Bullish or Bearish?

Roger G. Clarke and Meir Statman

The sentiment of newsletter writers, whether bullish or bearish, does not
forecast future returns, but past returns and the volatility of those returns
do affect sentiment. High returns over four-week periods are associated
with a migration of newsletter writers from the bearish camp into the
bullish camp. High returns over periods of 26 and 52 weeks are associated
with “nervous bullishness”—a migration of newsletter writers from the
bearish camp into both the bullish and the correction camps. High volatility,
instead of scaring newsletter writers into bearishness, reduces the effects
of both positive and negative returns on sentiment. Also, contrary to a
popular hypothesis, the crash of 1987 had no significant effect on the
pattern of forecasts.

I
nvestors are often divided into two groups— quent stock returns. Followers of newsletter writers
smart “information” traders and not-so- degrade their investment performance in two
smart “noise” traders. Noise traders commit ways. First, they pay money for newsletters whose
cognitive errors as they make their financial advice is no better than a free toss of a coin. Second,
decisions; information traders do not. Information they move away from the optimal trade-off
traders make the right decisions, and in the long between risk and return in their strategic asset allo-
run, they win at the expense of noise traders. But cation in favor of faulty tactical asset allocation.
who are the information traders and who are the The forecasts of newsletter writers are faulty.
noise traders? And how do they forecast stock But how do they form their faulty forecasts? The
returns? This article attempts to answer these nature of forecasts is important because forecast
questions. patterns affect returns, volatility, and trading vol-
ume. In the framework suggested by De Long,
The Issues Shleifer, Summers, and Waldmann (1990), noise
traders are positive-feedback traders, traders who
Noise traders lose because they use trading rules
forecast continuations of past returns. Shefrin and
that degrade investment performance rather than
Statman (1994) showed that the results of De Long
improve it. Few investors pin the noise trader label
et al. depended crucially on whether noise traders
on themselves; others call them that. Investors Intel-
are indeed positive-feedback traders. The effect is
ligence, for example, compiles the forecasts of writ-
quite different if noise traders expect reversals
ers of investment newsletters about future stock
rather than continuations of past returns. More-
returns into a Bullish Sentiment Index, which is
over, the pattern of forecasts may change from
promoted as a contrary indicator; that is, users of
continuations to reversals, perhaps as a result of
the index turn bullish as the Bullish Sentiment
dramatic, memorable events in the stock market.
Index turns bearish. So, users of the Bullish Senti-
The market crash of 1987 was dramatic and
ment Index pin the label “noise trader” on writers
memorable, and it seems to have exerted a pro-
of investment newsletters.
found effect on the prices of stock index options.
Solt and Statman (1988) found that writers of
Option prices in the post-1987 period exhibit a
investment newsletters deserve their noise trader
“skew” consistent with a high likelihood of another
label; they found no statistically significant rela-
crash. Rubinstein (1994) dubbed this phenomenon
tionship between the sentiment of investment
“crash-o-phobia.” Did the crash of 1987 affect fore-
newsletter writers, bullish or bearish, and subse-
casts in other ways? One argument is that the quick
recovery from the crash of October 1987 and the
even quicker recovery from the mini-crash of Octo-
Roger G. Clarke is chair of Analytic TSA, and Meir ber 1989 taught investors that declines in stock
Statman is a professor in the finance department at Santa prices should be viewed as buying opportunities,
Clara University. not as beginnings of long bear markets.

