- Email: [email protected]
Investor sentiment, stock mispricing, and long-term growth expectations
Accepted Manuscript Title: Investor sentiment, stock mispricing, and long-term growth expectations Author: Kotaro Miwa PII: DOI: Reference:
S0275-5319(15)30045-3 http://dx.doi.org/doi:10.1016/j.ribaf.2015.10.003 RIBAF 420
To appear in:
Research in International Business and Finance
Received date: Revised date: Accepted date:
20-2-2015 2-10-2015 5-10-2015
Please cite this article as: Miwa, K.,Investor sentiment, stock mispricing, and longterm growth expectations, Research in International Business and Finance (2015), http://dx.doi.org/10.1016/j.ribaf.2015.10.003 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Investor sentiment, stock mispricing, and long-term growth expectations
ip t
KOTARO MIWA1 Tokio Marine Asset Management Co., Ltd
cr
Abstract
us
I analyze whether or not market-wide investor sentiment induces stock mispricing, by affecting the boldness of predictions of firms’ long-term earnings growth. I predict that bullish market-wide sentiment induces investors to aggressively separate firms with high growth futures from others, and
an
that this excessive boldness results in a high level of mispricing. Consistent with my prediction, I observe an excessively large dispersion in consensus growth forecasts when proxies for investor sentiment are high at the beginning of the period. Furthermore, stocks with higher-predicted growth
M
experience more negative forecast revisions and lower subsequent stock returns, especially following periods of high investor sentiment.
te
stock mispricing
d
Key words: investor sentiment; long-term earnings growth forecast; cross-sectional stock return,
Ac ce p
Investor sentiment, stock mispricing, and long-term growth expectations Abstract
I analyze whether or not market-wide investor sentiment induces stock mispricing, by affecting the boldness of predictions of firms’ long-term earnings growth. I predict that bullish market-wide sentiment induces investors to aggressively separate firms with high growth futures from others, and that this excessive boldness results in a high level of mispricing. Consistent with my prediction, I observe an excessively large dispersion in consensus growth forecasts when proxies for investor sentiment are high at the beginning of the period. Furthermore, stocks with higher-predicted growth experience more negative forecast revisions and lower subsequent stock returns, especially following 1 1-3-1, Marunouchi, Chiyoda-ku, Tokyo, Japan, phone: + 81-3-3212-8186 - fax: +81-3-3212-7576, e-mail: [email protected]
1
Page 1 of 23
periods of high investor sentiment. Key words: investor sentiment; long-term earnings growth forecast; cross-sectional stock return,
ip t
stock mispricing
1. Introduction
Considerable financial studies debate the role and effect of investor sentiment on asset pricing.
cr
Investor sentiment is seen as crowd psychology, revealed through the activity and asset price
movements in the market. Traditional asset pricing theory suggests that rational arbitrage necessarily
us
forces prices closer to fundamentals and leaves no role for investor sentiment. However, several behavioral finance studies argue that investor sentiment drives stock prices away from fundamental values (De Long et al. 1990; Shleifer and Vishny 1997).
an
Consistent with this argument, studies show that time-varying investor sentiment affects stock prices. Lemmon and Portniaguina (2006) and Baker and Wurgler (2006, 2007; hereafter
M
“Baker/Wurgler”) use a measure of investor sentiment to show that difficult-to-value stocks (i.e., small, newer, volatile stocks) are overvalued, especially when sentiment is high. Furthermore, Hribar and McInnis (2012) find that those stocks receive optimistic earnings forecasts during such periods.
d
They argue that, since difficult-to-value stocks are highly influenced by the trading activity of noise traders, valuation of these stocks is considerably affected by investor sentiment2.
te
On the other hand, some studies analyze the sentiment effects on the overall accuracy of analysts’ earnings forecasts, which can be regarded as proxies for investor expectations of future
Ac ce p
payoffs. Walther and Willis (2012) show that financial optimism regarding quarterly earnings forecasts is affected by investor sentiment. These studies raise the possibility that investor sentiment induces stock mispricing by affecting market participants’ expectations about a firm’s future payoffs. However, while prior studies focus on the sentiment effects on the overall accuracy of forecasts of near term payoffs, few studies analyze the sentiment effect on long-term earnings growth expectations.
Long-term earnings growth expectations are crucial to stock price valuations. In many valuation
models, an estimated intrinsic value of a firm’s stock depends on long-term earnings growth (LTG) expectations (Frankel and Lee 1998; Gebhardt et al. 2001). For example, according to the Gordon growth model (1962), a price-to-dividend ratio of 20 implies that a 1% increase in long-term dividend growth translates into a 20% return. Consistent with the implication of the model, Copeland et al. (2004) report that analysts’ long-term earnings forecasts, a proxy for investor long-term growth 2
Other sentiment effects include the effect on the performance of technology IPO shares (Saade, 2015).
2
Page 2 of 23
expectations, exert a great influence on stock prices. Therefore, to further pinpoint the sentiment effect on asset pricing, it is essential to examine whether investor sentiment affects investor long-term earnings expectations and thereby induces
ip t
mispricing. The extant research reveals that investors’ long-term growth expectations, which are
represented by analysts’ long-term earnings forecasts, are, on average, too extreme, and this results
cr
in economically significant mispricing. La Porta (1996) and Chan et al. (2003) reveal that firms with
high predicted growth are aggressively distinguished from others. Research shows that this excessive
us
boldness regarding LTG expectations induces significant mispricing; subsequent stock returns are negatively associated with LTG forecasts (Lakonishok et al. 1994; La Porta 1996; Dechow and
an
Sloan 1997; Dechow et al. 2000).
Investors may more readily bet on firms’ growth potential when market-wide sentiment is more supportive of their preferences, i.e., market-wide sentiment is higher. Thus, I predict that higher
M
sentiment encourages investors to more aggressively distinguish firms with high-predicted growth from others. These excessively bold predictions are then corrected in a subsequent period, and this correction affects subsequent cross-sectional stock returns by reducing mispricing, which was
d
induced by the prior excessively bold predictions. Thus, I predict that firms with higher expected
te
growth experience more negative earnings forecast revisions and lower stock returns, especially after periods of bullish sentiment.
Ac ce p
In this study, I present empirical analyses to test these predictions. Following the study of
Baker/Wurgler, I use their market-wide investor sentiment index, and the Michigan Consumer Sentiment Index, to explore investor sentiment effects. In addition, following other prior studies (e.g. Mezrich et al. 2001; Chan et al. 2003), I utilize analyst LTG forecasts as a representative proxy for investor expectations of firm long-term earnings growth. The paper proceeds as follows. Section 2 details the development of my hypotheses. Section 3
presents the sample and descriptive statistics. Section 4 describes the methodologies and results. Finally, I summarize my findings in Section 5.
2. Theory To test the possibility that investor sentiment affects the boldness of LTG expectations and thereby induces stock mispricing, I examined the following hypotheses.
2.1 Investor sentiment and bold growth forecasts
3
Page 3 of 23
First, I examined the association between market-wide investor sentiment and investor excessive boldness in LTG predictions. I predict that bullish market-wide sentiment encourages investors to aggressively distinguish firms with high-predicted growth from others, and the boldness of their long-term growth forecasts lowers the accuracy of these growth forecasts. Hence, first I test
ip t
the following two hypotheses:
Hypothesis 1-1: Bullish investor sentiment induces bold LTG forecasts (boldness in LTG forecasts is
cr
positively associated with investor sentiment)
Hypothesis 1-2: The positive association between investor sentiment and forecast boldness lowers
us
the accuracy of these growth forecasts.
The Hypothesis 1-1 addresses the sentiment effect on boldness in LTG forecasts; Hypothesis
an
1-2 addresses whether the sentiment effect is aligned with forecast accuracy. If both are satisfied, I can say that bullish investor sentiment induces excessive boldness in LTG forecasts.
M
2.2 Correction of bold forecasts and mispricing
Second, I examined whether excessively bold LTG forecasts caused by bullish market-wide investor sentiment induces stock mispricing. The excessive boldness in the LTG forecasts could be
d
corrected in a subsequent period and that correction could affect cross-sectional stock returns by reducing mispricing, which was induced by the prior excessively bold predictions. If this is the case,
te
firms with high-predicted growth should experience more negative (downward) forecast revisions and lower subsequent stock returns than firms with low-predicted growth, especially when proxies
Ac ce p
for investor sentiment are high at the beginning of the period. Accordingly, the hypotheses to be tested are:
Hypothesis 2-1: There is a stronger negative association between the LTG forecasts and subsequent earnings forecast revisions when proxies for investor sentiment are higher. Hypothesis 2-2: There is a stronger negative association between the LTG forecasts and subsequent stock returns when proxies for investor sentiment are higher.
3. Sample and descriptive statistics As mentioned earlier, following previous studies (e.g., Mezrich et al. 2001; Chan et al. 2003), consensus Institutional Brokers Estimate System (IBES) forecasts are used for investor consensus expectations of firm earnings and earnings growth rates. I obtain the sample of earnings forecasts from an unadjusted file of the IBES summary.
4
Page 4 of 23
I collect data from stocks listed on the New York Stock Exchange (NYSE), American Stock Exchange (Amex), and NASDAQ that received LTG forecasts. I exclude the shares of non-US firms and stocks priced below $3 to ensure that the empirical findings are not driven by low prices. I utilized data from the beginning of 1986 to the end of 2011 (monthly). To evaluate realized growth, I
ip t
required five years of realized earnings growth. Thus, the final year of the sample is 2006. The number of eligible firms ranged from 1,593 to 3,457; on average, the sample comprised about 2,500
cr
firms.
For the main part of my analysis, I measure investor sentiment using the monthly time series of
us
the investor sentiment index constructed by Baker/Wurgler and the Michigan Consumer Sentiment Index, since these indices are used in extant investor-sentiment studies3. To ensure that both indices
an
are free of macroeconomic influences, following the study by Baker/Wurgler, I conduct my investigation using an orthogonal version of the indices, which is obtained by regressing the indices against a set of macroeconomic variables4. Figure 1 plots the two series. The two sentiment indices
M
rise during the mid-1980s and late 1990s, and fall during the early 1990s and early 2000s. These patterns are in line with the evidence for investor sentiment discussed by Baker/Wurgler.
d
[Figure 1]
te
4. Methods and results
4.1 Sentiment and dispersion in the LTG forecasts
Ac ce p
To test Hypothesis 1-1, I examine the association between investor sentiment and LTG forecast boldness. Chan et al. (2003) evaluate boldness in LTG forecasts by dispersion in consensus LTG forecasts. Following their methodology, each month firms are sorted into quintiles on consensus (mean) LTG forecasts, from LTG5 (High) to LTG1 (low). I calculate an average for the consensus LTG forecasts for each LTG quintile. I define the dispersion in the LTG forecasts as the spread in predicted growth between the top and bottom quintiles of the LTG forecasts. Next, I run the following regressions:
LTGX t = High, t − LTGX t = Low,t = ck + d k SENTIMENTk ,t + ε k ,t LTG X
t
= High , t
(1)
denotes an average of the consensus LTG forecasts for stocks included in the top
3
Practitioners often consider the volatility (VIX) index as the investor sentiment index. However, Whaley (2000) observes that the VIX index is an indicator of investor fear of the market downside and cannot capture investor optimism during a market boom. My study examines whether or not high investor sentiment (investor optimism) induces excessive boldness regarding long-term growth expectations. Thus, I do not use the VIX index, as it cannot capture market-wide optimism as well as the investor sentiment index in my study. 4 Please refer to Baker and Wurgler (2006, 2007) for more detail on the methodology to obtain macroeconomic free indices.
