects of weekly reporting. If, on the other hand, weekly data
had been more informative than quarterly data, this would presumably offset, but not

eliminate the negative effects shown in this study.ls

 

'5 As a result of holding real information content constant, the weekly data have visible jumps at quarter
end, potentially making the underlying quarterly seasonal pattern salient. However, this aspect of the data
would bias against finding the predicted results.

26

Section 3.3 Efixperimental Materials and Procedrm

Participants received a packet along with an instruction sheet. Each packet contained a
consent form, case materials, and a post questionnaire. The case materials informed
participants that the most recent quarter end for the target firm was December 31, 2003.
Participants were asked to predict earnings for the first quarter of 2004. In the more
frequent reporting condition, a special caution was added to ensure that subjects gave a
quarterly and not a weekly prediction. Participants received earnings data in both tabular
and graph form. In addition, a disk containing the tabular data was also given to the
participants to aid in calculations if they desired. Experimental sessions took place in a
computer lab, and all participants had access to spreadsheets. After making their
predictions, participants were asked to generate the equivalent of 90% confidence
intervals. See Appendix 1 for examples of instruments.

At the conclusion of the prediction task, a post-experiment questionnaire was
administered. Questions were included on participants’ (1) stock market experience, (2)
finance and statistical knowledge, (3) risk attitude, (4) judgments about the perceived
volatility of the data, and (5) reports of the level of importance and influence of recent
time periods in their judgments. The latter allowed for the identification of recency
effects. Questions targeted at assessing the participants’ comprehension of range and
variance was also included in the debriefing questionnaire. Finally participants were
required to describe in detail how they generated their prediction. Participants earned
class credit for participation in the experiment. In addition, participants with predictions

within $.04 of the actual earnings per share earned an additional $10.

27

Section 3. 4 Participants

Participants in the experiment were 84 first and second year MBA students as well as
masters students fi'om a large midwestem university. Results from 12 participants were
excluded because they generated weekly earnings predictions instead of quarterly
earnings predictions, resulting in 72 participants being used in the analysis. Participants
were recruited from a MBA financial statement analysis classes. 22 (31%) of the
participants indicated they had investing experience, which ranged from 1 month to 84
months. The average number of finance courses taken was three and the average number
of statistics courses taken by participants was two. The experiment was administered in

two sessions over the course of one day (morning and evening).

28

4.0 RESULTS: HYPOTHESES 1-3: EFFECTS OF MORE FREQUENT REPORTING

Hypotheses 1-3 test the between-subjects effects of more frequent reporting on
uncertainty, absolute prediction error and dispersion of predictions. Therefore, only the
data from the first task are used for analysis. Table 1 reports the correlations between
reporting frequency, accuracy, and confidence intervals for the first task. The correlation
between absolute prediction error and reporting frequency was marginally significant (r =
.21 , p < .08). In addition, there was a significant correlation between absolute prediction
error and uncertainty (r = .27, p < .02). ANOVAs used to test Hland H2 include
additional independent variables to test for the effects of different data sets participants
received (random error draw and direction of prediction) as well as reporting frequency. I
also tested for effects of prior investing experience and/or effects of session and include
these variables in the reported AN OVAs when they had significant main or interaction
effects (p < .05).

Analysis 01H]

H1 predicts that uncertainty, measured by the width of the confidence interval, will be
greater in the more frequent reporting condition. Cell means and standard deviations are
shown in Table 2. Task two will be analyzed in section 4.1. The mean confidence
intervals are .57 for both the less and more frequent reporting conditions. Results of the
ANOVA for task one,16 shown in Table 3, indicate no main effect of frequency (F = .09,
p < .76). There were no other main effects or interactions present (all F’s < 2.44, all p’s >

.13).

 

'6 Investing experience (in months) is used as a covariate.

29

Analysis ofHZ

H2 states that nonprofessional investors’ earnings predictions will be less accurate, as
measured by their absolute prediction errors, in the more frequent reporting condition
than in the less fiequent reporting condition. Table 4 reports the cell means and standard
deviations. As expected, the absolute prediction error was larger (mean = .14) with more
fiequent reporting than with less frequent reporting (mean = .09). This difference was
significant, as shown by the main effect of fiequency in the AN OVA reported in Table 5.
(F = 4.04, p < .05). There were also main effects of direction (F = 15.83, p < .00) and
random error draw (F = 22.16, p < .00). However, there were no interactions involving
these factors and frequency (all F’s < 1.54, p’s > .22). These results provide support for
H2.

Although the main effects of experience and the frequency ‘ experience interaction
are not significant, 1 compared the prediction errors of the more and less experienced
participants to allay concerns that the effect of fi'equency might be largely dependent on
less experienced participants. Qualitatively, the analysis reveals that the difference
between the fiequency groups was greater for those participants who indicated they had
investing experience. Here, experience is treated as a categorical, 0 — 1 variable where 1
represents previous investing experience and 0 otherwise. Previously, it was presented as
a continuous variable (# of months). For experienced participants the mean was .08 in
the less frequent reporting condition and .14 in the more frequent condition. For
inexperienced participants the mean was .10 in the less frequent reporting condition and
.14 in the more frequent condition. Thus the results cannot be explained a lack of

investing experience

30

Absolute prediction error can include both random error and systematic error (signed
error or bias). Mean signed error was .01 and not significantly different from zero (t =
.77, p < .44). Analysis of signed prediction errors (Appendix —1 Table 10) revealed a
main effect of random error draw (F = 14.75, p < .00) but no main effect of frequency (F
= .00, p < .95) and no significant interaction between fi'equency and random error draw
(F = .38, p < .54). Bias is significantly greater than fiom 0 (t = 4.82, p < .00) with one
random error draw, though not significantly less than 0 with the other (t = -.87, p < .39).

Analysis of H3

H3 states that the dispersion of predictions will be larger in the more frequent
reporting condition than in the less firequent reporting condition. Table 6 reports the
variances of participants’ earnings forecasts and results of the F-test for equal variances.
The variance for the more fiequent reporting condition was 0.016 and the variance for the
less fi'equent reporting condition was 0.006 (F = 2.56, p <.00). These results support H3.

Disc;u_ssion of Results/ram H1-3

Results of H1 were inconsistent with the theoretical discussion in section 2. The
participants were not affected by the fi'equency of fluctuations (switches between
negative and positive changes) brought on by more frequent reporting. One possible
explanation for the lack of support of H1 is that participants’ perception of volatility was
not affected by more fi'equent reporting. In the more frequent reporting group,
participants had a perceived volatility rating of 4.05 while those in the less frequent
reporting condition had a rating of 4.54 (F = 1.25, p < .27). The results are inconsistent
with the findings of Gneezy and Potters (1997) that suggest perceived risk should

increase with more frequent reporting. Another possible explanation of the lack of

3]

support for H1 could be that participants did not perceive the task to be more difficult
using more frequently reported data. In the more frequent reporting condition
participants had a perceived difficulty rating of 3.73, while those in the less fiequent
reporting had a rating of 3.33 (F = .57, p < .45). If participants had perceived the
volatility to be greater and perceived the task to be more difficult in the more frequent
reporting condition, they would have generated wider confidence intervals indicating
more uncertainty.

Results of a correlation analysis (Appendix 1 - Table 11) also confirm the results
obtained from hypothesis testing. While there was a significant positive correlation
between perceived difficulty and perceived volatility (.24, p < .04), there were no
significant correlations between absolute prediction error and perceived volatility (-.04, p
< .72) or absolute prediction error and perceived difficulty (-.04, p < .73). Similarly,
there were no significant correlations between confidence intervals and perceived
volatility (.07, p < .54) or confidence intervals and perceived difficulty (.03, p < .82).

