CONSUMER CONFIDENCE AND ASSET PRICES:
SOME EMPIRICAL EVIDENCE∗
Michael Lemmon
University of Utah
Evgenia Portniaguina
University of Oklahoma
∗
We are grateful to Avner Kalay, Uri Lowenstein, Michael Cliff, Nicholas Bollen, Scott
Linn, Bill Megginson, and the seminar participants at the University of Utah, University
of Houston, and University of Oklahoma for helpful comments and suggestions. We
thank Kenneth French, Malcolm Baker, Jeffrey Wurgler, Stefan Nagel, and Ludovic
Phalippou for providing data. All errors are ours. Corresponding author: Evgenia
Portniaguina, Michael F. Price College of Business, University of Oklahoma, 307 West
Brooks, room 205A, Adams Hall, Norman, OK 73019-4005, [email protected], (405) 3257727 (phone), (405) 325-7688 (fax).
Consumer Confidence and Asset Prices: Some Empirical Evidence
Abstract
We estimate fundamental and sentiment components of consumer confidence. In a timeseries framework, we model the returns of equity portfolios sorted on various
characteristics as a function of the market factor, allowing market beta to vary with the
fundamental component of confidence. After controlling for the time variation of betas,
we study the time variation of the pricing error with sentiment. Over the last 25 years
(which represent relatively active household participation in the equity markets),
consumer confidence forecasts returns in a manner consistent with the sentiment – based
behavioral hypothesis.
2
Introduction
Does investor sentiment affect stock prices? A number of papers provide
evidence on whether various measures of investor sentiment predict stock returns in an
attempt to explain various pricing ‘anomalies’ such as the equity premium puzzle, the
size premium, the closed-end fund puzzle, etc. Many of these papers focus on small
stock returns. The reason why small stocks are a natural candidate to be affected by
sentiment is that they are owned predominantly by individual investors, while large
stocks tend to be held primarily by institutions (e.g., Nagel (2003)). In addition, higher
trading costs and other frictions are more likely to limit arbitrage in small stocks, making
them more susceptible to variations in irrational investor sentiment.
Following this logic, Lee, Shleifer, and Thaler (1991) find that closed-end fund
discounts are correlated with small stock returns. However, Chen, Kan, and Miller
(1993) and Do ukas and Milonas (2002) provide evidence to the contrary. Neal and
Wheatley (1998) find that closed-end fund discounts and net mutual fund redemptions
predict the size premium, especially at long horizons. Brown and Cliff (2004) use the
data on investor sentiment from a survey conducted by the American Association of
Individual Investors and do not find evidence that fund discounts reflect investor
sentiment. They form a measure of sentiment using the survey data together with several
technical variables and find that sentiment has little forecasting power for near-term
returns. Baker and Wurgler (2003) design a measure of investor sentiment using several
market variables, which include fund discounts, number of IPO’s, NYSE turno ver, etc.
They show that the returns on equity portfolios with different “salient ” (e.g., age, size,
dividend payout, distress) investment characteristics are sensitive to changes in investor
1
sentiment. Baker and Wurgler report that periods of low sentiment are followed by high
returns on small, young, unprofitable, and dividend-nonpaying stocks. 1
All in all, the evidence on the issue remains controversial. This paper contributes
to the debate on the relationship between investor sentiment and equity returns using data
on consumer confidence as a measure of sentiment. Participation (either direct or
indirect) of individual households in financial markets has increased substantially over
recent years suggesting that measures of consumer confidence may be a useful barometer
of how individual investors feel about the economy and the financial markets.
Two surveys of consumer sentiment have been conducted in the United States.
One is collected by the Conference Board (the Index of Consumer Confidence) and the
other is independently conducted by the University of Michigan Survey Research Center
(the Index of Consumer Sentiment). Both surveys poll a large number of households on
their personal financial situation, their expectations regarding the U.S. economy, and
their propensity to consume major household items. The Index of Consumer Confidence
is included in the list of ten major leading economic indicators by the Conference Board,
having proven useful in predicting past recessions. Several papers (Acemoglu and Scott,
1994; Carroll et al., 1994; Bram and Ludvigson, 1998) find that consumer confidence
predicts future household spending. According to the University of Michigan Survey
Research Center, consumers anticipate changes in interest rate, unemployment, inflation,
real GDP, and house sales (from contemporaneous to nine months ahead) with
1
Baker and Wurgler (2003) report a “U-shaped difference” for value premium conditional on investor
sentiment. When sentiment is low, their evidence suggests that both extreme BM -sorted portfolios (high
growth and high value) have higher returns following a period of low sentiment, relative to the medium –
BM portfolios. We do not find that growth and value stocks respond differently to changes in consumer
confidence.
2
correlations ranging from 0.74 to 0.9. Figure 1 shows that consumer expectations are
good predictors of business cycle peaks and troughs.
While being well known as an economic indicator, consumer sentiment has not
received attention in the literature as a potential measure of investor optimism. 2 In this
paper, we adopt a time series framework and show that consumer confidence helps to
explain the time variation in equity portfolio returns, particularly the size premium. The
methodology we adopt is as follows. First, we regress consumer confidence on a set of
macroeconomic variables. Although the regression has a high R2 (around 0.6 – 0.8
depending on the specific confidence index), a substantial portion of confidence remains
unexplained. We treat the residual from this regression as our measure of excessive
sentiment (optimism or pessimism) unwarranted by fundamentals. One general criticism
that applies to our approach is that we have to take a stand on the information set used in
order to separate the fundamental component. Hence we try to make our information set
as large as feasible given the number of observations in our sample. Our information set
includes dividend yield, default spread, short-term interest rate, growth in real GDP,
consumption, labor income, unemployment rate, and inflation, as well as the lags (leads)
of these variables. We believe that this information set is as comprehensive as those
commonly used in the literature. 3
We employ the fundamental and sentiment components of consumer confidence
to explore the time series behavior of the conditional betas and the pricing errors for the
size-sorted Fama-French portfolios. Our main findings are as follows. After 1977, we
2
Charoenrook (2002) studies the predictive ability of the Index of Consumer Sentiment for the aggregate
market returns in the U.S.
3
See, e.g. Chen, Roll, and Ross (1986), Nelson (1976), Lee (1992), Chen (1991), Fama and French (1988),
Baker and Wurgler (2003), Lewellen and Nagel (2003).
3
find that the conditional market betas of small stocks exhibit larger time variation than
those of large stocks and tend to be higher following quarters of low confidence.
However, after we allow for the time variation of the conditional market betas, we still
find that the pricing errors vary over time with the sentiment component of confidence.
Consistent with the view that investor sentiment affects stock prices, we find that the
pricing errors of small stocks exhibit larger time variation than those of large stocks and
are higher following quarters of low sentiment. Thus we report evidence that investors
appear to overvalue small stocks relative to large stocks during periods when consumer
confidence is high, and vice versa. Our results are robust to the time period over which
we estimate the fundamental and sentiment components; to the choice of consumer
confidence measure; and to the inclusion of the sentiment measure from Baker and
Wurgler (2003) in our regression.
We also compare our residua l measure of behavioral sentiment to that used by
Baker and Wurgler (2003). Over the period from 1962 – 1976 the two sentiment
measures are actually negatively correlated. Over the period 1977 – 2002, and especially
towards the end of our sample period (the 1990s), we find that the two measures align
much better than in the earlier period. More interestingly, we find that prior to 1977 the
Baker and Wurgler sentiment measure forecasts the size premium, while consumer
confidence shows little predictive power. In contrast, in the post-1977 period, the
sentiment measure based on consumner confidence exhibits strong ability to forecast the
size premium, whereas the Baker and Wurgler measure exhibits essentially no forecasting
power over this time period. We conjecture that this finding might be due to the fact that
consumers may not have represented investors in the earlier years of our sample.