5
Page 5 of 23
LTG quintile (I denote these stocks as LTG5); LTGX =Low,t denotes an average of consensus LTG t
forecasts for stocks included in the bottom LTG quintile (I denote these stocks as LTG1). Thus, the dependent variable in the equation (1) is dispersion in a firm’s consensus long-term growth forecast between the top and bottom quintile; SENTIMENTk,t (k=1,2) denotes the Baker/Wurgler Sentiment
ip t
Index and the Michigan Consumer Sentiment Index at the end of month t, respectively.
Figure 2 plots the two sentiment index series and dispersion in the consensus LTG forecasts.
cr
Table 1 shows the regression results and descriptive statistics for growth forecast dispersion. The difference in LTG forecast between LTG5 and LTG1 is on average 24% and ranges from 18.4% to
us
38.9%. The regression result, shown in Table 1 (b), reveals that the dispersion in the LTG forecasts is significantly positively associated with the two sentiment indices. This result supports Hypothesis
an
1-1, which posits that bullish investor sentiment induces bold LTG forecasts. [Table 1]
M
[Figure 2]
4.2 Sentiment and the realized growth rate dispersion To test Hypothesis 1-2, I examine the association between investor sentiment and the realized
d
growth spread between predicted high-growth firms and firms with predicted low-growth. If the
te
realized growth spread is not positively associated with investor sentiment, the positive association between sentiment and LTG forecast dispersion is not aligned with the accuracy of the forecasts; in
Ac ce p
other words, the LTG forecasts boldness induced by high sentiment, lowers the accuracy of the forecasts.
The LTG forecasts refer to earnings growth, but realized growth in this variable is highly prone
to measurement problems (such as the exclusion of cases with negative earnings values). Therefore, in keeping with the study of Chan et al. (2003), I report not only realized EPS (earnings per share) growth but also realized growth in sales per share and operating income before depreciation (OIBD) per share. Growth rates in sales and OIBD are correlated with EPS growth, but are better behaved and are still available for a large fraction of the sample. Earnings/sales/OIBD per share is defined by the past four quarters’ earnings/sales/OIBD per share; I calculate the geometric average of the growth rate over three years (36 months) and five years (60 months). To analyze the relationship between sentiment and the realized growth dispersion, I run the following regression:
Growthh, X t = High,t ,t + s − Growthh, X t = Low,t ,t + s = ch.k , s + dh, k , s SENTIMENTk ,t + ε h, k , s,t
(2)
6
Page 6 of 23
Growthh , X t =High,t ,t +s (h=1,2,3) denotes a median value5 of EPS, OIBD, and sales growth over subsequent s months (s=36 or 60) for stocks included in the top LTG quintile (LTG5); on the other hand, Growthh , X t = Low,t ,t + s (h=1,2,3) indicates a median value of EPS, OIBD, and sales growth over subsequent s months for the bottom LTG quintile (LTG1). Thus, the dependent variable in
ip t
equation (2) is the spread in a firm’s realized growth rate between the top and bottom LTG quintiles.
Table 2 shows the regression results for equation (2) and descriptive statistics for realized
cr
growth dispersion. The mean spread in realized growth between LTG5 and LTG1 is about 11%
(three-year average) and 9% (five-year average). On the other hand, the spread in LTG forecast
us
(shown in Table 1) is on average 24%. The result suggests that investors are aggressively
distinguishing between firms with high and low growth prospects.
an
As shown in Table 2 (b), some regression results reveal that the dispersion in realized growth is significantly negatively associated with the two sentiment indices; the dispersion in three- and five-year realized EPS growth, five-year OIBD growth, and five-year sales growth are significantly
M
negatively associated with the Baker/Wurgler Index; and the dispersion in five-year realized EPS growth, five-year OIBD growth, and three- and five-year sales growth are significantly negatively associated with the Michigan Index. On the other hand, no result supports a positive association
d
between the dispersion in realized growth and market-wide investor sentiment. While dispersion in
te
investor LTG forecasts is greater following higher investor sentiment, the realized growth spread between them tends to be narrower following higher sentiment. Thus, my results support Hypothesis
Ac ce p
1-2, which posits that the positive association between sentiment and LTG boldness lowers the accuracy of the growth forecasts.
[Table 2]
4.3 Investor sentiment and correction of earnings forecasts To test Hypothesis 2-1, I examine the association between market-wide investor sentiment and
subsequent corrections (revisions) of the earnings forecasts. When bold LTG forecasts induced by bullish investor sentiment are corrected, stocks with higher-predicted growth should experience a more negative forecast revision. Thus, I evaluate the difference in the LTG forecast revisions between the top LTG quintile (LTG5) and the bottom LTG quintile (LTG1); I examine whether the difference in the subsequent one-month revisions in the LTG forecast is negatively associated with each sentiment index. In addition, for a robustness check, I also examine an association between sentiment and the difference in the revisions of next fiscal-year earnings forecasts between LTG5 5 Firms that grow from low positive values of base-year net income, introduce large outliers. To reduce the influence of these outliers, I evaluate a median value of firm growth rates instead of a mean growth rate.
7
Page 7 of 23
and LTG16. It is highly possible that stocks with higher-predicted growth also experience more negative revisions in their next fiscal-year earnings forecasts when the prior bold LTG forecasts induced by bullish investor sentiment are corrected. Thus, I examine whether the difference in revisions between the LTG5 and LTG1 in next fiscal-year earnings forecasts is negatively associated
ip t
with each market-wide sentiment index.
Re vh, X t = High,t − Re vh, X t = Low,t = ch.k + d h ,k SENTIMENTk ,t + ε h,k ,t
(3)
cr
To this end, I run the following regressions:
where Re vh , X t = High,t (h=1,2) denotes an average of revisions of next fiscal-year earnings and LTG
us
forecasts for LTG5 in the following month; Re vh , X t = Low, t denotes an average of the subsequent one-month revisions for LTG1. The dependent variable in equation (3) is the difference in earnings
an
forecast revisions between the top and bottom LTG quintile.
A subsequent one-month revision of next fiscal-year earnings and LTG forecast for firm i at the end of month t (I denote these as Rev1,i,t and Rev2,i,t, respectively) are defined as:
M
Rev1,i ,t ≡ ( fEPS i , yt + 2,t +1 − fEPS i , y t + 2 ,t ) Pi ,t
Rev 2 ,i ,t ≡ fLTGi ,t +1 − fLTGi ,t
d
where yt is the most recent fiscal year in which an earnings announcement was made at the end of
te
month t; fEPS i , y t + 2,t is firm i’s consensus annual EPS forecast for fiscal year yt+2 (next fiscal year) at the end of month t, fLTGi,t is the mean LTG forecast for firm i at the end of month t, and Pi,t is the
Ac ce p
stock price at the end of month t. Re vh , X t = High,t and Re v h , X t = Low,t are defined by mean Revh,i,t for LTG5 and LTG1, respectively.
Table 3 shows the regression results for equation (3) and descriptive statistics for the spreads in
earnings forecast revisions. The mean revisions in next fiscal-year earnings and LTG in the following month are negative for LTG5. Furthermore, the revisions in next fiscal-year earnings and LTG are lower for LTG5 than LTG1. Thus, stocks with high-predicted growth experience more negative earnings forecast revisions than those with low-predicted growth. The regression results, shown in Table 3(b), reveal that the difference in LTG forecast revisions between LTG5 and LTG1 is significantly negatively associated with the two sentiment indices. In addition, I find that the 6
I did not evaluate an association between market-wide sentiment and the difference in revisions in current fiscal-year earnings forecasts for the following reason: As the day of the earnings announcement for the fiscal-year ending draws closer, earnings results for the current fiscal year are gradually reported. At the end of the fiscal year, since earnings results for Q1, Q2, and Q3 of the fiscal year are already reported, a revision of the current fiscal-year earnings forecast reflects expectations regarding Q4 earnings. On the other hand, at the beginning of the fiscal year, since no earnings results are yet reported, a revision of the current fiscal-year earnings forecast could reflect expectations regarding earnings for Q1, Q2, Q3, and Q4 of the fiscal year. Thus, I cannot make a time series comparison in forecast revisions of current fiscal-year earnings.
8
Page 8 of 23
difference in the revision of next fiscal-year earnings forecast is also significantly negatively associated with these sentiment indices. These results indicate that stocks with higher-predicted earnings growth experience a more negative earnings forecast revision, especially following a period of bullish investor sentiment. In other words, these results support Hypothesis 2-1, which posits that
revisions when investor sentiment is higher at the beginning of the period.
cr
[Table 3]
ip t
there is a stronger negative association between the LTG forecasts and subsequent earnings forecast
us
4.4 Investor sentiment and subsequent price corrections
To test Hypothesis 2-2, I examine whether or not the stock return spread between LTG5 and LTG1 is negatively associated with investor sentiment. I should note that to examine the sentiment
an
effect on cross-sectional stock returns, it is necessary to distinguish sentiment effects from well-known co-movements: size, value (book-to-market), and momentum. Baker/Wurgler report that when the beginning period proxies for investor sentiment are low, stock returns are relatively high
M
for small stocks. They also predict that when stocks are sorted by book-to-market value, the subsequent stock returns at both extremes are influenced more by sentiment than other returns.
d
Furthermore, Antoniou et al. (2012) show that investor sentiment is positively associated with the profitability of price momentum strategies.
te
To distinguish the sentiment effect from these co-movements, I first calculate one-month characteristic-adjusted returns. Then, I examine whether differences in the adjusted returns between
Ac ce p
LTG5 and LTG1 are negatively associated with beginning period proxies for investor sentiment. I follow the characteristic-matching procedure in Daniel et al. (1997) to account for firm size,
momentum, and book-to-market effects in my analysis; I calculate one-month characteristic-adjusted returns as follows.
(i) I first divide the firms into five groups on the basis of firm size (log value of market
capitalization); (ii) within each group, the firms are again divided into five groups based on book-to-market ratio (book value for the most recent reported quarter to market); (iii) within each group, the firms are divided into five groups based on mid-term stock return (the return from t - 12 months to t - 2 months); (iv) after forming a set of 125 (5 X 5 X 5) groups, I subtract the return of the cap-weighted benchmark portfolio that stock belongs to from the return of that stock. Then, I run the following regression.