Consistent with my predictions, more frequent reporting resulted in larger prediction
errors and greater dispersion of predictions. Theoretical discussion in section 2 suggests
that more frequent reporting could obscure the underlying data pattern leading to greater
prediction errors and dispersion in predictions. Results of the post-experiment
questionnaire revealed that an equal proportion (80%) of participants across both
fiequency reporting groups expected earnings per share data to be seasonal. However, 28
of 31 (90%) participants correctly identified the data pattern as seasonal in the less
fi'equent reporting condition, while 31 of 41 (75%) of participants correctly identified the

data pattern as seasonal in the more fi'equent reporting condition. Thus it appears,

32

qualitatively, that more frequent reporting obscured the underlying data pattern making it
more difficult for investors to make accurate predictions.

Although participants were less accurate in the more frequent reporting condition,
they were not less confident, as measured by their confidence intervals. These results are
counterintuitive since one would expect more uncertainty (wider confidence intervals)
when earnings predictions were less accurate. One possible explanation could be the lack
of difference in perceived difficulty across conditions.

Although I did not hypothesize effects of direction of prediction, AN OVAs revealed
main effects of direction of prediction on prediction accuracy. Specifically, I find that
when the prediction period is positive (negative), the mean of the less frequent reporting
condition was .13 (.06) and the mean of the more frequent reporting condition was .18
(.10). This qualitatively suggests that participants performed worse when the prediction
period was positive.

ANOVA results also indicated main effects of random error. For random error draw
one (random error draw two), the mean of the less fiequent reporting group was .14 (.05)
and the mean for the more frequent reporting group was .19 (.08). This data qualitatively
suggests that participants performed worse with one random error draw than with the
other. These results are surprising since statistically, both random error draws had the
same statistical properties. Future research should investigate these phenomena further.

Section 4.1 Results: Research Questions 1-3: Eflects 01 Joint Presentation

Tests of the three research questions use the dependent variables from both tasks (i.e.,

 

prediction with single fi‘equency and joint frequency reporting). Table 2 reports cell

means and standard deviations of confidence intervals on the first and second tasks.

33

Table 7 shows the results of the repeated measure AN OVA with confidence interval as
the dependent variable. Between-subject variables are the same as in previous ANOVAs.
There were no significant main effect or interaction affects of session and experience,
which are therefore omitted fi'om the analysis shown in Table 7. Although the cell means
in Table 2 show that adding weekly reporting (cell 2) to quarterly reporting (cell 1)
appears to increase uncertainty, and adding quarterly (cell 4) to weekly reporting (cell 3)
appears to decrease uncertainty, these effects are not significant, as shown by the task‘.
frequency interaction (F = .69, p < .41). Additional analysis reveals no effects of
reporting frequency on uncertainty. '7

Tables 4 and 8 report the cell means and results of the repeated-measures AN OVA
with prediction error as the dependent variable. Analysis revealed no main or interaction
effects of session, thus session was omitted from the analysis. Examination of the
task‘f’requency interaction in Table 8 allows me to determine if providing both quarterly
and weekly data reduces or eliminates the differences in prediction errors shown in the
first task. There is significantly less difference in accuracy between the two frequency
groups in the second task than the first: (cell 1- cell 3) > (cell 2 — cell 4), (F = 4.76, p <
.03). Simple effects tests reveal that the addition of weekly data to quarterly data (cell 2
— cell 1) increases the prediction error, but the increase is not significant (F= 1.15, p <
.30). The addition of quarterly data to weekly data (cell 4 - cell 3) significantly reduces

prediction error (F = 4.51, p < .04).

 

l7'l‘able 7 also shows a task‘direction interaction and a task by random error interaction. Additional
analysis (Appendix 1 — Tables 12-16 reveals that there was a marginally significant difference between task
1 (.60) and task 2 (.54) when the expected change in direction from the current period to the forecasted
period was positive (F = 2.92, p < .10). Alternatively when the expected change in direction from the
current period to the forecasted period was negative, the difference between task 1 (.53) and task 2 (.57)
was not significant (F = .90, p < .35). Furthermore, for one random error draw, the difference between task
1 (.61) and task 2 (.44) was significant (F = 15.84, p < .00). For the other random draw, difference between
task 1 (.52) and task 2 (.66) was significant (F = 6.64, p < .01).

34

I also compare the means of cell 1 and cell 4, to investigate whether adding quarterly
to weekly data (cell 4) results in predictions that are as accurate as those based on
quarterly data only (cell 1). The absolute mean prediction error was in cell 4 was .11,
while the absolute mean prediction error in cell 1 was .09, (F = 1.30, p < .26). Finally, I
examine means of cell 2 and cell 4 to investigate if participants weighted the most
recently received data more heavily than the first data set. The mean absolute prediction
error for those that received less frequent reporting data in the first task was .17 while the
mean for those that received more frequent reporting data in the first task was .11. This
difference was not significant (F = 1.22, p < .27).”

I also conducted a similar analysis (Appendix 1 - Tables 17-19) using signed
prediction error as the dependent variable. There was no significant interaction between
task and frequency of reporting (F = .09, p < .76). Additional analysis shows that the
mean signed prediction error in the first task for those who received the less frequent
reporting data was -.004 while for these participants in the second task (joint
presentation), it was -.02 (F = .11, p < .74). The mean signed prediction error for those
who received more fi'equent reporting data in the first task was .03 while for those in the

second task (joint presentation), it was .01 (F = .78, p < .38).

 

" Table 8 reports a significant task ‘ direction interaction. Additional analysis (Appendix 1 — Tables 20-29)
reveals that there was a significant difference between task 1 (.16) and task 2 (.10) when the expected
change in direction Earn the current period to the forecasted period was positive (F = 7.83, p < .01).
Alternatively when the expected change in direction from the current period to the forecasted period was
negative, the difference between task 1 (.08) and task 2 (.17) was significant (F = 12.38, p < .00). Table 8
also reports a significant task * error interaction. For one random error draw, the difference between task 1
(.17) and task 2 (.08) was significant (F = 18.29, p < .00). For the other random draw, difference between
task 1 (.07) and task 2 (.19) was significant (F =17.25, p < .00). Finally, Table 8 reports a significant
task’experience interaction. There is no significant difference between task 1 (.11) and task 2 (.13) when
participants indicated they had prior investing experience (F = 1.89, p < .19). When participants indicated
they had prior investing experience, the difference between task 1 (.12) and task 2 (.14) was not significant
at the conventional level (F = 2.69, p < .11).

35

Analysis of the final dependent variable, prediction dispersion (Table 6), reveals that
the addition of weekly data (cell 2) to quarterly data (cell 1) results in a significant
increase in prediction dispersion (F = 4.73, p < .00). Referring to Table 6 it can be seen
tlmt the addition of quarterly data (cell 4) to weekly data (cell 3) results in a decrease in
dispersion (F = 1.87, p < .03). I also compare cell 2 and cell 4. The variance for those
participants that received less frequent reporting in the first task was .17 while the mean
for those participants that received more fi‘equent reporting in the first task was .09 (F =
3.27, p < .00). These results can suggest that participants weighted the most recently
received data set more heavily. Comparison of means from cell 1 and cell 4 show that
nonprofessional investors that use both weekly and quarterly data generate predictions

with dispersion equal to those that only receive quarterly data (F = 1.37, p < .19).

36

5.0 CONCLUSIONS

This study investigates the effects of more frequent reporting on nonprofessional
investors’ earnings predictions. As hypothesized, I find that more fi'equent reporting
of earnings per share results in less accurate and more dispersed predictions of
quarterly earnings per share. I hypothesize but do not find that more frequent
reporting significantly increases confidence interval widths among nonprofessional
investors.