4
Finally, we perform the same time-series analysis for other portfolio sorts. We
sort equity portfolios by dividends (payers versus non-payers), earnings growth, sales
growth, and institutional ownership. Our findings exhibit the same pattern for those
portfolio sorts as for the size-sorted portfolios. We find that, after controlling for time
variation in beta, quarters of high investor optimism are followed by lower returns on
dividend non-paying stocks, stocks with low earnings growth, stocks with low sales
growth, and stocks with low institutional ownership. The results based on institutional
ownership provide some additional evidence that is consistent with the idea that
individual investor sentiment is an important determinant of mispricing in stocks where
arbitrage may be more limited.
The rest of this paper is organized as follows. The next section gives a detailed
description of the consumer confidence survey data along with the other data used in our
study. Section 2 discusses the ability of consumer confidence to predict macroeconomic
activity. Section 3 presents the initial evidence on size premium. Section 4 describes the
fundamental and the sentiment components of consumer confidence and addresses
several robustness checks. Section 5 discusses the forecasting ability of the components
of consumer confidence with regards to various equity portfolios. Section 6 concludes.
1. Data
The University of Michigan survey of consumer sentiment started in 1947 on a
quarterly basis for months 2, 5, 8, and 11. Beginning in 1978, the index is available on a
monthly basis. The main reason for increasing the sampling frequency in 1978,
according to the University of Michigan Survey Research Center, was the increased
5
frequency in the reporting of other macroeconomic data around that time. The final
survey results for each month become available either at the end of that month or at the
beginning of the following month. The survey is sent to 500 households, and the
respondents are asked the following questions: (1) Would you say that you (and your
family living there) are better off or worse off financially than yo u were a year ago? (2)
Do you think that a year from now, you (and your family living there) will be better off
financially, or worse off, or about the same as now? (3) Now turning to business
conditions in the country as a whole – do you think that during the next twelve months,
we’ll have good times financially or bad times or what? (4) Looking ahead, which would
you say is more likely – that in the country as a whole we’ll have continuous good times
during the next five years or so or that we’ll have periods of widespread unemplo yment
or depression, or what? (5) Do you think now is a good or bad time for people to buy
major household items? The relative score for each question is then calculated as the
percent of favorable replies minus the percent of unfavorable replies, plus 100, rounded
to the nearest whole number. These relative scores for questions (1) and (5) compose the
Index of Current Economic Conditions, the relative scores for questions (2), (3), (4)
compose the Index of Consumer Expectations, and all five questions compose the overall
Index of Consumer Sentiment. We have the data on all five individual survey questions
(hereafter Q1 through Q5), two sub-indexes (hereafter CI for current conditions and EI
for expectations) and the composite sentiment index (hereafter ICS), starting in the
1940’s (the precise beginning of the sample varies for the individual questions).
6
The Conference Board survey is available on a shorter time interval. The survey
began on a bimonthly basis in 1967 and turned into a monthly poll in 1977. 4 However, a
clear advantage of this survey relative to the one conducted by the University of
Michigan is the much larger pool of respondents. The survey is mailed to 5,000
households. Bram and Ludvigson (1998) ran a horse race between the Conference Board
and the University of Michigan surveys and reported that overall, the Conference Board
index does a better job forecasting future household spending. The survey is similar in
spirit to that of the University of Michigan, although the questions are somewhat
different. Respondents are asked the following five questions: (1) How would you rate
present general business conditions in your area? (2) What would you say about
available jobs in your area right now? (3) Six months from now, do you think business
conditions in your area will be better, same or worse? (4) Six months from now, do you
think there will be more, same or fewer jobs available in your area? (5) Would you guess
your total family income to be higher, same, or lower six months from now? The scores
for each question are calculated as the percent of favorable replies divided by the sum of
favorable and unfavorable replies. The scores for questions (1) and (2) compose the
Current Index, the scores for questions (3), (4), (5) compose the Expectations Index, and
all five questions compose the overall Index of Consumer Confidence. The final results
for each survey month become available on the last Tuesday of the following month. For
this survey, we have the data on the overall Index of Consumer Confidence and the
Expectations Index only (hereafter CBIND and CBEXP). All our tables report the results
4
It is interesting to note that both surveys of consumer confidence switched to monthly sampling frequency
in 1977-78. Even though the survey questions stayed exactly the same as before (we could not find any
evidence to the contrary), still the increase in sampling frequency may have indicated that the index of
consumer confidence has become a more meaningful and more closely followed indicator.
7
for the major confidence indexes and omit those for the individual survey questions in the
interest of parsimony.
Despite the obvious similarities in the two survey methodologies, we should
mention several differences important for our study. First, the Current Index of the
University of Michigan is somewhat backward- looking, since the respondents are asked
to compare their present situation to that one year before. Second, the University of
Michigan survey asks specifically about current buying conditions, whereas the
Conference Board survey does not. Third, the Expectations Index of the University of
Michigan measures expectations over a longer term (one to five years) than the similar
index from the Conference Board (six months ). Fourth, the Conference Board index is
considerably more focused on job availability, whereas the other survey asks generally
about financial situation. Fifth, the University of Michigan survey asks about the
economic conditions in the country as a whole, while the Conference Board survey
focuses on the respondent’s specific area of residence. These differences substantiate the
need to test the robustness of our results to the use of either survey.
It is important to note that the results of consumer confidence surveys are not
fully observable at the end of the survey month and generally become fully observable
only during the month following the survey. This delay applies both to the University of
Michigan survey and (especially) to the Conference Board survey. Therefore, we lag all
confidence observations by one month. For example, we use consumer confidence level
reported for November to forecast returns over the first quarter of the following year
(January through March). This way, we use observations for both indexes for Februa ry,
May, August, and November of each year in the sample. This procedure leaves
8
negligibly few observations prior to 1977 for the Conference Board index, since it was
collected bi- monthly before then. Regressing the consumer confidenc e indexes on
monthly dummies does not reveal significant seasonality in any index, and none of the
indexes exhibit noticeable trends.
We employ several macroeconomic variables, either as controls or as dependent
variables, observed quarterly and measured in percent : default spread (DEF), measured
as the difference between the yields to maturity on Moody’s Baa-rated and Aaa-rated
bonds; yield on 3- month Treasury bill (YLD3); dividend yield (DIV), measured as the
total cash ordinary dividend of the CRSP value-weighted index over the last four
quarters, divided by the value of the index at the end of the current quarter (as in Fama
and French, 1988); GDP growth (GDP), measured as 100 times the quarterly change in
the natural logarithm of chained (1996 dollars) GDP; consumption growth (CONS),
measured as 100 times the quarterly change in the natural logarithm of personal
consumption expenditures; labor income growth (LABOR), 100 times the quarterly
change in the natural logarithm of labor income, computed as total personal income
minus dividend income, per capita, deflated by PCE deflator; unemployment rate
(URATE), seasonally adjusted, as reported by the Bureau of Labor Statistics, averaged
over the most recent three months; growth in unemployment rate (URCHG), measured as
the difference between the end-of-quarter and beginning-of-quarter level of
unemployment rate; inflation rate (CPIQ), from CRSP, averaged over the most recent
three months.
As our measure of the size premium in stock returns, we use the difference
between the returns on the smallest and the largest Fama-French size decile portfolios
9
(M110). We also look at the portfolios sorted according to other salient characteristics.
We sort equity portfolios into dividend payers and non-payers and consider the return on
dividend non-payers less the return on dividend payers (DV110). We also consider the
return on the lowest decile less the return on the highest decile in earnings growth
(ER110), sales growth (SG110), and institutional ownership (IO110). The monthly
returns on all equity portfolios are averaged over each quarter months and measured in
percent. All our observations are non-overlapping.
Tables 1 and 2 summarize the descriptive statistics and the correlations. Both
tables show the numbers for the pre-1977 and the post-1977 sub-sample s. Despite the
differences in the consumer confidence surveys conducted by the Conference Board and
the University of Michigan, their respective indexes are highly correlated. The means
and standard deviations of the confidence variables, as well as their correlations with
each other are comparable over the pre – 1977 and post – 1977 sub-periods. Most of the
macroeconomic variables show strong contemporaneous correlations with consumer
confidence, suggesting that consumer confidence is related to econo mic activity.
