The Monday Effect:

A Disaggregation Analysis
J. Clay Singleton
Rollins College

John R. Wingender, Jr.

Creighton University

In this research we employ two statistical approaches not previously used to

examine the Monday effect: exploratory data analysis and content analysis.
While most previous Monday effect studies examine confirmatory analysis
statistics (means and variances), exploratory data analysis employs counting
and visual techniques to describe the empirical distributions of daily returns.
The results indicate that Monday returns are highly affected by large,
infrequent, negative outliers in both large and small cap stocks, regardless of
the shift over time in the Monday effect for large cap stocks.
Content analysis is used to investigate whether there are any micro or macro
events newsworthy enough to be cited more often by the financial press (as
reported in the Wall Street Journal) to explain these large, negative outlier
Monday markets than are cited to explain the returns for other Mondays. Our
results indicate that economic or psychological factors may be responsible for
the Monday effect. This research points out new directions for researchers to
consider when examining the causes of the Monday effect.

Introduction
Many researchers have documented that average stock returns on Mondays are
statistically negative, while returns on other days of the week are positive. Some
authors have reported that average Mondays used to be negative for large stocks and
then recently changed to neutral or even positive. Most authors agree, however, that
average negative Mondays persist in small stocks. At the same time, researchers
have not offered a consistent and convincing explanation for this effect. In fact,
almost every study offers a different explanation, as the articles in this issue attest.
To date, however, no one has characterized the underlying distribution as we do or
used the analytical techniques we present here.
Our results are consistent with those of other researchers and demonstrate that
the distribution of Monday returns shows evidence of negative outliers. That is, the
91
0747–5535/03/1400 & 1500/0091/$2.00
Copyright University of Nebraska—Lincoln
92 Singleton and Wingender

negative average returns observed by other researchers appear to come from an

underlying distribution with a negative mean and median occasionally interspersed
with large negative returns. These outliers appear sporadically in both large and
small stock indices with the available CRSP daily data from mid-1963 through the
end of 2001. For large cap stocks the negative average Monday effect has lessened
recently, but the negative outliers remain. For small cap stocks Monday’s outliers are
more numerous than for large cap stocks in all time periods and contribute to the
negative average returns. The Monday effect, therefore, appears to be a mixture of at
least two causes—an underlying tendency for negative returns and occasionally large
negative outliers. We believe these negative outliers are an important part of the
Monday effect. The purpose of this article, therefore, is to document these negative
outliers and investigate their source.
Even with the insight gained from studying these occasional outliers, we have
not found a definitive explanation. We suspect the cause or causes will continue to
be difficult to detect because few daily relevant data are available to researchers.
While it is reasonable to suppose (as previous authors have) that the Monday effect
comes from information flows reflecting economic, behavioral, or technical factors,
few data series are available to track likely suspects (e.g., changes in inflation, nega-
tive investor psychology, or price momentum) on a daily basis.
Content analysis is one way of ferreting out relationships that would be difficult
to detect with standard economic or market data. Content analysis supposes text
material contains repeated concepts that, when properly analyzed, reveal an
underlying association with the treatment of interest. Content analysis parses text
material using a structured dictionary containing entries one would expect to be
associated with the treatment. The technique compares the frequencies of the words
in the dictionary between the treatment and control groups. In our case we selected
the Wall Street Journal columns that reported what market participants said on Tues-
day caused the previous day’s large negative market movements. Our null hypothesis
is that there are no systematic differences in the factors cited by the WSJ between
Mondays with negative outliers (defined below) and other Mondays.
The statistics from our content analysis tend to reject the null hypothesis and
suggest that economic and behavioral factors (and technical factors to a much lesser
extent) may be responsible. Statistically significant conclusions are difficult, how-
ever, because of the subjective nature of content analysis. If our suspicions are
correct, though, it is satisfying that most previous researchers were pursuing relevant
explanations but were hampered (and occasionally mislead) by previously unde-
tected outliers and the difficulty of measuring the contemporaneous relationship
between negative Mondays and their likely causes.
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 93

The Index Return Data

The index return data were drawn from the CRSP daily returns file. The Stan-
dard & Poor’s 500 Index and Nasdaq value-weighted indexes were used to represent
large and small cap stocks, respectively. Returns for the Standard & Poor’s 500
Index were available from July 2, 1962 through December 31, 2001 distributed
among the days of the week as shown below in Exhibit 1. Returns on the Nasdaq
Index were available from December 15, 1972 through December 31, 2001, as
shown below in Exhibit 2. In each of these exhibits Panel B contains the subset
January 1982 through December 2001.

Exhibit 1—S&P 500 Returns by Day of the Week

Weekday Monday Tuesday Wednesday Thursday Friday
Panel A: 1962—2001
Observations 1913 2030 2017 1998 1985
Mean -0.00053 0.00048 0.0084 0.00028 0.00062
p-value 0.030 0.020 < 0.001 0.150 0.002

Panel B: 1982—2001
Observations 962 1034 1032 1015 1007
Mean 0.00029 0.00076 0.0068 0.00032 0.00043
p-value 0.470 0.010 0.020 0.300 0.170

Exhibit 2—Nasdaq Returns by Day of the Week

Weekday Monday Tuesday Wednesday Thursday Friday
Panel A: 1972—2001
Observations 1399 1499 1500 1470 1465
Mean -0.00132 -0.00023 0.01181 0.00129 0.00139
p-value < 0.001 0.47 0.012 0.011 0.011