I find that the addition of weekly data to quarterly data increases prediction error
by 89%, but the effect is not significant. Conversely, I find that the addition of
quarterly data to weekly data significantly reduces prediction errors. Nonprofessional
investors who receive both weekly and quarterly data generate predictions of
quarterly earnings comparable to those who receive only quarterly data. The addition
of weekly data to quarterly significantly increases dispersion of predictions.
Additionally, I find that the addition of quarterly data to weekly data results in a
significant decrease in the dispersion of prediction. 1 find no such effects on
confidence intervals.

The results of this study have several implications for regulators to consider
regarding the adoption of increased frequency of reporting. Should more fi'equent
reporting be adopted, it appears that joint presentation could mitigate some but not all
of the negative effects created by sole dependence on more frequently reported
information.

This study is subject to several limitations. First, 1 limited the amount of

information participants received. Typically, when investors are evaluating the

37

frnancial performance of a firm, more information is available and the information
environment is more complex. However, reducing the complexity allows me to make
stronger inferences about the effects of more frequent reporting. Second, this
experiment uses MBA students as surrogates for nonprofessional investors. It is
likely that they have less investment experience than many nonprofessional investors
and therefore their actions may not accurately reflect the opinions of actual investors.
Finally, this experiment uses earnings data that contains fourth quarter seasonality
which may not be applicable to all firms. Future research should investigate the
efl’ects of more frequent reporting when earnings data has different properties (i.e.

random walk).

38

 

Table l-Descriptive Data Correlations

 

 

Frequency Absolute Confidence
Prediction Error Interval
Frequency 1 - -
Absolute .21" 1 -
Prediction Error
Confidence .00 .27“ 1
Interval

 

 

‘Correlation is significant at the 0.05 level (2-tailed)
*Correlation is significant at the 0.10 level (2-tailed)

Frequency = The frequency which the historical data is presented (high or low).
Accuracy = The absolute prediction error (difference between earnings predicted by the
participant and that generated by the Foster model.

Confidence interval = The uncertainty surrounding the prediction generated by
participants, measured by the difference between the upper and lower bounds.

39

Table 2-Cell Means and Standard Deviations for Confidence Intervals in Task 1 and 2

 

 

 

 

 

 

 

Reporting Task 1 Task 2
Frequency in Single Joint
1" Task Frequency Frequency
Less Cell 1 Cell 2
Frequent
Mean .57 .61
(s.d.) (.49) (.44)
[II] [311 [31]
More Cell 3 Cell 4
Frequent
Mean .57 .52
(s.d.) (.32) (.32)
[n] [41] [41]

 

 

Planned Pairwise Comparisons for Research Question 1
Dependent Variable = Confidence intervals

Less fiequent reporting in task 1 = Less frequent reporting in task 2.
Cell 2 = Cell 1, (F= .06, p < .81)

More fi'equent reporting in task 1 = More frequent reporting in task 2.
Cell 3 = Cell 4, (F = .87, p < .36)

Less fi'equent reporting = both more and less fiequent reporting.
Cell 1 = Cell 4, (F = .25, p < .62)

Less fiequent reporting received 1" = more frequent reporting received 1".
Cell 2 = Cell 4, (F = 1.27, p < .26)

4O

Table 3-Analysis of Effects of Reporting Frequency

on Confidence Intervals

Between-Subjects ANOVA with Confidence Interval as DV

 

Sauce df F-value p-vahrg
Corrected Model 25 1.51 .11
Intercept 1 33.04 .00
Frequency 1 .09 .76
Direction 1 .67 .42
Session 1 .02 .90
Rand.Error 1 1 .02 .32
Tirnelnvesting 1 .37 .55
F requency‘Direction 1 I .69 .20
Frequency‘Session 1 .00 .98
Frequency‘RandError 1 .16 .70
Frequency‘TimeInvesting 1 .42 .52
Direction“ Session 1 l .79 .1 9
Directin‘Error 1 .68 .41
Direction‘TimeInvesting 1 .35 .56
Session‘Error 1 .22 .64
Session‘TimeInvesting l 2.44 .13
Rand.Error*TimeInvesting 1 .75 .39
Frequency‘Direction“ Session 1 .81 .38
F requency‘Direction‘RandError l .1 l .75
Frequency‘Direction‘TimeInvesting l .40 .53
Frequency" Session’RandError 1 .12 .73
Frequency" Session*TimeInvesting 1 2.1 1 .1 5
Frequency’Error‘TimeInvesting 1 .01 .93
Direction*Session*Rand.Error 1 1.09 .30
Direction‘Session‘TimeInvesting I 2.08 .16
Direction‘Rand.Error*Timelnvesting 1 .04 .85
Session‘RandError‘TimeInvesting 1 1.42 .24
Error 46

Total 72

Corrected Total 71

Frequency = frequency conditions between subjects; Direction = direction of change-
upward or downward-between earnings in the last reported period and the forecasted
period; Session = morning or evening session of experiment; Rand.Error = two different
random error draws used to create earnings series: Tirnelnvesting = Amount of investing
experience in months.

41

Table 4-Cell means and Standard Deviations for
Absolute Prediction Errors in Task 1 and Task 2

 

 

 

 

 

 

Reporting Task 1 Task 2
Frequency in Single Joint
1" Task Frequency Frequency
Less Cell 1 Cell 2
Frequent
Mean .09 .17
(s.d.) (.08) (.17)
In] [31] [31]
More Cell 3 Cell 4
Frequent
Mean .14 .1 1
(s.d.) (.13) (.09)
[n] [41] ml

 

 

 

 

Planned Pairwise Comparisons For Research Question 2
Dependent Variable = Absolute Prediction Errors

Less frequent reporting in task 1 = both more and less fi'equent reporting in task 2.
Cell 2 = Cell l,(F=1.15, p < .30)

More frequent reporting in task 1 = both more and less frequent reporting in task 2.
Cell 3 = Cell 4, (F = 4.51, p < .04)

Less frequent reporting = both more and less frequent reporting.
Cell 1 = Cell 4, (F = 1.30, p < .26)

Both more and less frequent reporting received 2“‘l task = both more and less frequent
reporting received 2"" task. Cell 2 = Cell 4 (F = 1.22, p < .27)

42

Table 5-Analysis of Effects of Reporting Frequency

on Absolute Prediction Error

Between-Subjects ANOVA with Absolute Prediction Error as DV

 

Moe df F-value gm;
Corrected Model 15 3.11 .00
Intercept 1 100.89 .00
Frequency 1 4.04 .05
Direction 1 1 5.83 .00
Experience 1 1 .99 . 1 6
Rand.Error 1 25.16 .00
Frequency‘Direction 1 .20 .66
F requency‘Experience 1 1 .02 .32
Direction‘Experience 1 1 .04 .3 l
Frequency“Direction‘Experience 1 .37 .54
Frequency‘RandError 1 1.54 .22
Direction’RandError 1 3.95 .05
Frequency*Direction‘Rand.Error 1 .00 .97
Experience*Rand.Error 1 2.68 .1 1
F requency“ Experience*Rand.Error 1 1 .98 . 1 6
Dhection‘Experience‘RandError 1 .83 .37
Frequency’Direction“Experience“Rand.Error 1 .56 .46
Error 56

Total 72

Corrected Total 7]

Descriptions same as Table 3

43

Table 6—Cell Variances for Task 1 and Task 2

 

 

 

 

 

 

 