May/June 1998 63
Financial Analysts Journal

We investigated the patterns of forecasts that from within because, first and foremost, we
investment newsletter writers make, the effect of never second-guess our indicators. We have
historical returns and volatility on the patterns, and been publishing The Chartist since 1969 and,
the effects of the crash of 1987. quite frankly, this is not the first time that we
have been out of sync with the market.
Investors Intelligence classifies newsletter writ-
The Sentiment Data ers as bullish, bearish, or expecting a correction.
Our data for newsletter writer sentiment are from The last category consists of newsletter writers who
the Bullish Sentiment Index, which is included in expect a decline in stock prices soon but an increase
Investors Intelligence, a publication of the invest- in stock prices in the long term. Investors Intelligence
ment service company Chartcraft. A June 5, 1995, measures bullish versus bearish sentiment in its
report in Investors Intelligence states that the Bullish Sentiment Index, which is the percentage of
Bullish Sentiment Index is based on the analysis bullish newsletter writers relative to the sum of
of more than 140 independent (non-brokerage- bullish and bearish newsletter writers. For exam-
house) stock market newsletters. Examples of ple, on May 5, 1995, the S&P 500 was at 520.12 and
newsletters analyzed are the Zweig Forecast, the the DJIA was at 4343.40. On that date, 40.3 percent
Cabot Market Letter, and the Todd Market Timer. For of newsletter writers were bullish and 34.5 percent
more than 30 years, claims the report, Investors were bearish. The Bullish Sentiment Index thus
Intelligence has found the Bullish Sentiment Index stood at 53.9 percent. The Bullish Sentiment Index
to be a contrary opinion indicator; that is, high does not take forecasts of corrections into account,
bullish sentiment, as measured by the index, but Investors Intelligence reported that 25.2 percent
indicates that the market is making a top. of newsletter writers on that date expected a
Newsletter writers, like all investors, are correction.
influenced by many factors as they make their Investors Intelligence publishes the forecast data
forecasts of future returns. For example, the Zweig approximately one week after newsletter writers
Forecast newsletter, quoted by Investors Intelligence, express their forecasts in their newsletters. Because
considers the advance/decline (A/D) figures as our interest was in the forecasts of the newsletter
one factor. The May 19, 1995, newsletter noted that writers, we used the newsletter issue dates in our
the A/D figures were positive in 20 of the previous study. The sentiment figures were compiled by
22 weeks and related that factor to expectations Investors Intelligence every month in 1963, every two
about stock returns: weeks from 1964 to 1969, and every week since
We went back to 1928 and checked out all cases then. Our analysis began with the forecasts of news-
when the weekly A/D was positive in even 17 letter writers on February 7, 1964, and ended with
out of 20 weeks, a mark we hit this time on April the forecast of newsletter writers on May 19, 1995.
14. There were only 11 such prior instances.
After the 17-of-20 figure was struck, the S&P
500 Index gained 4.7 percent in the next 3 Value of the Forecasts
months, 7.6 percent in 6 months, and 13.6 per- Solt and Statman found, using the data through
cent in 12 months vs. the normal 12-month gain 1985, that the Bullish Sentiment Index is useless as
of just 5.2 percent. Indeed, since April 14, the a forecast, straight or contrary, of future stock
S&P has already tacked on some 3.5 percent. returns. They found no statistically significant rela-
For The Chartist newsletter, bond yields and tionship between the index and subsequent stock
price–dividend ratios of the DJIA are factors that returns. We extended the time period through 1995
influence forecasts. In the June 1, 1995, newsletter, and found that the passage of time since 1985 did
as quoted by Investors Intelligence, they noted that, nothing to improve the forecasting ability of news-
in retrospect, their recent forecasts had turned out letter writers. As Figure 1 and Table 1 show, we
to be too bearish: found no statistically significant relationship
We feel much like a weatherman telling every- between the level of the Bullish Sentiment Index
body that it is about to rain any minute; but (BSI) and subsequent S&P 500 returns for nonover-
when we open the door and look outside, it’s lapping periods of 4 weeks, 26 weeks, or 52 weeks.
85 degrees and not a cloud in the sky! Obvi-
Investors Intelligence has argued that the BSI is
ously, this market has gone against us. So much
useful as a contrary indicator when its level is very
so, that in a recent Hotline we indicated our
intention of going back to the drawing board in high or very low, but that argument is not consis-
an attempt to find out what went wrong. . . . In tent with the evidence. When we compared total
the meantime, we will admit that the pressure S&P 500 returns for the four-week periods follow-
to recommend stocks has been intense. How- ing the top 30 percent BSI readings with the S&P
ever, the pressure we have felt has not been 500 returns for the four-week periods following the

64 Association for Investment Management and Research

 Bullish or Bearish?

Figure 1. Scatterplots of Bullish Sentiment bottom 30 percent BSI readings, we found no evi-
Index and S&P 500 Returns in Subse- dence of the BSI’s usefulness as a contrary indica-
quent, Nonoverlapping Periods tor. As Table 2 shows, the mean four-week return
20 following the bottom BSI reading is 0.58 percent,
which is indeed higher than the 0.46 percent mean
return following the top BSI readings, but the stan-
Subsequent S&P 500 Returns (%)

dard deviations of returns within both the top and

10 the bottom groups are so high that the difference
between the two means is far from statistically
significant. Results for the 26-week and 52-week
0
periods were similar.