Ret X t = High,t − Ret X t = Low,t = ck + d k SENTIMENTk ,t + ε k ,t
(4)
9
Page 9 of 23
RetX = High,t ( RetX t
t
= Low,t
) denotes an average of characteristic-adjusted return over month t+1 for
LTG5 (LTG1). Thus, the dependent variable in equation (4) is the spread in a subsequent one-month characteristic-adjusted return between the top and bottom LTG quintiles.
ip t
The reason why I control for size, value, and momentum effects, by utilizing the characteristic-adjusted return, is that these effects on the return spread between LTG5 and LTG1 are
time-varying (I did not use the Carhert momentum and Fama-French factors as independent
cr
variables in the regression model (4), to control for these effects). Table 4 shows normalized values of size (log value of market capitalization), value (book value for the most recent reported quarter to
us
market), and momentum (the return from t - 12 months to t - 2 months) for each LTG quintile. Figure 3 plots the spreads in these variables between LTG5 and LTG1. As shown in the figure and the table, the spreads in these exposures are time-varying; on average, the size spread is negative,
an
however, during the dot-com bubble, there is little difference in size between LTG5 and LTG1. In addition, the spread in momentum highly fluctuates; during the dot-com bubble, the momentum
M
exposure is higher for LTG5 than for LTG1, whereas the momentum exposure is lower for LTG5 than for LTG1 after the bubble bursts. Thus, I should control for size, value, and momentum effects at each time point (at end of each month); therefore, I utilize characteristic-adjusted returns for
d
controlling these effects.
characteristic-adjusted
te
Table 5 shows the regression results for equation (4) and descriptive statistics for returns
for
each
LTG
quintile.
There
is
little
difference
in
Ac ce p
characteristic-adjusted returns between LTG5 and LTG1. However, the return spread between them highly fluctuates; the return spread ranges from -17.7% to 22.3%, whereas the average return spread is only -0.3%. The regression results shown in Table 5(b) reveal that the spread in characteristic-adjusted returns between LTG5 and LTG1 is significantly negatively associated with the two sentiment indices. Thus, the results indicate that stocks with high-predicted growth experience low subsequent stock returns, especially following periods of bullish investor sentiment; these support Hypothesis 2-2, which posits that there is a stronger negative association between the LTG forecasts and subsequent stock returns when proxies for investor sentiment are higher. In summary, my analyses reveal that stocks with higher-predicted growth experience more negative earnings revisions and lower stock returns, especially following periods of bullish sentiment. The findings support the inference that bullish investor sentiment induces excessively bold growth forecasts and these excessive bold forecasts cause stock mispricing, which is subsequently corrected through the correction of the bold earnings forecasts.
10
Page 10 of 23
While Baker/Wurgler report that there is no significant sentiment effect on the return spread between extremely high- and low-growth stocks, my results support the sentiment effect on the return spread between such stocks. A key difference between my study and theirs is that I sort out high-growth firms from low-growth ones on the basis of investor growth predictions, while
ip t
Baker/Wurgler sort out firms on the basis of past one-year sales growth. Although past firm growth
could be positively associated with firm expected growth rates (Chan et al., 2003), past firm sales
cr
growth is a naïve and indirect predictor for a firm’s expected earnings growth. Thus, the sentiment
[Figure 3]
an
[Table 4]
us
effect cannot be observed in the return spread between stocks with high and low past sales growth.
[Table 5]
5. Conclusions
M
Consistent with the considerable debate on the role and effect of investor sentiment on asset pricing, prior studies show that market-wide investor sentiment has an influence on stock prices by affecting investor preference for securities whose valuations are highly subjective and difficult to
d
arbitrage (Baker and Wurgler 2006, 2007; Hribar and McInnis 2012) and by affecting short-term
te
earnings expectations (Walther and Willis 2012). However, although LTG expectations are crucial to stock price valuations, no studies examine the possibility that investor sentiment induces stock
Ac ce p
mispricing by affecting the overall accuracy of investor LTG expectations. In this study, I contribute to the behavioral research by examining this possibility. I predict that
higher investor sentiment induces investors to aggressively distinguish high-growth firms from others, and these excessively bold LTG predictions induce stock mispricing, which is subsequently corrected through the correction of these bold predictions. To test this possibility, I examine whether bullish investor sentiment induces extreme dispersion
in consensus LTG forecasts. The analyses reveal that dispersion in LTG forecasts is larger following higher market-wide investor sentiment, although a realized growth spread between firms with high LTG forecasts and firms with low ones tends to be narrower following such sentiment. These findings support the inference that bullish investor sentiment induces investors to excessively separate firms with high-predicted growth from others. I then examine whether the excessively bold LTG forecasts induce stock mispricing. The analyses reveal that stocks with higher-predicted growth experience more negative forecast revisions
11
Page 11 of 23
and lower subsequent stock returns, especially following periods of bullish investor sentiment. These findings support the view that excessive dispersion in LTG forecasts is corrected in a subsequent period and that the correction affects cross-sectional stock returns by reducing the mispricing.
investors to make excessively bold LTG forecasts, induces stock mispricing.
ip t
Therefore, my findings support the inference that market-wide investor sentiment, by inducing
cr
The studies most closely related to mine are those of Baker and Wurgler (2006, 2007) and
Hribar and McInnis (2012); they focus on the impact of investor sentiment on difficult to value (hard
us
to arbitrage) stocks. A key difference between my study and theirs is that I focus on impact of market-wide sentiment on the return spread between stocks with high and low growth, while extreme high and low-growth stocks are regarded as difficult to value in their study. In addition, the
an
effect found by Baker/Wurgler is confined to small, newer, and volatile stocks. On the other hand, through investor expectations of firm long-term cash flow, I see the sentiment effect on the prices of
M
a wide-spread of stocks.
My evidence has potential implications regarding the debate between laissez-faire policy and government activism in the market. Since my analysis shows the wide-spread sentiment effect, the
d
evidence emphasizes the importance of controlling excess market optimism or pessimism. Daniel et
te
al. (2002) states that the market anomaly, which can be explained by investors’ sentiment, does not always support government activism because governments might have no ability in determining
Ac ce p
when asset prices are significantly distorted by investor sentiment. However, my results reveal that the sentiment effect on asset prices through long-term growth expectations can be predicted by the sentiment indices, which are objective and observable. Thus, by utilizing the sentiment indices, rules can be established to control excess optimism or pessimism, and the rules could reduce the irrational sentiment effect on asset pricing.
References
Antoniou, C., Doukas, J., Subrahmanyam, A., 2012. Cognitive dissonance, sentiment, and momentum. J. Financ. Quant. Anal, Forthcoming. Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns. J. Financ., 61, 1645-1680. Baker, M., Wurgler, J., 2007. Investor sentiment in the stock market. J. Econ., 21, 129-151. Copeland, T., Dolgoff, A., Moel, A., 2004. The role of expectations in explaining the cross-section of
12
Page 12 of 23
stock returns. Rev. Account. Stud., 9, 149-188. Chan, L., Karceski, J., Lakonishok, J., 2003. The level and persistence of growth rates. J. Financ., 58, 643-684.
characteristic-based benchmarks. J. Financ., 52, 1035-1058.
ip t
Daniel, K., Grinblatt, M., Titman, S., Wermers. R., 1997. Measuring mutual fund performance with
Daniel, K., Hirshleifer, D., Teoh, S., 2002. Investor psychology in capital markets: evidence and
cr
policy implications. J. Monetary. Econ., 49,139-209.
us
De Long, B., Shleifer, A.,. Summers, L. H., Waldmann, R. J. 1990. Noise trader risk in financial markets. J. Polit. Econ., 98, 703-738.
Dechow, M., Sloan, G., 1997. Returns to contrarian investment strategies: tests of naive expectations
an
hypotheses. J. Financ. Econ., 43, 3-27.
Dechow, M., Hutton, P., Sloan, G., 2000. The relation between analysts’ forecasts of long-term
M
earnings growth and stock price performance following equity offerings. Contemp. Account. Res., 17, 1-32.
Frankel, R., Lee, C., 1998. Accounting valuation, market expectation, and cross-sectional stock
d
returns. J. Account. Econ., 25, 283-319.
Res., 39, 135-176.
te
Gebhardt, W., Lee, C., Swaminathan, B., 2001. Toward an implied cost of capital. J. Accounting
Ac ce p
Gordon, M., 1962. The investment, financing, and valuation of the corporation. Irwin, Homewood. Hribar, P., McInnis, J., 2012. Investor sentiment and analysts' earnings forecast errors. Manage. Sci., 58, 293-307.
La Porta, R., 1996. Expectations and the cross-section of stock returns. J. Financ., 51, 1715-1742. Lakonishok, J., Shleifer, A., Vishny, R., 1994. Contrarian investment, extrapolation, and risk. J. Financ., , 49, 1541-1578.
Lemmon, M., Portniaguina, E., 2006. Consumer confidence and asset prices: some empirical evidence. Rev. Financ. Stud., 19, 1499-1539. Mezrich, J., Zeng, Q., Nordquist, D., Seshadri, L., 2001. Rebroadcast: Changing of the guard. Equity Research Quantitative Strategy, Morgan Stanley, New York. Saade, S., 2015. Investor sentiment and the underperformance of technology firms initial public offerings. Res. Int. Bus. Financ., 34, 205-232.
13
Page 13 of 23
Shleifer, A., Vishny, R., 1997. The limits of arbitrage. J. Financ., 52, 35-55. Walther, B., Willis, R. 2012. Do investor expectations affect sell-side analysts' forecast bias and forecast accuracy? Rev. Account. Stud., 17, 1-21.
Ac ce p
te
d
M
an
us
cr
ip t
Whaley, E., 2000. The Investor Fear Gauge. J. Portfolio Manage., 26, 12-17.
14
Page 14 of 23
Table 1 LTG forecast dispersion Each month, from 1986 to 2006, I collect consensus (mean) forecasts of long-term earnings growth (LTG) from IBES.
ip t
Each month, firms are sorted into quintiles on consensus LTG forecast from LTG5 (High) to LTG1 (low). I calculate an average of consensus LTG forecast for each LTG quintile portfolio. I take the difference in the extreme quintiles (1
cr
and 5) to construct forecast dispersion portfolios (denoted as “LTG5-LTG1”). Panel (a) shows descriptive statistics for each portfolio’s LTG forecast. In panel (b), I show ordinary least squares regression results for Eq. (1) in the text.
us
The column “B&W Index” shows a time-series average of coefficient of the index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series average of coefficient of Michigan Consumer Sentiment Index. The figures in parentheses are t-statistics based on Newey-West standard errors.
