OREGON RESEARCH INSTITUTE
Analyzing the Expert Judge: A Descriptive Study
of a Stockbroker's Decision Processes
Paul Slovic
ORI Research Bulletin
Vol. 8 No. 3
June, 1968
ANALYZING THE EXPERT JUDGE: A DESCRIPTIVE STUDY
OF A STOCKBROKER'S DECISION PROCESSES
. 2
Paul Slovic
Oregon Research Institute, Eugene
Abstract
This study illustrates an analysis-of-variance technique for describ
ing the use of information by persons making complex judgments. Ss were
two stockbrokers who rated the growth potential of stocks on the basis
of 11 factors taken from Standard S Poor reports. The technique proved
capable of providing aprecise quantitative description of configural
and nonconfigural information utilization. Each broker exhibited a
substantial amount of configural processing. The technique appears to
have promise for providing expercs with insight into their own processes
and for teaching and evaluating "student" judges.
ANALYZING THE EXPERT JUDGE: A DESCRIPTIVE STUDY
OF A STOCKBROKER'S DECISION PROCESSES1
Paul Slovic
Oregon Research Institute, Eugene
The task of the expert judge, be he military officer, detective,
businessman, physician, clinical psychologist, financial analyst, etc.,
requires him to combine items of information from a number of different
sources into a decision or judgment. The key to the expert's success
resides in his ability to interpret and integrate information appro
priately. This means he must weight items of information differentially,
according to their relevance, and must be able to qualify his inter
pretations of a given fact when other considerations make such qualification
necessary.
There is no need to dwell upon the tremendous importance of being
able to understand and describe how the expert uses information.
However, such understanding does not come easily. All too often
expert judgment is regarded as a mysterious, intuitive phenomenon—
incapable of being described precisely. For example, Lusted (1960)
relates a story about a radiologist famed for his diagnostic ability.
Once, when he was questioned as to why he thought a particular shadow
on an X-ray was a metastatic lesion, the physician replied, "Because
it looks like it!" At the other extreme, we're all familiar with the
expert who instructs others in the art of emulating his judgments by
Slovic
reeling off the dozens of factors that he takes into consideration,
each accompanied by an elaborate rationale. Information of this sort
is quite difficult for the student of expertise to use and, in addition,
may not accurately represent what the expert is really doing.
Only in the past 20 years has there been any extensive study of
the judgment process, and this study has been primarily within the
context of clinical psychology. The earliest research efforts focused
on the accuracy of judgments and the degree to which experts agreed
with one another in their evaluations. The results of these studies
have indicated a distressing lack of accuracy and interjudge agree
ment both in medicine (Garland, 1959, 1960) and in clinical psychology
(Goldberg, in press).
As a result of these findings, the emphasis has shifted from
research on the validity and reliability of judgments to attempts to
understand the judgment process itself. This recent research aims to
"simulate" or "model" the hidden cognitive processes of the judge.
Hopefully, by understanding these processes we will learn why some judges
are more accurate than others, and this knowledge will, in turn, help
us to train persons to make better judgments.
Some of the first models for quantitatively describing the judgment
process were developed by Hoffman (1960) and by Hammond and his associ
ates (see Hammond, Hursch, & Todd, 1964, for example). While their
techniques have been quite successful in describing how individual
items of information are weighted and combined by a judge, they have
Slovic
not been successful in describing complex patterned or configural
use of information; i.e., the process whereby an item of information
is interpreted differently from one time to the next, depending on
the nature of other available information. Since experts generally
claim that they use information configurally, it is important that
techniques used to describe judgment be sensitive to such processes.
One technique that analyzes the judgment process in all its
complexity has been described by Kleinmuntz (1968), who had clinical
psychologists and neurologists 'think aloud" into a tape recorder as
they made diagnostic judgments. Kleinmuntz utilized these rich intro
spective reports to construct a computer program simulating the
diagnosticians'thought processes. The resulting programs were complex
sequential (e.g., hierarchical or "tree") representations of the
diagnosticians' verbal reports. At the present time it is not clear
whether the failure of investigators other than Kleinmuntz to find
experimental evidence for configurality stems from lack of configurality
in the processes themselves or from deficiencies in the models and
procedures employed to evaluate those processes (Goldberg, in press).