Panel B: 1982—2001
Observations 962 1034 1032 1015 1007
Mean -0.00101 -0.00002 0.00145 0.00130 0.00116
p-value 0.03 0.95 < 0.001 0.001 0.003

Over the entire time period large cap stock returns (Exhibit 1—Panel A) on Mondays
are, on average, negative and statistically different than zero. Over the twenty year
period ending in 2001 (Exhibit 1—Panel B), however, the average Monday was
positive and statistically insignificant. As has been observed by other authors1 the
negative Monday effect disappeared from large stocks. The Nasdaq tells a different
story. For small cap stocks (Exhibit 2), Monday returns are negative on average and
statistically different than zero for the entire time period and (although a bit less sig-
nificant) for the last twenty years. Notice that the average Tuesday is also negative,
but not statistically significant.

1
Pettengill (2003) reviews the reported changes in the negative Monday effect in the section
entitled “Shifts in the Monday Effect”.
94 Singleton and Wingender

Exploratory Data Analysis

Confirmatory analysis uses parametric statistics to summarize the data while
exploratory analysis uses counting and visual techniques to present a more descrip-
tive picture of the underlying data. Exploratory data analysis focuses on the
empirical distribution of returns and highlights outliers. Tukey (1977) is the source
for the exploratory techniques in this article.

Monday Outliers
The unique properties of Mondays can best be seen by comparing exploratory
and confirmatory data analysis. Confirmatory statistics (such as the means and t-tests
in Exhibits 1 and 2) can be overly influenced by outliers (they are non-resistant,
while exploratory measures are generally resistant). The characteristic of resistance
suggests a comparison of confirmatory and exploratory measures to detect the pres-
ence or absence of outliers. Exhibits 3 and 4 contain a variety of exploratory
measures, while Exhibits 5 and 6 contain Box and Whisker plots of the data.

Exhibit 3—Exploratory Data Analysis—Outliers

S&P 500 Returns by Day of the Week
Weekday Monday Tuesday Wednesday Thursday Friday
Panel A: 1962—2001
Median -0.00024 0.00011 0.00083 0.00029 0.00082
Skewness -4.17460 0.49687 0.87286 0.17402 -0.66288
Lower Inner Fence -0.01499 -0.01453 -0.01235 -0.01258 -0.01252
Observations below LIF 103 94 116 121 109
Lower Outer Fence -0.01991 -0.1940 -0.01676 -0.01687 -0.01696
Observations below LOF 53 33 46 51 56

Panel B: 1982—2001
Median 0.00073 0.00007 0.00063 0.00015 0.00074
Skewness -5.39448 0.56254 1.12043 0.08293 -0.86326
Lower Inner Fence* -0.01381 -0.01568 -0.01351 -0.01506 -0.01479
Observations below LIF 66 47 61 54 60
Lower Outer Fence** -0.01864 -0.02091 -0.01822 -0.02005 -0.01995
Observations below LOF 33 17 20 23 28
*The Lower Inner Fence (LIF) is the median less 1.5 times the interquartile range
**The Lower Outer Fence (LOF) is the median less 2.0 times the interquartile range

In Panel A, Monday is the only day of the week with a negative median.
Because the median is negative, over half the Monday returns must have been
negative, which is consistent with the negative mean over this period. The positive
median for the more recent observations in Panel B, however, shows that eliminating
the early years changes the center point of the distribution. Other authors have also
documented this positive shift, mainly by looking at Monday’s negative average
return. Our primary interest, however, is the difference in outliers between these two
time periods.
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 95

Fences are commonly used to designate outliers. A fence is defined by the

median and the interquartile range. The lower inner fence (LIF) is the median less
1.5 times the interquartile range, while the lower outer fence (LOF) is 2.0 times the
interquartile range. Similar fences can be defined above the median, although they
are not of interest here.2 These logical boundaries allow us to consider the prevalence
and strength of outliers in the daily return data. Panel A of Exhibit 3 shows that
Monday’s LIF and LOF are set at about the same level as the other days of the week
and that Mondays have about as many outliers as the other days of the week. A few
strong outliers are also consistent with Monday’s strongly negative skew. In Panel B
of Exhibit 3 using the last twenty years Monday’s negative skew is larger than for
1962 through 2001. Monday’s lower fences are smaller (a function of the median
and interquartile range). Even though Monday’s mean and median return over this
period were positive, the returns themselves are characterized as having more nega-
tive skewness (-5.39—the most negative of any weekday) with more LOF outliers
than the other weekdays (33 for Mondays versus Fridays with the second most at
28). As before with the confirmatory analysis, the exploratory perspective suggests
the negative Monday effect has changed in recent years. The center of the distribu-
tion has shifted from negative to positive. At the same time, however, the Monday’s
skewness and the returns themselves continue to evidence the presence of more
negative outliers than the other days of the week.