Forecast Patterns
–10 Newsletter writers have pronounced opinions
about the future direction of the stock market, even
though no statistically significant relationship
exists between their forecasts and future returns.
–20 Why don’t they recognize that the patterns they see
0 20 40 60 80 100
BSI at Beginning of 4-Week Period
in stock prices are, in fact, illusory?
Investors and newsletter writers who perceive
40 patterns in stock returns that are, in fact, random,
are not alone. The tendency to identify patterns in
random data has been observed in many settings.
30
Subsequent S&P 500 Returns (%)

As Gilovich, Vallone, and Tversky (1985) noted,

intuitive perceptions of randomness depart sys-
20 tematically from the laws of statistics that underlie
randomness. People apparently expect that if a
10 series is random, the essential characteristics of
randomness will manifest themselves not only in
large samples but also in small ones. So, when they
0
find patterns in a small sample of the series, they
reject the possibility that the series is random. For
–10 example, knowing that coin tosses, given a large
series of tosses, generate roughly half heads and
–20
half tails, people expect short sequences of coin
20 40 60 80 100 tosses to contain roughly half heads and half tails.
BSI at Beginning of 26-Week Period Large deviations from the half-and-half propor-
tions are common in short sequences of coin tosses,
40 however; much more common in fact than in peo-
ple’s perceptions of fact. Tversky and Kahneman
(1971) described these common perceptions as a
Subsequent S&P 500 Returns (%)

belief in the “law of small numbers,” an erroneous

20
belief that the law of large numbers also applies to
small numbers.
People see patterns in random series, but which
0 patterns do they see? Do they predict continuations
of past numbers, or do they predict reversals?
Researchers have found that predictions are highly
sensitive to context. For example, when told that a
–20
coin has come up heads in three tosses in a row, most
people predict a reversal for the next toss (this pre-
diction tendency is known as “gambler’s fallacy”).
–40 When told that a basketball player has hit three
20 40 60 80 100
baskets in a row, however, most people predict
BSI at Beginning of 52-Week Period
continuation, in the belief that the player has a “hot

May/June 1998 65
Financial Analysts Journal

Table 1. Relationship between the BSI and S&P 500 Returns in Subsequent, Nonoverlapping Periods
(t-statistics in parentheses)
Coefficient of
Dependent Variable: Independent Independent Durbin–Watson Number of
S&P 500 Returns Intercept Variable Variable: BSI Adjusted R2 Statistic Observations
In the subsequent 0.98 –0.01 At beginning 0.00 1.99 409
4-week period (1.28) (–0.59) of a 4-week period
In the subsequent 5.20 –0.03 At beginning 0.00 2.15 62
26-week period (1.03) (–0.32) of a 26-week period
In the subsequent 16.88 –0.15 At beginning 0.00 2.17 32
52-week period (1.78) (–1.01) of a 52-week period

hand.” Gilovich, Vallone, and Tversky found that imply that one group of investors is forecasting
basketball players, coaches, and fans believe in the continuation of past stock returns while the other
“hot hand” in basketball even though series of bas- is forecasting reversal. It may be, for example, that
ketball shots follow a random walk. Why do most all investors recognize that there is no link between
people predict reversals in the case of coin tosses past stock returns and future stock returns and they
and predict continuations in the case of basketball are bullish or bearish on the basis of other informa-
shots? Gilovich, Vallone, and Tversky attributed tion, such as the state of the business cycle or mea-
differences between predictions of continuation and sures of valuation. Still, past returns seem to affect
predictions of reversal to differences in perceptions forecasts of future returns.
about the underlying process in the series. For Experiments provide most of the evidence on
example, it is hard to imagine a credible factor that perceptions of the link between past and future
could create a link between successive coin tosses, stock returns. The experiments demonstrate that
but many factors—confidence, fatigue, and so on— perceptions of the process that underlies stock
could create a link between successive shots by a returns are highly sensitive to setting. In one exper-
basketball player. iment, Andreassen (1988) presented some subjects
Do forecasters’ perceptions about the underly- with a series of past levels of stock prices and pre-
ing processes of stock markets lead them to predict sented other subjects with a series of past stock
reversals or continuations? Common investment returns (i.e., changes in stock prices). Subjects pro-
proverbs provide no answers because they reflect vided with levels of stock prices traded as though
diametrically opposite perceptions of the underly-
they expected reversals in stock prices; subjects
ing process. For each proverb that implies one
provided with stock returns traded as though they
should expect reversals (e.g., “trees don’t grow to
expected continuations.
the sky”), there is a proverb implying that continu-
ations are the rule (e.g., “don’t fight the tape”). In another study, Andreassen (1987) presented
There is no necessary link between being bull- past returns to one group of subjects and presented
ish or bearish and the tendency to forecast contin- the same returns to another group but added to the
uations or reversals. At equilibrium price, the return data news stories—positive news accompa-
quantity of stocks demanded is always equal to the nying positive returns and negative news accom-
quantity supplied and the number of bullish dol- panying negative returns. He reported that the
lars that change hands is equal to the number of tendency to forecast continuations was higher in
bearish dollars. The fact that some investors are the group that received news stories than in the
bullish at each point and others are bearish does not group that received no news stories.