Hoffman, Slovic, and Rorer (1968) introduced a technique
based upon the analysis of variance (ANOVA) for quantitatively des
cribing both configural and nonconfigural use of information in judgment
They employed this technique to study the processes whereby radiologists
diagnose the malignancy of gastric ulcers on the basis of roentgeno
logical signs. Although the radiologists were found to process
Slovic
information configurally in many instances, the overall influence of
such nonlinear processing was slight. Most of the variability in
the diagnoses could be predicted from a linear combination of signs.
Because the ANOVA technique proved quite capable of describing
the use of information by individual radiologists and because it was
sensitive to configural processing it appeared to merit further use.
The purpose of the present paper was to test the adequacy of the ANOVA
technique for describing the way that a stockbroker employs inform
ation as he evaluates the attractiveness of a company's stock. The
stockmarket was selected as the domain in which to study expertise
for several reasons. First, the task of predicting the future market
price of a security is an important one. Hundreds of thousands of
decisions, involving many millions of dollars, are made daily in the
market. Second, this task is interesting because it is extremely
difficult and complex. There are hundreds of factors which may be
relevant, some of them economic, some of them financial, and some of
them psychological in nature. In addition, introspective reports
by financial analysts indicate that they believe that the relevant
factors should be interpreted in a complex configural manner. For
example, many analysts claim that one cannot interpret recent price
changes of a stock without taking into account the volume of sales
that accompanied those changes.
Slovic
Method
Subjects. The Ss were two young brokers. Each had about three
years' experience with aprominent brokerage firm. While these brokers
may, on occasion, merely fill a client's order, they frequently are
called upon for advice, and in some instances have complete responsi
bility for managing aclient's portfolio. These men are quite concerned
about their ability to judge stocks and spend several hours each day
studying the market, attempting to glean information from avariety
of sources such as newspapers, the ticker tape, company reports,
financial analysts' reports, etc.
Procedure. The application of ANOVA to the study of judgment is
simple and direct; one first selects a set of presumably relevant
factors (i.e., items of information or dimensions along which a stimulus
can be described) and then constructs stimuli such that all possible
combinations of these factors are represented. When the judgments
that are made about each of these stimuli are analyzed in terms of
an ANOVA model, a significant main effect for Factor 1 indicates that
the judge's responses varied systematically with Factor 1 independent
of the levels of the other factors. This implies that Factor 1 was im
portant to the judge. Asignificant interaction between Factors 1and 2
implies that the judge was interpreting particular patterns of these factors
in a configural manner; that is, the interpretation of Factor 1 upon
judgment differed as a function of the value taken by Factor 2.
The present task was constructed with the assistance of Broker A.
When asked to list the minimum number of factors upon which he could
Slovic
comfortably base a recommendation about a stock, Broker A selected
11 variables commonly provided in Standard & Poor's reference reports.
These variables were:
a. Yield (YLD). The cash dividend income for the past year
as a percentage of the market price.
b. Near Term Prospects (NTP). A one- or two-paragraph forecast
concerning sales, profits, dividends, earnings, etc., for the coming
year. Included is pertinent information concerning new products,
political or economic factors bearing on the company's future, etc.
c. Earnings Quarterly Trend (EQT). A comparison of quarterly
earnings over the past 4-5 years.
d. Past Year's Performance (PYP). A synopsis of relevant statistics
for the past year. Includes revenues, earnings and dividends, and
political and economic factors that influenced them.
e. Profit Margin Trend (PMT). A yearly comparison indicating the
trend in percentage of profit from company operations per sales dollar.
Presumably this relates to the efficiency with which the company is
managed and has implications for future earnings.
f. Earnings/share Yearly Trend (EYT).
g. Price/Earnings Ratio (PER). The ratio of market price to net
earnings per share over the past 12 months.
h. Shares Outstanding (SO). The number of shares of common stock
issued by the company.
i. Resistance Trend (RES). Trend of a line connecting several
recent high points on the chart of daily price action.
Slovic
j. Support Trend (SUPP). Trend of a line connecting several recent
low points on the daily price chart.
k. Sales Volume Trend (VOL). Trend of the number of shares
traded per day over a recent period of time.