Exhibit 4—Exploratory Data Analysis—Outliers

Nasdaq Returns by Day of the Week
Weekday Monday Tuesday Wednesday Thursday Friday
Panel A: 1972—2001
Median -0.00049 0.00011 0.00189 0.00182 0.00201
Skewness -1.57952 0.31919 0.80264 0.27965 -0.38858
Lower Inner Fence -0.01665 -0.01489 -0.01256 -0.01254 -0.01125
Observations below LIF 100 115 124 117 111
Lower Outer Fence -0.02197 -0.01987 -0.01737 -0.01731 -0.01566
Observations below LOF 62 71 69 70 68

Panel B: 1982—2001
Median -0.00041 0.00035 0.00188 0.00184 0.00176
Skewness -1.61514 0.32148 0.90471 0.43373 -0.33612
Lower Inner Fence -0.01757 -0.01603 -0.01381 -0.01417 -0.01239
Observations below LIF 68 87 80 81 75
Lower Outer Fence -0.02323 -0.02146 -0.01901 -0.01948 -0.01705
Observations below LOF 42 55 51 52 43

2
Monday has the smallest number of outliers beyond the upper outer fence (the median plus
1.5 times the interquartile range) in both time periods. Given the negative skewness, we
investigated the negative outliers.
96 Singleton and Wingender

The Nasdaq returns in Exhibit 4 shows that the median is negative in both the full
and more recent time periods. Monday’s skewness is the most negative of the week-
days for both time periods. (Friday’s skewness is also negative but by much less.)
Monday also has the lowest lower fences (again a function of the median and inter-
quartile range), but the fewest number of observations below these fences. Even so it
is the only day with a statistically significant negative average return over both peri-
ods.
The Box and Whisker plots in Exhibits 5 and 6 further suggest that Mondays are
characterized by occasional large outliers. Box and whisker plots highlight the out-
liers in a distribution. The left and right vertical edges of the boxes correspond to the
third and first quartiles, respectively. Each box, therefore, contains 50 percent of the
observations for that day, and the medians are marked within each box by a vertical
line. The mean is indicated by the diamond within the box and lies to the right or left
of the median, as appropriate. The whiskers are drawn from highest to the lowest
observation. Each lower whisker is intersected by the two lower fences. Observations
at and beyond the fences are plotted as squares (adjacent squares overlap). Only the
lower fences are marked.
Exhibit 5 contains plots for the S&P 500. These box and whisker plots suggest
that while every day has outliers, Mondays have more large negative outliers
whether considering Panel A (1962-2001) or Panel B (1982-2001). Both plots of
Monday’s returns are anchored by a single extraordinarily negative return (-20.5
percent on October 19, 1987). Even without this extraordinary value the difference
between Mondays and the other days of the week would still appear to be heavily
influenced by a relatively few more strongly negative outliers.3
Exhibit 6 shows the same analysis for the Nasdaq. In both time periods (Panels
A and B) Mondays display outliers that are a bit more negative than for other days of
the week, and again the difference appears to be in the strength of a relatively few
more outliers.

Monday Outliers in the Combined Observations

In the previous section we compared outliers by day of the week. In this section
we consider the entire set of S&P 500 data (9,943 observations). Outliers here, there-
fore, are defined by all observations. We calculated one first quartile (Q1), one LIF,
and one LOF for each time period and then counted the outliers. In this respect
Exhibits 7 and 8 differ from Exhibits 3 and 4 where the definitions of outliers were
specific to each day of the week. Here with a whole sample perspective, we can
compare the number of outliers each day contributes to the total. Exhibit 7 displays

3
We ran the confirmatory and exploratory analyses without this observation and the
conclusions were the same—Monday’s average return was still statistically significantly
negative for the whole time period and the skewness, while reduced, was still negative.
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 97

Exhibit 5—Box and Whisker Plots

S&P 500 Returns by Day of the Week
Panel A: 1962—2001

Friday

Thursday

Wednesday

Tuesday

Monday

-0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10

Panel B: 1982—2001

Friday

Thursday

Wednesday

Tuesday

Monday

-0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10

98 Singleton and Wingender

Exhibit 6—Box and Whisker Plots

Nasdaq Returns by Day of the Week
Panel A: 1972—2001

Friday

Thursday

Wednesday

Tuesday

Monday

-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15

Panel B: 1982—2001

Friday

Thursday

Wednesday

Tuesday

Monday

-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15

Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 99

counts of the S&P 500 returns falling in the intervals defined by the fences calcu-
lated from the combined observations.
The final column of the Exhibit shows the percentage of the entire sample that
qualified for that row’s definition of outlier. Of course 25 percent of the days fall
below Q1. Panels A and B both show that about 5.5 percent and just over 2 percent
of the samples fell below the LIF and the LOF, respectively.

Exhibit 7—Outliers in the Whole Sample

S&P 500 Returns by Day of the Week
Outlier Designation All Days -
Monday Tuesday Wednesday Thursday Friday % of sample
Panel A: 1962—2001
Below Q1 545 518 472 486 454 2475 – 25.0%
% below Q1 22.0% 20.9% 19.1% 19.6% 18.3%
Below LIF 136 113 99 96 97 541 – 5.5%
% below LIF 25.1% 20.9% 18.3% 17.7% 17.9%
Below LOF 68 40 37 42 50 237 – 2.4%
% below LOF 28.7% 16.9% 15.6% 17.7% 21.1%

Panel B: 1982—2001
Below Q1 228 269 250 253 263 1263 – 25.0%
% below Q1 18.1% 21.3% 19.8% 20.0% 20.8%
Below LIF 58 59 51 56 60 284 – 5.6%
% below LIF 20.4% 20.8% 18.0% 19.7% 21.1%
Below LOF 32 18 15 23 28 116 – 2.3%
% below LOF 27.6% 15.5% 12.9% 19.8% 24.1%