Regression results
M
LTG3 13.9% 16.2% 12.0% 16.1% 12.1% 1.5%
B&W Index 0.0675 (7.35)
LTG dispersion
LTG2 10.8% 12.5% 9.3% 12.5% 9.4% 1.0%
LTG1 6.4% 8.2% 4.7% 7.8% 5.0% 0.9%
LTG5-LTG1
24.0% 38.9% 18.4% 29.2% 19.7% 5.0%
Michigan Index 0.0037 (4.70)
Ac ce p
b)
LTG4 17.9% 22.9% 15.2% 22.1% 15.6% 2.6%
d
Mean Max Min 90th 10th Std.dev.
LTG5 30.4% 46.9% 24.0% 36.9% 25.6% 5.7%
an
Descriptive statistics
te
a)
15
Page 15 of 23
Table 2 Realized growth spread Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (High) to LTG1 (low). I calculate an average of realized growth rate for each LTG quintile. The row of “3-year EPS growth,” “5-year EPS growth,”
ip t
“3-year OIBD growth,” “5-year OIBD growth,” “3-year sales growth,” and “5-year sales growth” shows the regression results for when 3-year EPS growth, 5-year EPS growth, 3-year OIBD growth, 5-year OIBD growth,
cr
3-year sales growth, and 5-year sales growth are used for evaluating realized growth dispersion. Panel (a) shows descriptive statistics for each portfolio’s realized growth and the difference in realized growth between LTG5 and
us
LTG1. In panel (b), I show ordinary least squares regression results for Eq. (2) in the text. The column “B&W Index” shows a time-series average of the coefficient of the index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series average of the coefficient of the Michigan Consumer Sentiment Index. The
an
figures in parentheses are t-statistics based on Newey-West standard errors.
Mean 11.5% 8.8% 11.7% 8.5% 10.9% 8.7%
Max 25.5% 19.3% 22.6% 15.6% 19.0% 14.6%
(b) Regression results
d
Ac ce p
3-year EPS growth 5-year EPS growth 3-year OIBD growth 5-year OIBD growth 3-year sales growth 5-year sales growth
Min -4.7% -1.8% 2.5% 1.6% 0.6% 1.7%
M
3-year EPS growth 5-year EPS growth 3-year OIBD growth 5-year OIBD growth 3-year sales growth 5-year sales growth
te
(a) Descriptive statistics
B & W Index -0.0371 (2.07) -0.0253 (2.02) -0.0177 (0.84) -0.0161 (1.68) -0.0214 (0.72) -0.0237 (2.12)
90th 18.2% 12.1% 17.5% 12.0% 15.5% 12.5%
10th 3.7% 5.7% 6.7% 5.0% 6.1% 4.5%
Std.dev. 5.6% 3.1% 4.1% 2.8% 4.0% 3.0%
Michigan Index -0.0015 (0.97) -0.0011 (2.76) -0.0008 (0.97) -0.0014 (5.04) -0.0014 (1.77) -0.0021 (6.91)
16
Page 16 of 23
Table 3 Earnings forecast revisions Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (high) to LTG1 (low). I calculate an average of earnings forecast revisions in the following month for each LTG quintile. The rows of the “EPS
ip t
forecast revision” and the “LTG forecast revision” show the result for the difference in forecast revisions with regard to EPS for the next fiscal year and LTG, respectively. Panel (a) shows descriptive statistics for the portfolio’s forecast
cr
revisions and the difference in earnings forecast revisions between LTG5 and LTG1. Panel (b) shows ordinary least
squares regression results for Eq. (3). The column “B&W Index” shows a time-series average of the coefficient of the
us
index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series average of the coefficient of Michigan Consumer Sentiment Index. The figures in parentheses are t-statistics based on White's
an
heteroscedasticity consistent standard errors.
(a) Descriptive statistics
LTG forecast revision Mean LTG5-LTG1 -0.46% LTG5 -0.35% LTG1 0.10%
Max 0.36% 0.43% 0.49%
Min -0.71% -0.96% -0.83%
M
Max 0.61% 0.01% 0.15%
Ac ce p
te
d
EPS forecast revison Mean LTG5-LTG1 -0.08% LTG5 -0.18% LTG1 -0.10%
Min -1.85% -1.65% -0.21%
90th 0.06% -0.06% 0.01%
10th -0.23% -0.31% -0.26%
Std.dev. 0.14% 0.14% 0.12%
90th -0.17% -0.13% 0.22%
10th -0.82% -0.63% 0.00%
Std.dev. 0.29% 0.26% 0.10%
(b) Regression results
EPS forecast revison LTG forecast revision
B&W Index -0.0008 (4.72) -0.0019 (3.81)
Michigan Index -0.00003 (2.60) -0.00004 (1.80)
17
Page 17 of 23
Table 4 Size, book-to-market, and momentum exposures Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (high) to LTG1 (low). I calculate an average for the normalized book-to-market ratio, log value of market capitalization (size), and the return from t -
ip t
12 months to t - 2 months (momentum) for each LTG quintile portfolio. I take the difference in the extreme quintiles
(denoted as “LTG5-LTG1”). The tables show descriptive statistics for the portfolio’s book-to-market ratio, size, and
cr
momentum, and the difference in those variables between LTG5 and LTG1.
LTG5 -0.37 0.10 -0.58 -0.26 -0.47 0.11
LTG4 -0.13 0.03 -0.30 -0.04 -0.22 0.07
te
Ac ce p
Mean Max Min 90th 10th Std.dev.
LTG2 0.26 0.56 -0.01 0.44 0.09 0.12
LTG1 0.50 0.86 0.11 0.68 0.29 0.15
LTG5-LTG1
LTG3 0.16 0.38 -0.11 0.29 0.01 0.11
LTG2 0.25 0.48 0.05 0.36 0.14 0.09
LTG1 0.23 0.40 -0.04 0.35 0.05 0.11
LTG5-LTG1
LTG3 -0.03 0.26 -0.23 0.10 -0.15 0.10
LTG2 -0.02 0.41 -0.41 0.20 -0.24 0.17
LTG1 0.00 0.46 -0.39 0.26 -0.25 0.19
LTG5-LTG1
d
b) Size
LTG3 0.03 0.27 -0.27 0.15 -0.14 0.11
an
LTG4 -0.17 0.08 -0.42 0.00 -0.33 0.13
M
Mean Max Min 90th 10th Std.dev.
LTG5 -0.47 0.01 -0.75 -0.21 -0.66 0.17
us
a) Book-to-market ratio
-0.96 -0.15 -1.48 -0.56 -1.29 0.30
-0.60 0.11 -0.91 -0.33 -0.82 0.19
c) Momentum
Mean Max Min 90th 10th Std.dev.
LTG5 0.05 0.90 -1.09 0.51 -0.42 0.37
LTG4 -0.03 0.36 -0.33 0.10 -0.15 0.11
0.05 1.25 -1.46 0.73 -0.67 0.54
18
Page 18 of 23
Table 5 Return spreads between high-growth and low-growth stocks Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (high) to LTG1 (low). I calculate an average of characteristic-adjusted returns in the following month for each LTG quintile portfolio. I take the
ip t
difference in the extreme quintiles (denoted as “LTG5-LTG1”). Panel (a) shows descriptive statistics for the
portfolio’s returns and the difference in characteristic-adjusted returns between LTG5 and LTG1. Panel (b) shows
cr
ordinary least squares regression results for Eq. (4). The column “B&W Index” shows a time-series average of the coefficient of the index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series
based on White's heteroscedasticity consistent standard errors.
Stock return
M
LTG3 -0.1% 3.9% -6.5% 1.1% -1.3% 1.1%
d
B&W Index -0.0144 (1.94)
Ac ce p
(b) Regression results
LTG4 -0.1% 13.0% -6.0% 1.9% -2.3% 2.0%
te
Mean Max Min 90th 10th Std.dev.
LTG5 -0.3% 20.7% -15.4% 4.0% -5.0% 4.1%
an
(a) Descriptive statistics
us
average of the coefficient of the Michigan Consumer Sentiment Index. The figures in parentheses are t-statistics
LTG2 -0.1% 3.3% -4.5% 0.9% -1.2% 0.9%
LTG1 -0.1% 4.4% -3.1% 1.1% -1.4% 1.1%
LTG5-LTG1
-0.3% 22.3% -17.7% 4.8% -5.9% 4.6%
Michigan Index -0.00059 (1.91)
19
Page 19 of 23
Figure 1
ip t
Sentiment Indices
30.00
3
Michigan Index (left axis)
2.5
20.00
cr
Baker & Wurgler Index (right axis)
us
10.00
an
0.00
-10.00
M
-20.00
1.5 1 0.5 0 -0.5 -1 -1.5
te
d
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
-30.00
2
Monthly values of the investor sentiment index constructed by Baker and Wurgler (Baker & Wurgler Index) and the
Ac ce p
Michigan Consumer Sentiment Index (Michigan Index). To ensure that both indices are free of macroeconomic influences, following Baker and Wurgler (2006, 2007), I conduct the investigation using an orthogonal version of the indices, which is obtained by regressing the indices on a set of macroeconomic variables.
20
Page 20 of 23
Figure 2 LTG dispersion and sentiment indices 3 65% 2.5
Dispersion in long-term growth forecasts (left axis) Baker & Wurgler Index (right axis)
45%
2
ip t
55%
1.5
35%
cr
1
25%
us
15% 5%
an
-5%
0 -0.5 -1 -1.5
M
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
-15%
0.5
A. Baker & Wurgler Index and LTG dispersion
35%
Ac ce p
30%
te
40%
d
45%
30
20
10
25%
0
20% 15% 10% 5%
-10
Dispersion in long-term growth forecasts (left axis) Michigan Index (right axis)
-20
-30
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
0%
B. Michigan Consumer Sentiment Index and LTG dispersion Panel A (B) shows monthly values of the investor sentiment index constructed by Baker and Wurgler (the Michigan Consumer Sentiment Index) and the dispersion between the top and bottom quintile of IBES long-term forecasts, denoted by “Dispersion in long-term growth forecasts.”