Next, Broker A was asked whether, in the interests of simplification,
he could still make a reasonable evaluation of a company's stock
if information about the 11 factors was presented in dichotomous
form (for example, yield being described as either high or low, trends
as either up or down, etc.). The broker said that he could. Further
questioning indicated that there would be no combination of these
factors so unreasonable as to make the company seem unreal and, therefore,
impossible to judge.
The next step involved the construction of hypothetical companies.
Ideally it would have been desirable to combine the 11 dichotomous
factors in all possible ways, but in this case that would have resulted
in 211 or 2048 companies, clearly an unmanageable number to judge.
However, if one is willing to assume that the higher order interactions
are negligible, it is possible, by means of a fractional replication
design (Cochran & Cox, 1957), to evaluate the main effects and lower
order interactions with a considerably reduced number of stimuli.
Previous work on judgment (Goldberg, in press) suggested that
the assumption that higher order interactions would be negligible
was not too unreasonable. Therefore, hypothetical companies were
constructed by combining the levels of the 11 factors according to
Slovic
a1/16 fractional replication of a211 factorial ANOVA design. This
produced a set of 128 companies. This reduction of stimuli results
in the confounding of main effects and two-way interactions with certain
of the higher order interactions. Other higher order interactions serve
to estimate the error term in the ANOVA. Thus, if configural use of
three or more factors does occur, the error term will be inflated.
Fig. 1 illustrates the way in which information about a company
was displayed to the brokers. The spatial format of the variables
was designed to approximate the layout of a Standard 6 Poor report as
closely as possible. The stimuli were bound in a notebook which
Insert Figure 1 about here
the brokers took home. The brokers worked on the judgments in
their leisure time over a three-week period. Broker A reported spend
ing 10-1/2 hours making his judgments. Broker B spent about 9 hours
at the task. Although they knew the companies were hypothetical, both
brokers reported that the task was extremely interesting to them and
that they were able to conjure up images of real companies as they
read the stimulus information.
The brokers were asked to make a recommendation about each company
based on their judgment of the likelihood that the market price of
that company's stock would increase substantially in the next 6-12
months. The recommendation was made on a nine-category rating scale
Slovic
where Category 1was labeled "strong recommendation not to buy,"
Category 4 was a "slight recommendation not to buy,"
Category 5 was a "neutral" evaluation, and Categories 6 and 9 were
labeled slight and strong "recommendations to buy" respectively.
Results
The mean rating given the 128 companies by Broker A was 5.62
with a standard deviation of 1.94. Broker B was less favorably inclined
towards the companies' stocks (mean = 3.96) and more variable in his
ratings (standard deviation = 2.96).
Despite the fact that Broker B was recruited as a subject by
Broker A on the grounds that his approach to selecting stocks was
relatively similar to that of Broker A, there was rather poor
agreement between the two with regard to their ratings. The cor
relation between the two brokers' judgments, across the 128 companies,
was only .32.
In order to isolate the factors influencing the recommendations,
a separate ANOVA was performed on each broker's responses. Sums of
squares and mean squares were computed for each of the 11 main effects
(individual factors), each of the two-way interactions, and each of
the few three-way interactions that were confounded only with four-
way or higher order interactions. In addition, two indices of the
importance of a factor or interaction were computed for each effect.
One was simply the standard calculation of the magnitude of an effect,
based upon the degree to which the mean judgment shifted as the levels
of a factor were varied. In this regard, the magnitude of a two-way
Slovic 10
interaction effect indicates the degree of change in the mean judgments
as a function of variation in the levels of a pair of factors after
the main effects have been partialed out. The second index, called
u>2, is a function of the squared magnitudes of effect and provides
an estimate of the proportion of the total variance in the broker's
judgments that could be attributed to a particular main effect or
interaction (Hays, 1963).
Tables 1 and 2 present the results of the ANOVAs for the two
brokers. The ratings of Broker A changed significantly with variation
in the levels of each of six factors (main effects), the most influential
of these being Near Term Prospects, Price-Earnings Ratio, and Earnings
Quarterly Trend. In addition, six interactions were significant, one of
these (Resistance Trend x Support Trend) being the fourth strongest
effect. Broker B exhibited seven significant main effects, the strongest
of which were due to the Earnings Yearly Trend, Price Earnings Ratio,
and Profit Margin Trend. In addition, five two-way interactions were
significant.
