Вы находитесь на странице: 1из 47

A conditional characteristics model of stock returns

Malcolm Baker Harvard Business School and NBER mbaker@hbs.edu Jeffrey Wurgler NYU Stern School of Business jwurgler@stern.nyu.edu August 25, 2003

Abstract Traditional asset pricing models maintain that differences in expected returns across securities reflect differences in systematic risk. We develop a model in which crosssectional differences in expected returns reflect the pattern of mispricing created by time variation in investor sentiment and cross-sectional variation in the difficulty of valuation and arbitrage. As predicted, we find that when investor sentiment is high, stocks whose salient characteristics can sustain optimistic valuations and at the same time are hard to arbitrage stocks characterized by young age, small size, high return variance, no profits, no dividends, extreme growth, or distress have relatively low expected returns. When sentiment is low, these patterns attenuate or fully reverse. Further results also suggest that the cross-section of expected returns may be better described by a sentiment-based conditional characteristics model than an equilibrium systematic risk model.

We thank Martin Lettau, Anthony Lynch, and Jeremy Stein for helpful comments, as well as participants of seminars at Harvard Business School, Norwegian School of Economics and Business, Norwegian School of Management, University of Amsterdam, and University of British Columbia. Baker gratefully acknowledges financial support from the Division of Research of the Harvard Business School.

I.

Introduction Traditional asset pricing models attribute cross-sectional differences in expected returns

to differences in systematic risk. As in Markowitz (1959), investors are assumed to care about the mean and variance of their portfolio, and think about individual securities in statistical terms, as how their respective expected returns, variances, and covariances affect the portfolio return. Sharpe (1964), Merton (1973), Ross (1976), and Breeden (1979) show that in an equilibrium of diversified or Markowitz-type investors, individual securities are priced so that their expected returns depend only on their systematic risks, i.e. their covariances with relevant market aggregates or state variables. The systematic risk view is elegant, but its empirical relevance remains uncertain. Market beta does not explain cross-sectional differences in expected returns, while firm characteristics, such as size and book-to-market, have robust effects (Banz (1981), Basu (1983), and Fama and French (1992)). Fama and French (1993) suggest that these characteristics proxy for covariances with systematic risk factors, but Daniel and Titman (1997) find that they have cross-sectional explanatory power that is separate from the covariance structure of returns. Recent research has focused on conditional variation in systematic risks, but again no consensus is near. For instance, Lettau and Ludvigson (2001) find that conditional covariances with consumption can address some cross-sectional regularities, but Lewellen and Nagel (2003) argue that conditional models require implausibly large variation in risks and risk premia. Perhaps most unsettling, even as research on systematic risk models proceeds, episodes like the Internet bubble seem to challenge their basic assumptions. In this paper, we develop and test different model of the cross-section of expected returns. Its key ingredients are that typical investors view individual securities, and the firms on

which they are claims, as bundles of salient investment characteristics, not as variances and covariances; that investor sentiment varies over time; and that securities vary cross-sectionally in the difficulty of valuation and arbitrage. These ingredients lead to a conditional characteristics model in which expected returns reflect predictable corrections of mispricing, not compensation for systematic risk. In particular, the model predicts that when sentiment is optimistic, stocks whose salient characteristics can sustain optimistic valuations, especially those that are difficult to arbitrage, are relatively overpriced and hence have low expected returns. Conditional on low sentiment, the same stocks are predicted to be relatively underpriced, and have high expected returns. We motivate the approach and these specific predictions with a discussion of historical bubbles and crashes, as well as theoretical and psychological considerations. To test these predictions, we gather a set of investor sentiment proxies to use as time series conditioning variables. We use the closed-end fund discount, NYSE turnover, the volume and first-day returns on IPOs, the equity share, the dividend premium, and a composite index based on the common variation in these variables. To reduce the likelihood that these proxies are connected to systematic risks, we orthogonalize them to a range of macroeconomic variables. The sentiment proxies are highly correlated and visibly line up with historical bubbles, suggesting that they are valid sentiment measures. We then examine how sentiment affects the expected returns associated with various characteristics. We use monthly stock returns between 1963 and 2001. Our first method is to sort firm-month observations according to the level of sentiment, first, and then the decile rank of a given firm characteristic, second. We find that when sentiment is low (pessimistic), small stocks earn particularly high subsequent returns but when sentiment is high (optimistic), there is no size effect at all. Even more striking, conditional on low sentiment, subsequent returns are much

higher on young stocks than older stocks, unprofitable stocks than profitable ones, and nonpayers than dividend payers. When sentiment is high, however, these patterns completely reverse. Thus, variables that do not have any unconditional explanatory power actually reveal a striking signflip pattern, in the predicted direction, conditional on sentiment. The sorts also reveal U-shaped conditional patterns involving salient characteristics of growth opportunities and distress. When sentiment is low, expected returns are high on both highest- and lowest-sales growth deciles relative to firms in between. These extreme deciles represent ultra-growth stocks and distressed stocks with contracting sales, respectively. When sentiment is high, this U-shaped pattern flips upside down, so both ultra-growth and distressed stocks now expect lower returns relative to those in the middle. There are U-shaped conditional patterns in external finance and book-to-market as well. These patterns are consistent with the prediction that hard-to-value firms are particularly prone to underpricing (overpricing) when investors are pessimistic (optimistic). Again, note that these U-shaped conditional patterns, like the monotonic conditional patterns involving age, profitability, and dividend payment, are averaged away in unconditional studies. We verify these patterns with two other methods. We run monthly cross-sectional regressions to predict individual stock returns with various characteristics, and then examine how sentiment variables affect the time series of the characteristics coefficients. We also consider regressions that use sentiment proxies to predict returns on long-short portfolios that isolate a characteristic. The results correspond closely to those from the sorts. For several reasons, including that the sentiment conditioning variables are orthogonal to macroeconomic conditions, the results are more consistent with a conditional characteristics model than a systematic risk model. The results of two more exercises further support this