Table 2. Mean S&P 500 Returns in Periods Following High and Low BSI Readings
4-Week Period 26-Week Period 52-Week Period
S&P 500 returns following top 30 percent BSI readings
Mean (%) 0.46 1.99 5.33
Standard deviation (%) 3.08 8.37 15.79
Number of observations 122 18 9
S&P 500 returns following bottom 30 percent BSI readings
Mean (%) 0.58 3.15 7.41
Standard deviation (%) 4.95 12.88 17.02
Number of observations 122 18 9
t-Statistic of the difference between means –0.23 –0.32 –0.27

66 Association for Investment Management and Research

 Bullish or Bearish?

Andreassen and Krause (1990) used another casting a “bounce,” bullish in their outlook for the
set of experiments to focus on the “salience” (i.e., short term but bearish for the longer term. Experi-
visibility or conspicuousness) of stock returns and ments by De Bondt (1991) are consistent with ner-
the effect of variance of stock returns on that vous bullishness and nervous bearishness. Subjects
salience. They found that high variance reduces the who predicted increases in stock prices acknowl-
salience of returns because it obscures both high edged a high likelihood of a correction, and subjects
and low returns. The reduction in the salience of who predicted decreases in stock prices acknowl-
returns reduces the tendency to forecast continua- edged a high likelihood of a bounce. We find that
tions. Subjects who were presented with a period newsletter writers, as a group, display nervous
when the overall return was positive but this posi- bullishness, although we lack the data to examine
tive overall return was accompanied by high vari- whether they also display nervous bearishness.
ance of returns tended to be less bullish than High returns over a four-week period are asso-
subjects presented with the same total returns over ciated with a migration of newsletter writers from
a period when the returns were accompanied by the bearish camp into the bullish camp, with almost
low variance within the period. Similarly, subjects no effect on the correction camp. This finding is
presented with a series of negative returns accom- consistent with forecasts of future stock returns as
panied by high variance were less bearish than continuations of past returns. As Table 3 and Figure
subjects presented with the same overall negative 2 indicate, a 1 percent increase in S&P 500 returns
returns but the returns were accompanied by low over a four-week period was associated with a 1.23
variance. percentage point increase in the number of bulls.
Investors in real markets have access to The increase in the number of bulls came almost
information that is not controlled as it is controlled entirely from a 1.18 percentage point decrease in
in experiments. Investors in real markets have the number of bears. Only a tiny portion, 0.05 per-
information on prices, returns, variances, and news centage point, came from a decrease in the number
stories. What affects the forecasts of investors in of newsletter writers expecting a correction. (The
real markets? sum of the three must, of course, be zero.) When
longer time periods are considered, however, the
pattern of forecasts is different.
Forecasts: Continuations or
High returns over 26-week and 52-week peri-
Reversals? ods are associated with “nervous bullishness,” a
Research indicates that some investors expect migration of newsletter writers out of the bearish
continuation of past stock returns and some don’t. camp not only into the bullish camp but also into
De Bondt (1993) found that individual investors the correction camp.
surveyed by the American Association of Individ- An increase of 1 percent in S&P 500 returns in
ual Investors (AAII) forecast future stock returns as a 26-week period is associated with a 0.30 percent-
though they expect continuations of past stock age point increase in the number of bulls. That 1
returns. In another study, De Bondt (1991) found percent increase in the S&P 500 returns is also
that the economists in the Livingston survey, associated with a 0.70 percentage point decrease in
however, forecast future stock returns as though the number of bears. These two pieces of data are
they expect reversals of the past stock returns. The consistent with a picture of newsletter writers fore-
reversal forecasts of the Livingston economists, by casting future returns as continuations of past
the way, proved to be no better than the continua- returns. When the newsletter writers in the correc-
tion forecasts of the AAII investors; De Bondt (1991) tion camp are considered, the picture changes. That
found no statistically significant relationship 1 percent increase in the S&P 500 returns for a
between the forecasts of the Livingston economists previous 26-week period is associated with ner-
and subsequent stock returns. vous bullishness in the form of a 0.40 percentage
Imagine newsletter writers who have watched point increase in the number of newsletter writers
the stock market go up for some time. Will they turn expecting a correction.
bullish or bearish? The categories of bullish and The effect of returns for a previous 52 weeks on
bearish are not sufficiently fine to capture all fore- newsletter writers’ forecasts of future returns is
casts. Some newsletter writers might be “nervous similar to the effect of returns of the previous 26
bulls,” forecasting a “correction.” Such writers are weeks. High returns are associated with a migra-
bearish in their outlook for the short term but bull- tion of newsletter writers from the bearish camp
ish for the longer term. Similarly, some newsletter into both the bullish camp and the correction camp.
writers who have watched the stock market go An increase of 1 percent in the S&P 500 returns is
down for some time might be “nervous bears” fore- associated with a 0.60 percentage point decrease in