Insert Tables 1 and 2 about here
Since the 11 factors studied here were specifically selected
by Broker A as the most important ones from among a much larger set,
the fact that his judgments were not influenced significantly by a
number of these factors is especially noteworthy. During the process
r1 . 11
Slovic
of selecting these factors the broker was able to give an elaborate
rationale for including each one. Perhaps it was too difficult for him
to use all of the factors simultaneously.
Summing the w2 index over the statistically significant factors
indicated that about 72% of the variance in Broker A's responses
was predictable from knowledge of six main effects and an additional
7% could be attributed to configural use of cues (significant interactions).
Comparable figures for Broker B were 80% (main effects) and 5% (inter
actions). These percentages could be interpreted as evidence
for the negligibility of configural cue utilization as were the comparable
percentages found in the study of radiologists by Hoffman, Slovic,
and Rorer (1968). However, the use of variance percentages as
descriptive indicators may be more meaningful statistically than
psychologically, and the magnitude of effect index, based upon the
influence of a factor upon the mean judgments, might well be a more
appropriate gauge for assessing the relative importance of configural
effects. This index indicates that configurality was substantial,
accounting for 27% of the total effects on Broker A and 19% of the
effects on Broker B. Even this is a conservative estimate of the
degree of configurality. Extrapolating from the excellent discussions
of linear and configural models presented by Green (1968) and Hayes
(1968), one could argue that whenever the interaction between two
factors was significant the variance accounted for by the main effects
for these factors should be counted as configural variance.
Slovic 12
Following this rule would boost the percentage of configural variance
to 36% for Broker A and 85% for Broker B. Additional evidence for
the argument that meaningful configural information processing was
taking place here is the fact that two interactions (RES x VOL and
SUPP x VOL) were common to both brokers. Detailed analysis of these
interactions showed each of them to be almost identical in form for
the two brokers.
An index of the overall importance of a given factor was cal
culated by summing the magnitude of effect index for the main effect
of that factor with the magnitude of effect indices of all significant
interactions containing that factor. The summed effect of a given
factor was divided by the sum of the effects of all factors. This
index of importance was thus a percentage score where the sum
of all percentages totaled 100.
Fig. 2 illustrates the relative importance of the 11 factors for
each broker based on the index just described. Despite the fact
that the brokers viewed themselves as quite similar in orientation,
there was a considerable difference in their use of information.
These differences undoubtedly indicate why they disagreed so often
in their rating of a particular stock. Broker A considers himself to
be a "technical analyst" (i.e., one who weights information from price
and volume charts especially heavily),and in this regard it is noteworthy
that the ANOVA model showed him to be using the three chart variables,
Resistance, Support, and Volume Trends, to a greater extent than did
13
Slovic
Broker B, who views himself, and appropriately so, as more of a
"fundamentalist" (i.e., one who relies on traditional balance sheet
and income indicators).
Insert Figure 2 about here
Validity of subjective weights. How closely would the brokers'
subjective impressions of the relative importance of the 11 factors con
form to the importance indices calculated from the ANOVA model?
To provide an answer to this question, each broker was asked, after
completing his ratings, to distribute 100 points over the 11 factors
proportionally to his feelings about their importance in determining
his judgments. These subjective weightings were compared with the
magnitude of effect indices pictured in Figure 2and with the to2 index,
the latter also being combined over both main effects and interactions,
and normed to sum to 100 over the 11 factors, The results of this
comparison are depicted in Figures 3 and 4. They show that the subjective
weightings of Broker A were extremely close to the magnitude of effect
index while Broker B had less accurate insight into his use of the
various factors. The a>2 index was very discrepant from the subjective
weights of both brokers. This index tended to exaggerate the dif
ferences between the most important factors and the lesser ones.
To the extent that one feels that expert judges should have some insight
about their own weighting system, this result implies that the magnitude
Slovic 14
of effect index is a better measure of a factor's relative importance
than the go2 index.