The percent rows display the percentage of observations falling below Q1,
below the LIF, or below the LOF for each day of the week based on definitions
derived from the whole sample. Panel B shows the results of the same analysis for
the 5,050 observations from 1982 through 2001. These three designations: below Q1,
the LIF and the LOF, are progressively more conservative definitions of outliers. In
Panel A Monday has only a slightly larger share of observations below Q1 (22.0 per-
cent). Recall that for this time period only the average Monday return was negative.
Once we set the bar at the lower fences, however, we see that Monday contributed a
disproportionate share of these outliers in the whole sample (25.1 percent of all
observations below the LIF and 28.7 percent of all those below the LOF). The LIF
and LOF rows both provide statistically significant chi-squares beyond the 0.01 level
when these rows are tested individually against the null hypothesis of outlier fre-
quency being the same as frequency of the days themselves in the sample.4 The Q1
row does not produce a statistically significant chi-square statistic. These statistical
tests suggest that Monday has a significantly larger share of the outliers in the sam-

4
Using the sample frequency rather than allocating 20 percent to each weekday adjusts for
trading holidays.
100 Singleton and Wingender

ple than the other days of the week. In Panel B with the more recent data we see the
negative Monday effect only when the lower outer fence is used. In fact, Monday
does not contribute the highest percentage of the outliers below Q1 or the LIF. In this
sample the average Monday return was positive. Only the LOF row produces a sta-
tistically significant chi-square statistic for Monday’s contribution of outliers to all
the outliers below the LOF in the sample. For large cap stocks, Monday contributed
more strong negative outliers (below the LOF) than the other days. In the more
recent period, Monday continues to show the most outliers below the LOF despite
the documented positive shift in the underlying distribution of Monday returns.
The Nasdaq results in Exhibit 8 display a different pattern than the S&P 500
results in Exhibit 7. First, the “All Days” column shows that small caps stocks
produce proportionally more outliers below the LIF and LOF. Second, Monday has
the highest percentage of outliers for each progressively more conservative
definition. Comparing Panels A and B suggests that, unlike large cap stocks, Monday
contributes the most small cap outliers of any weekday. At the same time the average
small cap Monday returns are negative on average for both time periods. Chi-square
tests of the rows in these two panels provide statistically significantly differences
(beyond the 0.01 level) between the frequency of outliers and the frequency of the
days of the week in the samples. In contrast, for large cap stocks (Exhibit 7), only the
LOF produced a statistically significant difference in outliers by day of the week.

Exhibit 8—Outliers in the Whole Sample

Nasdaq Returns by Day of the Week
Outlier All Days—
Designation Monday Tuesday Wednesday Thursday Friday % of Sample
Panel A—1972—2001
Below Q1 449 433 333 336 282 1833 – 25%
% below Q1 24.5% 23.6% 18.2% 18.3% 15.4%
Below LIF 149 129 110 107 84 579 – 8%
% below LIF 25.7% 22.3% 19.0% 18.5% 14.5%
Below LOF 86 80 62 65 56 349 – 5%
% below LOF 24.6% 22.9% 17.8% 18.6% 16.0%

Panel B—1982—2001
Below Q1 292 289 235 242 205 1263 – 25%
% of below Q1 23.1% 22.9% 18.6% 19.2% 16.2%
Below LIF 98 91 72 77 64 402 – 8%
% below LIF 24.4% 22.6% 17.9% 19.2% 15.9%
Below LOF 61 54 44 51 46 256 – 5%
% below LOF 23.8% 21.1% 17.2% 19.9% 18.0%

Our day-by-day analysis suggests that Mondays exhibit more outliers than other
days of the week. As a proportion of the whole sample, the number of Mondays’
outliers is statistically significantly larger than the other days of the week. While the
other days of the week also exhibit outliers, they do not have the same intensity, and
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 101

their average and median returns are positive. Strong negative outliers, therefore, are
a relatively unique characteristic of Monday returns. We now turn to content analysis
to uncover some clues as to what might be responsible for these occasional large
negative outliers on Mondays. While this research may not fully explain the under-
lying negative distribution of Monday returns or the documented recent shift in large
cap stocks (Exhibit 3), it should contribute to our understanding of this complex
phenomenon.

Content Analysis
Most researchers who have investigated the negative Monday effect have
offered economic, psychological, or technical explanations. Whatever the root
causes, if Mondays are characterized by occasional outliers, then the associated
events should be newsworthy. Content analysis is an exploratory tool that looks for
an underlying commonality or structure in text.5 In this case we have used the tech-
nique to parse the Wall Street Journal column that reports what Wall Street
professionals are saying explained the previous day’s market movement. We assume
that whatever causes negative outlier Mondays should be newsworthy enough to be
mentioned more frequently on the following Tuesday and less frequently on Tues-
days following other Mondays.
Content analysis cannot operate in a vacuum. The technique looks for words and
phrases within a structure that is described by the researcher. This structure is
expressed as a hierarchical list of terms (a dictionary in the content analysis
literature) that the investigator expects to be associated with the effect. In our case
we needed to decide which terms and phrases commonly used to describe market-
related economic, psychological, and technical events should be included in our dic-
tionary.
Obviously the way we built our dictionary would largely have determined any
structure we might find and is a potential source of bias in our results. The content
analysis literature suggests structured dictionaries should not be devised by the pri-
mary researchers, especially those who have studied the text to be analyzed. Our
alternative was to convene a panel of experts to suggest concepts and factors they
would expect to be associated with the general phenomenon, in this case daily stock
market movements.6 These concepts could then be converted into dictionary entries
and the analysis directed to look for words and phrases that represented these entries.
To help us, we invited a panel of experts made up of the authors of the other
papers in this edition. We asked our panelists to assume they had agreed to write a