2. Forecasting future economic activity
Previous evidence suggests that consumer confidence indicators forecast future
economic activity. For example, Bram and Ludvigson (1998) perform a detailed
comparison between the two U.S. consumer sentiment surveys and report that consumer
sentiment forecasts various categories in future household spending, with the Conference
Board indicator doing a better job overall. The University of Michigan Survey Research
Center reports that consumers assess future economic conditions rather accurately. The
10
corresponding correlation coefficients are: 0.74 for changes in interest rate (six months
ahead); 0.80 for unemployment rate (nine months ahead); 0.90 for CPI (three months
ahead); 0.90 for real GDP growth (contemporaneous); 0.77 for house sales (six months
ahead); 0.73 for vehic le sales (six months ahead). 5
We test the forecasting ability of consumer confidence for quarterly GDP growth,
consumption growth, labor income growth, the unemployment rate, and the change in the
unemployment rate. Table 3 reports the results. It is worth noting that the predictive
ability of consumer confidence is strong in the post – 1977 period only. The components
of consumer confidence that measure expectations (EI, CBEXP) do a particularly good
job predicting future economic activity, especially consumption growth and labor income
growth, as measured by the incremental adjusted R2 . 6 On the other hand, the index of
current economic conditions (CI) has poor predictive ability for future economic activity,
either in terms of statistical significanc e or in terms of the incremental R2 .
Another way to look at the relationship between consumer confidence and the
economy is to regress the level of confidence on a set of macroeconomic variables. The
results are in Table 4. As explanatory variables, we use dividend yield, default spread, 3month Treasury bill yield, GDP growth, consumption growth, labor income growth, CPI
growth, and change in unemployment rate. All explanatory variables are measured
contemporaneously with consumer confidence. For robustness, we also estimate a
similar regression with the independent variables leading by one quarter. The results are
5
This information is from the University of Michigan Survey Research Center website. The reported
correlations are based on separate questions asked specifically about inflation, interest rates, etc., which are
not a part of the Index of Consumer Sentiment.
6
We used our data to run tests similar to Bram and Ludvigson (1998). Our results are similar to theirs.
Adding four lags of confidence improves forecasting ability for future consumption substantially when we
use the Conference Board index; and insignificantly if we use the University of Michigan index.
11
qualitatively the same and are not reported here. The adjusted R2 in Table 4 is high
enough to suggest that at the very least, consumer confidence contains a certain
fundamental component that is rooted in true economic fundamentals.
3. Consumer confidence and the size premium
If consumer confidence reflects individual investor beliefs, then both the rational
and the behavioral hypotheses of asset pricing suggest a negative relationship between
consumer confidence and subsequent size premium. For example, Chan and Chen (1991)
and Jagannathan and Wang (1996) argue that small stocks contain a larger proportion of
distressed firms and hence their conditional market betas should be more responsive to
changes in the business cycle than those of large stocks. Hence to the extent that
consumer confidence reflects variation in (a portion of) the market risk premium, we
expect a higher size premium following a decline in the confidence level.
On the other hand, according to Daniel, Hirshleifer, and Subrahmanyam (1998),
overconfidence in one’s own private information may lead one to overvalue assets in
times of high confidence and to undervalue them in times of low confidence because
investors fail to properly consider public information. Since small stocks are mostly
individually owned the limits to arbitrage make such stocks more likely to sustain misvaluation. To the extent that consumer confidence reflects irrational investor sentiment,
we expect a widening in size premium after a drop in confidence.
After 1977, our findings are consistent with both hypotheses. The results are in
Table 5. We consistently observe a strong negative relationship between consumer
confidence and the subsequent quarter’s size premium. The incremental R2 ranges from
12
0.03 to 0.10. Interestingly, in contrast to the results obtained for the macroeconomic
variables, this time the predictive power is strong for the index of current conditions (CI)
with an incremental R2 of 0.07. An improvement in consumers’ assessment of the
current conditions by one standard deviation leads to a drop of about 1.25% per month in
the average monthly size premium over the next three months. This finding suggests that
the forecasting ability for the future equity returns may go beyond the mere reflection by
consumers’ responses of their expectations of the fundamental economic activity.
We find no relationship between confidence and size premium in the pre – 1977
sub-period. The results are strikingly different for the two sub-periods in terms of sign,
statistical significance and R2 . A possible explanation for these different results may be
that consumer confidence was not representative of investor beliefs prior to the mid1970s. For example, according to the Survey of Consumer Finances (Aizcorbe,
Kennickell, and Moore, 2003), the percent of households with either direct or indirect
stock ownership (including mutual funds and retirement accounts) has increased from
31.8% in 1989 to 51.9% in 2001. In 1983, only 20.4 percent of families reported public
corporate stock or mutual fund ownership, and 24.2 percent reported retirement accounts
(Kennickell and Shack-Marquez, 1992). Out of those who reported stock ownership in
1983, only 40 percent said they owned shares of more than one company (Avery et. al.,
1984). Figure 2 shows the net acquisition of mutual funds and pension funds by
households and non-profit organizations. After the mid-1970s, pension fund ownership
started growing at a rapid rate (owing, perhaps, to the Employee Retirement Income
13
Security Act of 1974), while mutual fund ownership started increasing in the early
1980’s. 7
4. Fundamental and sentiment components of confidence
The findings reported in the previous sections prompt us to explore the
forecasting ability of consumer confidence further. We find it reasonable to conjecture
the presence of two different components in the consumer confidence index: one rooted
in fundamentals, and one reflecting sentiment. To separate consumer confidence index
into the two components, we estimate the following regressions for each index:
CONF t = a + b1 DIVt + b 2 DEFt + b3YLD 3 t + b 4 GDPt + b5 CONS t + b6 LABOR
+ b 7URATE t + b8 CPI t + b9 DIVt −1 + b10 DEFt −1 + b11YLD3 t −1 + b12 GDPt −1
(1)
+ b13 CONS t −1 + b14 LABOR t −1 + b15 URATEt −1 + b16 CPI t −1 + ηt
The predicted value from this regression is our measure of the fundamental
component of confidence, and the residual represents sentiment. For robustness, we also
regress confidence on contemporaneous and leading values of the explanatory variables.
The predicted and the residual components obtained in the two ways are very closely
aligned for each confidence measure (the correlations are around 0.85 to 0.95 both for the
predictions and for the residuals), so we do not consider these results further here.
Figure 3 shows the predicted and the residual compone nts of the Conference
Board index of consumer confidence (CBIND) versus the University of Michigan index
7
One reason that may have contributed to the increased indirect participation of households in the equity
markets through mutual funds was the banning of fixed brokerage commissions on May 1, 1975, which has
decreased commission rates for mutual funds and increased trading volumes, while not reducing trading
costs for small individual investors (Blum and Lewellen, 1983; Edmister, 1978).
14
of consumer sentiment (ICS). Table 6 reports the correlations between the predicted and
the residual components from (1) over 1977 - 2002. The correlation between the
composite indexes (ICS and CBIND) is 0.86 for the predictions, and 0.77 for the
residuals. The values for the expectations indexes (EI and CBEXP) are the same. Thus
the time-series behavior of the predicted and the residual components is robust to the
choice of confidence measure.
We compare our measures of sentiment to other measures used in the literature.