conclusion. We directly test, and reject, the basic systematic risk alternative explanation in which the conditionally varying expected returns of young, small, unprofitable, nonpaying, etc. firms are tied to conditionally varying covariances with the stock market or consumption growth. Also, we find evidence of conditional patterns in returns around earnings announcements that roughly match those in the sorts, indicating that at least a portion of those results are due to the correction of too-optimistic or too-pessimistic expectations. The results have broad implications, which we highlight at the end of the paper. Here we point out that the conditional characteristics model builds upon and ties together several themes in the recent literature. Lettau and Ludvigson (2001) study conditional systematic risks, much as we condition on sentiment. Daniel and Titman (1997) introduce a firm characteristics model for expected returns. We extend this approach and give it a specific, conditional motivation. Shleifer (2000) reviews early research on sentiment and limited arbitrage, two key ingredients in the model. Barberis and Shleifer (2003) and Barberis, Shleifer, and Wurgler (2003) develop and test models of category-level trading. Conditional shifts in demand for groups of securities with the same salient characteristics are at the core of the model. The paper proceeds as follows. Section II motivates the conditional characteristics approach. Section III describes the data. Section IV presents empirical results. Section V discusses implications.

II. A.

A conditional characteristics model: Specification and motivation Specification By a conditional characteristics model of expected returns, we mean an empirical model

of the general form:

E[Rit ] = a + b 1 ' x it 1 + b 2 'Tt1x it 1 where i indexes firms or securities, t is time, x is a vector of firm or security characteristics, and T is a time series conditioning variable. Formally, this model adds conditional effects to the unconditional characteristics model of Daniel and Titman (1997). In the remainder of this section, we argue that Eq. (1) is appropriate for capturing the correction of mispricings that result from broad fluctuations in investor sentiment, i.e. optimism or pessimism, in the presence of cross-sectional differences in the difficulty of valuation and arbitrage. We motivate more specific aspects of Eq. (1) with historical observations, economic considerations, and psychological considerations. We summarize the discussion with empirical predictions regarding appropriate characteristics and conditioning variables. B. Historical motivation Descriptions of historical market bubbles and crashes offer a number of useful insights. Kindleberger (2001) draws general lessons from bubbles that have occurred over the past few hundred years, while Brown (1991), Dreman (1979), Malkiel (1990, 1999), Shiller (2000), and Siegel (1998) focus more specifically on recent U.S. stock market episodes. These accounts help us to identify the timing of major shifts in investor sentiment and the characteristics of firms most affected. We summarize these accounts here because historical perspective on market bubbles is hard to find in the academic literature. We take any one account with a grain of salt, and emphasize the themes that appear repeatedly. Our own data starts in 1961, so we summarize these accounts from then onward. Malkiel (1990) and Brown (1991) note a strong demand for small, young, growth stocks in 1961. Malkiel writes of a new-issue mania that was concentrated on new tronics firms. The tronics

(1)

boom came back to earth in 1962 Growth stocks took the brunt of the decline, falling much further than the general market (p. 54 - 57). Dreman (1979, p. 70) confirms this account. The next widely noted episode of speculation occurred in 1967 and 1968. Brown writes that scores of franchisers, computer firms, and mobile home manufactures seemed to promise overnight wealth. [while] quality was pretty much forgotten (p. 90). Malkiel and Dreman also note this pattern a focus on firms with strong earnings growth or potential, while an avoidance of the major industrial giants, buggywhip companies, as they were sometimes contemptuously called (Dreman 1979, p. 74-75). Another characteristic viewed as out of favor was dividends. According to the New York Times, during the speculative market of the late 1960s many brokers told customers that it didnt matter whether a company paid a dividend just so long as its stock kept going up (9/13/1976). However, after 1968, as it became clear that capital losses were possible, investors came to value dividends (10/7/1999). Anecdotal accounts describe the early 1970s as a bear market, with general investor sentiment at a low level. The only stocks with notably high valuations were a set of old, large, stable, consistently profitable stocks known as the nifty fifty. Brown, Malkiel, and Siegel (1998) each point out this pattern. Siegel writes, All of these stocks had proven growth records, continual increases in dividends and high market capitalization (p. 106). This is a striking mirror image of the speculative episodes described above (and below). They focus on small, young, unprofitable growth stocks in periods of high sentiment, whereas the nifty fifty episode is characterized as a bubble in firms with the virtually opposite set of characteristics old age, large size, and continuous earnings and dividends in a period of low sentiment. The late 1970s through mid-1980s are described as a period of generally high sentiment that saw a series of speculative episodes. Dreman argues for a bubble in gambling issues in 1977