5
Content analysis has a wide variety of uses in other disciplines (e.g., discovering whether all
of Shakespeare’s plays had the same author).
6
Avoiding the investigator bias was discussed in the content analysis literature as early as
1959; see Pool (1959) pp. 214-216. For the expert approach see Saris-Gallhofer et al. (1978)
and a discussion of their work in Weber (1985) pp. 19-20.
102 Singleton and Wingender

daily market commentary for the business section of their local paper. Their charge
was to describe and explain, in two or three paragraphs, the factors behind the mar-
ket’s behavior since the previous day’s close. We asked them to be even-handed in
attributing market movements to both economic and behavioral factors. Each panel-
ist listed five economic and behavioral factors they would expect to regularly invoke
in their columns. We entered these factors into a lexicographical program that pro-
vided grammatical variations on the panel’s suggestions, expanding the dictionary
entries without, we believe, changing the panel’s intent. We then organized the
words and phrases hierarchically into major categories. The resulting categories are
shown in Exhibit 9, and the complete dictionary is contained in the appendix.
Throughout the remainder of our analysis we have suppressed the grammatical
variations without, we trust, a substantial loss of clarity.

Exhibit 9—Major Categories in the Content Analysis Dictionary

Economic Categories
Interest Rates Bonds
Price Levels Fiscal Policy
Fed Policy Oil
Trade Exchange Rates
Corporate Profits Technology
US Economy US Politics
Regulation
Psychological Categories
Positive Sentiment Negative Sentiment
Technical Categories
Market Actors Technical Factors

As shown in the appendix, each of these categories had a variety of words and
phrases associated with it that directed the content analysis. For example the cate-
gory “Price Levels” was further defined by the words and phrases: Consumer Price
Index, Purchaser Price Index (and their abbreviations), Overheated Economy, Infla-
tion, Wholesale Prices, and Deflation. Combinations or grammatical permutations of
these words counted as a hit.
The panel’s most significant contribution to the dictionary was the suggestion to
add technical factors. While we had not originally considered technical factors, our
subsequent analysis revealed that the Wall Street Journal occasionally invokes tech-
nical explanations such as momentum, resistance levels, and computer-generated
trading. Economic and psychological factors were anticipated and, as shown below,
useful.
The final dictionary combined the panel’s suggestions and the words and
phrases from the lexicographical program with our additions based on preliminary
tests of the text against the dictionary. We added words and adjusted phrases to make
the dictionary more consistent with the Wall Street Journal’s choice of words while,
in our opinion, maintaining the panel’s intent. While we recognize that our additions
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 103

may have been a source of bias toward finding significant categories in the text, we
believe the need to conform our dictionary to the Wall Street Journal’s style was
worth more in results than it cost in potential bias. Unfortunately, without these
adjustments, content analysis found too few occurrences of the words in the original
dictionary to draw any conclusions.7 As a final check on our dictionary, we let con-
tent analysis suggest words that it found differentiated between the two sets of
Monday articles. These suggestions did not add significantly to the results, so only
our dictionary results are reported here.
Validity is a concern to any researcher. Weber (1985, p. 18) gives the following
example of validity in content analysis: a valid content analysis assumes “there is a
correspondence between the category (economic) and the abstract concept that it
represents (concern with economic matters)” (parentheses in the original). In our
research, we assume there is a correspondence between the WSJ’s explanation of
market movements and the categories listed in Exhibit 9 and their constituent words
listed in the appendix. As the WSJ’s purpose in writing the column is to report mar-
ket professionals’ explanations of market movements, the assumption of association
is reasonable. Content analysis, however, is not capable of discerning cause and
effect; that is, the mention of the word does not necessarily mean professionals
attributed the market’s movement to the abstract concept that word or category was
designed to represent. With this cause and effect caveat, we believe that our applica-
tion of content analysis, guided by the expert panel, is valid.

Text Data
Once the dictionary was complete, we collected articles from the Wall Street
Journal (WSJ) for all available Mondays below the LIF and an identically-sized
sample of regular Mondays chosen at random. We used separate samples for the
S&P 500 and Nasdaq. Our electronic WSJ files started in 1988. Using the LIF defi-
nition of outliers, our sample consisted of 46 negative Mondays for the S&P 500 (79
percent of the 58 since 1982; see Exhibit 7, Panel B row labeled “Below LIF”) and
61 negative Mondays for the Nasdaq (62 percent of the 98 since 1982; see Exhibit 8,
Panel B, row labeled “Below LIF”). The full samples were then 92 Mondays for the
S&P 500 and 122 Mondays for the Nasdaq. We do not know if our conclusions
would have been any different if we had more articles to analyze. Our analysis of the
return data suggests, however, that a more recent sample is biased against finding
any significant association with outlier Mondays in large cap stocks because the
average over the last twenty years has been positive. Despite these drawbacks the
analysis proved compelling.