Insert Figures 3 and 4 about here
Analysis of interactions. The finding of a significant main
effect or interaction is only a first step in understanding how a judge
interprets information. It should be viewed as a signal that
something interesting is going on. Graphical representation of an
effect followed by interrogation of the judge concerning the rationale
behind his behavior can be used to further one's understanding of the
effect. To illustrate, three of the significant interactions found
in the judgments of Broker A are pictured in Figure 5. Broker A
was shown these figures and was asked to provide an explanation for
each one. A paraphrased version of his explanation for each effect
is as follows.
Insert Figure 5 about here
(a) The YLD x PMT effect. Why is high yield a more favorable indi
cator than low yield when PMT is down while the reverse is true when
PMT is up?
"Because when PMT is down, earnings probably are down, and
en I 15Slovic
accordingly the price of the stock should decline. A low dividend
yield would make the stock even less attractive while ahigh yield
would tend to compensate for the poor earnings prognosis. When PMT
is up, earnings are probably up and the outlook for price appreci
ation is good. A quality company whose earnings portend good growth
doesn't usually offer a large dividend, so low yield in conjunction
with a rising PMT suggests that the stock has a very promising future.
A high yield in this case suggests that the company is probably not
putting enough of its capital into growth or perhaps that the outlook
for future price appreciation is not really so promising, hence the
need for a larger dividend to make the stock attractive to the investor."
(b) The EYT x SUPP effect. Why should a rising trend in yearly
earnings be a better sign than a declining earnings trend when the
support trend (price) is down while the reverse is true when the support
trend is up?
"When both support and earnings trends are down, the stock has
nothing going for it. But if the support trend is down despite the
fact that the earnings are going up, the market may be generally bad
and this may be a good time to buy the stock. In contrast, when the
support and earnings trends are both rising, the stock may have al
ready made its move and thus may be overpriced, while a rising support
trend in conjunction with declining earnings may indicate that the
smart money knows the earnings will be up next year and the stock
may be a very good buy."
Slovic 16
(c) The VOL x SUPP effect. Why is rising volume viewed as a
favorable indicator when it occurs in conjunction with a rising support
level and viewed as a relatively unfavorable sign when it occurs
with a stock whose price is declining?
"A stock that is declining on relatively low volume is considered
to be strong. People have enough confidence in it to hang on to it,
rather than sell. If price declines on high volume the story is dif
ferent. Everyone is selling and the prospects are thought to be very
poor. Similarly, if volume is down on a stock that has been appreciating
in price, confidence in that stock's future must be low, in contrast
with a stock that is rising because many people are buying it (high
volume)."
Discussion
The results of the present study indicate that the ANOVA technique
has considerable promise as a device for describing and furthering
the understanding of complex judgment processes. It is likely that
this technique can provide even the expert with new insight into his
inferential processes. Furthermore it might also be a valuable teach
ing device that would enable "trainees" to see exactly how their own
processes differ from that of their expert model (see Todd £ Hammond,
1965, for a related idea). Imagine the difficulty of asking the expert
to describe his judgment process in detail, obtaining a series of des
criptive paragraphs such as those given above to describe interactions,
and then trying to fit all these together in a way that would enable
Slovic 17
you to emulate his judgments. The task is extremely difficult if
not impossible—yet this is a common way in which expertise is communicated
However, such introspective comments become considerably more helpful
when they are accompanied by the precise quantitative descriptions
provided by the ANOVA technique.
The present results are important in another way. They provide
experimental evidence to support the commonly believed notion that
judges use information configurally. The results of previous studies,
most of which used less direct methods and percentage of variance
indices to infer the importance of configural processes, have led a
number of workers to assert that humans are predominantly linear
information processors (see discussions of this issue by Hoffman, 1968;
and Goldberg, in press). It is now clear that substantial configural
processing of information does occur and can readily be detected by
the ANOVA technique.
Slovic 18
References
Cochran, W. G., & Cox, G. M. Experimental designs. (2nd ed.) New
York: Wiley, 1957.
Garland, L. H. Studies on the accuracy of diagnostic procedures.
American Journal of Roentgenology, Radium Therapy, and Nuclear
Medicine, 1959, 82, 25-38.
Garland, L. H. The problem of observer error. Bulletin of the New
York Academy of Medicine, 1960, 36, 569-584.
Goldberg, L. R. Simple models or simple processes?: Some research
on clinical judgments. American Psychologist, in press.