21
Page 21 of 23
Figure 3 Size, book-to-market, and momentum exposures
ip t
2.00
1.50
B/M
MOMENTUM
SIZE
cr
1.00
us
0.50
0.00
an
-0.50
-1.00
M
-1.50
d
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
-2.00
The graph shows monthly values of differences in normalized Book-to-Market ratio (B/M), log value of market
Ac ce p
LTG1.
te
capitalization (SIZE), and the return from t - 12 months to t - 2 months (MOMENTUM), between LTG5 and
22
Page 22 of 23
d
te
Ac ce p
25%
20% 0
ip t
15%
cr
10%
5%
Dispersion in long-term growth forecasts (left axis) Michigan Index (right axis)
0%
us
an
M
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
45% 30
40%
35% 20
30% 10
-10
-20
-30
23
Page 23 of 23
S0275-5319(15)30045-3 http://dx.doi.org/doi:10.1016/j.ribaf.2015.10.003 RIBAF 420
To appear in:
Research in International Business and Finance
Received date: Revised date: Accepted date:
20-2-2015 2-10-2015 5-10-2015
Please cite this article as: Miwa, K.,Investor sentiment, stock mispricing, and longterm growth expectations, Research in International Business and Finance (2015), http://dx.doi.org/10.1016/j.ribaf.2015.10.003 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Investor sentiment, stock mispricing, and long-term growth expectations
ip t
KOTARO MIWA1 Tokio Marine Asset Management Co., Ltd
cr
Abstract
us
I analyze whether or not market-wide investor sentiment induces stock mispricing, by affecting the boldness of predictions of firms’ long-term earnings growth. I predict that bullish market-wide sentiment induces investors to aggressively separate firms with high growth futures from others, and
an
that this excessive boldness results in a high level of mispricing. Consistent with my prediction, I observe an excessively large dispersion in consensus growth forecasts when proxies for investor sentiment are high at the beginning of the period. Furthermore, stocks with higher-predicted growth
M
experience more negative forecast revisions and lower subsequent stock returns, especially following periods of high investor sentiment.
te
stock mispricing
d
Key words: investor sentiment; long-term earnings growth forecast; cross-sectional stock return,
Ac ce p
Investor sentiment, stock mispricing, and long-term growth expectations Abstract
I analyze whether or not market-wide investor sentiment induces stock mispricing, by affecting the boldness of predictions of firms’ long-term earnings growth. I predict that bullish market-wide sentiment induces investors to aggressively separate firms with high growth futures from others, and that this excessive boldness results in a high level of mispricing. Consistent with my prediction, I observe an excessively large dispersion in consensus growth forecasts when proxies for investor sentiment are high at the beginning of the period. Furthermore, stocks with higher-predicted growth experience more negative forecast revisions and lower subsequent stock returns, especially following 1 1-3-1, Marunouchi, Chiyoda-ku, Tokyo, Japan, phone: + 81-3-3212-8186 - fax: +81-3-3212-7576, e-mail: [email protected]
1
Page 1 of 23
periods of high investor sentiment. Key words: investor sentiment; long-term earnings growth forecast; cross-sectional stock return,
ip t
stock mispricing
1. Introduction
Considerable financial studies debate the role and effect of investor sentiment on asset pricing.
cr
Investor sentiment is seen as crowd psychology, revealed through the activity and asset price
movements in the market. Traditional asset pricing theory suggests that rational arbitrage necessarily
us
forces prices closer to fundamentals and leaves no role for investor sentiment. However, several behavioral finance studies argue that investor sentiment drives stock prices away from fundamental values (De Long et al. 1990; Shleifer and Vishny 1997).
an
Consistent with this argument, studies show that time-varying investor sentiment affects stock prices. Lemmon and Portniaguina (2006) and Baker and Wurgler (2006, 2007; hereafter
M
“Baker/Wurgler”) use a measure of investor sentiment to show that difficult-to-value stocks (i.e., small, newer, volatile stocks) are overvalued, especially when sentiment is high. Furthermore, Hribar and McInnis (2012) find that those stocks receive optimistic earnings forecasts during such periods.
d
They argue that, since difficult-to-value stocks are highly influenced by the trading activity of noise traders, valuation of these stocks is considerably affected by investor sentiment2.
te
On the other hand, some studies analyze the sentiment effects on the overall accuracy of analysts’ earnings forecasts, which can be regarded as proxies for investor expectations of future
Ac ce p
payoffs. Walther and Willis (2012) show that financial optimism regarding quarterly earnings forecasts is affected by investor sentiment. These studies raise the possibility that investor sentiment induces stock mispricing by affecting market participants’ expectations about a firm’s future payoffs. However, while prior studies focus on the sentiment effects on the overall accuracy of forecasts of near term payoffs, few studies analyze the sentiment effect on long-term earnings growth expectations.
Long-term earnings growth expectations are crucial to stock price valuations. In many valuation
models, an estimated intrinsic value of a firm’s stock depends on long-term earnings growth (LTG) expectations (Frankel and Lee 1998; Gebhardt et al. 2001). For example, according to the Gordon growth model (1962), a price-to-dividend ratio of 20 implies that a 1% increase in long-term dividend growth translates into a 20% return. Consistent with the implication of the model, Copeland et al. (2004) report that analysts’ long-term earnings forecasts, a proxy for investor long-term growth 2
Other sentiment effects include the effect on the performance of technology IPO shares (Saade, 2015).
2
Page 2 of 23
expectations, exert a great influence on stock prices. Therefore, to further pinpoint the sentiment effect on asset pricing, it is essential to examine whether investor sentiment affects investor long-term earnings expectations and thereby induces
ip t
mispricing. The extant research reveals that investors’ long-term growth expectations, which are
represented by analysts’ long-term earnings forecasts, are, on average, too extreme, and this results
cr
in economically significant mispricing. La Porta (1996) and Chan et al. (2003) reveal that firms with
high predicted growth are aggressively distinguished from others. Research shows that this excessive
us
boldness regarding LTG expectations induces significant mispricing; subsequent stock returns are negatively associated with LTG forecasts (Lakonishok et al. 1994; La Porta 1996; Dechow and
an
Sloan 1997; Dechow et al. 2000).
Investors may more readily bet on firms’ growth potential when market-wide sentiment is more supportive of their preferences, i.e., market-wide sentiment is higher. Thus, I predict that higher
M
sentiment encourages investors to more aggressively distinguish firms with high-predicted growth from others. These excessively bold predictions are then corrected in a subsequent period, and this correction affects subsequent cross-sectional stock returns by reducing mispricing, which was
d
induced by the prior excessively bold predictions. Thus, I predict that firms with higher expected
te
growth experience more negative earnings forecast revisions and lower stock returns, especially after periods of bullish sentiment.
Ac ce p
In this study, I present empirical analyses to test these predictions. Following the study of
Baker/Wurgler, I use their market-wide investor sentiment index, and the Michigan Consumer Sentiment Index, to explore investor sentiment effects. In addition, following other prior studies (e.g. Mezrich et al. 2001; Chan et al. 2003), I utilize analyst LTG forecasts as a representative proxy for investor expectations of firm long-term earnings growth. The paper proceeds as follows. Section 2 details the development of my hypotheses. Section 3
presents the sample and descriptive statistics. Section 4 describes the methodologies and results. Finally, I summarize my findings in Section 5.
2. Theory To test the possibility that investor sentiment affects the boldness of LTG expectations and thereby induces stock mispricing, I examined the following hypotheses.
2.1 Investor sentiment and bold growth forecasts
3
Page 3 of 23
First, I examined the association between market-wide investor sentiment and investor excessive boldness in LTG predictions. I predict that bullish market-wide sentiment encourages investors to aggressively distinguish firms with high-predicted growth from others, and the boldness of their long-term growth forecasts lowers the accuracy of these growth forecasts. Hence, first I test
ip t
the following two hypotheses:
Hypothesis 1-1: Bullish investor sentiment induces bold LTG forecasts (boldness in LTG forecasts is
cr
positively associated with investor sentiment)
Hypothesis 1-2: The positive association between investor sentiment and forecast boldness lowers
us
the accuracy of these growth forecasts.
The Hypothesis 1-1 addresses the sentiment effect on boldness in LTG forecasts; Hypothesis
an
1-2 addresses whether the sentiment effect is aligned with forecast accuracy. If both are satisfied, I can say that bullish investor sentiment induces excessive boldness in LTG forecasts.
M
2.2 Correction of bold forecasts and mispricing
Second, I examined whether excessively bold LTG forecasts caused by bullish market-wide investor sentiment induces stock mispricing. The excessive boldness in the LTG forecasts could be
d
corrected in a subsequent period and that correction could affect cross-sectional stock returns by reducing mispricing, which was induced by the prior excessively bold predictions. If this is the case,
te
firms with high-predicted growth should experience more negative (downward) forecast revisions and lower subsequent stock returns than firms with low-predicted growth, especially when proxies
Ac ce p
for investor sentiment are high at the beginning of the period. Accordingly, the hypotheses to be tested are:
Hypothesis 2-1: There is a stronger negative association between the LTG forecasts and subsequent earnings forecast revisions when proxies for investor sentiment are higher. Hypothesis 2-2: There is a stronger negative association between the LTG forecasts and subsequent stock returns when proxies for investor sentiment are higher.
3. Sample and descriptive statistics As mentioned earlier, following previous studies (e.g., Mezrich et al. 2001; Chan et al. 2003), consensus Institutional Brokers Estimate System (IBES) forecasts are used for investor consensus expectations of firm earnings and earnings growth rates. I obtain the sample of earnings forecasts from an unadjusted file of the IBES summary.
4
Page 4 of 23
I collect data from stocks listed on the New York Stock Exchange (NYSE), American Stock Exchange (Amex), and NASDAQ that received LTG forecasts. I exclude the shares of non-US firms and stocks priced below $3 to ensure that the empirical findings are not driven by low prices. I utilized data from the beginning of 1986 to the end of 2011 (monthly). To evaluate realized growth, I
ip t
required five years of realized earnings growth. Thus, the final year of the sample is 2006. The number of eligible firms ranged from 1,593 to 3,457; on average, the sample comprised about 2,500
cr
firms.
For the main part of my analysis, I measure investor sentiment using the monthly time series of
us
the investor sentiment index constructed by Baker/Wurgler and the Michigan Consumer Sentiment Index, since these indices are used in extant investor-sentiment studies3. To ensure that both indices
an
are free of macroeconomic influences, following the study by Baker/Wurgler, I conduct my investigation using an orthogonal version of the indices, which is obtained by regressing the indices against a set of macroeconomic variables4. Figure 1 plots the two series. The two sentiment indices
M
rise during the mid-1980s and late 1990s, and fall during the early 1990s and early 2000s. These patterns are in line with the evidence for investor sentiment discussed by Baker/Wurgler.
d
[Figure 1]
te
4. Methods and results
4.1 Sentiment and dispersion in the LTG forecasts
Ac ce p
To test Hypothesis 1-1, I examine the association between investor sentiment and LTG forecast boldness. Chan et al. (2003) evaluate boldness in LTG forecasts by dispersion in consensus LTG forecasts. Following their methodology, each month firms are sorted into quintiles on consensus (mean) LTG forecasts, from LTG5 (High) to LTG1 (low). I calculate an average for the consensus LTG forecasts for each LTG quintile. I define the dispersion in the LTG forecasts as the spread in predicted growth between the top and bottom quintiles of the LTG forecasts. Next, I run the following regressions:
LTGX t = High, t − LTGX t = Low,t = ck + d k SENTIMENTk ,t + ε k ,t LTG X
t
= High , t
(1)
denotes an average of the consensus LTG forecasts for stocks included in the top
3
Practitioners often consider the volatility (VIX) index as the investor sentiment index. However, Whaley (2000) observes that the VIX index is an indicator of investor fear of the market downside and cannot capture investor optimism during a market boom. My study examines whether or not high investor sentiment (investor optimism) induces excessive boldness regarding long-term growth expectations. Thus, I do not use the VIX index, as it cannot capture market-wide optimism as well as the investor sentiment index in my study. 4 Please refer to Baker and Wurgler (2006, 2007) for more detail on the methodology to obtain macroeconomic free indices.