and 78. Ritter (1984) studies the hot issue market of 1980, finding greater initial returns on IPOs of natural resource start-ups than on larger, mature, profitable offerings. Of 1983, Malkiel (p. 74-75) writes that the high-technology new-issue boom of the first half of 1983 was an almost perfect replica of the 1960s episodes The bubble appears to have burst early in the second half of 1983 the carnage in the small company and new-issue markets was truly catastrophic. Brown confirms this account. Of the mid-1980s, Malkiel suggests that What electronics was to the 1960s, biotechnology became to the 1980s. new issues of biotech companies were eagerly gobbled up. having positive sales and earnings was actually considered a drawback (p. 77-79). By 1987 and 1988, however, market sentiment had changed from an acceptance of an exciting story to a desire to stay closer to earth with low-multiple stocks that actually pay dividends (p. 79). The late 1980s and early 1990s are not viewed as remarkable by these sources. The mid- through late-1990s bubble in Internet and technology stocks will be familiar. By all accounts, investor sentiment was generally high before the bubble popped in 2000. This episode is analyzed in detail by Ofek and Richardson (2002). Malkiel draws parallels to episodes in the 1960s, 1970s, and 1980s, and Shiller (2000) compares the Internet to the late 1920s. Like earlier speculative episodes that occurred in high sentiment periods, demand for dividend payers appears to have been low (New York Times (1/6/1998)). In terms of the demand for other characteristics, Ljungqvist and Wilhelm (2003) find that 80% of 1999 and 2000 IPOs had negative earnings per share and the median age of 1999 IPOs was 4 years. This compares to 9-10 years from a few years earlier or the 12 to 14 years by 2001-02 reported by Ritter (2003). This brief summary suggests that a conditional characteristics model might be useful for capturing the reversion of mispricings caused by speculative episodes. In particular, conditions

of generally high sentiment appear to have been associated with a tendency for firms with characteristics such as small size, young age, no dividends, low or negative current profits, and apparently strong growth potential to be overpriced relative to firms with the opposite characteristics. Conditions of low sentiment, on the other hand, are associated with the opposite pattern in relative prices. C. Economic motivation Economic considerations tend to reinforce the historical impression that mispricings are more frequent and severe in young, small, unprofitable, nonpaying, and/or high perceivedgrowth firms. The first is that the characteristics of no earnings history and apparently unlimited growth opportunities combine to allow unsophisticated investors to plausibly defend a wide range of valuations, from (objectively) much too low to much too high. In bubble periods, the same characteristics allow investment bankers or, worse, swindlers, to plausibly defend the higher range in selling securities. Thus, stocks with these characteristics are naturally more sensitive to sentiment swings. By contrast, the valuation of a mature firm, with an earnings history and high dividends, is less arbitrary. The second consideration is that it is more difficult for would-be arbitrageurs to enforce efficient pricing in small growth firms. Their higher idiosyncratic risk makes arbitrage trades fundamentally more risky (Shleifer and Vishny (1997), Wurgler and Zhuravskaya (2002)). The fact that noise traders can convince themselves of a wider range of valuations itself generates an unattractive noise-trader risk (De Long, Shleifer, Summers, and Waldmann (1990)).1 Finally, the relatively low liquidity and higher costs of trading and shorting small growth stocks also tend to limit arbitrage (Amihud and Mendelsohn (1986), DAvolio (2002)).
1

We note but do not incorporate the equilibrium prediction of DeLong, Shleifer, Summers, and Waldmann (1990), which holds that securities with more exposure to sentiment have higher unconditional expected returns.

D.

Psychological motivation Financial economists have tended to view firm characteristics as merely proxies for

deeper fundamentals, but some psychological principles suggest that investor demands may be based directly on firm characteristics. To see how, some comments on the cognitive process of categorization may be useful.2 Psychologists suggest that an objects category is determined by a set of essential features, or characteristics, that are shared by the members of the category. As a result, categorization is useful for making inferences. For example, for the category birds some of the defining characteristics include can fly, has wings, and has feathers. Knowing that an object has two of these features allows it to be categorized as a bird; and then one can reasonably infer the third feature. In this spirit, and the spirit of Lancaster (1966, 1971), we propose that typical investors view individual securities as bundles of salient investment characteristics for example no earnings, young age, or high dividends. We hypothesize that typical investors determine the merit of investing in a security by processing its salient characteristics. Investors may evaluate the salient characteristics directly. More interestingly, they may use characteristics to categorize a security into value stocks or Internet stocks, implicitly infer the remaining characteristics, and then determine investment merit at the category level. This view of the investment process clearly differs from Markowitz (1959), where investors view individual securities as lists of statistics and do not categorize as a shortcut. Popular investment advice often involves a discussion of salient characteristics. If investors view securities as bundles of characteristics, then their demands, and hence any mispricing and subsequent returns, are necessarily closely tied to characteristics. As
2

The implications of categorization for finance have recently been explored by Baker and Wurgler (2003), Barberis and Shleifer (2003), Barberis, Shleifer, and Wurgler (2003), Greenwood and Sosner (2003), and Peng and Xiong (2002).