Green, B. F. Descriptions and explanations: A comment on papers by
Hoffman and Edwards. In B. Kleinmuntz (Ed.), Formal representation
of human judgment. New York: Wiley, 1968. Pp. 91-98.
Hammond, K. R., Hursch, C. J., 6 Todd, F. J. Analyzing the components
of clinical inference. Psychological Review, 1964, 71, 438-456.
Hayes, J. R. Strategies in judgmental research. In B. Kleinmuntz (Ed.),
Formal representation of human judgment. New York: Wiley, 1968.
Pp. 251-259.
Hays, W. L. Statistics for psychologists. New York: Holt, Rinehart, £
Winston, 1963.
Hoffman, P. J. The paramorphic representation of clinical judgment.
Psychological Bulletin, 1960, 57, 116-131.
cl . 19
Slovic
Hoffman, P. J. Cue-consistency and configurality in human judgment.
In B. Kleinmuntz (Ed.), Formal representation of human judgment.
New York: Wiley, 1968. Pp. 53-90.
Hoffman, P. J., Slovic, P., £ Rorer, L. G. An analysis of variance
model for the assessment of configural cue utilization in clinical
judgment. Psychological Bulletin, 1968, in press.
Kleinmuntz, B. The processing of clinical information by man and
machine. In B. Kleinmuntz (Ed.), Formal representation of human
judgment. New York: Wiley, 196 8. Pp. 149-186.
Lusted, L. B. Logical analysis and roentgen diagnosis. Radiology,
1960, 74, 178-193.
Todd, F. J., £ Hammond, K. R. Differential feedback in two multiple-
cue probability learning tasks. Behavioral Science, 1965, 10,
429-435.
20
Slovic
Footnotes
1. This research was supported by Grants MH 04439 and MH 12972
from the United States Public Health Service. Computing assistance
was obtained from the Health Sciences Computing Facility, UCLA.
Portions of this work were presented at the meetings of the Western
Psychological Association, San Diego, March, 1968.
2. The author wishes to thank Terry Ashwill for his invaluable
assistance in the design of the study and for his participation as a
subject, Robert Kraus for serving as the second subject, Jerry Solomon
for his assistance in analyzing the data, and Sarah Lichtenstein and
Leonard Rorer for their comments on the manuscript.
T
a
b
l
e
1
The
Re
lat
ive
Im
por
tan
ce
of
the
11
Fa
ct
or
s
an
dT
hei
rS
ign
ifi
can
tI
nt
er
ac
tio
ns
for
Br
oke
rA
De
sc
ri
pt
io
n
o
f
L
e
v
e
l
s
M
e
a
n
J
u
d
g
m
e
n
t
M
a
gn
it
ud
e
o
f
F
a
c
t
o
r
Le
ve
l
1
Le
ve
l
2
Le
ve
l
1
Le
ve
l
2
Ef
fe
ct
M
e
a
n
Sq
ua
re
M
a
i
n
E
f
f
e
c
t
s
Y
ie
ld
(Y
LD
)
Ne
ar
Te
rm
Pr
os
pe
ct
s
(N
TP
)
Ea
rn
in
gs
Qu
ar
te
rl
y
Tr
en
d
(E
QT
)
Pa
st
Ye
ar
's
Pe
rf
or
ma
nc
e
(P
YP
)
Pr
of
it
Ma
rg
in
Tr
en
d
(P
MT
)
Ea
rn
in
gs
Ye
ar
ly
Tr
en
d
(E
YT
)
Pr
ic
e/
Ea
rn
in
gs
Ra
ti
o
(P
ER
)
Nu
mb
er
of
Sh
ar
es
Ou
ts
ta
nd
in
g
(S
O)
Re
si
st
an
ce
Tr
en
d
(R
ES
)
Su
pp
or
t
Tr
en
d
(S
UP
P)
Sa
le
s
Vo
lu
me
Tr
en
d
(V
OL
)
I
n
t
e
r
a
c
t
i
o
n
s
E
Y
T
x
S
U
P
P
Y
L
D
x
P
M
T
R
E
S
x
S
U
P
P
R
E
S
x
V
O
L
S
U
P
P
x
V
O
L
R
E
S
x
S
U
P
P
x
V
O
L
e
r
r
o
r
_
_
.