5
Page 5 of 23
LTG quintile (I denote these stocks as LTG5); LTGX =Low,t denotes an average of consensus LTG t
forecasts for stocks included in the bottom LTG quintile (I denote these stocks as LTG1). Thus, the dependent variable in the equation (1) is dispersion in a firm’s consensus long-term growth forecast between the top and bottom quintile; SENTIMENTk,t (k=1,2) denotes the Baker/Wurgler Sentiment
ip t
Index and the Michigan Consumer Sentiment Index at the end of month t, respectively.
Figure 2 plots the two sentiment index series and dispersion in the consensus LTG forecasts.
cr
Table 1 shows the regression results and descriptive statistics for growth forecast dispersion. The difference in LTG forecast between LTG5 and LTG1 is on average 24% and ranges from 18.4% to
us
38.9%. The regression result, shown in Table 1 (b), reveals that the dispersion in the LTG forecasts is significantly positively associated with the two sentiment indices. This result supports Hypothesis
an
1-1, which posits that bullish investor sentiment induces bold LTG forecasts. [Table 1]
M
[Figure 2]
4.2 Sentiment and the realized growth rate dispersion To test Hypothesis 1-2, I examine the association between investor sentiment and the realized
d
growth spread between predicted high-growth firms and firms with predicted low-growth. If the
te
realized growth spread is not positively associated with investor sentiment, the positive association between sentiment and LTG forecast dispersion is not aligned with the accuracy of the forecasts; in
Ac ce p
other words, the LTG forecasts boldness induced by high sentiment, lowers the accuracy of the forecasts.
The LTG forecasts refer to earnings growth, but realized growth in this variable is highly prone
to measurement problems (such as the exclusion of cases with negative earnings values). Therefore, in keeping with the study of Chan et al. (2003), I report not only realized EPS (earnings per share) growth but also realized growth in sales per share and operating income before depreciation (OIBD) per share. Growth rates in sales and OIBD are correlated with EPS growth, but are better behaved and are still available for a large fraction of the sample. Earnings/sales/OIBD per share is defined by the past four quarters’ earnings/sales/OIBD per share; I calculate the geometric average of the growth rate over three years (36 months) and five years (60 months). To analyze the relationship between sentiment and the realized growth dispersion, I run the following regression:
Growthh, X t = High,t ,t + s − Growthh, X t = Low,t ,t + s = ch.k , s + dh, k , s SENTIMENTk ,t + ε h, k , s,t
(2)
6
Page 6 of 23
Growthh , X t =High,t ,t +s (h=1,2,3) denotes a median value5 of EPS, OIBD, and sales growth over subsequent s months (s=36 or 60) for stocks included in the top LTG quintile (LTG5); on the other hand, Growthh , X t = Low,t ,t + s (h=1,2,3) indicates a median value of EPS, OIBD, and sales growth over subsequent s months for the bottom LTG quintile (LTG1). Thus, the dependent variable in
ip t
equation (2) is the spread in a firm’s realized growth rate between the top and bottom LTG quintiles.
Table 2 shows the regression results for equation (2) and descriptive statistics for realized
cr
growth dispersion. The mean spread in realized growth between LTG5 and LTG1 is about 11%
(three-year average) and 9% (five-year average). On the other hand, the spread in LTG forecast
us
(shown in Table 1) is on average 24%. The result suggests that investors are aggressively
distinguishing between firms with high and low growth prospects.
an
As shown in Table 2 (b), some regression results reveal that the dispersion in realized growth is significantly negatively associated with the two sentiment indices; the dispersion in three- and five-year realized EPS growth, five-year OIBD growth, and five-year sales growth are significantly
M
negatively associated with the Baker/Wurgler Index; and the dispersion in five-year realized EPS growth, five-year OIBD growth, and three- and five-year sales growth are significantly negatively associated with the Michigan Index. On the other hand, no result supports a positive association
d
between the dispersion in realized growth and market-wide investor sentiment. While dispersion in
te
investor LTG forecasts is greater following higher investor sentiment, the realized growth spread between them tends to be narrower following higher sentiment. Thus, my results support Hypothesis
Ac ce p
1-2, which posits that the positive association between sentiment and LTG boldness lowers the accuracy of the growth forecasts.
[Table 2]
4.3 Investor sentiment and correction of earnings forecasts To test Hypothesis 2-1, I examine the association between market-wide investor sentiment and
subsequent corrections (revisions) of the earnings forecasts. When bold LTG forecasts induced by bullish investor sentiment are corrected, stocks with higher-predicted growth should experience a more negative forecast revision. Thus, I evaluate the difference in the LTG forecast revisions between the top LTG quintile (LTG5) and the bottom LTG quintile (LTG1); I examine whether the difference in the subsequent one-month revisions in the LTG forecast is negatively associated with each sentiment index. In addition, for a robustness check, I also examine an association between sentiment and the difference in the revisions of next fiscal-year earnings forecasts between LTG5 5 Firms that grow from low positive values of base-year net income, introduce large outliers. To reduce the influence of these outliers, I evaluate a median value of firm growth rates instead of a mean growth rate.
7
Page 7 of 23
and LTG16. It is highly possible that stocks with higher-predicted growth also experience more negative revisions in their next fiscal-year earnings forecasts when the prior bold LTG forecasts induced by bullish investor sentiment are corrected. Thus, I examine whether the difference in revisions between the LTG5 and LTG1 in next fiscal-year earnings forecasts is negatively associated
ip t
with each market-wide sentiment index.
Re vh, X t = High,t − Re vh, X t = Low,t = ch.k + d h ,k SENTIMENTk ,t + ε h,k ,t
(3)
cr
To this end, I run the following regressions:
where Re vh , X t = High,t (h=1,2) denotes an average of revisions of next fiscal-year earnings and LTG
us
forecasts for LTG5 in the following month; Re vh , X t = Low, t denotes an average of the subsequent one-month revisions for LTG1. The dependent variable in equation (3) is the difference in earnings
an
forecast revisions between the top and bottom LTG quintile.
A subsequent one-month revision of next fiscal-year earnings and LTG forecast for firm i at the end of month t (I denote these as Rev1,i,t and Rev2,i,t, respectively) are defined as:
M
Rev1,i ,t ≡ ( fEPS i , yt + 2,t +1 − fEPS i , y t + 2 ,t ) Pi ,t
Rev 2 ,i ,t ≡ fLTGi ,t +1 − fLTGi ,t
d
where yt is the most recent fiscal year in which an earnings announcement was made at the end of
te
month t; fEPS i , y t + 2,t is firm i’s consensus annual EPS forecast for fiscal year yt+2 (next fiscal year) at the end of month t, fLTGi,t is the mean LTG forecast for firm i at the end of month t, and Pi,t is the
Ac ce p
stock price at the end of month t. Re vh , X t = High,t and Re v h , X t = Low,t are defined by mean Revh,i,t for LTG5 and LTG1, respectively.
Table 3 shows the regression results for equation (3) and descriptive statistics for the spreads in
earnings forecast revisions. The mean revisions in next fiscal-year earnings and LTG in the following month are negative for LTG5. Furthermore, the revisions in next fiscal-year earnings and LTG are lower for LTG5 than LTG1. Thus, stocks with high-predicted growth experience more negative earnings forecast revisions than those with low-predicted growth. The regression results, shown in Table 3(b), reveal that the difference in LTG forecast revisions between LTG5 and LTG1 is significantly negatively associated with the two sentiment indices. In addition, I find that the 6
I did not evaluate an association between market-wide sentiment and the difference in revisions in current fiscal-year earnings forecasts for the following reason: As the day of the earnings announcement for the fiscal-year ending draws closer, earnings results for the current fiscal year are gradually reported. At the end of the fiscal year, since earnings results for Q1, Q2, and Q3 of the fiscal year are already reported, a revision of the current fiscal-year earnings forecast reflects expectations regarding Q4 earnings. On the other hand, at the beginning of the fiscal year, since no earnings results are yet reported, a revision of the current fiscal-year earnings forecast could reflect expectations regarding earnings for Q1, Q2, Q3, and Q4 of the fiscal year. Thus, I cannot make a time series comparison in forecast revisions of current fiscal-year earnings.
8
Page 8 of 23
difference in the revision of next fiscal-year earnings forecast is also significantly negatively associated with these sentiment indices. These results indicate that stocks with higher-predicted earnings growth experience a more negative earnings forecast revision, especially following a period of bullish investor sentiment. In other words, these results support Hypothesis 2-1, which posits that
revisions when investor sentiment is higher at the beginning of the period.
cr
[Table 3]
ip t
there is a stronger negative association between the LTG forecasts and subsequent earnings forecast
us
4.4 Investor sentiment and subsequent price corrections
To test Hypothesis 2-2, I examine whether or not the stock return spread between LTG5 and LTG1 is negatively associated with investor sentiment. I should note that to examine the sentiment
an
effect on cross-sectional stock returns, it is necessary to distinguish sentiment effects from well-known co-movements: size, value (book-to-market), and momentum. Baker/Wurgler report that when the beginning period proxies for investor sentiment are low, stock returns are relatively high
M
for small stocks. They also predict that when stocks are sorted by book-to-market value, the subsequent stock returns at both extremes are influenced more by sentiment than other returns.
d
Furthermore, Antoniou et al. (2012) show that investor sentiment is positively associated with the profitability of price momentum strategies.
te
To distinguish the sentiment effect from these co-movements, I first calculate one-month characteristic-adjusted returns. Then, I examine whether differences in the adjusted returns between
Ac ce p
LTG5 and LTG1 are negatively associated with beginning period proxies for investor sentiment. I follow the characteristic-matching procedure in Daniel et al. (1997) to account for firm size,
momentum, and book-to-market effects in my analysis; I calculate one-month characteristic-adjusted returns as follows.
(i) I first divide the firms into five groups on the basis of firm size (log value of market
capitalization); (ii) within each group, the firms are again divided into five groups based on book-to-market ratio (book value for the most recent reported quarter to market); (iii) within each group, the firms are divided into five groups based on mid-term stock return (the return from t - 12 months to t - 2 months); (iv) after forming a set of 125 (5 X 5 X 5) groups, I subtract the return of the cap-weighted benchmark portfolio that stock belongs to from the return of that stock. Then, I run the following regression.