7
Our experience with content analysis reinforced one of the drawbacks cited in the literature—
replicability. Different researchers with different dictionaries might have reached different
conclusions. Despite this drawback we believe our dictionary is plausible and our results
credible.
104 Singleton and Wingender

Results
The content analysis results are summarized in the two panels of Exhibit 10.
This exhibit shows the frequencies with which the words within each category were
mentioned in the WSJ, each row’s statistical test and p-value, as well as the overall
statistical significance of the combined categories. This exhibit is a fundamental part
of our conclusions, so we cover it here in some detail.

Exhibit 10—Dictionary Categories

Frequencies and Statistical Significance
Frequency Counts
Non- Negative
Outlier Outlier
Category Mondays Mondays Total χ2 p-value
Panel A: S&P 500
Bonds 198 399 597 67.67 0.00
US Economy 11 23 34 4.24 0.04
Exchange Rates 251 475 726 69.11 0.00
Fed Policy 90 130 220 7.27 0.01
Fiscal Policy 4 11 15 3.27 0.07
Interest Rates 43 112 155 30.72 0.00
Market Actors 8 31 39 13.56 0.00
Oil 27 39 66 2.18 0.14
Price Levels 37 57 94 4.26 0.04
Corporate Profits 26 135 161 73.80 0.00
Regulation 0 1 1 1.00 0.32
Sentiment 71 222 293 77.82 0.00
Technical Factors 6 18 24 6.00 0.01
Technology 110 260 370 60.81 0.00
Trade 1 3 4 1.00 0.32
US Politics 14 40 54 12.52 0.00
Totals 897 1956 2853 393.09 0.00

Panel B: Nasdaq
Bonds 355 303 658 4.11 0.04
US Economy 30 17 47 3.60 0.06
Exchange Rates 466 316 782 28.77 0.00
Fed Policy 191 112 303 20.60 0.00
Fiscal Policy 11 7 18 0.89 0.35
Interest Rates 91 72 163 2.21 0.14
Market Actors 18 21 39 0.23 0.63
Oil 33 57 90 6.40 0.01
Price Levels 64 42 106 4.57 0.03
Corporate Profits 51 169 220 63.29 0.00
Regulation 0 0 0 NMF* NMF*
Sentiment 123 220 343 27.43 0.00
Technical Factors 14 21 35 1.40 0.24
Technology 180 458 638 121.13 0.00
Trade 5 0 5 5.00 0.03
US Politics 23 41 64 5.06 0.02
Totals 1655 1856 3511 11.51 0.00
* Not meaningful
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 105

In a paired sample such as ours, the most reasonable test is whether each cate-
gory is mentioned more often on outlier Mondays than non-outlier Mondays. We
tested this hypothesis with a chi-square statistic.8
Both Panels in Exhibit 10 show that differences between the total negative out-
liers and non-outliers for large and small cap stocks are both significant in our χ2
tests beyond the 0.001 level. Most of the individual categories also have significant
differences in frequencies.
Before we analyze Exhibit 10, it is instructive to look behind these categories to
their constituents. As an example, consider the Fed Policy category for large cap
stocks. Exhibit 10 shows that overall the words and phrases in this category were
mentioned 130 times on negative outlier Mondays versus 90 times on non-outlier
Mondays. Exhibit 11 shows that in this category only two of the constituent words
and phrases—“Fed” and “Federal Reserve” —had significantly different frequencies
between outlier and non-outlier Mondays. “Tighten” was also significantly different,
but only contributed three observations. The other words and phrases were not men-
tioned or mentioned with such similar frequencies as to be insignificantly different.
These results were typical in the sense that for Exhibit 10’s categories with signifi-
cant p-values, less than half of their individual words and phrases showed significant
differences. The categories themselves, however, showed significant differences
because the predominant frequency favored either outlier or non-outlier Mondays.
Exhibit 11, therefore, is a more detailed view of the “Fed Policy” row in Panel A of
Exhibit 10 and typical of the detail behind the remainder of the rows. We believe that
reporting statistical significance at the category level as in Exhibit 10 best reflects the
intent of our content analysis.

8
Because we have an equal number of outlier and non-outlier Mondays our null hypothesis for
each row is that the frequency counts are equal between outlier and non-outlier Mondays.
Formally we used the statistic:
2
[N i − E(N i )]2
Q= ∑
i =1
E(N i )
where Ni is the observed cell frequency and E(Ni) is the expected cell frequency (i.e., row
total/2). For example the statistic for the Bonds category in Exhibit 10 was computed using
E(Ni) = 597/2 = 298.5, N1 = 198, and N2 = 399.
(198 − 298.5) 2 (399 − 298.5) 2
The χ2 = + = 67.67
298.5 298.5
which is approximately a χ2 with 1 degree of freedom and significant beyond the 0.001 level.
106 Singleton and Wingender

Exhibit 11—Frequencies and Statistical Significance

Federal Reserve Policy Category Words and Phrases
S&P 500
Frequency Count
Non- Negative
Outlier Outlier
Word or Phrase Mondays Mondays Total χ2 p-value
Ease 6 6 12 0.00 1.00
Fed 47 73 120 5.63 0.02
Fed Policy 2 2 4 0.00 1.00
Federal Reserve 21 37 58 4.41 0.04
Intervention 3 4 7 0.14 0.71
Looser Fed Policy 0 0 0 NMF* NMF*
Loosen 0 0 0 NMF* NMF*
Monetary 2 2 4 0.00 1.00
Monetary Policy 4 2 6 0.67 0.41
Refunding 2 4 6 0.67 0.41
Reserve 0 0 0 NMF* NMF*
Tighter Fed Policy 0 0 0 NMF* NMF*
Tighten 3 0 3 3.00 0.08
Total 90 130 220 7.27 0.01
* Not meaningful