Baker and Wurgler (2003) collect a sample of several commonly used sentiment
variables: value-weighted dividend premium, number of IPO’s, average first-day IPO
return, value-weighted closed-end fund discount, equity share in new issues, and NYSE
turnover. They extract the first principal component of these sentiment measures and
regress it on several macroeconomic variables, including industrial production index,
personal consumption expenditures, and NBER recession indicator. The residual from
the regression (the variable they name SENTIMENT) is their main measure of investor
sentiment unwarranted by economic expectations. All their data is observed at year-end
and is available over 1962 – 2002. For fairness of the comparison, we re-estimated the
residual components of the confidence indexes over the same time interval and retained
the year-end observations. 8
Table 7 shows the correlations of the confidence residuals with the Baker and
Wurgler (2003) sentiment measure. The correlations over 1962 – 1976 are in panel A
and those for 1977 – 2002 are in panel B. The difference between the two sub-periods is
8
For robustness, we also estimate confidence residuals using year-end observations of macroeconomic
variables such that our residual data arre available on an annual basis to begin with. Comparing these
residuals to the measures used by Baker and Wurgler (2003) provides the same qualitative pattern.
15
rather striking. After 1977, the correlations mostly have the expected sign. However, in
the pre-1977 sub-period, they reverse sign. This sign reversal pattern is clear in Figure 4,
which shows SENTIMENT plotted against the residual from the University of Michigan
Expectations Index. Clearly, in the beginning of the sample, the correlation between the
two is negative, while later in the sample they exhibit a marked positive correlation,
which becomes especially strong in the 1990s.
5. Time series tests for size -sorted portfolios
We now turn to the main question of our paper. In a time-series framework, we
model the behavior of the conditional market betas and the pricing errors of the sizesorted Fama-French portfolios. We estimate the following time-series regression for the
portfolio long in stocks representing the smallest size decile and short in stocks
representing the largest size decile (M110).
Rt = ( a1 + a 2 RES t−1 ) + (b1 + b2 PREDt −1 ) Rm, t
(2)
In (2), RES and PRED are the residual (sentiment) and the predicted
(fundamental) components of consumer confidence, respectively, and Rm is the return on
the CRSP value-weighted index. All returns are excess, net of the return on the onemonth T-bill. Thus we make the conditional market beta of each portfolio a function of
the fundamental compone nt of consumer confidence, while the portion of return
unexplained by the conditional market beta (the pricing error) is a function of the
sentiment component. 9
9
As a robustness check we also estimate the regression in equation (2) using the raw confidence measure
instead of its predicted value to condition the beta estimates. The results are identical to those reported in
the paper.
16
Table 8A compares the results of (2) for the pre – 1977 and the post – 1977 subperiods. In the post – 1977 regression, we use the confidence components estimated
using only post-1977 data. For robustness, we ran (2) over 1977 – 2002 with the
confidence components estimated first on the entire available dataset. The results are
qualitatively similar. Once again, we observe a striking difference across the two subperiods. Prior to 1977, neither component of consumer confidence has any ability to
explain the time variation in the size premium. The slope coefficients are
indistinguishable from zero, and the incremental R2 is zero or negative. In the second
sub-period, the slope coefficients for both confidence components are significantly
negative, most at 95% to 99% level of significance. The incremental adjusted R2 ’s are
high, up to 0.13.
Table 8A also reports the average conditional market betas and the average
pricing errors: (1) for each sub-period; (2) for the observations from each sub-period
where the corresponding confidence component is below median; and (3) for the
observations from each sub-period where the corresponding confidence component is
above median. The pattern is clear. The conditional market betas of our ‘small minus
large’ strategy are higher in times when consumer confidence is low, which is consistent
with the argument that the conditional market betas of small stocks are more sensitive to
the business cycle than those of large stocks. However, even after controlling for the
conditional market betas, we still observe that the pricing errors of the ‘small minus
large’ strategy covary negatively with the component of confidence related to investor
sentiment, conforming to the behavioral hypothesis that overconfidence induces
individual investors to overvalue small stocks relative to large stocks, while excessive
17
pessimism leads them to do the opposite. A t-test (not reported) confirms that the
observed differences in the conditional betas and the pricing errors are statistically
significant. Again, the fact that we do not observe a similar pattern prior to 1977 may
have to do with the negligible participation of consumers in the equity markets during the
prior years.
Next, we amend the above regression by including SENTIMENT from Baker and
Wurgler (2003) as a second sentiment variable. So we run the following regression for
the M110 portfolio:
Rt = ( a1 + a 2 RES t −1 + a 3 BWSENTt−1 ) + (b1 + b2 PREDt −1 ) Rm,t
(3)
The results are reported in Table 8B. Since BWSENT is available annually, we
assign the year-end observation of this variable to the first 11 months of the following
year, in order to run the quarterly regressions. The pattern in Table 8B resembles our
previous results. Over 1962 – 1977, the confidence components have very little
additional predictive power. The only exceptions are the index of current economic
conditions (CI). However, after 1977, consumer confidence has significantly negative
slope coefficients even after controlling for BWSENT, while BWSENT loses its statistical
significance. The incremental adjusted R2 ’s are also impressive.
As a final robustness check, we perform the same analysis as in the previous
section for four other portfolios: (1) the difference in returns between dividend nonpayers and dividend payers (DV110); (2) the difference in returns between the lowest and
the highest earnings growth deciles (ER110); (3) the difference in returns between the
lowest and the highest sales growth deciles (SG110); and (4) the difference in returns
between the lowest and the highest institutional ownership deciles (IO110). The results
18
for 1977 – 2002 are in Table 9. The pattern observed is similar to that found for the
‘small minus large’ strategy. After 1977, we find that the quarters of relatively high
excessive optimism are followed on average by lower returns on dividend non-paying
stocks, stocks with low earnings growth, stocks with low sales growth, and stocks with
low institutional ownership. Prior to 1977, we find nothing, similar to the size-sorted
portfolios. Those results are not reported.
6. Concluding remarks
We find that consumer confidence predicts time variation in returns of
equity portfolios sorted on size and other characteristics that are likely to reflect crosssectional differences in the sensitivity of returns to the business cycle as well as the
potential for mispricing due to investor sentiment. Consistent with the view that investor
sentiment affects stock prices, we find that the pricing errors of small stocks exhibit
larger time variation than those of large stocks and are higher following quarters of low
sentiment. Thus we report evidence that investors appear to overvalue small stocks
relative to large stocks during periods when consumer confidence is high and ,and vice
versa. Our results are robust to the time period over which we estimate the fundamental
and sentiment components; to the choice of consumer confidence measure; and to the
inclusion of the sentiment measure from Baker and Wurgler (2003) in our regression.
An important element of our findings is that we find very different results for the
pre – 1977 and the post – 1977 time intervals. The findings in the post-1977 period
parallel the rapid increase of household participation in the equity markets. Obtaining a
19
more thourough understanding of the extent to which investor sentiment affects stock
prices remains an interesting topic for future research.
20
References
Acemoglu, D., and A. Scott, 1994, “Consumer confidence and rational expectations: Are
agents’ beliefs consistent with the theory?” Economic Journal, 104, 1 – 19.
Aizcorbe, A. M., A. B. Kennickell, and K. B. Moore, 2003, “Recent changes in U.S.
family finances: Evidence from the 1998 and 2001 Survey of Consumer Finances,”
Federal Reserve Bulletin, 89, 1 – 32.
Avery, R. B., G. E. Elliehausen, G. B. Canner, and T. A. Gustafson, 1984, “Survey of
Consumer Finances, 1983,” Federal Reserve Bulletin, 70, 679 – 692.
Baker, M., and J. Wurgler, 2003, “A conditional characteristic model of stock returns,”
working paper, Harvard University, NBER, and New York University.
Barber, B. M., and T. Odean, 2001, “The Internet and the investor,”, Journal of Economic
Perspectives, 15, 41-54.
Blum, G. A., and W. G. Lewellen, 1983, “Negotiated brokerage commissions and the
individual investor,” Journal of Financial and Quantitative Analysis, 18, 331 – 343.
Bram, J., and S. Ludvigson., 1998, “Does consumer confidence forecast household
expenditure? A sentiment index horse race,” FRBNY Policy Review, June 1998, 59 – 78.
21
Brown, G. W., and M. T. Cliff, 2002, “Investor sentiment and the near-term stock
market,” Journal of Empirical Finance, 2004, 11, 1 – 27.