Ret X t = High,t − Ret X t = Low,t = ck + d k SENTIMENTk ,t + ε k ,t
(4)
9
Page 9 of 23
RetX = High,t ( RetX t
t
= Low,t
) denotes an average of characteristic-adjusted return over month t+1 for
LTG5 (LTG1). Thus, the dependent variable in equation (4) is the spread in a subsequent one-month characteristic-adjusted return between the top and bottom LTG quintiles.
ip t
The reason why I control for size, value, and momentum effects, by utilizing the characteristic-adjusted return, is that these effects on the return spread between LTG5 and LTG1 are
time-varying (I did not use the Carhert momentum and Fama-French factors as independent
cr
variables in the regression model (4), to control for these effects). Table 4 shows normalized values of size (log value of market capitalization), value (book value for the most recent reported quarter to
us
market), and momentum (the return from t - 12 months to t - 2 months) for each LTG quintile. Figure 3 plots the spreads in these variables between LTG5 and LTG1. As shown in the figure and the table, the spreads in these exposures are time-varying; on average, the size spread is negative,
an
however, during the dot-com bubble, there is little difference in size between LTG5 and LTG1. In addition, the spread in momentum highly fluctuates; during the dot-com bubble, the momentum
M
exposure is higher for LTG5 than for LTG1, whereas the momentum exposure is lower for LTG5 than for LTG1 after the bubble bursts. Thus, I should control for size, value, and momentum effects at each time point (at end of each month); therefore, I utilize characteristic-adjusted returns for
d
controlling these effects.
characteristic-adjusted
te
Table 5 shows the regression results for equation (4) and descriptive statistics for returns
for
each
LTG
quintile.
There
is
little
difference
in
Ac ce p
characteristic-adjusted returns between LTG5 and LTG1. However, the return spread between them highly fluctuates; the return spread ranges from -17.7% to 22.3%, whereas the average return spread is only -0.3%. The regression results shown in Table 5(b) reveal that the spread in characteristic-adjusted returns between LTG5 and LTG1 is significantly negatively associated with the two sentiment indices. Thus, the results indicate that stocks with high-predicted growth experience low subsequent stock returns, especially following periods of bullish investor sentiment; these support Hypothesis 2-2, which posits that there is a stronger negative association between the LTG forecasts and subsequent stock returns when proxies for investor sentiment are higher. In summary, my analyses reveal that stocks with higher-predicted growth experience more negative earnings revisions and lower stock returns, especially following periods of bullish sentiment. The findings support the inference that bullish investor sentiment induces excessively bold growth forecasts and these excessive bold forecasts cause stock mispricing, which is subsequently corrected through the correction of the bold earnings forecasts.
10
Page 10 of 23
While Baker/Wurgler report that there is no significant sentiment effect on the return spread between extremely high- and low-growth stocks, my results support the sentiment effect on the return spread between such stocks. A key difference between my study and theirs is that I sort out high-growth firms from low-growth ones on the basis of investor growth predictions, while
ip t
Baker/Wurgler sort out firms on the basis of past one-year sales growth. Although past firm growth
could be positively associated with firm expected growth rates (Chan et al., 2003), past firm sales
cr
growth is a naïve and indirect predictor for a firm’s expected earnings growth. Thus, the sentiment
[Figure 3]
an
[Table 4]
us
effect cannot be observed in the return spread between stocks with high and low past sales growth.
[Table 5]
5. Conclusions
M
Consistent with the considerable debate on the role and effect of investor sentiment on asset pricing, prior studies show that market-wide investor sentiment has an influence on stock prices by affecting investor preference for securities whose valuations are highly subjective and difficult to
d
arbitrage (Baker and Wurgler 2006, 2007; Hribar and McInnis 2012) and by affecting short-term
te
earnings expectations (Walther and Willis 2012). However, although LTG expectations are crucial to stock price valuations, no studies examine the possibility that investor sentiment induces stock
Ac ce p
mispricing by affecting the overall accuracy of investor LTG expectations. In this study, I contribute to the behavioral research by examining this possibility. I predict that
higher investor sentiment induces investors to aggressively distinguish high-growth firms from others, and these excessively bold LTG predictions induce stock mispricing, which is subsequently corrected through the correction of these bold predictions. To test this possibility, I examine whether bullish investor sentiment induces extreme dispersion
in consensus LTG forecasts. The analyses reveal that dispersion in LTG forecasts is larger following higher market-wide investor sentiment, although a realized growth spread between firms with high LTG forecasts and firms with low ones tends to be narrower following such sentiment. These findings support the inference that bullish investor sentiment induces investors to excessively separate firms with high-predicted growth from others. I then examine whether the excessively bold LTG forecasts induce stock mispricing. The analyses reveal that stocks with higher-predicted growth experience more negative forecast revisions
11
Page 11 of 23
and lower subsequent stock returns, especially following periods of bullish investor sentiment. These findings support the view that excessive dispersion in LTG forecasts is corrected in a subsequent period and that the correction affects cross-sectional stock returns by reducing the mispricing.
investors to make excessively bold LTG forecasts, induces stock mispricing.
ip t
Therefore, my findings support the inference that market-wide investor sentiment, by inducing
cr
The studies most closely related to mine are those of Baker and Wurgler (2006, 2007) and
Hribar and McInnis (2012); they focus on the impact of investor sentiment on difficult to value (hard
us
to arbitrage) stocks. A key difference between my study and theirs is that I focus on impact of market-wide sentiment on the return spread between stocks with high and low growth, while extreme high and low-growth stocks are regarded as difficult to value in their study. In addition, the
an
effect found by Baker/Wurgler is confined to small, newer, and volatile stocks. On the other hand, through investor expectations of firm long-term cash flow, I see the sentiment effect on the prices of
M
a wide-spread of stocks.
My evidence has potential implications regarding the debate between laissez-faire policy and government activism in the market. Since my analysis shows the wide-spread sentiment effect, the
d
evidence emphasizes the importance of controlling excess market optimism or pessimism. Daniel et
te
al. (2002) states that the market anomaly, which can be explained by investors’ sentiment, does not always support government activism because governments might have no ability in determining
Ac ce p
when asset prices are significantly distorted by investor sentiment. However, my results reveal that the sentiment effect on asset prices through long-term growth expectations can be predicted by the sentiment indices, which are objective and observable. Thus, by utilizing the sentiment indices, rules can be established to control excess optimism or pessimism, and the rules could reduce the irrational sentiment effect on asset pricing.
References
Antoniou, C., Doukas, J., Subrahmanyam, A., 2012. Cognitive dissonance, sentiment, and momentum. J. Financ. Quant. Anal, Forthcoming. Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns. J. Financ., 61, 1645-1680. Baker, M., Wurgler, J., 2007. Investor sentiment in the stock market. J. Econ., 21, 129-151. Copeland, T., Dolgoff, A., Moel, A., 2004. The role of expectations in explaining the cross-section of
12
Page 12 of 23
stock returns. Rev. Account. Stud., 9, 149-188. Chan, L., Karceski, J., Lakonishok, J., 2003. The level and persistence of growth rates. J. Financ., 58, 643-684.
characteristic-based benchmarks. J. Financ., 52, 1035-1058.
ip t
Daniel, K., Grinblatt, M., Titman, S., Wermers. R., 1997. Measuring mutual fund performance with
Daniel, K., Hirshleifer, D., Teoh, S., 2002. Investor psychology in capital markets: evidence and
cr
policy implications. J. Monetary. Econ., 49,139-209.
us
De Long, B., Shleifer, A.,. Summers, L. H., Waldmann, R. J. 1990. Noise trader risk in financial markets. J. Polit. Econ., 98, 703-738.
Dechow, M., Sloan, G., 1997. Returns to contrarian investment strategies: tests of naive expectations
an
hypotheses. J. Financ. Econ., 43, 3-27.
Dechow, M., Hutton, P., Sloan, G., 2000. The relation between analysts’ forecasts of long-term
M
earnings growth and stock price performance following equity offerings. Contemp. Account. Res., 17, 1-32.
Frankel, R., Lee, C., 1998. Accounting valuation, market expectation, and cross-sectional stock
d
returns. J. Account. Econ., 25, 283-319.
Res., 39, 135-176.
te
Gebhardt, W., Lee, C., Swaminathan, B., 2001. Toward an implied cost of capital. J. Accounting
Ac ce p
Gordon, M., 1962. The investment, financing, and valuation of the corporation. Irwin, Homewood. Hribar, P., McInnis, J., 2012. Investor sentiment and analysts' earnings forecast errors. Manage. Sci., 58, 293-307.
La Porta, R., 1996. Expectations and the cross-section of stock returns. J. Financ., 51, 1715-1742. Lakonishok, J., Shleifer, A., Vishny, R., 1994. Contrarian investment, extrapolation, and risk. J. Financ., , 49, 1541-1578.
Lemmon, M., Portniaguina, E., 2006. Consumer confidence and asset prices: some empirical evidence. Rev. Financ. Stud., 19, 1499-1539. Mezrich, J., Zeng, Q., Nordquist, D., Seshadri, L., 2001. Rebroadcast: Changing of the guard. Equity Research Quantitative Strategy, Morgan Stanley, New York. Saade, S., 2015. Investor sentiment and the underperformance of technology firms initial public offerings. Res. Int. Bus. Financ., 34, 205-232.
13
Page 13 of 23
Shleifer, A., Vishny, R., 1997. The limits of arbitrage. J. Financ., 52, 35-55. Walther, B., Willis, R. 2012. Do investor expectations affect sell-side analysts' forecast bias and forecast accuracy? Rev. Account. Stud., 17, 1-21.
Ac ce p
te
d
M
an
us
cr
ip t
Whaley, E., 2000. The Investor Fear Gauge. J. Portfolio Manage., 26, 12-17.
14
Page 14 of 23
Table 1 LTG forecast dispersion Each month, from 1986 to 2006, I collect consensus (mean) forecasts of long-term earnings growth (LTG) from IBES.
ip t
Each month, firms are sorted into quintiles on consensus LTG forecast from LTG5 (High) to LTG1 (low). I calculate an average of consensus LTG forecast for each LTG quintile portfolio. I take the difference in the extreme quintiles (1
cr
and 5) to construct forecast dispersion portfolios (denoted as “LTG5-LTG1”). Panel (a) shows descriptive statistics for each portfolio’s LTG forecast. In panel (b), I show ordinary least squares regression results for Eq. (1) in the text.
us
The column “B&W Index” shows a time-series average of coefficient of the index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series average of coefficient of Michigan Consumer Sentiment Index. The figures in parentheses are t-statistics based on Newey-West standard errors.
Regression results
M
LTG3 13.9% 16.2% 12.0% 16.1% 12.1% 1.5%
B&W Index 0.0675 (7.35)
LTG dispersion
LTG2 10.8% 12.5% 9.3% 12.5% 9.4% 1.0%
LTG1 6.4% 8.2% 4.7% 7.8% 5.0% 0.9%
LTG5-LTG1
24.0% 38.9% 18.4% 29.2% 19.7% 5.0%
Michigan Index 0.0037 (4.70)
Ac ce p
b)
LTG4 17.9% 22.9% 15.2% 22.1% 15.6% 2.6%
d
Mean Max Min 90th 10th Std.dev.