suggested above, this will be true even if there is an intervening process of categorization and even if all trading is done at the category level, so long as psychologists are correct that categories are defined by a common bundle of characteristics. It is also empirically convenient to be able to boil categories down to their defining characteristics. This allows us to study such episodes as the 1968 growth stocks bubble and the late-1990s Internet bubble in terms of a roughly similar set of characteristics. The meaning of specific category labels, on the other hand, are often ephemeral casual observation suggests that the connotation of Internet stocks changed drastically from 1999 to 2001. E. Predictions This discussion helps to flesh out a predictive model of the form of Eq. (1). Both historical and economic considerations imply that in conditions of general investor optimism (call it high sentiment, a bubble, or a bull market), firms whose salient characteristics suggest an opportunity for extreme capital gains are more likely to be overpriced relative to firms with the opposite characteristics. The reverse applies when investors are more pessimistic (low sentiment, a crash, a bear market). In addition, psychological considerations suggest that investor demand may be based directly on the characteristics themselves, as opposed to deeper sources of perceived investment value that are hidden to the econometrician. After describing the data, we test the conditional characteristics model using Eq. (1) as an organizing framework. It can be examined through two-way sorts (averages of returns for given x and T) or regressions. Either way, the strategy is to use proxies for general investor sentiment as conditioning variables, and then to see whether the way that characteristics spread future returns depends on the conditioning variable.

10

III. A.

Data Firm characteristics and returns The firm-level data is from the merged CRSP-Compustat database. The sample includes

all common stock (share codes 10 and 11) of nonfinancial firms (excluding SIC code 6) between 1962 through 2001. We match accounting data for fiscal year-ends in calendar year t -1 to (monthly) returns from July t through June t +1. Table 1 shows summary statistics. Panel A summarizes returns variables. Following common practice, momentum MOM is defined as the cumulative raw return for the elevenmonth period between 12 and two months prior to the observation return. We will consider momentum as a control or robustness variable, not as a salient firm characteristic. The remaining panels summarize the firm and security characteristics that we consider. The discussion of the previous section points us directly to several variables. To that list, we add a few more characteristics that, by introspection, seem likely to be salient to investors. Overall, we roughly group characteristics as most directly pertaining to firm size and age, profitability, dividends, growth opportunities, and distress. Panel B summarizes size and age characteristics. Market equity ME from June of year t , measured as price times shares outstanding from CRSP, is matched to monthly returns from July of year t through June of year t +1. Age is the number of years since the firms first appearance on CRSP, measured to the nearest month. Sigma is the standard deviation of monthly returns over twelve months ending in June of year t . If there are at least nine returns, Sigma is matched to monthly returns from July of year t through June of year t +1. Panel C summarizes profitability characteristics. The return on equity E+/BE is positive for profitable firms and zero for unprofitable firms. Earnings (E) is income before extraordinary

11

items (Item 18) plus income statement deferred taxes (Item 50) minus preferred dividends (Item 19), if earnings are positive; book equity (BE) is shareholders equity (Item 60) plus balance sheet deferred taxes (Item 35). E>0 is a dummy variable for profitable firms. Panel D summarizes dividend payment characteristics. Dividends to equity D/BE is dividends per share at the ex date (Item 26) times Compustat shares outstanding (Item 25) divided by book equity. D>0 is a dummy for a firm with positive dividends per share by the ex date. The decline in the percentage of firms paying dividends is noted by Fama and French (2001) is apparent. As they point out, this is partly explained by the increasing proportion of unprofitable firms. Panel E summarizes characteristics that could serve as salient indictors of growth opportunities, distress, or both. The elements of book-to-market equity BE/ME are defined above. External finance activity EF/A is defined as the change in assets (Item 6) minus the change in retained earnings (Item 36) divided by assets. Sales growth (GS) is the change in net sales (Item 12) divided by prior-year net sales. We measure and report sales growth GS/10 as the decile of the firms sales growth in the prior year relative to NYSE firms decile breakpoints. It is important to bear in mind that the variables in Panel E, in particular, may capture multiple effects. For instance, book-to-market wears at least three separate hats. Extremely high values are likely to indicate distress; extremely low values are likely to indicate high growth opportunities; and, as a scaled-price variable, book-to-market also serves as a generic valuation indicator, varying with mispricing or rational expected returns. The sales growth and external finance variables wear at least two hats: extremely low values (which are negative) are likely to indicate distress; extremely high values are likely to indicate high growth opportunities. If

12

market timing motives are important to external finance, that variable could serve as a generic misvaluation indicator as well. In Panels C, D, and E, the accounting data for fiscal years ending in t -1 are matched to monthly returns from July of year t through June of year t +1. To reduce the influence of outliers and data errors, all characteristics variables, plus momentum, are Winsorized each year at their 0.5 and 99.5 percentiles. B. Investor sentiment Prior research suggests a number of plausible proxies for investor sentiment to use as time-series conditioning variables. We consider six proxies the average closed-end fund discount, NYSE share turnover, the number and average first-day returns on IPOs, the equity share in new issues, and the dividend premium as well as a composite sentiment index that takes the first principal component of these proxies. Each variable is measured annually from 1962 through 2000. To isolate the sentiment component of these proxies from any business cycle components, we orthogonalize each of them with respect to several macroeconomic variables. All of our results are based on these cleaned proxies. Specifically, we regress each of the raw variables (which are described next) on growth in the industrial production index (Federal Reserve Statistical Release G.17), growth in consumer durables, nondurables, and services (all from BEA National Income Accounts Table 2.10), and a dummy variable for NBER recessions. We take the residuals from these regressions as a cleaner proxy that is independent of major business cycle effects. Table 2 summarizes the raw sentiment proxies in Panel A and the cleaned versions in Panel B. Figure 1 plots both versions. The closed-end fund discount CEFD is the average