Brown, J. R., Z. Ivkovich, P. A. Smith, and S. J. Weisbenner, 2004, “The geography of
stock market participation: The influence o f communities and local banks,” Working
Paper #W10235, NBER.
Carroll, C. D., J. C. Fuhrer, and D. W. Wilcox, 1994, “Does consumer sentiment forecast
household spending? If so, why?” American Economic Review, 84, 1397 – 1408.
Chan, K. C., and N. F. Chen, 1991, “Structural and return characteristics of small and
large firms,” Journal of Finance, 46, 1467 – 1484.
Charoenrook, A., 2002, “Change in consumer sentiment index and aggregate stock
returns,” working paper, Vanderbilt University.
Chen, N. F., 1991, “Financial investment opportunities and the macroeconomy,” Journal
of Finance, 46, 529 – 554.
Chen, N. F., R. Kan, and M. H. Miller, 1993, “Are the discounts on closed-end finds a
sentiment index?” Journal of Finance, 48, 795 – 800.
22
Chen, N. F., R. Roll, and S. A. Ross, 1986, “Economic forces and the stock market,”
Journal of Business, 59, 383 – 403.
Daniel, K., D. Hirshleifer, and A. Subrahmanyam, 1998, “Investor psychology and
security market under- and overreactions,” Journal of Finance, 53, 1839 – 1867.
Doukas, J. A., and N. T. Milonas, 2002, “Investor sentiment and the closed-end find
puzzle: out-of-sample evidence,” working paper, New York University.
Edmister, R. O., 1978, “Commission cost structure: Shifts and scale economies,” Journal
of Finance, 33, 477 – 486.
Fama, E. F., and K. French, 1988, “Dividend yields and expected stock returns,” Journal
of Financial Economics, 22, 3 -25.
Fama, E. F., and K. French, 1989, “Business conditions and expected returns on stocks
and bonds,” Journal of Financial Economics, 25, 23 -50.
Heaton, J., and D. Lucas, 2000, “Portfolio choice and asset prices: The importance o f
entrepreneurial risk,” Journal of Finance, 55, 1163 – 1198.
Hong, H., J. D. Kubik, and J. C. Stein, 2004, “Social interaction and stock- market
participation,” Journal of Finance, 59, 137 – 163.
23
Jagannathan, R., and Z. Wang, 1996, “The conditional CAPM and the cross-section of
expected returns,” Journal of Finance, 51, 3 -53.
Kennickell, A., and J. Shack-Marquez, 1992, “Changes in family finances from 1983 to
1989: Evidence from the Survey of Consumer Finances,” Federal Reserve Bulletin,
January 1992, 1 – 18.
Lee, B. S., 1992, “Casual relations among stock returns, interest rates, real activity, and
Inflation,” Journal of Finance, 47, 1592 – 1603.
Lee, C. M. C., A. Schleifer, and R. H. Thaler, 1991, “Investor sentiment and the closedend fund puzzle,” Journal of Finance, 46, 75 – 109.
Lewellen, J., and S. Nagel, 2003, “The conditional CAPM does not explain asset-pricing
anomalies,” working paper, MIT.
Neal, R., and S. M. Wheatley, 1998, “Do measures of investor sentiment predict market
returns?” Journal of Financial and Quantitative Analysis, 33, 523 – 547.
Nelson, C. R., 1976, “Inflation and rates of return on common stocks,” Journal of
Finance, 31, 471 – 483.
24
Table 1. Descriptive statistics
CBIND
CBEXP
ICS
CI
EI
DIV
DEF
YLD3
GDP
CONS
LABOR
URATE
CPIQ
MKTQ
SMBQ
HMLQ
M110
DV110
ER110
SG110
IO110
NOBS
21
21
89
90
86
205
205
174
120
120
120
117
205
203
203
203
203
55
55
55
n/a
MEAN
101.34
101.02
88.62
89.30
88.09
4.37
1.16
2.48
0.91
0.93
0.63
5.05
0.20
0.87
0.23
0.41
0.82
0.44
0.57
0.47
n/a
PRE – 1977
STDEV
24.88
18.80
10.44
10.24
13.47
1.41
0.84
2.23
1.16
1.02
1.04
1.40
0.50
3.40
2.29
2.56
6.12
3.09
2.55
1.65
n/a
MIN
54.50
53.90
57.60
60.20
49.40
1.33
0.32
0.01
-2.73
-3.00
-2.50
2.57
-1.39
-13.14
-4.42
-9.84
-12.71
-5.71
-4.86
-2.88
n/a
MAX
138.20
123.70
105.40
109.00
105.20
11.12
5.53
8.33
4.06
4.99
6.16
8.87
2.96
23.89
16.80
18.05
41.40
7.46
6.70
4.19
n/a
NOBS
102
102
102
102
102
103
103
103
103
103
102
104
102
102
102
102
102
102
102
102
86
MEAN
97.71
94.77
87.81
99.43
80.34
3.10
1.09
6.52
0.75
0.78
0.45
6.28
0.36
1.11
0.22
0.21
0.15
0.02
-0.23
0.56
0.23
POST - 1977
STDEV
MIN
23.54
47.30
15.04
50.00
12.92
51.70
12.46
61.70
14.00
45.30
1.25
1.10
0.45
0.55
3.00
1.17
0.81
-2.06
0.62
-2.30
0.77
-2.20
1.44
3.93
0.30
-0.30
2.78
-7.54
1.76
-3.79
2.25
-9.72
3.00
-6.00
2.92
-11.37
3.19
-13.54
1.85
-6.09
4.69
-14.70
MAX
144.70
124.30
111.30
121.10
107.80
5.54
2.61
15.02
3.78
2.14
1.95
10.67
1.46
6.75
5.57
7.44
9.16
9.75
12.17
6.95
16.08
25
Notes for Table 1. The descriptive statistics are estimated over the available observations
during 1926 – 2002. All observations are as of March, June, September, and December,
except for the confidence data, which is as of February, May, August, and November.
CBIND: the Conference Board Index of Consumer Confidence; CBEXP: the Conference
Board Expectations Index; ICS: the University of Michigan Index of Consumer
Sentiment; CI: the University of Michigan Index of Current Economic Conditions; EI:
the University of Michigan Index of Consumer Expectations; DEF: default spread (the
difference between Baa- and Aaa- rated bond yields); DIV: dividend yield on the CRSP
value-weighted index, as in Fama and French (1988); YLD3: yield on 3-month Treasury
bill; GDP: 100 times the change in the natural log of chained GDP; CONS: 100 times the
change in the natural log of personal consumption expenditures; LABOR: 100 times the
change in the natural log of labor income, measured as total personal income net of
dividend income, per capita, deflated by the PCE deflator; URATE: unemployment rate
as reported by the Bureau of Labor Statistics, averaged over three months; CPIQ:
inflation rate from CRSP, averaged over three months; MKTQ, SMBQ, HMLQ: Fama –
French factor returns, averaged over three months; M110: difference in returns between
the smallest and the largest size deciles (small minus large); DV110: difference in returns
between dividend non-payers and dividend payers (non-payers minus payers); ER110:
difference in returns between the lowest and the highest earnings growth deciles (low
minus high); SG110: difference in returns between the lowest and the highest sales
growth deciles (low minus high); IO110: difference in returns between the lowest and the
highest institutio nal ownership deciles (low minus high). All data is in percent, except for
the confidence measures.