Reporting Task 1 Task 2
Frequency in Single Joint
1’t Task Frequency Frequency
Less Cell 1 ‘ Cell 2
Frequent
Variance .006 .028
In] [31] [31]
More Cell 3 Cell 4
Frequent
Variance .016 .008
[ll [41] [41]

 

‘

 

Result of' Between-Subjects Hypotheses Test-H3
Less Frequent reporting in task 1 < More frequent reporting in task 2
(F = 2.56, p < .00)

Planned Pairwise Comparisons For of Research Question 3
Dependent Variable = Variance

Less frequent reporting in task 1 > Less fi'equent reporting in task 2.
Cell 2 = Cell 1, (F = 4.73, p < .00)

More frequent reporting in task 1 > More frequent reporting in task 2.
Cell 3 = Cell 4, (F = 1.87, p < .03)

Less frequent reporting = both more and less frequent reporting.
Cell 1 = Cell 4, (F= 1.37, p < .19)

Less frequent reporting received 1" = more fiequent reporting received 1".
Cell 2 = Cell 4 (F = 3.27, p < .00)

Table 7-Analysis of Effects of Frequency on Confidence Intervals

 

 

Panel A

Repeated Measure ANOVA with Confidence Interval as DV in Task 1 and Task 2
§_o_urce df F-vzdue p-value
Intercept 1 1 69. 16 .00
Frequency 1 .357 .55
Direction 1 .273 .60
Rand.Error 1 .431 .51
Frequency*Direction 1 2.05 .16
Frequency’RandError 1 .243 .62
Direction’RandError l 1 .27 .26
Frequency‘Direction‘RandError 1 .01 .92
Task 1 .28 .60
Task*Frequency 1 .69 .41
Task‘Direction 1 3.45 .04
Task‘Rand. Error 1 19.53 .00
Task*Frequency"Direction 1 3.32 .07
Task‘Frequency*Rand.Error 1 1.32 .26
Task‘Direction*Error 1 .62 .43
Task’Frequency"Direction‘RandError 1 .08 .78
Error 64

Task = 2 tasks consisting of 2 predictions (confidence intervals).

Freq = Frequency conditions between subjects: more and less reporting in forecasting
tasks; Direction = Direction of change-upward or downward-between earnings in the last
reported period and the forecasted period; Rand-Error = Two different random error
draws used to create earnings series.

45

Table 8-Analysis of Effects of Frequency on Absolute Prediction Error“

 

Panel A

Repeated Measure ANOVA with Absolute Prediction Error in Task 1 and Task 2
_S_QL_II‘CC df F-vaflue p—value
Interecept 1 143.36 .00
Frequency 1 .16 .69
Direction 1 .53 .47
Error 1 .56 .46
Experience I .04 .84
Frequency‘Direction 1 .33 .57
Frequency‘RandError 1 3.58 .06
Direction*Rand.Error 1 3.98 .06
Frequency‘Direction*Rand.Error 1 .44 .51
Frequency‘Experience 1 1 .83 . I 8
Direction’Experience l .02 .88
Frequency‘Direction”Experience 1 .34 .56
Rand.Error‘Experience 1 3.49 .07
Frequency‘RandError‘Experience 1 .49 .49
Direction’RandError‘Experience 1 .16 .69
Freqency*Direction"‘Rand.Error*Experience l .24 .63
Task 1 .16 .69
Task‘Frequency 1 4.76 .03
Task’Direction 1 19.32 .00
Task’RandError 1 32.67 .00
Task‘Experience 1 4. 1 8 .05
Task‘Experience‘Direction 1 1 .40 .24
Task‘Frequency" Rand.Error 1 .15 .70
Task‘Direction‘RandError 1 .22 .64
Task"Frequency‘Direction‘RandError 1 .42 .52
Task‘Frequency‘Experience l .01 .91
Task‘Direction*Experience 1 2.20 .14
Task‘Frequency‘Direction"Experience 1 .03 .86
Task*Rand.Error‘Experience 1 .02 .88
Task‘Frequency‘RandErroflExerience l 1.19 .28
Task‘Direction‘RandError“Experience 1 .57 .45
Task‘Frequency‘Direction" Rand.Error*Experience 1 2.20 . 14

Error

56

‘Description of variables is the same as in Table 7

46

1"" .

APPENDIX 1

Appendix 1-Example of Less Frequent Reporting Instrument

The current date is December 31, 2003. You own stock in ALPHA, Inc. and want to
predict earnings per share for the next quarter (January — March 2004, a total of 13
weeks). The following pages contain a history of earnings per share for the past three
years along with graphs to aid in your decision You can review the pages in any order
you desire and you can revisit pages if you desire. The disk contains a copy of the data
tables presented below. The earnings data in the worksheet is protected and you cannot
make changes to it. However, you can use the adjacent areas in the worksheet for any
calculations you wish to make.

v -m‘arfi

Based on the information provided on the following pages, ..
please provide your best prediction of earnings per share for Quarter 1, 2004
[January 1, 2004 - March 31, 2004].

1. Your prediction of earnings per share is: $

 

2. The actual earnings per share may not turn out to be exactly the same as you
predicted in your answer to the previous question. But you probably have
some idea of the range within which earnings per share is likely to fall.

With this in mind, please answer the following questions:

a. “I would be extremely surprised (no more than a 5% chance of occurring)
if next quarter’s earnings were higher than S

 

b. “I would be extremely surprised (no more than a 5% chance of occurring)
if next quarter’s earnings were lower than S

 

 

47

 

 

Quarterly Earnings

EPS
$2.50 «e—w—v-W— -—~----—~ --—---—---— —

$2.00 A /\~ e

$1.50

 

$1.00

 

$0.50

 

 

sO-m I T T f 7 Y 7 Y r Y Y
1 2 3 4 5 . 6 . 7 8 9 1o 11 12
Trme Penod - Quarter

 

 

 

 

 

 

 

 

 

 

 

 

48

2001 2002 2003 .
Qtr. 1 $1.95 $2.04 $2.32
Qtr. 2 $1.72 $1.81 $2.03
Qtr. 3 $1.65 $1.69 $1.91
00.4 $1.66 $1.74 $2.00

Appendix l-Example of More F reguent Reporting Instrument '

The current date is December 31, 2003. You own stock in ALPHA, Inc. and want to
predict earnings per share for the next quarter (January — March 2004, a total of 13
weeks). The following pages contain a history of earnings per share for the past three
years along with graphs to aid in your decision. You can review the pages in any order
you desire and you can revisit pages if you desire. In the past, we have discovered that
some students like to make calculations. Because we value your time, we have included a
disk that contains a copy of the data tables presented below. The earnings data in the
worksheet is protected and you cannot make changes to it. Although it is not required,
you can use the adjacent areas in the worksheet for any calculations you wish to make.

Based on the information provided on the following pages,
please provide your best prediction of earnings per share for Quarter 1, 2004
[January 1, 2004 - March 31, 2004].

3. Your prediction of earnings per share is: $
(Please make sure you give a prediction for 13 weeks of earnings, not a
single week)

 

4. The actual earnings per share may not turn out to be exactly the same as you
predicted in your answer to the previous question. But you probably have
some idea of the range within which earnings per share is likely to fall.