Su
m
o
f
e
ff
ec
ts
o
v
e
r
t
he
s
ta
ti
st
ic
al
ly
s
ig
ni
fi
ca
nt
fa
ct
or
s
L
o
w
P
o
o
r
D
o
w
n
P
o
o
r
D
o
w
n
D
o
w
n
P
o
o
r
F
e
w
D
o
w
n
D
o
w
n
D
o
w
n
H
ig
h
G
o
o
d
U
p
G
o
o
d
U
p U
p
G
o
o
d
M
a
n
y
U
p
U
p U
p
5
.
5
6
4
.
5
3
4
.
8
7
5
.
4
0
5
.
4
4
5
.
5
6
4
.
8
0
5
.
7
0
5
.
5
0
5
.
3
9
5
.
6
9
5
.
6
7
6
.
7
0
6
.
3
6
5
.
8
3
5
.
8
0
5
.
6
7
6
.
4
4
5
.
5
3
5
.
7
3
5
.
8
4
5
.
5
5
.
1
1
2
.
1
7
1
.
4
9
.
4
3
.
3
6
.
1
1
1
.
6
4
.
1
7
.
2
3
.
4
5
.
1
4
.
3
6
.
3
9
.
4
8
.
3
6
.
3
9
.
4
2
.
4
1
5
1
.
0
*
*
7
0
.
5
*
*
5
.
7
*
4
.
1
*
.
4
8
6
.
1
*
*
1
.
0
1
.
8
6.
6
*
*
.
6
4
.
1
*
4
.
9
*
7.
5
*
*
4
.
1
*
4
.
9
*
5
.
7
*
1
.
0
(m
ai
n
e
ff
ec
ts
)
6.
54
(7
3%
)
(i
nt
er
ac
ti
on
s)
2.
40
(2
7%
)
8
.
9
4
"
Ba
se
d
up
on
the
deg
ree
to
wh
ich
the
me
an
jud
gme
nt
ch
an
ge
s
as
the
fa
ct
or
cha
nge
s.
*
n
<
.
0
5
*
*
D
<
.
0
1
%
o
f
V
a
r
i
a
n
c
e
(u
>2
)
.
0
0
1
.
3
3
4
.
1
5
6
.
0
1
3
.
0
0
9
.
0
0
1
.
1
9
0
.
0
0
2
.
0
0
4
.
0
1
5
.
0
0
1
.
0
0
9
.
0
1
1
.
0
1
7
.
0
0
9
.
0
1
1
.
0
1
3
T
a
b
l
e
2
Th
e
Re
la
ti
ve
Im
po
rt
an
ce
of
th
e
11
Fa
ct
or
s
an
d
Th
ei
r
Si
gn
if
ic
an
t
In
te
ra
ct
io
ns
fo
r
Br
ok
er
B
F
a
c
t
o
r
M
a
i
n
E
f
f
e
c
t
s
Y
ie
ld
(Y
LD
)
Ne
ar
Te
rm
Pr
os
pe
ct
s
(N
TP
)
Ea
rn
in
gs
Qu
ar
te
rl
y
Tr
en
d
(E
QT
)
Pa
st
Ye
ar
's
Pe
rf
or
ma
nc
e
(P
YP
)
Pr
of
it
Ma
rg
in
Tr
en
d
(P
MT
)
Ea
rn
in
gs
Ye
ar
ly
Tr
en
d
(E
YT
)
Pr
ic
e/
Ea
rn
in
gs
Ra
ti
o
(P
ER
)
Nu
mb
er
o
f
Sh
ar
es
Ou
ts
ta
nd
in
g(
SO
)
Re
si
st
an
ce
Tr
en
d
(R
ES
)
Su
pp
or
t
Tr
en
d
(S
UP
P)
Sa
le
s
Vo
lu
me
T
r
e
n
d
(V
OL
)
I
n
t
e
r
a
c
t
i
o
n
s
Y
L
D
x
P
E
R
E
Y
T
x
P
E
R
P
Y
P
x
P
M
T
R
E
S
x
V
O
L
S
U
P
P
x
V
O
L
e
r
r
o
r
S
u
m
o
f
e
f
f
e
c
t
s
o
v
e
r
t
he
s
ta
ti
st
ic
al
ly
s
ig
ni
fi
ca
nt
fa
ct
or
s
M
e
a
n
J
u
d
g
m
e
n
t
L
e
v
e
l
1
L
e
v
e
l
2
3
.9
1
4
.0
2
3
.4
1
4
.5
2
3
.7
5
4
.1
7
3
.8
9
4
.0
3
3
.2
6
4
.6
6
2
.9
1
5
.0
2
3
.1
2
4
.8
0
4
.0
3
3
.8
9
3
.5
0
4
.4
2
3
.6
1
4
.3
1
3
.8
3
4
.0
9
M
a
g
n
it
ud
e
o
f
Ef
fe
ct
3
.