Returning to Exhibit 10, we see that most of the categories in both panels are
significantly different. From these results, we conclude that outlier and non-outlier
Mondays are different. That is, our content analysis has identified categories of
words and phrases that differentiate between outlier and non-outlier Mondays for
both large and small cap stocks. These results suggest that the market may have dif-
ferent explanations for movements on outlier and non-outlier Mondays. This is the
first dimension of our content analysis results.
We also observe some interesting differences between large and small cap
stocks. In Panel A, all the categories are mentioned much more often on negative
outlier Mondays. For half of the categories in Panel B, the opposite is true. Of the
significant categories for Panel A’s large cap stocks (all of which are predominant on
negative outlier Mondays), only Sentiment, Technical Factors, Technology, Trade,
and US Politics are also mentioned more frequently on negative outlier Mondays for
small cap stocks. (Oil is significant for small caps, but insignificant for large caps).
The other categories (Bonds, US Economy, Exchange Rates, Fed Policy, Fiscal Pol-
icy, Interest Rates, Market Actors, Price Levels, and Corporate Profits) are all more
frequently mentioned on non-outlier Mondays for small cap stocks. Apparently the
market associates these categories with large negative movements in large cap
stocks, but not with similar movements in small cap stocks. This is the second
dimension of our content analysis results. One could speculate that most of these
categories are relevant on negative Mondays for large stocks because they are usu-
ally associated with macroeconomic factors, but that supposition, while attractive, is
beyond the capabilities of the technique. Content analysis only reports that the words
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 107

and phrases in these categories are mentioned significantly more often on negative
outlier Mondays.
Like other statistical procedures, content analysis does not reveal causation. We
cannot say that because, for example, the Fed Policy category was mentioned more
often on large cap negative outlier Mondays that concerns over Fed policy caused the
large cap market to drop sharply on those days. Without analyzing the full text of the
articles, it is not possible to tell why the words and phrases in the Fed Policy cate-
gory were mentioned more often on outlier Mondays.9 We can only infer from the
preponderance of statistically significant categories what sorts of terms were associ-
ated by the Wall Street Journal reporters—by attribution to market professionals—
with the proximate events of the day. This example also illustrates how rumors about
important economic variables might (and in an efficient market probably should)
move markets but not show up in any economic time series.
The third dimension of our results concerns the only categories—Sentiment,
Corporate Profits, Oil, Technology, and US Politics—that are associated with nega-
tive outlier Mondays in small cap stocks. A tenable hypothesis is that small and large
cap stocks differ in their susceptibility to macro and microeconomic trends. Small
caps might not be influenced much by the macroeconomic categories that affected
large caps, but would respond on a microeconomic level to diminished prospects for
corporate profits and specific industries such as technology. If true, this hypothesis
could explain our results in both panels of Exhibit 10 and would be consistent with
our speculation above about large cap stocks. Again, however, this speculation is not
testable with the data we have collected or the content analysis technique.
Technical factors were suggested by our expert panel. They were mentioned by
the WSJ more frequently on outlier Mondays for both large and small cap stocks.
While the difference in frequency is significant for large cap stocks, the total number
of observations is small for both large and small cap stocks. These results do not
mean that technical factors or, for that matter, any of the other insignificant factors
are not a potential explanation. The WSJ reporters may have had a personal bias or
an editorial policy against technical analysis. The market participants consulted by
these reporters may have had their own biases. Content analysis, like other tech-
niques, cannot make inferences if the data do not contain the information, whether by
accident or design.
While we cannot claim to have found an explanation for the negative outlier
Mondays, we believe our content analysis results suggest market professionals

9
We did review the context of each occurrence of all these terms. During the time period
studied the WSJ repeatedly reported that the market was concerned that the Fed would raise
rates and deflate the rally. Hints or rumors of pending Fed action (which, in fact, did not
materialize) appeared to be enough to cause the market to fall. The point is that while we can
infer this connection (given our biases), it is unfair to expect a statistical analysis to reveal
causation.
108 Singleton and Wingender

believe economics and investor psychology are responsible. While large cap stocks
did exhibit outlier Mondays during the period for which the WSJ articles were avail-
able, our earlier evidence suggests that not as many outliers occurred during this
period as earlier. Small cap stocks, however, should provide a better test because
they exhibited a more consistent pattern of outliers over the whole period studied.
If small cap stocks are a better laboratory for investigating outliers because out-
liers occur here more regularly than for large cap stocks, our content analysis results
suggest previous researchers were on the right track. Factors that economists believe
should matter, such as corporate profits and investor behavior, are associated with
outlier Mondays. Whether these factors caused the market to decline on these days is
beyond the ability of the statistics to tell us. One way to interpret these results is to
take comfort in their familiarity—in the sense that we do not need to invoke expla-
nations beyond efficient markets or investor psychology. At the same time, some
things that we expected to be major contributors to the difference between Mondays
(such as Technical Factors and Regulation) were not. We do not know whether those
results stem from our inability to craft our dictionary skillfully enough or from the
relatively short time span covered by our WSJ articles. From a broader perspective,
however, these results are consistent with previous research and may help explain
why some of the prior researchers reached the conclusions they did.