With this in mind, please answer the following questions:

a. “I would be extremely surprised (no more than a 5% chance of occurring)
if next quarter’s earnings were higher than S

 

b. “I would be extremely surprised (no more than a 5% chance of occurring)
if next quarter’s earnings were l_ower than S

 

49

Year
2001

 

 

1/1

01!. 1

Week1
Week 2
Week 3
Week4
Week 5
Week 6
Week 7
Week 6
Week 9
Week 10
Week 11
Week 12
Week 13

0.16
0.16
0.14
0.14
0.16
0.15
0.15
0.15
0.15
0.15
0.13
0.16
0.15

00.2

Week 14
Week 15
Week 16
Week 17
Week 18
Week 19
Week 20
Week 21
Week 22
Week 23
Week 24
Week 25
Week 26

0.13
0.14
0.14
0.13
0.12
0.13
0.14
0.13
0.1 1
0.12
0.14
0.14
0.15

01!. 3

Week 27
Week 28
Week 29
Week 30
Week 31
Week 32
Week 33
Week 34
Week 35
Week 36
Week 37
Week 38
Week 39

0.1 1
0.1 3
0.14
0.1 2
0.1 3
0.14
0.12
0.13
0.1 1
0.16
0.13
0.1 1
0.12

017.4

Week 40
Week 41
Week 42
Week 43
Week 44
Week 45
Week 46
Week 47
Week 48
Week 49
Week 50
Week 51
Week 52

0.13
0.12
0.13
0.14
0.14
0.13
0.11
0.12
0.13
0.13
0.14
0.13
0.11

12/31

 

Year
2002

 

 

1/1

00.1

Week 1
Week 2
Week 3
Week 4
Week 5
Week 6
Week 7
Week 8
Week 9
Week 10
Week 11
Week 12
Week 13

0.16
0.15
0.16
0.15
0.15
0.16
0.15
0.17
0.17
0.15
0.17
0.16
0.14

017.2

Week 14
Week 15
Week 16
Week 17
Week 18
Week 19
Week 20
Week 21
Week 22
Week 23
Week 24
Week 25
Week 26

0.15
0.14
0.15
0.15
0.13
0.14
0.13
0.14
0.14
0.14
0.14
0.13
0.13

Qtr. 3

Week 27
Week 28
Week 29
Week 30
Week 31
Week 32
Week 33
Week 34
Week 35
Week 36
Week 37
Week 38
Week 39

0.14
0.14
0.12
0.13
0.13
0.13
0.13
0.12
0.13
0.1 1
0.14
0.14
0.13

Week 40
Week 41
Week 42
Week 43
Week 44
Week 45
Week 46
Week 47
Week 48
Week 49
Week 50
Week 51
Week 52

0.15
0.13
0.11
0.15
0.13
0.13
0.11
0.13
0.14
0.14
0.14
0.15
0.13

12/31

 

50

 

 

Year
2003

 

 

Qtr. 1 Qtr. 2 Qtr. 3 WA

1/1 Week 1 0.17 Week 14 0.13 Week 27 0.14 Week 40 0.15
Week 2 0.16 Week 15 0.14 Week 28 0.15 Week 41 0.16
Week 3 0.20 Week 16 0.17 Week 29 0.15 Week 42 0.15
Week 4 0.18 Week 17 0.17 Week 30 0.15 Week 43 0.15
Week 5 0.17 Week 18 0.15 Week 31 0.14 Week 44 0.16
Week 6 0.18 Week 19 0.14 Week 32 0.14 Week 45 0.14
Week 7 0.19 Week 20 0.15 Week 33 0.15 Week 46 0.14
Week 8 0.18 Week 21 0.17 Week 34 0.15 Week 47 0.14
Week 9 0.17 Week 22 0.16 Week 35 0.15 Week 48 0.17

Week 10 0.19 Week 23 0.16 Week 36 0.16 Week 49 0.16
Week 11 0.17 Week 24 0.15 Week 37 0.15 Week 50 0.16 '
Week 12 0.18 Week 25 0.16 Week 38 0.13 Week 51 0.16 4
Week 13 0.18 Week 25 0.18 Week 39 0.15 Week 52 0.16 12/31

 

 

 

51

 

 

EPS

0250
0200
0150
0100
0050
0000

9

17 25

33 41

Weekly Earnings

49 57 65 73 81 as 97
Time Period (Weekly)

 

105 113 121 129 137 145 153

 

 

52

Table 9-Means and Standard Deviations,First Task

Categorized by Experience in Investing

 

 

 

 

 

 

Less Frequent More Frequent
Experienced .08 .14
(.07) 4.14)
No experience .10 .14
(.09) (13)

 

Table IO-Analysis of Effects of Frequency on Signed Prediction Error

 

Panel A

Between Subjects ANOVA with Signed Prediction Error as DV

So_uge df F-value p-va_lqg
Corrected Model 1 1.92 .04
Intercept 1 1.54 .22
Frequency 1 .00 .95
Direction 1 3.65 .06
Rand.Error 1 14.75 .00
Experience 1 4.32 .04
Frequency‘Direction 1 .00 .92
Frequency‘RandError I .38 .54
Direction’RandError l 12.04 .00
Frequency‘Experience*Rand.Error 1 .06 .80
F requency*Experience 1 2.81 .09
Direction‘Experience 1 .92 .34
Frequency*Direction“Experience 1 .06 .80
Rand.Error*Experience 1 3.73 .06
Frequency*Rand.Error*Experience 1 1.81 .18
Dhcction‘Rand.Error*Experience 1 .99 .33
Frequency“Direction“Rand.Error*Experience 1 .27 .61
Error 56

Total

Frequency = frequency conditions between subjects,First task; Direction = direction of
change-upward or downward-between earnings in the last reported period and the
forecasted period; Rand.Error = two different random error draws used to create earnings
series; Experience = yes or no to prior investing experience

53

 

Table ll-Correlations, Post-Experiment Question and Judgment Performance

 

Perceived
Volatility

Perceived

Difficulty

Absolute
Prediction
Error

Absolute
Uncertainty

 

Perceived
Volatility

 

Perceived

Difficulty

.24*

 

Absolute
Prediction
Error

-.04

-.04

 

 

Absolute

Uncertainty
(Confidence

Interval)

 

.07

 

.03

 

.27

 

 

*Significance < .05
Perceived Difficulty captures the perceived difficult in the prediction task.
Perceived Volatility captures how volatile investors perceived the reporting earnings

54

 

 

Table 12-Means and Standard Deviations for Uncertainty in Task 1 and Task 2
Controlling for Direction of Change in Earnings in Prediction Period

 

 

 

 

Less Frequent More Frequent
Positive Direction .60 .54
(.43) (.40)
Negative Direction .53 .57
(.38) (.35)

 

 

 

Table 13-Analysis of Effects of Reporting Frequency on Confidence Intervals When the
Prediction Period Change in Earnings is Positive

Repeated Measures ANOVA with Confidence Interval as DV

 

Source df F-vafilue p-va_hl§
Task 1 2.92 .10
Task‘Frequency 1 .50 .48
Task *Rand.Error l 6.73 .02
Task‘Frequency‘Error 1 .39 .54
Error 29

Description same as in Table 7

Table 14-Analysis of Effects of Reporting Frequency on Confidence Intervals When the
Prediction Period Change in Earnings is Negative

Repeated Measures ANOVA with Confidence Interval as DV

 

Moe df F—value ivm
Task 1 .90 .35
Task‘Frequency 1 3.61 .07
Task *RandError 1 13.93 .00
Task‘Frequency’Error 1 1 .04 .31
Error 29

Description same as in Table 7

55

 

Table lS-Means and Standard Deviations of Confidence Intervals in Task 1 and Task 2
Controlling for Random Error Draw

 

 

 

 

Less Frequent More Frequent
Random Error Draw One .61 .44
(.38) (.29)
Random Error Draw Two .52 .66
(.42) (.41)

 

 

 

Table 16-Analysis of Effects of Reporting Frequency on Confidence Intervals For
Random Error Draw One

Repeated Measures ANOVA with Confidence Interval as DV

 