May/June 1998 67
 68

Financial Analysts Journal

Table 3. Relationship between S&P 500 Returns and Changes in Newsletter Writers’ Forecasts
(t-statistics in parentheses)
Coefficient of S&P 500 Durbin–Watson Number of
Dependent Variable Intercept Returns for the Same Period Adjusted R2 Statistic Observations
4-week period
Change in percentage of bullish writers 0.64 1.23 0.22 2.45 408
(–1.36) (10.75)***
Change in percentage of bearish writers 0.69 –1.18 0.38 2.28 408
(2.27) (–15.88)
Change in percentage of writers expecting a correction –0.05 –0.05 0.00 2.43 408
(–0.12) (–0.49)

26-week period
Association for Investment Management and Research

Change in percentage of bullish writers –1.14 0.30 0.04 2.71 62

(–0.65) (1.89)*
Change in percentage of bearish writers 3.15 –0.70 0.23 2.89 62
(1.78) (–4.33)***
Change in percentage of writers expecting a correction –2.01 0.40 0.10 2.75 62
(–1.32) (2.82)***

52-week period
Change in percentage of bullish writers –1.80 0.24 0.02 1.88 31
(–0.62) (1.22)
Change in percentage of bearish writers 5.56 –0.60 0.23 2.41 31
(2.02) (–3.23)***
Change in percentage of writers expecting a correction –3.76 0.36 0.13 2.60 31
(–1.63) (2.31)**
* = significant at the 10 percent level.
** = significant at the 5 percent level.
*** = significant at the 1 percent level.
 Bullish or Bearish?

Figure 2. Changes in the Proportions of Bulls, drive newsletter writers out of the bearish camp
Bears, and Corrections Associated into the bullish camp, is the drive also affected by
with a 1 Percent Increase in S&P 500 the volatility of returns? One hypothesis is that,
Returns when returns are held constant, high volatility
1.5 scares investors into bearishness. To test the
hypothesis, we regressed the change in the BSI on
1.0 returns and on the volatility of returns. The depen-
dent variable was the change in the BSI over a four-
Change in Proportion (%)

week period, and the independent variables were

0.5
the total return on the S&P 500 over the four-week
period and the standard deviation of daily returns
0 within the four-week period.
Volatility does not scare newsletter writers
–0.5 into bearishness. If anything, as Table 4 reveals,
high volatility appears to be associated with an
–1.0
increase in bullishness. The coefficient of the stan-
dard deviation is positive, although it is not statis-
tically significant.
–1.5
If volatility does not scare newsletter writers
Over 4 Weeks Over 26 Weeks Over 52 Weeks
into bearishness, does it affect newsletter writers in
Bears Correction Bulls
other ways? Andreassen and Krause found in their
experiments on salience of stock returns that vola-
the number of bears, a 0.24 percentage point tility obscures the visibility of returns and, there-
increase in the number of bulls, and a 0.36 percent- fore, reduces the effect of past returns on forecasts
age point increase in the number of newsletter of future returns. Positive total returns over a
writers expecting a correction. period together with high daily volatility of returns
In sum, high returns for the short period are within that period made subjects less bullish than
associated with increased bullishness, decreased the same positive total returns with low volatility.
bearishness, and almost no change in expectations Similarly, negative total returns accompanied by
for a correction. In contrast, high returns for the high volatility made subjects less bearish than the
longer periods are associated with nervous same negative returns accompanied by low volatil-
bullishness: Bullishness increases and bearishness ity. In sum, high volatility obscured the absolute
decreases, as in the four-week case, but a significant magnitude of returns, positive or negative.
portion of newsletter writers migrate to the camp
We set out to examine whether the Andreassen
of those who, although bullish for the long run,
and Krause results hold in real-life settings. A test
expect a correction in the short run.
of the salience hypothesis called for a separation of
periods into two groups: the group of four-week
Effect of Volatility periods with negative returns and the group of
Given that high returns over a four-week period four-week periods with positive returns. Consider,