13

difference between the net asset value of closed-end fund shares and their market prices. Prior work suggests that the discount is inversely related to investor sentiment. Zweig (1973) uses the closed-end discount to forecast reversion in Dow Jones stocks, and Lee, Shleifer, and Thaler (1991) examine several attributes of the discount. We take the value-weighted average discount on closed-end stock funds for 1962 through 1993 from Neal and Wheatley (1998), for 1994 through 1998 from CDA/Wiesenberger, and for 1999 and 2000 from turn-of-the-year issues of the Wall Street Journal. NYSE share turnover is based on the ratio of reported share volume to average shares listed from the NYSE Fact Book . Baker and Stein (2002) motivate turnover (and more generally liquidity) as a sentiment index. In a market with short-sales constraints, irrational investors participate and add liquidity only when they are optimistic, and hence high liquidity coincides with overvaluation. Consistent with this interpretation, Jones (2001) finds that high turnover forecasts low market returns. Turnover displays an exponential positive trend over our period, however, and the May 1975 elimination of fixed commissions have a visible effect. As a partial solution, the raw turnover ratio TURN is detrended by the five-year moving average (specifically, the log of the ratio is detrended by its moving average). This helps to identify sharp changes in turnover, but we will clearly still pick up sharp changes caused by market structure. The IPO market has long been viewed as sensitive to investor sentiment, and high firstday returns on IPOs may reflect investor enthusiasm. The low idiosyncratic returns on IPOs are consistent with successful market timing (Stigler (1964), Ritter (1991)). The number of IPOs NIPO and the average first-day returns RIPO in a given year using data from Jay Ritters website, which updates the sample in Ibbotson, Sindelar, and Ritter (1994). Of course, there are

14

non-sentiment explanations for why IPO volume and underpricing vary over time, but they do not imply predictive relationships in the cross-section of stock returns. The share of equity issues in total equity and debt issues is another measure of financing activity that may capture an aspect of investor sentiment. Baker and Wurgler (2000) find that when the equity share is in its bottom (top) historical quartile, the next years equal-weighted market return averages 27% (-8%). The equity share is defined as gross equity issuance divided by gross equity plus gross long-term debt issuance using data from the Federal Reserve Bulletin. The equity share S is adjusted for the apparent regime change in debt issues in the early 1980s, which coincided with the emergence of junk debt markets and the drop in inflation.3 While both measures reflect equity issuance, the number of IPOs and the equity share have some important differences. The equity share includes SEOs, has strong predictive power for market returns, and scales by total external finance to isolate the composition of finance from the overall level. On the other hand, the IPO variables may better reflect demand for certain regions of the cross-section including young firms, small firms, and extreme-growth firms that Section II suggests may be particularly sensitive to sentiment. The dividend premium PD-ND is defined as the log ratio of the average market-to-book ratios of payers and nonpayers. Baker and Wurgler (2002, 2003) propose the dividend premium as a proxy for relative investor demand for dividend payers. Since Fama and French (2001) find that payers are larger, more profitable, and have weaker growth opportunities, the dividend premium may also pick up investor demand for this correlated bundle of characteristics. Intuitively, to detect the hypothesized patterns in predictability, we want to condition on the turning points in investor sentiment. But the sentiment proxies above have certain lead-lag
3

To adjust the series, we assume that the equity share moves proportionally with the unadjusted share from 1983 forward, thus removing the one-time change in the share from 1983 to 1984.

15

relationships, which raises the possibility that different variables may reflect sentiment sooner or later. For instance, Ibbotson and Jaffe (1975) and Lowry and Schwert (2002), and Benveniste, Wilhelm, Ljungqvist, and Yu (2003) find that IPO volume lags the first-day returns on IPOs. A natural interpretation is that sentiment is in part behind the high first-day returns, and this in turn attracts additional IPO volume with some lag. Put more generally, one suspects that sentiment measures involving firm supply responses (S or NIPO) are likely to lag sentiment measures that are based directly on stock prices or investor behavior (RIPO, PD-ND, TURN , and CEFD). We form a composite sentiment index SENTIMENT in order to capture the common factor in the sentiment proxies and to help us identify the most effective timing of the individual proxies. The process is as follows. We start by estimating the first principal component of the six sentiment proxies and their lags. This gives us a first-stage index with twelve loadings, one for each of the current and lagged proxies. We then compute the correlation between the first-stage index and the current and lagged values of each of the proxies. We then construct SENTIMENT as the first principal component of six terms the lead or lag of each proxy, whichever has the higher correlation with the first-stage index. This leads to the final, more parsimonious index: SENTIMENTt = 0.09CEFDt + 0.11TURN t1 + 0.32 NIPOt
D ND + 0.20 RIPOt 1 + 0.18S t 0.32 Pt 1

(2)

where each of these components have been cleaned of macroeconomic conditions and standardized. The correlation between the twelve-term first-stage index and the SENTIMENT index is 0.96, suggesting that little information is lost in dropping six terms. SENTIMENT is intuitively appealing for two reasons. First, each individual sentiment proxy enters with the expected sign. Second, five of the six proxies enter with the expected timing with the exception of CEFD, price and investor behavior variables lead firm supply