Spurfce df F-value p-va_lr§
Task 1 1 5.84 .00
Task‘Frequency 1 2.52 .123
Task *Rand.Error 1 .74 .397
Task‘Frequency*Error 1 2.85 . 1 02
Error 30

Description same as in Table 7

Table 17-Analysis of Effects of Reporting Frequency on Confidence Intervals For
Random Error Draw Two

Repeated Measures ANOVA with Confidence Interval as DV

 

Sprgce df F dime p-galie
Task ' 1 6.64 .01
Task‘Frequency l .04 .83
Task ’Rand.Error 1 3.08 .08
Task‘Frequency‘Error 1 1 .05 .31
En'or 30

Description same as in Table 7

56

 

Table 18-Analysis of Effects of Reporting Frequency on Signed Prediction Error

Repeated Measures ANOVA with Signed Prediction Error as DV

 

Sgnce df F -va_11u_e p—value
Task 1 .69 .41
Task*Frequency 1 .09 .76
Task *Rand.Error l 14.66 .00
Task‘Direction 1 1 .70 .20
Task‘Experience 1 4.73 .03
Task‘Frequency’RandError 1 .27 .60
Task‘Frequency‘Direction 1 .16 .69
Task‘Error‘Direction 1 .48 .49
Task‘Frequency*Rand.Error.‘Direction 1 .74 .34
Task‘Frequency*Experience 1 . I 8 .67
Task‘Rand.Error’Experience 1 .99 .32
Task‘Frequency‘RandError" Direction 1 2.42 . 1 3
Task*Direction‘Experience 1 .34 .56
Task‘Frequency*Direction* Experience 1 .00 .95
Task’Rand.Error*Direction‘Experience 1 .67 .43
Task’Frequency“Rand.Error*Direction*

Experience 1 . 12 .73

Error 56
Description same as in Table 7

Table 19-Analysis of Effects of Reporting Frequency on Signed Prediction Error for
Those Who Received Less Frequent Reporting in the 1" Task

Repeated Measures ANOVA with Signed Prediction Error as DV

 

flee df F-value p—vm
Task 1 .1 l .74
Task'Direction 1 .32 .58
Task‘RandError 1 4.33 .05
Task*Experience 1 1 .22 .28
Task‘Direction‘RandError 1 .95 .34
Task‘Direction*Experience 1 .1 1 .75
Task‘Rand.Error*Experience l . 1 2 .73
Task*Direction*Rand.Error*Experience 1 .54 .47
Error 23

Description same as in Table 7

57

Table 20-Analysis of Effects of Reporting Frequency on Signed Prediction Error for
Those Who Received More Frequent Reporting in the 1" Task

Repeated Measures ANOVA with Signed Prediction Error as DV

 

Source df F -value p-val_u_e
Task 1 .78 .38
Task‘Direction 1 1.78 .19
Task’RandError 1 1 1.64 .00
Task’Experience 1 4. 1 5 .05
Task’Direction’RandError 1 .02 .90
Task*Direction*Experience l .26 .61
Task‘Rand.Erl'or*Experience 1 4.00 .05
Task‘Direction‘RandError"Experience 1 . 1 3 .72
Error 23

Description same as in Table 7

58

Table 21-Means and Standard Deviations for Prediction Error in Task 1 and Task 2
Controlling for Direction of Prediction Period Change in Earnings

 

 

 

 

 

 

Less Frequent More Frequent
.~‘ Random Error Draw One .16 . .10
(.12) (. 12)
Random Error Draw Two .08 .17
(.09) (.14)

 

Table 22-Analysis of Effects of Reporting Frequency on Prediction Error When the
Prediction Period Change in Earnings is Positive

Repeated Measures ANOVA with Prediction Error as DV

 

_S_pr.rrce df F-value p-vm
Task 1 7.83 .01
Task*Frequency 1 3.86 .06
Task "RandError 1 13.05 .00
Task‘Experience l 4.24 .05
Task‘Frequency‘RandError 1 .02 .88
Task’Frequency1‘Experience 1 .00 .96
Task‘Rand.Error*Experience 1 .28 .60
Task“Frequency*Rand.Error*Experience 1 2.26 . I 5
Error 25

Description same as in Table 7

Table 23-Analysis of Effects of Reporting Frequency on Prediction Error When the
Prediction Period Change in Earnings is Negative

Repeated Measures ANOVA with Prediction Error as DV

 

S_o_urce df F-value mvalue
Task ' 1 12.38 .00
Task‘Frequency 1 .77 .39
Task ’Rand.Error 1 21.28 .00
Task‘Experience l .24 .66
Task‘Frequency*Rand.Error l .83 .37
Task*Frequency*Experience 1 .07 .80
Task‘RandError‘Experience 1 .28 .60
Task‘Frequency" Rand.Error*Experience 1 . 12 .73
Error 25

Description same as in Table 7

59

 

Table 24-Means and Standard Deviations for Prediction Error in Task 1 and Task 2
Controlling for Random Error Draw

 

 

 

Less Frequent More Frequent
Random Error Draw One .17 .08
,‘ (.13) - (.06)
Random Error Draw Two .07 .19
(.06) (.16)

 

 

 

 

Table 25-Analysis of Effects of Reporting Frequency on Prediction Error for Random
Error Draw One

Repeated Measures ANOVA with Prediction Error as DV

 

wee df F-value p-vglu_e
Task 1 18.29 .00
Task*Frequency 1 1 .58 .22
Task *Experience 1 2.37 .14
Task‘Direction 1 1 1.60 .00
Task‘Frequency*Experience 1 .46 .50
Task*Frequency*Direction 1 1 .65 .21
Task*Experience*Direction 1 2.45 .13
Task“Frequency‘Experience‘Direction 1 1 .34 .26

Error 25
Description same as in Table 7

Table 26-Analysis of Effects of Reporting Frequency on Prediction Error for Random
Error Draw Two

Repeated Measures ANOVA with Prediction Error as DV

 

Source df fialue gym;
Task 1 1 7.25 .00
Task‘Frequency 1 4.03 .05
Task *Experience 1 2.18 .15
Task‘Direction 1 9.39 .00
Task*Frequency‘Experience 1 .89 .35
Task‘Frequency‘Direction 1 .1 7 .68
Task’Experience‘Direction 1 .33 .57
Task‘Frequency*Experience*Direction 1 1 .05 .32
Error 25

Description same as in Table 7

60

 

Table 27-Means and Standard Deviations for Prediction Error in Task 1 and Task 2

Controlling for Experience

 

 

 

 

 

 

Less Frequent More Frequent
Experienced . l 1 .13
(. 12) (. 1 5)
Non-Experienced . 12 . l 4
(.1 1) (. 1 3)

 

Table 28-Analysis of Effects of Reporting Frequency on Prediction Error for Investors
with Investing Experience

Repeated Measures ANOVA with Prediction Error as DV

 

Source df F -value p-val_ue_
Task 1 l .89 . 1 9
Task‘Frequency 1 1 .35 .26
Task *Direction 1 10.96 .00
Task‘RandError 1 10.93 .00
Task*Frequency*Direction 1 .59 .46
Task‘Frequency‘RandError 1 .16 .70
Task*Experience*Direction*Rand.Error 1 .48 .50
Task*Frequency*Direction‘RandError 1 1 .45 .25
2

5
Description same as in Table 7

Error

Table 29-Analysis of Effects of Reporting Frequency on Prediction Error for Investors
with No Investing Experience

Repeated Measures ANOVA with Prediction Error as DV

 

Som'ce df F-value p-vaJpe
Task 1 2.70 .1 1
Task‘Frequency 1 5.27 .03
Task *Direction 1 8.43 .00
Task‘RandError l 30.77 .00
Task‘Frequency‘Direction 1 1 .02 .32
Task*Frequency*Rand.Error 1 2.1 7 .1 5
Task‘Experience“Direction‘RandError 1 .08 .78
Task*Frequency*Direction‘RandError 1 .69 .41
2

5
Description same as in Table 7

Error

61

 

REFERENCES
Abarbanell, J., and V. L. Bernard, 1992, “Test of Analysts Overreaction/Underreaction
to Earnings Information as an Explanation for Anomalous Stock Price Behavior.”
Journal of Fin_ance 47 (July): pp. 1181-1208.