1
1
1
.
1
1
.
4
2
.
1
4
1
.
4
0
2
.
1
1
1
.
6
8
.
1
4
.
9
2
.
7
0
.
2
6
.
3
0
.
5
5
.
3
3
.
3
9
.
4
2
(m
ai
n
e
ff
ec
ts
)
8.
34
(8
1%
)
(i
nt
er
ac
ti
on
s)
1.
99
(1
9%
)
1
0
.
3
3
M
e
a
n
S
q
u
a
r
e
.
4
4
0
.
0
*
*
5
.
7
*
*
.
6
6
1
.
9
*
*
1
4
2
.
4
*
*
8
9
.
4
*
*
.
6
2
7
.
2
*
*
1
5
.
8
*
*
2
.
3
2
.
8
*
9.
6
*
*
3
.
4
*
4.
9*
*
5
.
7
*
*
0
.
6
aB
as
ed
up
on
th
e
de
gr
ee
to
wh
ic
h
th
e
me
an
jud
gme
nt
ch
an
ge
s
as
th
e
fa
ct
or
ch
an
ge
s.
*
<
.
0
5
*
*
<
.
0
1
%
o
f
V
a
r
i
a
n
c
e
(0
)2
)
.
0
0
0
.
0
8
2
.
0
1
1
.
0
0
0
.
1
2
9
.
2
9
9
.
1
8
7
.
0
0
0
.
0
5
6
.
0
3
2
.
0
0
3
0
0
5
0
1
9
,
0
0
6
,
0
0
9
,
0
1
1
7
9
9
,
0
5
0
,
8
4
9
Resistance Level
--Aqw?!
Support level
..J»
Volume Trend
up
Past Year's Performance
good
Common Share Earnings
Quarterly Trend
up
COMPANY NUMBER
Yield
_high
Prospects
Near Term
good
Profit Margin Trend
down
INCOME STATISTICS
Earnings/Share Yearly Trend PE Ratio Comparison
up good
CAPITALIZATION
Common Stock Outstanding
few shares
Figure 1. Example of a typical stimulus company
I
25
20—
o
o
15--
O
UJ
V
Z
<
ac
O
a.
2
10--
5 —
« • BROKER A
• . BROKER B
+ f i —+ +
YLD NTP EOT PYP PMT EYT PER SO RES SUPP VOL
FACTOR
Figure 2. The relative importance of each factor for Brokers
A and B.
o
o
O
LU
u
z
<
2
40-r
30--
20--
10-
YLD NTP EOT PYP PMT EYT PER
FACTOR
STRENGTH OF EFFECT
SUBJECTIVE WEIGHT
— • U)2INDfX
SUPP VOL
Figure 3. Comparison between the strength of effect index,
subjective weight, and w2 for Broker A.
40 T
O 30
o
z
<
ac
O
5
20 ••
10 -•
YLD NTP EQT
STRENGTH OF EFFECT
SUBJECTIVE WEIGHT
CD2 INDEX
RES SUPP VOL
Figure 4. Comparison between the strength of effect index,
subjective weight, and u>2 for Broker B.
z
ui
u*
O
o
D
Z
<
z
Ui
ui
o
o
D
z
<
z
Ui
o
Q
3
z
<
6.0-r
5.5--
5.0
6.0-r
5.5 —
5.0
6.0
5.5
5.0
H—
LOW
YLD
DOWN
EYT
PMT DOWN
PMT UP
+•
HIGH
SUPP TREND UP
SUPP TREND DOWN
UP
SUPP TREND UP
SUPP TREND DOWN
DOWN UP
VOL TREND
Figure 5. Graphical representation of selected interaction
effects for Broker A.