Table 4. Relationship between Changes in the BSI and S&P 500 Returns and Standard Deviation
of Returns
(t-statistics in parentheses)
Coefficient of the
Coefficient of Standard Deviation of
Dependent Variable: S&P 500 Returns in Daily S&P 500 Returns Durbin–Watson Number of
Change in BSI Intercept the Same Period within the Same Period Adjusted R2 Statistic Observations
In four-week periods –1.89 1.55 1.32 0.34 2.38 408
(–2.17) (14.59)*** (1.37)
In four-week periods 2.14 1.47 –3.19 0.18 1.84 245
of positive returns (1.70) (7.47)*** (–1.97)**
In four-week periods –4.73 1.52 3.74 0.14 1.90 162
of negative returns (–3.55) (5.25)*** (2.63)***
** = significant at the 5 percent level.
*** = significant at the 1 percent level.

May/June 1998 69
Financial Analysts Journal

as before, a regression of the change in the BSI over The question we examined is whether newslet-
four-week periods on total returns over the periods ter writers, as this argument would suggest,
and the volatility of daily returns within the peri- changed their ways of forecasting in the post-1989
ods. The salience hypothesis predicts that the coef- period. We compared the relationship between
ficient of standard deviation will be negative for the changes in the BSI and S&P 500 returns for the 71
group of periods with positive S&P 500 returns four-week observations in the post-1989 period
because, with returns held constant, high volatility (1990 through 1995) with the relationship for the 71
should obscure the magnitude of positive returns four-week observations in the pre-October 1987
and their otherwise bullish effect on the forecasts of period (1982 through 1987). As Table 5 shows, we
future returns. We found (see the middle rows in found the relationship between the changes in the
Table 4) that the coefficient of the standard devia- BSI and S&P 500 returns for both periods to be
tion for the group of positive S&P 500 observations positive and statistically significant. The implica-
is indeed negative and is statistically significant, a tion is that, on the whole, newsletter writers did not
finding consistent with the salience hypothesis. change their ways of forecasting; they did not see
The salience hypothesis also predicts that the declines in stock prices as buying opportunities
either before or after the crash of 1987. Moreover, a
coefficient of standard deviation will be positive for
Chow test revealed the coefficient of the post-1989
the group of periods with negative S&P 500 returns
period to be not statistically different from the
because, with returns held constant, high volatility
coefficient of the pre-1987 period.
should obscure the magnitude of negative returns
We found earlier that the pattern of forecasting
and their otherwise bearish effect on the forecasts
varies with the period of past returns, so we also
of future returns. We found (see the last rows in
investigated whether the crash of 1987 affected the
Table 4) a positive and statistically significant coef-
ways newsletter writers forecasted when postcrash
ficient of the standard deviation, a finding that is periods were longer than four weeks. We did not
also consistent with the salience hypothesis. have enough postcrash data for an examination of
In sum, volatility does not scare newsletter 26-week or 52-week periods, but an examination of
writers into bearishness. Instead, volatility obscures 12-week periods, as shown in Table 5, indicated
both positive and negative returns. that the pattern of forecasts after the crash of 1987
was not different from the pattern before the crash.
Effect of the Crash of 1987 on In addition to studying changes in bullish/
bearish predictions between pre- and postcrash
Forecasts periods, which can be observed from the BSI data,
The argument that the quick recoveries from the we also wanted to study changes in writers fore-
crash of October 1987 and the mini-crash of October casting a correction, which the BSI does not include
1989 drastically changed the way investors forecast but which Investors Intelligence does report. We
stock returns posits that before October 1987, bull found that the crash of 1987 did not affect the
markets were long and so were bear markets; there- pattern of forecasts of corrections. A Chow test
fore, forecasting continuations of past returns made revealed no statistically significant difference
sense. However, the argument goes, the rapid between the patterns of correction forecasts before
recoveries after 1987 and 1989 taught investors that the crash and after it.
declines in stock prices should be regarded as buy- In sum, we found that the crash of 1987 exerted
ing opportunities, not as beginnings of protracted no statistically significant effect on the pattern of
bear markets. For example, McGee (1997) wrote: newsletter writers’ forecasts.
Three weeks ago, when the Japanese Prime
Minister Ryutaro Hashimoto rattled U.S. mar- Conclusion
kets with comments that some analysts inter- A study of the forecasts of investors is important
preted as a veiled threat to dump Japanese
for two reasons. First, the forecasts affect stock
holdings of U.S. bonds and stocks, the indus-
prices, volatility, and trading volume. Second, the
trial average plunged 192 points, its biggest
trades that investors make based on their forecasts
point-size decline since “Black Monday” in the
1987 correction. But even as the correction was
affect the welfare of the traders themselves. From
unfolding, money managers with cash to put our study of the forecasts of investment newslet-
to work were rubbing their hands with glee, ters, we conclude that the writers are unable to
adding to some of their favorite holdings with reliably forecast stock returns. We found no statis-
each drop. The following day, the average tically significant relationship between the fore-
regained the vast majority of its loss, rising casts of newsletter writers and subsequent stock
153.80 points. (p. C1) returns.