Andreassen, P.B., 1987, “On the Social Psychology of the Stock Market: Aggregate
Attributional effects and Regressive of Prediction.” Journflf Personality and
Social Psychology 53, 490-496.

1990, “Judgmental Extrapolation and Market Overreaction on The Use and Disuse
of News.” Jouraal of Behavioral Decision Making 3, 153-174.

and Kraus S., 1990, “Judgmental Extrapolation and the Salience of
Change.” Journala of Forecasting, 9, pp. 347-372.

Ball, R., and E. Bartov, 1996, “How Naive Is the stock Market’s Use of Earnings
Information?” Journal of Accounting and Economics, pp. 319-37.

Barberis, N., A. Shleifer, R. Vishny, 1998, “A Model of Sentiment.” Journal of
Financial Economics 49, pp. 307-343.

Bartov, E., S. Radhakrishnan, & I. Krinsky, 2000, “Investor Sophistication and Patterns
in Stock Returns after Earnings Announcements.” The Accounting Review, 75,

(January): pp. 43-63.

Benard, V. and J. Thomas, 1989, “Post-Earnings-Announcement Drifi: Delayed Price
Response or Risk Premium?” Journal of Accounting Research (Supplement):
pp.l-36.

Benartzi, S. & R. Thaler, 1995, “Myopic Loss Aversion and the Equity Premium
Puzzle,” Quarterly Journal of Economics, CX, pp. 73-92.

Bloomfield, R., and J. Hales, 2002, “Predicting the next step of a random walk:
experimental evidence of regime-shifting beliefs.” Jouraal of Financial
Economics, 65, pp. 397-414.

Bloomfield, R., W. Libby, M. Nelson, 2000, “The effects of Over-Weighting Old
Elements of the Earnings Time Series.” Unpublished working paper, Cornell
University.

Bouwman, M. 1982, “The Use of Accounting Information: Expert vs. Novice Behavior”
In Deci_sion Making: An Interdiciplmagg Ingug.” edited by G. Ungson and D.
Braunstein Boston, MA: Kent: pp. 134-167.

Burgstahler, D. and I. Dichev, 1997, “ Earnings Management to Avoid Earnings

62

Decreases and Losses.” Jourgl of Accountipg and Econ_omics 24 (1) pp. 99-126.

Burns, B. 2003, “When is it Adaptive to Follow Streaks: Variability and Stocks.” In
press.

Burns B., B. Corpus, 2003, “Randomness and Introductions fi‘om Streaks: “Gamblers
Fallacy vs, Hot hand.” In press.

Byrnes, N., 2002 “The Downside of Disclosure Too much data can be a bad thing. It's
quality of information that counts, not quantity.” Business Weak; New York;

(August 26).

Camerer, C., G. Loewenstein, and M. Weber, 1989, “ The curse of knowledge in
economic settings: An experimental analysis.” Journal of Political Economy
(October): pp. 1232-1254.

Commission of the European Communities, 2003, Proposal for a Directive of the
European Parliament and of the council on the harmonization of transparency
requirements with regard to the information about issuers whose securities are
admitted to trading on a regulated market and amending Directive 2001/24/EC.

Degeorge, F., J. Patel, and R. Zeckhauser, 1999, “Earnings Management to exceed
thresholds.” Journal of Business 72: pp. 1-33.

De Bondt, F .W. and R. Thaler, 1985, “Does the stock market overreact.” The Journal of
Finance, No. 3, pp. 793-805.

1993, “Betting on Trends: Intuitive Forecast of Financial Risk and
Return.” lntemational Jouanal of Forecasting, 9, pp. 355-371.

Editorial Staff, 2002, “New Feature: Reversal of Fortune: Institutional Ownership
Declining,” Investor Relations Business, pp. 8-9.

Eggleton, 1. RC, 1982, “Intuitive Time Series Extrapolation.” Jourrlal of Accountiag
Research, Vol. 20, pp. 68-102.

Eggleton, 1. RC, 1976, “Patterns Protoypes and Prediction: An Exploratory Study”
Selected studies on Human Information Processing in Accounting.” Supplement
to Journal of Accounting Research. Vol. 14 pp. 68-131.

Elroy, D. & A. Jackson, 2002, “High Frequency Performance and Monitoring.” Working
Paper.

Evans, C., 2003, “Quarterly Reporting - Companies face quarterly reporting nightmare.”
Accountancy, vol. 132, pp. 36.

63

F ischhoff, B. 1977, “Perceived informativeness of facts.” Journal of Exxrimental '
Psychology: Human Perception and Performance, (May): pp. 349-358.

Financial Accounting Standards Board (FASB), 2000, “Electronic Distribution of
Business Reporting Information.” Steering Committee Report Series. Business
Reporting Research Project.”

Freudenthal, H. 1974, “The Empirical Law of Large Numbers or The Stability of
Frequencies.” Educational Studies in Mathematics, 4, pp. 484-490.

Foster, 6., 1977, “Quarterly Accounting Data: Time-Series Properties and Predictive-
Ability Results.” The Accounting Review, vol. 52, pp. 1-21.

Gilovich, T., R. Vallone, & A. Tversky. (1985). “The Hot Hand in Basketball: On the
Misperception of Random Sequences.” Cogaitive Psycholagy, 17, pp. 295-314.

Gneezy U., & J. Potters, 1997, “An Experiment on Risk Taking and Evaluation Periods.”
The Quarterly low] of Ecoaomics, CXII pp. 631-645.

Healy, P. M. and J.M. Wahlen, 1999, “A Review of the Earnings Management Literature
and its Implications for Standard Setting.” Accounting Horizens, 13 (4): pp. 365-
383.

Hoge, RD, 1970, “Confidence and Decision as an Index of Perceived Accuracy of
Information Processing.” Paychonomic Science, 18, pp. 351-353.

Hopewood, W.S., J.C. Mckeown, & P. Newbold, 1982, “The Additional Information
Content of Quarterly Earnings Reports: Intertemporal Disaggregation.” Journal of
Accounting Research. 20, No. 2 Pt. 1, Autumn, pp. 343-349.

Hunton J. A., and R.A. McEwen, 1997, “An assessment of the relation between analysts’
earnings forecast accuracy, motivational incentives and cognitive information
search strategy.” The Accounting Review, Vol. 72, No. 4, pp. 497-515.

Hunton, J., E. Wright, and S. Wright. 2003, “Assessing the Impact of More Frequent
External Financial Statement Reporting and Independent Auditor Assurance of
Quality of Earnings and Stock Market Effects.” Working Paper, Bentley College,
Boston University and Unversity of Massachusetts, Boston.

Jacob, J., and T. Lys, 1996, “Determination ands Implications of the Serial Correlation in
Analysts’ Earnings forecasts. Working Paper, Northwestern University, Evanston
IL.

Kahneman, D., & A. Tversky, 1972, “Subjective probability: A judgment of
representativeness.” Cogpa' ive Psychology, 3, pp. 430-454.