16

variables. We will make use of this pattern in the subsequent analysis, i.e. we will condition on the first predetermined value of CEFD, NIPO, S, and SENTIMENT, and the lag of the first predetermined value of TURN , RIPO, and PD-ND. Figure 1 shows that the sentiment proxies line up well with anecdotal accounts of investor sentiment. Most of the proxies point to low sentiment in the first few years of the sample (which follow a 1961 crash in growth stocks) the closed-end fund discount and the dividend premium are relatively high, and turnover and equity issuance-related variables are low. Each variable identifies a spike in sentiment in 1968 and 1969, matching anecdotal accounts. Sentiment then tails off until, by the mid-1970s, it is low by most measures, although for the turnover this pattern is confounded by deregulation. The late 1970s through mid-1980s sees generally rising sentiment, and according to the composite index, sentiment has not dropped below a medium level since 1980. In 1999, the peak of the Internet bubble, sentiment peaks again by most proxies. This correspondence with anecdotal accounts is encouraging. It confirms, to the extent possible, that the proxies capture the intended variation. Also noteworthy is that cleaning the sentiment proxies of macroeconomic conditions has little qualitative effect on their time-series properties. Moreover, Table 2 suggests that the cleaned proxies are equally if not more correlated with each other than are the raw proxies, indicating that sentiment can be distinguished from macroeconomic conditions. As mentioned above, we consider only these cleaned proxies in our subsequent empirical work, so that we can better rule out explanations that involve business cycles.

17

IV. A.

Empirical tests Sorts Table 3 gives a simple, nonparametric look at the conditional characteristics model. We

place each monthly returns observation into a bin according to the decile rank that a given characteristic takes at the beginning of that month, and then according to the level of a sentiment proxy from the beginning of that year. We then compute the average monthly return for that bin and look for patterns. We report sorts on CEFD in Table 3a and SENTIMENT in Table 3b. To be clear, in light of the timing discussion above, we condition returns from calendar year t on the December of year t -1 values of CEFD and SENTIMENT. Also, to keep the meaning of the deciles relatively constant over time, we define them based on NYSE firms only. The tradeoff is that there is not a smooth distribution of firms across bins in any given month. For brevity, we omit sorts on the five other sentiment proxies. They give similar results, which are available upon request, and they are broken out separately in all other tables, where they fit more compactly. But it seems worth showing results for CEFD alongside the overall index, however, because CEFD is the cleanest general sentiment indicator. It is not mechanically connected to any one characteristic or segment of the cross-section. In contrast, the volume of IPOs or the dividend premium could in theory be driven by a specific sentiment for firms with IPO characteristics or dividends, respectively. While this seems unlikely economic considerations and historical anecdote suggest that these variables are more exaggerated reflections of general sentiment than separate influences on their own we can more explicitly verify a general sentiment effect by comparing results for CEFD to those for other sentiment proxies or the composite index.

18

The first rows of Table 3 show the effect of size conditional on sentiment. They reveal that the cross-sectional size effect exists in low-sentiment conditions only, i.e. when CEFD is positive or when SENTIMENT is negative (a final reminder that here, and in all subsequent tables, the sentiment proxies are net of macroeconomic effects and standardized). Specifically, Table 3b shows that when SENTIMENT is negative, returns average 2.59 percent per month for the bottom ME decile and 0.84 for the top decile. This pattern puts some already known results into a broader perspective, namely that the size effect is largely a January effect (Keim (1983), Blume and Stambaugh (1983)), and that the January effect is in turn stronger after a period of low returns (Reinganum (1983)), which is precisely when sentiment is likely to be low. As an aside, the average returns across the first two rows of Table 3 show that expected returns tend to be higher across the board when sentiment is low. This is consistent with prior results that the equity share and turnover, for example, forecast market returns. More generally, it supports our premise that investor sentiment broadly affects stocks, and so the existence of richer conditional patterns in the cross-section should not be entirely surprising. The conditional cross-sectional effect of Age is even more striking. Investors demand young stocks when SENTIMENT is positive and old stocks when sentiment is negative. This is clear from the conditional difference in subsequent returns. When SENTIMENT is pessimistic, top-decile Age firms return 0.69 percent less than bottom-decile Age firms, but they subsequently return 0.73 percent more when SENTIMENT is optimistic. This is more intriguing than the conditional effect of size. In the case of Age, there is no unconditional effect at all. It averages out across high and low sentiment periods. Yet this indicates a strong conditional effect. The next rows indicate that the cross-sectional effect of returns volatility is conditional on sentiment in the hypothesized manner. In particular, high Sigma stocks appear to be intensely out

19

of favor when SENTIMENT is low they earn huge subsequent returns of 3.04 percent per month. However, as with ME, the cross-sectional effect of Sigma appears only in low sentiment conditions. This suggests that the two effects are closely related. The next rows examine profitability and dividends. The simplest and perhaps most salient comparison for investors is between profitable and unprofitable (E<0) firms and between payers and nonpayers (D=0). These contrasts are summarized in the extreme right columns of Table 3, where we average returns across profitable firms and payers and compare them to unprofitable firms and nonpayers, respectively. They again demonstrate a conditional sign-flip pattern. Table 3b shows that when SENTIMENT is positive, monthly returns are 0.33 percent higher on profitable firms than unprofitable firms and 0.42 percent higher on payers. When it is negative, however, returns are 1.08 percent per month lower on profitable firms and 0.95 percent lower for payers. The leftmost column shows that this pattern is driven mostly by variation in the returns of unprofitable and nonpaying firms. Since these firms are hard to value and hard to arbitrage, it is not surprising that they are especially sensitive to swings in sentiment. The remaining variables book-to-market, external finance, and sales growth also display intriguing patterns. Most simply, each of them appears to have some unconditional explanatory power. Expected returns are generally higher for high BE/ME stocks, low EF/A stocks, and low GS decile stocks. The EF/A result is reminiscent of Loughran and Ritter (1995) and Spiess and Affleck-Graves (1995, 1999). But there are more interesting conditional patterns at work in these variables. The most prominent pattern is the inverted U-shaped difference in expected returns that is conditional on sentiment. The most striking case is GS in Table 3b. When SENTIMENT is high, there is an inverted U-shaped pattern across GS, as summarized in the 5-1 and 10-5 decile contrasts. When