26
Table 2. Correlations between confidence indicators and macroeconomic variables
A. Pre-1977
CBIND
CBIND
1.00
CBEXP
0.73
ICS
0.84
CI
0.74
EI
0.84
DIV
-0.50
DEF
-0.58
YLD3
-0.20
GDP
0.39
CONS
0.31
LABOR
0.65
URATE
-0.71
CPIQ
-0.37
B. Post-1977
CBIND
CBIND
1.00
CBEXP
0.74
ICS
0.85
CI
0.84
E1
0.81
DIV
-0.51
DEF
-0.47
YLD3
-0.16
GDP
0.30
CONS
0.31
LABOR
0.38
URATE
-0.75
CPIQ
-0.15
CBEXP
ICS
CI
E1
DIV
DEF
YLD3
GDP
CONS
LABOR
URATE
CPIQ
1.00
0.89
0.72
0.91
-0.36
-0.10
-0.66
0.75
0.66
0.83
-0.16
-0.67
1.00
0.83
0.97
-0.40
-0.70
-0.45
0.48
0.41
0.51
-0.35
-0.63
1.00
0.69
-0.79
-0.35
0.10
0.38
0.34
0.40
0.04
-0.22
1.00
-0.26
-0.69
-0.62
0.41
0.36
0.42
-0.37
-0.73
1.00
0.47
-0.57
-0.10
-0.18
-0.19
-0.17
-0.19
1.00
-0.31
-0.25
-0.14
-0.20
0.60
-0.38
1.00
-0.10
-0.04
0.07
0.09
0.20
1.00
0.56
0.65
0.00
-0.11
1.00
0.35
0.09
-0.16
1.00
-0.06
0.01
1.00
0.03
1.00
CBEXP
ICS
CI
EI
DIV
DEF
YLD3
GDP
CONS
LABOR
URATE
CPIQ
1.00
0.78
0.63
0.82
-0.25
-0.22
-0.02
0.50
0.53
0.53
-0.20
-0.23
1.00
0.94
0.98
-0.69
-0.54
-0.42
0.40
0.47
0.40
-0.57
-0.47
1.00
0.85
-0.66
-0.65
-0.46
0.36
0.41
0.35
-0.69
-0.40
1.00
-0.66
-0.45
-0.37
0.40
0.48
0.40
-0.47
-0.48
1.00
0.63
0.76
-0.09
-0.18
-0.03
0.67
0.58
1.00
0.62
-0.25
-0.16
-0.15
0.67
0.28
1.00
-0.06
-0.14
0.11
0.40
0.64
1.00
0.62
0.63
-0.04
-0.05
1.00
0.53
0.03
-0.29
1.00
-0.10
-0.02
1.00
0.10
1.00
Notes for Table 2. All variables and time periods are as defined in Table 1.
27
Table 3. Consumer confidence as a predictor of economic activity
A. Pre-1977
ICS
CI
EI
slope
-0.0220
-1.15
-0.0324
-1.60
-0.0155
-1.11
GDP
R2
0.00
0.02
0.00
slope
0.0009
0.07
-0.0012
-0.09
-0.0028
-0.27
CONS
R2
-0.01
-0.01
-0.01
slope
-0.0008
-0.05
-0.0028
-0.20
-0.0066
-0.50
LABOR
R2
-0.01
-0.01
-0.01
slope
-0.0008
-0.15
0.0051
0.72
0.0009
0.26
URCHG
R2
-0.01
0.00
-0.01
slope
-0.0170
-2.42
0.0029
0.34
-0.0115
-2.03
URATE
R2
0.00
0.00
0.00
B. Post - 1977
CBIND
CBEXP
ICS
CI
EI
slope
0.0048
1.05
0.0238
3.68
0.0203
2.57
0.0042
0.49
0.0245
3.27
GDP
R2
-0.01
0.12
0.02
-0.02
0.06
slope
0.0053
1.76
0.0184
3.97
0.0220
3.66
0.0059
0.70
0.0255
4.61
CONS
R2
0.01
0.11
0.06
-0.01
0.12
slope
0.0071
1.62
0.0230
3.48
0.0312
2.96
0.0154
1.52
0.0327
3.75
LABOR
R2
0.01
0.13
0.10
0.01
0.14
slope
0.0007
0.41
-0.0065
-2.58
-0.0065
-2.20
0.0002
0.05
-0.0085
-2.84
URCHG
R2
-0.01
0.04
0.01
-0.01
0.04
slope
-0.0082
-4.70
-0.0088
-5.71
-0.0131
-4.34
-0.0134
-3.66
-0.0114
-4.51
URATE
R2
0.01
0.01
0.01
0.01
0.01
Notes for Table 3. The table reports the results of the regression of macroeconomic
variables on lagged confidence indicators, controlling for the lagged dependent variable,
default spread, dividend yield, and 3- month T-bill yield. All variables are as defined in
Table 1, except URCHG, the difference between end-of-quarter and beginning-of-quarter
unemployment rate. Newey-West t-statistics are in italics. R2 is the incremental adjusted
R2 relative to the base regression that includes only lagged dependent variable and other
controls but no confidence.
28
Table 4. Regression of consumer confidence on macroeconomic variables
B. Pre-1977
Intercept
DIV
DEF
YLD3
GDP
CONS
LABOR
URCHG
CPIQ
R2
ICS
120.61
-4.18
-11.75
-1.92
0.62
-1.37
3.49
-0.73
-5.06
0.74
26.85
-3.40
-3.85
-3.45
0.78
-1.35
3.08
-0.29
-1.11
CI
130.47
-9.52
-6.74
-1.18
1.34
0.05
0.74
1.11
4.30
23.18
-7.48
-2.21
-1.92
1.76
0.06
0.73
0.61
1.25
123.64
-3.20
-13.90
-2.58
0.19
-1.70
4.10
-1.16
-12.60
20.19
-1.73
-3.70
-2.65
0.18
-1.33
2.86
-0.36
-1.85
Intercept
DIV
DEF
YLD3
GDP
CONS
LABOR
URCHG
CPIQ
R2
123.64
-14.32
-16.59
4.63
-1.86
4.71
5.37
-4.88
3.76
0.51
15.63
-3.7
-2.56
2.38
-0.67
1.67
2.23
-0.78
0.62
94.17
-4.59
-1.14
2.24
0.94
3.53
3.35
-14.39
-10.50
20.7
-2.4
-0.28
2.24
0.44
1.3
1.77
-4.05
-2.02
104.88
-7.11
-4.69
1.45
-0.05
2.94
2.14
-7.38
-7.84
35.95
-5.15
-1.89
2.03
-0.04
1.97
1.54
-2.54
-2.34
118.80
-5.07
-9.92
0.92
-0.79
2.72
1.89
-6.34
-3.93
39.31
-3.47
-3.58
1.17
-0.65
2.05
1.37
-1.99
-1.18
95.94
-8.42
-1.30
1.79
0.42
3.08
2.31
-8.07
-10.33
27.98
-5.68
-0.49
2.44
0.28
1.55
1.37
-2.49
-2.55
EI
0.71
0.75
B. Post-1977
CBIND
CBEXP
ICS
CI
EI
0.48
0.69
0.63
0.67
Notes for Table 4. All explanatory variables are measured contemporaneously with
consumer confidence. All measurements are quarterly. Newey-West t-statistics are in
italics. Adjusted R2 is in the last column. All variables are as defined in Table 1.
29
Table 5. Regression of size premium on consumer confidence
A. Pre – 1977
Intercept
CONF
LAGDEP
DIV
DEF
YLD3
R2
Incr. R2
-1.81
0.02
-0.13
-0.02
1.36
-0.19
-0.03
-0.01
-0.25
0.47
-0.77
-0.02
1.33
-0.70
-0.86
0.02
-0.13
-0.16
1.31
-0.24
-0.02
-0.01
-0.11
0.36
-0.79
-0.23
1.22
-1.23
-1.86
0.02
-0.13
0.17
1.12
-0.17
-0.02
-0.01
-0.28
0.46
-0.80
0.24
1.15
-0.53
Intercept
CONF
LAGDEP
DIV
DEF
YLD3
R2
Incr. R2
2.64
-0.02
0.06
0.04
0.97
-0.25
0.03
0.00
1.36
-1.33
0.54
0.06
1.23
-1.36
5.09
-0.05
0.05
0.07
1.08
-0.24
0.07
0.04
2.59
-2.88
0.41
0.11
1.28
-1.41
9.48
-0.09
0.02
-0.33
0.83
-0.25
0.08
0.05
3.33
-3.24
0.19
-0.50
1.07
-1.55
12.69
-0.10
-0.01
-0.22
0.21
-0.27
0.10
0.07
4.04
-3.87
-0.07
-0.37
0.29
-1.82
6.51
-0.06
0.04
-0.26
1.18
-0.26
0.06
0.04
2.65
-2.68
0.39
-0.38
1.44
-1.55
ICS
CI
EI
B. Post – 1977
CBIND
CBEXP
ICS
CI
EI
Notes for Table 5. The quarterly regressions of size premium on consumer confidence
control for lagged dependent variable, dividend yield, default spread, and 3- month T-bill
yield. All independent variables are lagged by one quarter. Column CONF reports the
slope for the corresponding confidence indicator. Column LAGDEP reports the slope for
lagged dependent variable. As a measure of size premium we used M110, the difference
between the smallest and the largest size decile returns. The two last columns report the
regression’s adjusted R2 and the incremental adjusted R2 relative to the base regression
that includes no confidence. Newey-West t-statistics are in italics. All variables are as
defined in Table 1.