Relationship to Previous Research

If we believe the explanation for the Monday effect lies in rational pricing
responses to information flows [in the lexicon of Pettengill (2003)], then our
evidence is consistent with an announcement effect. Firms may wait until the market
closes on Friday to release bad news. Government agencies often do the same.
Unfortunately, as Pettengill (2003, p. 12) points out “… attempts to link such
releases with the weekend effect have been only partially successful.” Damodaran
(1989), for example, fails to find a significant link between two likely suspects—
earnings and dividend announcements—and the weekend effect. Other researchers
have come to similar conclusions.
Our research suggests that while these investigators were on the right track, they
were not—and probably could not—measure the variables that moved either indi-
vidual stocks or the market. Efficient markets attempt to anticipate information flows
and should respond to rumors—on the chance they turn out to be true. If so, meas-
uring these effects will be difficult. We suspect, however, that macro effects and
rumors about them will be a fertile area for future research.

Summary and Conclusion

Previous research has reported that only on Mondays have average returns been
negative, but has not been able to identify a source of these negative returns. We
have presented evidence that Monday’s negative average return comes not only from
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 109

consistently negative returns, but also from large infrequent outliers. These outliers
have been present in both large and small cap stock returns for at least forty years.
Large cap stock averages for the last twenty years have been positive, but the out-
liers, although somewhat weaker, remain. Small cap stocks average returns on
Mondays continue to be negative and characterized by outliers.
Apparently, information that arrives in the capital markets either over the week-
end or during the day on Monday is occasionally so strongly negative that it results
in large negative returns. We have identified the dates of these negative outliers for
both large and small cap stocks. They did not appear to follow any regular pattern
and, given the rigorous efforts of previous researchers, we did not believe a search
for an association with existing economic time series would be fruitful, largely
because so few daily economic time series are available. As others before us, we
hypothesized that economic, psychological, or technical factors might be at work.
We then turned to the Wall Street Journal’s daily explanation of what moved the
market to look for an explanation. We used content analysis to parse the text of these
articles with a dictionary suggested by a panel of experts in the field. The content
analysis suggested that the causes offered by previous researchers based on eco-
nomic or psychological factors might be responsible. While we may not have
identified the cause, our results certainly suggest that previous researchers were on
the right track. Further research might profitably focus on the variables suggested
here as the search for the source of the Monday effect continues.

References
1. Damodaran, Aswath, “The Weekend Effect in Information Releases: A Study of Earnings and
Dividend Announcements,” Review of Financial Studies, 2 (1989), pp. 607-623.
2. Pettengill, Glenn, “A Survey of the Monday Effect Literature,” Quarterly Journals of Business
and Economics, 42, nos. 3 and 4 (2003), pp. 3-28.
3. Pool, Ithiel, deS., Trends in Content Analysis (Urbana: University of Illinois Press, 1959).
4. Saris-Gallhofer, I.N., W.E Saris, and E.L. Morton, “A Validation Study of Holsti’s Content
Analysis Procedure,” Quality and Quantity, 12 (1978), pp. 131-145.
5. Tukey, John W., Exploratory Data Analysis (New York: Addison-Wesley Publishing Com-
pany, 1977).
6. Weber, Robert P., Basic Content Analysis (Beverly Hills: Sage Publications, 1985).
110 Singleton and Wingender

Appendix—Dictionary Structure for Content Analysis

The dictionary allowed the content analysis routine to look for permutations and
combinations of the basic words and phrases listed here. For example any mention of
interest discount, or prime rates counted as a hit in the interest rate category. The
program also counted plurals and different parts of speech of these words.

Bonds Market Actors

Convertible Bond Traders
Corporate Computer Driven
Coupon Arbitrageurs
Debenture Oil
Government Oil Prices
Junk Petroleum
High Yield Fossil Fuel
Municipal Crude Oil
Debenture Price Levels
Treasury Obligation Consumer Price Index
Corporate Profits Purchaser Price Index
Earnings Overheated Economy
Exchange Rates Inflation
Central Bank Intervention Wholesale Prices
Dollar Deflation
Euro Regulation
Yen SEC
Fed Policy Sentiment
Ease Positive
Tighten Comfortable
Intervention Confident
Monetary Policy Fearless
Fiscal Policy Glad
Budget Deficit Good News
Budget Surplus Happy
Interest Rates Healthy
Discount Rate No Fear
Prime Rate Fearless
Bond Yields Positive
30 year Yields & Price Reassure
Long Bond Yields & Price Robust
Sunny
Upbeat
Welcome News
Quarterly Journal of Business & Economics, Vol. 42, Nos. 3 and 4 111

Negative US Economy
Bad News Bankruptcy
Bleak GDP
Confused GNP
Cringe Industrial Production
Depressing Leading Indicators
Disappointing Payroll
Fragile Housing Starts
Jitters Employment
Negative Unemployment
Nervous Jobs
Fear US Politics
Gloom Congress
Spook President
Skittish Senate
Worry House
Technical Factors
Buying Pressure
Inflection Point
Momentum
Point of Support
Selling Pressure
Point of Resistance
Technology
Compaq
Dell
IBM
Hewlett Packard
Intel
Cisco
Lucent
Trade
Balance of Payments
Trade Deficit
Purchasing Power Parity
Trading
Program Trading
Profit Taking
Stock Index Arbitrage