20

sentiment reverses, this actually flips to an upward U shape. As a result, the difference between high and low sentiment regimes becomes a pronounced inverted U. Similar conditional U patterns are apparent, but somewhat less striking, in EF/A and BE/ME . Their shape suggests that investors are more willing to demand both high growth and distressed firms when they are optimistic, while they avoid these extremes when they are pessimistic. This may reflect the notion that extremely fast-growing or shrinking firms are hardest to value, and therefore most sensitive to sentiment. Overall, the sorts suggest that the cross-section of expected stock returns is highly sensitive to investor sentiment. The directions are roughly consistent with theoretical predictions and historical accounts. A highly stylized summary is as follows. When investors are optimistic, firms with salient characteristics that cause them to be particularly hard to value and same time hard to arbitrage young age, high volatility, no profits, no dividends, extreme growth or distress are more likely to be overpriced in relative terms. They therefore have lower expected returns. Likewise, firms with these characteristics are prone to being relatively underpriced when investors are pessimistic. They earn higher subsequent returns at such times. The conditional characteristics framework is helpful for thinking about these effects in a unified way. B. Predictive regressions for individual stocks We turn to cross-sectional predictive regressions. We run monthly cross-sectional predictive regressions and then study how the coefficients change with investor sentiment. Compared to the sorts, this approach allows us to conduct more formal inference, to determine which characteristics have conditional predictive ability that is distinct from known unconditional effects, and to present results for each sentiment proxy in a compact fashion.

21

As a baseline, we start by estimating the unconditional predictive ability of each characteristic. Each month, we run cross-sectional univariate predictive regressions: Rit = a + bt X it 1 + it , where X is a given characteristic. As earnings and dividends characteristics, we simply consider the profitability and payer dummies, as the sorts suggest that these will capture the main effects. Figure 2 reports the time series of the coefficients, and Panel A of Table 4 reports the average monthly coefficient and t-statistics based on the standard deviation of the coefficients. The results confirm prior evidence that size, book-to-market, profitability, and external finance have univariate, unconditional predictive power. High volatility and low sales growth are, on average, also associated with higher returns. These mean effects are apparent in Figure 2. Panel B runs multivariate cross-sectional regressions to distinguish novel unconditional effects from the known effects of size, book-to-market, and momentum: Rit = a + bt X it 1 + s t log (ME )it 1 + ht log ( BE ME )it1 + mt MOM it1 + it , where X denotes the characteristic of interest. (There is no X variable when we consider size and book-to-market themselves.) We find that age is modestly significant, while the unconditional effect of profitability disappears. Sales growth also loses some of its unconditional effect, while external finance and volatility retain very strong independent effects. Indeed, volatility actually knocks out size, a noteworthy result in itself. The key question is whether and how these coefficients change with investor sentiment. Are the fluctuations in Figure 2 connected to those in Figure 1? In Table 5, we address this question by regressing the monthly regression coefficients on sentiment proxies. The first several (4) (3)

22

columns regress coefficients from (3) on each of the sentiment proxies; the last column regresses coefficients from (4) on the composite index:4 = c + dSENTIMENT + . b t t 1 t To be clear, the monthly coefficients from cross-sectional regressions in calendar year t are regressed on the December of year t -1 value of CEFD, NIPO, S, and the overall SENTIMENT index, and the December of t -2 value of TURN , RIPO, and PD-ND. The sentiment proxies are standardized. Standard errors are bootstrapped to correct for the bias induced if the autocorrelated sentiment proxies have innovations that are correlated with innovations in the coefficients, as in Stambaugh (1999). The results confirm the sorts. The first rows show that as sentiment increases (lower CEFD or PD-ND or higher TURN , NIPO, RIPO, S, or SENTIMENT), expected returns tend to decrease on small firms, young firms, firms with volatile returns, unprofitable firms, and nondividend-paying firms. In all cases the results are in the expected direction, and those using IPO volume as the sentiment proxy are particularly strong. The last column shows that the connection between the cross-sectional effect of these variables and SENTIMENT can often be distinguished from arbitrarily time-varying cross-sectional effects of size, book-to-market, and momentum. It takes a bit of calculating to understand the magnitude of the effects. As an example, in the second-to-last columns of Table 5, the 0.2 coefficient of SENTIMENT on ME means that a one-SD increase in the sentiment index raises the size coefficient by 0.2 percentage points per month. Table 4 shows that such a change would bring the conditional size effect from negative to zero, which is in turn consistent with the Table 3b finding that the size effect does not exist in conditions of high sentiment. (5)

Intuitively, in terms of equation (1), this regresses estimates of (b 1 + b 2 Tt -1 ) on sentiment proxies Tt -1 .