Kahneman, D., & A. Tversky, 1979, “Prospect Theory: An Analysis of Decision Under
Risk.” Econometrica, XLVII (1979), pp. 263-91.

Kahneman, D. & A. Tversky, 1984, “Choices, Values and Frames.” American
Psychologist, XXXIX, pp. 341-50.

Klayman, J., “On The How and Why (Not) of Learning From Outcomes.”
Human Judgment: Th; SJT View. North Holland, 1988.

 

, & Burt, R. S., 1998, “Individual Differences in Confidence and Experiences
in Social Networks.” Working paper, Center for Decision Research, Graduate
School of Business, University of Chicago.

, 8011 JB, Gonzalez-Vallejo C, Barlas S., 1999, “Overconfidence: It Depends
on How, What, and Whom You Ask.” Organizational BehaviorJand Human
Decision Processes 79 pp. 216-247.

Laplace, RS. (1951). A Philosophical Essay On Probabilities (F.W. Truscott & F. L.
Emory Trans). New York: Dover. (Original work published in 1814).

Lawrence, M.J. and M.J. O’Connor, 1992, “Exploring Judgmental Forecasting,”
International Journal of Forecasting, 8, pp. 15-26.

, Edmundson R.H., and O’Connor M.J. 1985, “An examination of the
Accuracy of Judgmental Extrapolation of Time Series.” International Journal of

Forecasting, 1, pp. 25—35.

, Edmundson R.H., and O’Connor M.J. 1988, “The Accuracy of
Combining Judgmental and Statistical Forecast.” Managementa Science. 12, pp.
1521-1532.

and Makridakis S., 1989, “Factors Affecting Judgmental Forecast and
Confidence Intervals.” Organizational Behavior and Human Decision Processes,
42, pp. 172-187.

 

and O’Connor M., 1993, “Scale, Variability and the Calibration of
Judgemental Prediction Intervals.” Organizational Behavior and Human Decision
Processes, 56, pp 441-458.

Levitt, A., Take on the Street: What Wall Street and Corporate America Don’t
Want You to Know. What You Can Do to Fight Back. New York: Random
House, 2002.

 

Lim, J.S., and M. O’Connor, 1996, “Judgmental Forecasting with Time Series and Causal
Information.” International JoumaL of Forecasting 2, pp. 139-153.

65

Lipe, M., 1998, “Individual investors’ risk judgments and investment decisions: The

impact of accounting and market data” Accounting Organizations and Society.
Vol. 23, No. 7, pp. 625-640.

Maines, L., & J.R. Hand, 1996, “Individuals’ Perceptions and Misperceptions of Time
Series Properties of Quarterly Earnings.” The Accounting Review, 71, (July):
pp. 317-36.

Maines, L. & L. S. McDaniel, 2000, “Effects of Comprehensive-Income Characteristics
on Nonprofessional Investors’ Judgments: The Role of Financial Statement
Presentation Format.” The Accounting Review, 75, (April): pp. 179-207.

McNamee, M., 2001, “New Yardsticks for Investors.” Business Week; New York;
(November 5).

Miller, G. 1956, “The Magical Number Seven, Plus or Minus Two: Some Limits on Our
Capacity for Processing Information” The Psychological Review, 1956, vol. 63,
pp. 81-97.

Mosteller, F. A.F. Siege], E. Trapido and C. Youtz, 1981, “Eye Fitting Straight Lines.”
The Americaa Statistician, 35, pp. 150-152.

O’Conner, M. Remus, W., K. Griggs, 1993, “Judgmental Forecasting in Times of
Change.” nternational Journal of Forecastmg' 9, pp. 163-172.

and Lawrence M., 1989, “An Examination of the Accuracy of Judgmental
Confidence Intervals in Time Series Forecasting.” Journal of Forecasting, 2, pp
141-155.

and Lawrence M., 1989, “Time Series Characteristics and the Widths of
Judgemental Confidence Intervals.” International Journal of Forecasting, 7, pp.
413-420.

, Remus, W. and Griggs K., 1993, “Judgmental Forecasting in Times of
Change.” International Jouanal of Forecasting, 9, pp. 163-172.

Oskamp, S. 1965, “Overconfidence in Case Study Judgments.” ournal of Consulting
Psychology, 29, pp. 261-265.

Palepu, K.G., P.M. Healy and V.L. Bernard, Business Analysis and Valuation Using
Financial Statemenas: Text and CM. Mason, Ohio: Thomson Southwestern,

1 996.

 

Peterson, D., & G.F. Pitz, 1986, “The Effects of Amount of Information on Predictions of
Uncertain Quantities.” Acta Psycholigjga, 61, pp. 229-241.

66

 

Peterson, D., & G.F. Pitz, 1988, “Confidence, Uncertainty and The Use of Information.”
Jouml of Experimengl Psychology: Learning Memory and Cognition. 1, pp. 85-
92.

Pitz, G.F. 1974, “Subjective Probability Distributions for Imperfectly Known Quantities.”
in L. W. Gregg (Ed), Knowledge and Cograition (pp. 29—41). New York: Wiley.

Securities and Exchange Commission (SEC). 2000, Release No. 34-42728.
Interpretation: Use of Electronic Media.

Sedhneier, P. 1998, “The Distribution Matters: Two Types of Sample-size Tasks.”
Journal of Behavioral Decisions Making, Vol. 11, pp. 281 -301.

Sedlmeier, P. & G. Gigerenzer, 1997, “In Intuitions About Samples Size: The Empirical
Law of Large Numbers.” Journal of Behavioral Decision Making Vol. 10, 33-51.

Schroder, H. M., M.J. Driver and S. Streufert, Human Information Processing,
New York: Holt, 1967.

Shapira, Zur. Risk Taking: A Managerial Perspective. New York: Russell Sage
Foundation, 1995.

Slovic, P. 1973, “Behavioral Problems of Adhering to a Decision Policy.” Working
paper, Oregon Research Institute, Eugene Oregon.

Tasker, S., 1999, “Bridging the Information Gap: Quarterly Conference Calls as a
Medium for Voluntary Disclosure.” Review of Accounting Studies. Vol. 3, pp.
137-167.

Tversky, A. & D. Kahneman, “ Advances in Prospect Theory: Cumulative
Representation of Uncertainty,” Journal of Risk and Uncertainty, V(1992), pp.
297-323.

Thaler, R. 1985, “Mental Accounting and Consumer Choice,” Marketing Science, IV,
pp. 199-214.

Thaler, T., Tversky, A., D. Kahneman & A. Schwartz, 1997, “The Effect of Myopia and
Loss Aversion on Risk Taking: An Experimental Test.” The Qlanefly Journal of
Economics CXII, 647-661.

 

Wallman, S. M. H., 1995, “The future of accounting and disclosure in an evolving world:
The need for dramatic change.” Accountmg’ Horizons, September, pp. 81-91.

Wagenaar, W.A., 1991, “ Randomness and Randomizers: Maybe the Problem is Not So
Big.” Journal of Behayiora Decision Making, 4, 220-22.

67

Watson, L., McGuire, L., & E. Cohen, 2000 “Looking at business reports through
XBRL-tinted glasses.” Strategic Finance, 40-45.

Webby, R. and M. O’Conner, 1996, “Judgmental and statistical time series forecasting: a
review of the literature.” International Jouraal of Forecastin , 12, pp. 91-118.

Yates, J. F., McDaniel, L. S., and Brown, E. S. (1991), “Probabilistic Forecasts of Stock

Prices and Earnings: The Hazards of Nascent Expertise.” Organizational
Behavior and Human Decision Processes, 49, pp. 60-79.

68

   

lllllllllllTill"