LTG5 30.4% 46.9% 24.0% 36.9% 25.6% 5.7%
an
Descriptive statistics
te
a)
15
Page 15 of 23
Table 2 Realized growth spread Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (High) to LTG1 (low). I calculate an average of realized growth rate for each LTG quintile. The row of “3-year EPS growth,” “5-year EPS growth,”
ip t
“3-year OIBD growth,” “5-year OIBD growth,” “3-year sales growth,” and “5-year sales growth” shows the regression results for when 3-year EPS growth, 5-year EPS growth, 3-year OIBD growth, 5-year OIBD growth,
cr
3-year sales growth, and 5-year sales growth are used for evaluating realized growth dispersion. Panel (a) shows descriptive statistics for each portfolio’s realized growth and the difference in realized growth between LTG5 and
us
LTG1. In panel (b), I show ordinary least squares regression results for Eq. (2) in the text. The column “B&W Index” shows a time-series average of the coefficient of the index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series average of the coefficient of the Michigan Consumer Sentiment Index. The
an
figures in parentheses are t-statistics based on Newey-West standard errors.
Mean 11.5% 8.8% 11.7% 8.5% 10.9% 8.7%
Max 25.5% 19.3% 22.6% 15.6% 19.0% 14.6%
(b) Regression results
d
Ac ce p
3-year EPS growth 5-year EPS growth 3-year OIBD growth 5-year OIBD growth 3-year sales growth 5-year sales growth
Min -4.7% -1.8% 2.5% 1.6% 0.6% 1.7%
M
3-year EPS growth 5-year EPS growth 3-year OIBD growth 5-year OIBD growth 3-year sales growth 5-year sales growth
te
(a) Descriptive statistics
B & W Index -0.0371 (2.07) -0.0253 (2.02) -0.0177 (0.84) -0.0161 (1.68) -0.0214 (0.72) -0.0237 (2.12)
90th 18.2% 12.1% 17.5% 12.0% 15.5% 12.5%
10th 3.7% 5.7% 6.7% 5.0% 6.1% 4.5%
Std.dev. 5.6% 3.1% 4.1% 2.8% 4.0% 3.0%
Michigan Index -0.0015 (0.97) -0.0011 (2.76) -0.0008 (0.97) -0.0014 (5.04) -0.0014 (1.77) -0.0021 (6.91)
16
Page 16 of 23
Table 3 Earnings forecast revisions Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (high) to LTG1 (low). I calculate an average of earnings forecast revisions in the following month for each LTG quintile. The rows of the “EPS
ip t
forecast revision” and the “LTG forecast revision” show the result for the difference in forecast revisions with regard to EPS for the next fiscal year and LTG, respectively. Panel (a) shows descriptive statistics for the portfolio’s forecast
cr
revisions and the difference in earnings forecast revisions between LTG5 and LTG1. Panel (b) shows ordinary least
squares regression results for Eq. (3). The column “B&W Index” shows a time-series average of the coefficient of the
us
index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series average of the coefficient of Michigan Consumer Sentiment Index. The figures in parentheses are t-statistics based on White's
an
heteroscedasticity consistent standard errors.
(a) Descriptive statistics
LTG forecast revision Mean LTG5-LTG1 -0.46% LTG5 -0.35% LTG1 0.10%
Max 0.36% 0.43% 0.49%
Min -0.71% -0.96% -0.83%
M
Max 0.61% 0.01% 0.15%
Ac ce p
te
d
EPS forecast revison Mean LTG5-LTG1 -0.08% LTG5 -0.18% LTG1 -0.10%
Min -1.85% -1.65% -0.21%
90th 0.06% -0.06% 0.01%
10th -0.23% -0.31% -0.26%
Std.dev. 0.14% 0.14% 0.12%
90th -0.17% -0.13% 0.22%
10th -0.82% -0.63% 0.00%
Std.dev. 0.29% 0.26% 0.10%
(b) Regression results
EPS forecast revison LTG forecast revision
B&W Index -0.0008 (4.72) -0.0019 (3.81)
Michigan Index -0.00003 (2.60) -0.00004 (1.80)
17
Page 17 of 23
Table 4 Size, book-to-market, and momentum exposures Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (high) to LTG1 (low). I calculate an average for the normalized book-to-market ratio, log value of market capitalization (size), and the return from t -
ip t
12 months to t - 2 months (momentum) for each LTG quintile portfolio. I take the difference in the extreme quintiles
(denoted as “LTG5-LTG1”). The tables show descriptive statistics for the portfolio’s book-to-market ratio, size, and
cr
momentum, and the difference in those variables between LTG5 and LTG1.
LTG5 -0.37 0.10 -0.58 -0.26 -0.47 0.11
LTG4 -0.13 0.03 -0.30 -0.04 -0.22 0.07
te
Ac ce p
Mean Max Min 90th 10th Std.dev.
LTG2 0.26 0.56 -0.01 0.44 0.09 0.12
LTG1 0.50 0.86 0.11 0.68 0.29 0.15
LTG5-LTG1
LTG3 0.16 0.38 -0.11 0.29 0.01 0.11
LTG2 0.25 0.48 0.05 0.36 0.14 0.09
LTG1 0.23 0.40 -0.04 0.35 0.05 0.11
LTG5-LTG1
LTG3 -0.03 0.26 -0.23 0.10 -0.15 0.10
LTG2 -0.02 0.41 -0.41 0.20 -0.24 0.17
LTG1 0.00 0.46 -0.39 0.26 -0.25 0.19
LTG5-LTG1
d
b) Size
LTG3 0.03 0.27 -0.27 0.15 -0.14 0.11
an
LTG4 -0.17 0.08 -0.42 0.00 -0.33 0.13
M
Mean Max Min 90th 10th Std.dev.
LTG5 -0.47 0.01 -0.75 -0.21 -0.66 0.17
us
a) Book-to-market ratio
-0.96 -0.15 -1.48 -0.56 -1.29 0.30
-0.60 0.11 -0.91 -0.33 -0.82 0.19
c) Momentum
Mean Max Min 90th 10th Std.dev.
LTG5 0.05 0.90 -1.09 0.51 -0.42 0.37
LTG4 -0.03 0.36 -0.33 0.10 -0.15 0.11
0.05 1.25 -1.46 0.73 -0.67 0.54
18
Page 18 of 23
Table 5 Return spreads between high-growth and low-growth stocks Each month, firms are sorted into quintiles on consensus LTG forecasts from LTG5 (high) to LTG1 (low). I calculate an average of characteristic-adjusted returns in the following month for each LTG quintile portfolio. I take the
ip t
difference in the extreme quintiles (denoted as “LTG5-LTG1”). Panel (a) shows descriptive statistics for the
portfolio’s returns and the difference in characteristic-adjusted returns between LTG5 and LTG1. Panel (b) shows
cr
ordinary least squares regression results for Eq. (4). The column “B&W Index” shows a time-series average of the coefficient of the index constructed by Baker and Wurgler (2007); the column “Michigan Index” shows a time-series
based on White's heteroscedasticity consistent standard errors.
Stock return
M
LTG3 -0.1% 3.9% -6.5% 1.1% -1.3% 1.1%
d
B&W Index -0.0144 (1.94)
Ac ce p
(b) Regression results
LTG4 -0.1% 13.0% -6.0% 1.9% -2.3% 2.0%
te
Mean Max Min 90th 10th Std.dev.
LTG5 -0.3% 20.7% -15.4% 4.0% -5.0% 4.1%
an
(a) Descriptive statistics
us
average of the coefficient of the Michigan Consumer Sentiment Index. The figures in parentheses are t-statistics
LTG2 -0.1% 3.3% -4.5% 0.9% -1.2% 0.9%
LTG1 -0.1% 4.4% -3.1% 1.1% -1.4% 1.1%
LTG5-LTG1
-0.3% 22.3% -17.7% 4.8% -5.9% 4.6%
Michigan Index -0.00059 (1.91)
19
Page 19 of 23
Figure 1
ip t
Sentiment Indices
30.00
3
Michigan Index (left axis)
2.5
20.00
cr
Baker & Wurgler Index (right axis)
us
10.00
an
0.00
-10.00
M
-20.00
1.5 1 0.5 0 -0.5 -1 -1.5
te
d
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
-30.00
2
Monthly values of the investor sentiment index constructed by Baker and Wurgler (Baker & Wurgler Index) and the
Ac ce p
Michigan Consumer Sentiment Index (Michigan Index). To ensure that both indices are free of macroeconomic influences, following Baker and Wurgler (2006, 2007), I conduct the investigation using an orthogonal version of the indices, which is obtained by regressing the indices on a set of macroeconomic variables.
20
Page 20 of 23
Figure 2 LTG dispersion and sentiment indices 3 65% 2.5
Dispersion in long-term growth forecasts (left axis) Baker & Wurgler Index (right axis)
45%
2
ip t
55%
1.5
35%
cr
1
25%
us
15% 5%
an
-5%
0 -0.5 -1 -1.5
M
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
-15%
0.5
A. Baker & Wurgler Index and LTG dispersion
35%
Ac ce p
30%
te
40%
d
45%
30
20
10
25%
0
20% 15% 10% 5%
-10
Dispersion in long-term growth forecasts (left axis) Michigan Index (right axis)
-20
-30
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
0%
B. Michigan Consumer Sentiment Index and LTG dispersion Panel A (B) shows monthly values of the investor sentiment index constructed by Baker and Wurgler (the Michigan Consumer Sentiment Index) and the dispersion between the top and bottom quintile of IBES long-term forecasts, denoted by “Dispersion in long-term growth forecasts.”
21
Page 21 of 23
Figure 3 Size, book-to-market, and momentum exposures
ip t
2.00
1.50
B/M
MOMENTUM
SIZE
cr
1.00
us
0.50
0.00
an
-0.50
-1.00
M
-1.50
d
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
-2.00
The graph shows monthly values of differences in normalized Book-to-Market ratio (B/M), log value of market
Ac ce p
LTG1.
te
capitalization (SIZE), and the return from t - 12 months to t - 2 months (MOMENTUM), between LTG5 and
22
Page 22 of 23
d
te
Ac ce p
25%
20% 0
ip t
15%
cr
10%
5%
Dispersion in long-term growth forecasts (left axis) Michigan Index (right axis)
0%
us
an
M
1/1986 9/1986 5/1987 1/1988 9/1988 5/1989 1/1990 9/1990 5/1991 1/1992 9/1992 5/1993 1/1994 9/1994 5/1995 1/1996 9/1996 5/1997 1/1998 9/1998 5/1999 1/2000 9/2000 5/2001 1/2002 9/2002 5/2003 1/2004 9/2004 5/2005 1/2006 9/2006
45% 30
40%
35% 20
30% 10
-10
-20
-30
23
Page 23 of 23