70 Association for Investment Management and Research

May/June 1998

Table 5. Relationship between S&P 500 Returns and Changes in Newsletter Writers’ Forecasts: Comparison of the Pre-October 1987 Period
with the Post-1989 Period
(t-statistics in parentheses)
Coefficient of S&P 500 Durbin–Watson Number of
Dependent Variable Intercept Returns for the Same Period Adjusted R2 Statistic Observations
Change in the BSI over 4-week Pre periods (5/7/82–9/18/87) 0.47 0.70 0.26 1.69 71
(0.57) (5.03)***
Change in the BSI over 4-week Post periods (1/5/90–5/19/95) –0.20 0.64 0.15 2.35 71
(–0.25) (3.70)***

Change in the BSI over 12-week Pre periods (6/4/82–9/18/87) –4.50 1.22 0.22 2.95 24
(–1.35) (2.76)**
Change in the BSI over 12-week Post periods (1/15/90–4/21/95) –3.55 1.65 0.61 1.95 24
(–2.72) (6.18)***

Change in percentage of writers expecting a correction over 4-week Pre periods 0.40 –0.20 0.00 2.37 71
(5/7/82– 9/18/87) (0.50) (–1.09)
Change in percentage of writers expecting a correction over 4-week Post periods 0.25 –0.17 0.00 2.35 71
(1/5/90– 5/19/95) (0.47) (–1.04)

Change in percentage of writers expecting a correction over 12-week Pre periods –2.72 0.71 0.16 2.90 24
(6/4/82– 9/18/87) (–1.17) (2.30)**
Change in percentage of writers expecting a correction over 12-week Post periods –0.33 0.13 0.00 2.47 24
(1/5/90– 4/21/95) (–0.21) (0.40)
Note: “Pre” is the pre-October 1987 period; “Post” is the post-1989 period.
** = significant at the 5 percent level.
*** = significant at the 1 percent level.