30
Table 6. Correlations between the predicted and the residual components of different
consumer confidence indicators.
A. Predicted components
CBIND
CBEXP
ICS
CI
CBIND
1.00
CBEXP
0.73
1.00
ICS
0.86
0.83
1.00
CI
0.89
0.73
0.96
1.00
EI
0.81
0.86
0.99
0.89
EI
1.00
B. Residual components
CBIND
CBEXP
ICS
CI
CBIND
1.00
CBEXP
0.87
1.00
ICS
0.77
0.70
1.00
CI
0.54
0.40
0.84
1.00
EI
0.79
0.77
0.95
0.63
EI
1.00
Notes for Table 6. The correlations are estimated over 1977 – 2002. All data is quarterly.
The predicted and the residual components are estimated on the same time interval, from
the following regression of consumer confidence indicators on macroeconomic
variables:
CONF t = a + b1 DIVt + b 2 DEFt + b3YLD 3 t + b 4 GDPt + b5 CONS t + b6 LABOR
+ b 7URATE t + b8 CPI t + b9 DIVt −1 + b10 DEFt −1 + b11YLD3 t −1 + b12 GDPt −1
+ b13 CONS t −1 + b14 LABOR t −1 + b15 URATEt −1 + b16 CPI t −1 + ηt
31
Table 7. Correlations between confidence residuals and the sentiment measure from
Baker and Wurgler (2003)
A. 1962 – 1976
SENTIMENT
CBIND
CBEXP
ICS
CI
EI
n/a
n/a
-0.48
-0.60
-0.36
CBIND
CBEXP
ICS
CI
EI
0.37
0.34
0.32
-0.10
0.52
B. 1977 – 2002
SENTIMENT
Notes for Table 7. The table presents the correlations of the confidence residuals with the
sentiment measure of Baker and Wurgler (2003). For more details on the SENTIMENT
variable refer to Baker and Wurgler (2003). The data from Baker and Wurgler (2003) are
available as of year-end over 1962-2002. Only the year-end observations for confidence
residuals were retained for the purpose of estimating these correlations. Originally,
confidence residuals were estimated quarterly from regressing confidence on several
macroeconomic variables over 1962 – 2002; see equation (1) in the paper.
32
Table 8. Regression of ‘small minus large’ strategy on the confidence components
A1
a2
b1
b2
R2
0.4231
0.0739
0.3731
-0.0001
-0.01
1.29
1.15
0.57
-0.01
0.3991
0.0561
0.1291
0.0027
1.16
0.70
0.17
0.28
0.4770
0.0549
0.3152
0.0010
1.44
1.05
0.56
0.15
-0.1012
-0.0558
1.0388
-0.0081
-0.29
-3.14
2.12
-1.62
CBEXP
-0.0637
-0.0687
1.1998
-0.0106
-0.18
-3.13
1.92
-1.52
ICS
-0.1031
-0.1746
1.7973
-0.0177
-0.32
-4.70
2.67
-2.19
CI
-0.0671
-0.1960
2.0126
-0.0179
-0.22
-4.42
2.65
-2.25
0.1174
-0.1183
1.5613
-0.0165
-0.35
-4.01
2.72
-2.16
A.
ALL
LOW
HIGH
a
0.42
0.16
0.68
ß
0.36
0.37
0.36
a
0.40
0.20
0.59
ß
0.36
0.34
0.39
a
0.48
0.24
0.72
ß
0.40
0.40
0.41
a
-0.10
0.25
-0.45
ß
0.25
0.38
0.11
a
-0.06
0.39
-0.51
ß
0.20
0.30
0.10
a
-0.10
0.57
-0.76
ß
0.24
0.40
0.09
a
-0.07
0.74
-0.86
ß
0.23
0.38
0.09
a
0.12
0.65
-0.40
ß
0.24
0.40
0.08
Pre – 1977
ICS
CI
EI
-0.01
-0.01
Post – 1977
CBIND
EI
0.04
0.03
0.10
0.13
0.07
a1
a2
a3
b1
b2
R2
0.46
0.09
-0.59
-0.82
0.02
0.08
1.09
1.37
-1.98
-1.25
1.96
0.37
0.02
-0.74
-0.42
0.01
0.57
0.20
-2.49
-0.48
0.96
0.26
0.07
-0.62
-0.63
0.02
0.76
1.21
-2.15
-1.21
2.18
CBIND
-0.0162
-0.04
-0.0418
-1.79
-0.0432
-0.09
1.0862
2.21
-0.0087
-1.70
0.02
CBEXP
-0.0173
-0.0631
0.0155
1.1763
-0.0104
0.02
-0.05
-2.66
0.03
1.98
-1.56
0.1203
-0.1339
-0.0518
1.8574
-0.0191
0.33
-2.87
-0.10
2.90
-2.42
0.2918
-0.1156
-0.3092
2.1639
-0.0200
0.88
-3.65
-0.67
2.72
-2.34
0.0040
-0.0849
0.0407
1.5337
-0.0168
0.01
-2.09
0.07
3.03
-2.44
B.
1962-1976
ICS
CI
EI
0.04
0.09
1977 – 2002
ICS
CI
EI
0.08
0.08
0.05
33
Notes for Table 8. Panel A reports the results of the following time-series regression for
the portfolio long in the smallest size decile and short in the largest size decile (M110):
Rt = ( a1 + a 2 RES t−1 ) + (b1 + b2 PREDt −1 ) Rm, t , where
RES and PRED are the residual (sentiment) and the predicted (fundamental) components
of consumer confidence, respectively, and Rm is the return on the CRSP value-weighted
index. All returns are averaged over three months, net of the return on the one- month Tbill, in percent. The components of consumer confidence are estimated over the pre 1977 and post – 1977 intervals, respectively. Also reported are the average pricing error a
and the average conditional market beta ß, computed as follows: a = a1 + a2 RESt-1 , ß =
b1 + b2 PREDt-1 , for the entire estimation period (ALL), for the observations from the
same period where the corresponding confidence components are below median (LOW),
and for the observations from the same period where the corresponding confidence
components are above median (HIGH).
Panel B reports the results of the following regression for the M110 portfolio:
Rt = ( a1 + a 2 RES t −1 + a 3 BWSENTt−1 ) + (b1 + b2 PREDt −1 ) Rm,t , where BWSENT is the
measure of investor sentiment from Baker and Wurgler (2003) available annually. We
assign the year-end observation of this variable to the first 11 months of the following
year, in order to run quarterly regressions. The confidence components were estimated on
1962-2002.
In both panels, Newey-West t-statistics are in italics. R2 is the incremental
adjusted R2 relative to the base regression that includes the market factor only.