Bullish or Bearish?
71
Financial Analysts Journal

We also found that newsletter writers cannot As for the argument that the rapid recoveries
be classified merely as forecasting continuations of from the crash of 1987 and the mini-crash of 1989
stock returns or as forecasting reversals. High stock taught investors that declines in stock prices should
returns are associated with a move of some news- be regarded as buying opportunities, not as begin-
letter writers from the bearish camp into the bullish nings of bear markets, we found no effect of the 1987
camp, which is consistent with the forecasting of crash on the pattern of newsletter writers’ forecasts.
stock returns as continuations of past returns. But No statistically significant difference exists between
high stock returns are also associated with a move the pattern of forecasts before the 1987 crash and
of some newsletter writers into the correction the pattern of forecasts after the 1987 crash.
camp, which is consistent with forecasts of stock Our study of the forecasts of writers of invest-
returns as reversals of past returns. Time affects the ment newsletters contributes to the body of work
pattern of movement. High returns over four-week on the accuracy and influence of stock market fore-
periods are associated with a move from the bear- casts. Earlier work by De Bondt (1991, 1993) exam-
ish camp into the bullish camp; the effects on the ined the forecasts of economists in the Livingston
correction camp are minor. High returns over 26- survey and the forecasts of members of the AAII.
week and 52-week periods are associated with Lee, Shleifer, and Thaler (1991) studied “small
major moves into the bullish and correction camps
investors” who invest in closed-end funds, and
and away from the bearish camp.
Bernstein and Pradhuman (1994) studied brokerage
Past stock returns are not the only factor that firm professionals. Further studies of the forecasts
affects forecasts of stock returns. Volatility of stock
of investors would be useful in contributing not
returns also matters. We found that volatility does
only to our understanding of the behavior of inves-
not scare newsletter writers into bearishness.
tors in financial markets but also to our understand-
Rather, volatility affects the sentiment of newsletter
ing of the process by which the interactions of
writers through its effect on the salience, or visibil-
investors in financial markets determine stock
ity, of returns: High volatility reduces the salience
returns, volatility, and volume of trading.
of both positive and negative returns. Thus, posi-
tive total returns over a period accompanied by
high volatility within the period make newsletter
writers less bullish than the same total positive We thank Jennifer Riehl and Roopa Trivedi for assistance
returns accompanied by low volatility. Similarly, and Peter Bernstein, Richard Bernstein, Michele
negative returns accompanied by high volatility LaPlante, Hersh Shefrin, and Steven Thorley for
make newsletter writers less bearish than the same comments. Meir Statman acknowledges support from
negative returns accompanied by low volatility. the Dean Witter Foundation.

References
Andreassen, P. 1987. “On the Social Psychology of the Stock Rational Expectations.” Journal of Finance, vol. 45, no. 2
Market: Aggregate Attributing Effects and the Regressiveness (June):379–95.
of Prediction.” Journal of Personality and Social Psychology, vol. 53,
Gilovich, T., R. Vallone, and A. Tversky. 1985. “The Hot Hand
no. 3 (September):490–96.
in Basketball: On the Misperception of Random Sequences.”
———. 1988. “Explaining the Price–Volume Relationship.” Cognitive Psychology, vol. 17, no. 1 (January):295–314.
Organizational Behavior and Human Performance, vol. 41, no. 3
Lee, Charles M., Andrei Shleifer, and Richard H. Thaler. 1991.
(June):371–89.
“Investor Sentiment and the Closed-End Fund Puzzle.” Journal
Andreassen, P., and S. Krause. 1990. “Judgmental Extrapolation of Finance, vol. 46, no. 1 (March):75–109.
and the Salience of Change.” Journal of Forecasting, vol. 9, no. 4
(July–September):347–72. McGee, Suzanne. 1997. “Stocks Make Quicker Comebacks from
Slumps.” Wall Street Journal (July 14):C1.
Bernstein, R., and Satya D. Pradhuman. 1994. “A Major Change
in Our Work II: ‘Sell Side’ Indicator Gives a ‘Buy’ Signal.” Merrill Rubinstein, M. 1994. “Presidential Address: Implied Binomial
Lynch Quantitative Viewpoint:1–4. Trees.” Journal of Finance, vol. 49, no. 3 (July):771–818.
De Bondt, W. 1991. “What Do Economists Know about the Stock Shefrin, H., and M. Statman. 1994. “Behavioral Capital Asset
Market?” Journal of Portfolio Management, vol. 17, no. 2 Pricing Theory.” Journal of Financial and Quantitative Analysts,
(Winter):84–90. vol. 29, no. 3 (September):323–49.
———. 1993. “Betting on Trends: Intuitive Forecasts of Solt, M.E., and M. Statman. 1988. “How Useful Is the Sentiment
Financial Risk and Return.” International Journal of Forecasting, Index?” Financial Analysts Journal, vol. 44, no. 5 (September/
vol. 9, no. 3 (November):355–71. October):45–55.
De Long, J., A. Shleifer, L. Summers, and R. Waldmann. 1990. Tversky, A., and D. Kahneman. 1971. “Belief in the Law of Small
“Positive Feedback Investment Strategies and Destabilizing Numbers.” Psychological Bulletin, vol. 76, no. 2:105–10.

72 Association for Investment Management and Research