34
Table 9. Regressions for alternative portfolio sorts
a1
a2
b1
b2
R2
-0.2550
-0.0343
0.3005
0.0022
-0.00
-0.96
-1.33
0.71
0.49
CBEXP
-0.2849
-0.0415
0.6870
-0.0018
-1.07
-1.54
1.07
-0.26
ICS
-0.2595
-0.1198
0.1778
0.0040
-1.06
-2.63
0.31
0.59
CI
-0.2664
-0.1084
0.1697
0.0036
-1.07
-2.47
0.26
0.53
EI
-0.2541
-0.0926
0.1870
0.0042
-1.01
-2.36
0.37
0.64
-0.2984
-0.0481
-0.5341
0.0084
-1.02
-1.81
-0.99
1.48
-0.3239
-0.0526
-0.7606
0.0112
-1.07
-1.95
-1.04
1.39
-0.3078
-0.1251
-1.1255
0.0163
-1.13
-2.74
-1.78
2.09
CI
-0.3418
-0.1167
-1.1444
0.0147
-1.22
-2.75
-1.53
1.81
EI
-0.2867
-0.0935
-1.0392
0.0167
-1.04
-2.24
-1.85
2.20
CBIND
0.7087
3.79
-0.0358
-1.71
-0.4365
-1.21
0.0022
0.60
CBEXP
0.7565
-0.0239
-1.4383
0.0128
4.08
-1.18
-2.53
2.28
0.7051
-0.0700
-0.5720
0.0041
3.93
-2.02
-1.35
0.80
0.6926
-0.0716
-0.4269
0.0022
3.98
-2.32
-0.80
0.39
0.7158
-0.0476
-0.6247
0.0050
3.94
-1.62
-1.76
1.09
0.1603
-0.1015
0.1645
-0.0008
0.40
-2.33
0.24
-0.11
0.2553
-0.1134
-0.8509
0.0094
0.61
-2.47
-0.93
0.93
ICS
0.2663
-0.2554
0.1009
-0.0002
0.70
-2.64
0.10
-0.02
CI
0.2759
-0.2304
0.5350
-0.0046
0.71
-2.76
0.50
-0.39
EI
0.2470
-0.1974
-0.1556
0.0029
0.64
-2.36
-0.17
0.24
ALL
LOW
HIGH
a
-0.26
-0.04
-0.47
ß
0.52
0.48
0.55
a
-0.28
-0.01
-0.55
ß
0.52
0.53
0.50
a
-0.26
0.20
-0.71
ß
0.53
0.49
0.56
a
-0.27
0.18
-0.70
ß
0.53
0.50
0.56
a
-0.25
0.16
-0.66
ß
0.52
0.48
0.56
a
-0.30
0.01
-0.60
ß
0.29
0.15
0.43
a
-0.32
0.02
-0.66
ß
0.30
0.20
0.40
a
-0.31
0.17
-0.78
ß
0.31
0.16
0.45
a
-0.34
0.14
-0.81
ß
0.32
0.19
0.44
a
-0.29
0.13
-0.70
ß
0.30
0.14
0.46
0.02
a
ß
0.71
-0.22
0.94
-0.26
0.49
-0.19
0.06
a
0.76
0.91
0.60
ß
-0.23
-0.35
-0.11
A. DV110
CBIND
0.00
0.03
0.02
0.02
B. ER110
CBIND
CBEXP
ICS
0.03
0.02
0.07
0.05
0.06
C. SG110
ICS
CI
EI
0.03
0.02
0.02
a
0.71
0.97
0.44
ß
-0.21
-0.25
-0.18
a
0.69
0.99
0.40
ß
-0.21
-0.23
-0.19
a
0.72
0.93
0.51
ß
-0.22
-0.27
-0.18
a
0.16
0.81
-0.47
ß
0.09
0.10
0.07
a
0.26
1.00
-0.48
ß
0.04
-0.05
0.13
a
0.27
1.25
-0.69
ß
0.08
0.09
0.08
a
0.28
1.22
-0.65
ß
0.08
0.12
0.04
a
0.25
1.13
-0.62
ß
0.08
0.05
0.10
D. IO110
CBIND
CBEXP
0.01
0.03
0.05
0.04
0.03
35
Notes for Table 9. The table reports the results of the following time-series regression for
the difference in returns between dividend non-payers and dividend payers (A); the
lowest and the highest earnings growth deciles (B); the lowest and the highest sales
growth deciles (C); the lowest and the highest institutional ownership deciles (D):
Rt = ( a1 + a 2 RES t−1 ) + (b1 + b2 PREDt −1 ) Rm, t , where RES and PRED are the residual and
the predicted components of consumer confidence, respectively, and Rm is the return on
the CRSP value-weighted index. All returns are three- month averages, net of the return
on the one-month T-bill, in percent. All differences in returns are computed as “low
minus high.” The components of consumer confidence and the regressions are estimated
over the post – 1977 interval. Newey-West t-statistics are reported in italics. The R2 is the
incremental adjusted R2 relative to the base regression that includes the market factor
only. Also reported are the average pricing error a and the average conditional market
beta ß, computed as follows: a = a1 + a2 RESt-1 , ß = b1 + b2 PREDt-1 , for the entire
estimation period (ALL), for the observations from that period where the corresponding
confidence components are below median (LOW), and for the observations from the
same period where the corresponding confidence components are above median (HIGH).
36
Figure 1
140
1.5
120
1
100
0.5
80
0
60
-0.5
40
-1
20
CBEXP
EI
2002
2000
1999
1997
1996
1994
1993
1991
1990
1988
1987
1985
1984
1982
1981
1979
1978
1976
1975
1973
1972
1970
1969
1967
1966
1964
1963
1961
1960
1958
1957
1955
1954
1952
-1.5
1951
0
NBER
Figure 1. The indexes of consumer expectations from the University of Michigan (EI)
and the Conference Board (CBEXP), left scale, versus the NBER indicator of economic
cycle peaks and troughs, right scale.
37
Figure 2
300
250
200
$ billion
150
100
50
2002
2000
1998
1996
1994
1992
1990
1988
1986
1984
1982
1980
1978
1976
1974
1972
1970
1968
1966
1964
1962
1960
1958
1956
1954
1952
1950
1948
1946
0
-50
mutual funds
pension reserves
Figure 2. Acquisition of mutual funds and pension reserves by households and nonprofit
organizations. Source: Flow of Funds Accounts of the U.S., Federal Reserve statistical
release 1/15/2004.
38
Figure 3a
160
140
120
100
80
60
40
20
0
-20
-40
CBIND, pred
CBIND, res
ICS, pred
2002
2001
2001
2000
1999
1998
1998
1997
1996
1995
1995
1994
1993
1992
1992
1991
1990
1989
1989
1988
1987
1986
1986
1985
1984
1983
1983
1982
1981
1980
1980
1979
1978
1977
-60
ICS, res
Figure 3a. The predicted (upper) and the residual (lower) components of confidence.
CBIND is the Index of Consumer Confidence from the Conference Board. ICS is the
Index of Consumer Sentiment from the University of Michigan.
39
Figure 3b
140
120
100
80
60
40
20
0
-20
CBEXP, pred
CBEXP, res
EI, pred
2002
2001
2001
2000
1999
1998
1998
1997
1996
1995
1995
1994
1993
1992
1992
1991
1990
1989
1989
1988
1987
1986
1986
1985
1984
1983
1983
1982
1981
1980
1980
1979
1978
1977
-40
EI, res
Figure 3b. The predicted (upper) and the residual (lower) components of consumer
expectations indexes. CBEXP is the Expectations Index from the Conference Board. EI is
the Index of Consumer Expectations from the University of Michigan.
40
Figure 4
15
3.00
2.50
10
2.00
1.50
5
1.00
2002
2000
1998
1996
1994
1992
1990
1988
1986
1984
1982
1980
1978
1976
1974
1972
1970
1968
1966
1964
1962
0
0.50
0.00
-5
-0.50
-10
-1.00
-1.50
-15
-2.00
-20
-2.50
EI
SENTIMENT
Figure 4. The residual component of the Index of Consumer Expectations, University of
Michigan (EI), left scale, versus the investor sentiment measure (SENTIMENT) from
Baker and Wurgler (2003), right scale.
41