23

Like the sorts, the regressions show that the cross-sectional effects of book-to-market, external finance, and sales growth do not have a strong linear relationship with sentiment. To capture their U-shaped conditional effects, we re-run (3) and (4) for these characteristics but limit the sample to observations within the top or bottom seven deciles of these variables.5 The idea is to bite off separate sides of the U. For instance, the GS (1-7) row examines how the crosssectional effect of sales growth varies with sentiment among firms in the lower range of sales growth, where a marginal increase may predominantly reflect a marginal decrease in distress, not an increase in growth opportunities. The GS (4-10) row examines the cross-sectional effect of sales growth in a range where the most salient effect of a marginal increase is an increase in growth opportunities, not a decrease in distress. Effectively, this procedure views GS (1-7) and GS (4-10) as if they were distinct characteristics. Figure 2 plots the cross-sectional coefficient from these regressions, which indeed suggests that these characteristics have different crosssectional effects in the bottom and top of their ranges. This procedure brings out the U-shaped conditional effects in book-to-market, external finance, and sales growth. As in the sorts, expected returns on stocks with characteristics of either high growth or distress are negatively related to sentiment proxies. One interpretation of this pattern is that sentiment is correlated with demand for the characteristics of extreme growth or distress, relative to more staid characteristics. Some calculations will confirm that the magnitude of the effects implied here are similar to those in the sorts. C. Predictive regressions for long-short portfolios Another way to look for conditional characteristics effects is to use sentiment to forecast portfolios that are long on stocks with high values of a characteristic and short on stocks with
5

The decision of how many deciles to cut out is obviously somewhat arbitrary. The choice of three follows Fama and French (1993), who identify the top three BE/ME deciles with value and the bottom three with growth.

24

low values. We have just seen that the average payer, for example, earns higher returns than the average nonpayer when sentiment is high, so sentiment must forecast a long-short portfolio formed on dividend payment. But it is useful to show explicitly how sentiment affects portfolios like SMB and HML, for example, because a large literature has used them as proxies for systematic risks. Also, the portfolio method is less parametric than the regressions for individual stocks, so to the extent it delivers similar results it indicates that the results are not driven by changes in the cross-sectional distribution of firm characteristics. The first several columns of Table 6 use individual sentiment proxies to predict longshort portfolios:6
R Xit = High, t RX it = Low, t = c + dSENTIMENTt1 + it .

(6)

So the dependent variable is the monthly return on a long-short portfolio, such as SMB, and these monthly returns from January through December of t are regressed on the December t -1 value of SENTIMENT or a sentiment proxy. The last column again separates novel comovement effects from known ones using a multivariate prediction:

R Xit = High, t RX it = Low, t = c + dSENTIMENTt1 + RMKTt + sSMB t + hHMLt + mUMDt + it

(7)

Obviously, we exclude SMB and HML from the right side when those are the portfolios being forecast. RMRF is the excess return of the value-weighted market over the risk-free rate. UMD is the return on high-momentum stocks, where momentum is measured over the period from 12 months prior through 2 months prior, minus the return on low-momentum stocks. As described in Fama and French (1993), SMB is constructed to be the return on portfolios of small and big

Intuitively, in terms of equation (1), this regresses (b 1 X + b 2 Tt -1 X) on sentiment proxies Tt -1 , where X is the difference between high and low levels of a characteristic.
6

25

ME stocks that is separate from returns on HML. HML is constructed to isolate the difference between high- and low- BE/ME portfolios.7 For profitability and dividend payment status, the coefficients from the portfolio approach are identical to those for the individual stock approach by construction. The other results are very similar (the sign for predicting SMB is opposite to the earlier ME sign). Both SMB and HML are significantly related to sentiment in some specifications. But for HML, as with long-short portfolios formed on external finance and sales growth, the more interesting story is the U-shaped conditional effect. We separate growth opportunities and distress effects by constructing High, Medium, and Low portfolios for these characteristics based on the top three, middle four, and bottom three NYSE decile breakpoints, respectively. Using sentiment to forecast the High-Medium portfolio for sales growth, for example, is analogous to using it to predict the cross-sectional effect of GS (4-10). D. Systematic risk The conditional characteristics effects are unlikely to be compensation for systematic risk, for a number of reasons. First, the sentiment variables have been orthogonalized with respect to macroeconomic conditions. Second, the patterns match theoretical predictions about where investor sentiment should have its most pronounced effects in the cross-section. Third, the patterns fit well with historical accounts of bubbles and crashes. Fourth, it is hard to envision an economy in which the closed-end fund discount, or the first-day returns on IPOs, would serve as crucial state variables even before being orthogonalized to macroeconomic conditions. Fifth, the systematic risk explanation requires that older, profitable, dividend-paying firms are (when sentiment is high, i.e. in half the sample) actually somewhat riskier than younger, unprofitable,

These portfolios are taken from Ken Frenchs website and are described there.

26

nonpaying firms, and are recognized as such by the marginal investor. This is counterintuitive. Sixth, the effects seem large. This is clearest in the sorts, which hint that conditional expected returns on (clearly risky) stocks are around zero or even negative for extreme combinations of characteristics and sentiment. Table 7 investigates the systematic risk explanation directly, asking whether sentiment coincides with time-variation in market betas in a way that would at least qualitatively reconcile the results with a conditional asset pricing model. Specifically, we predict returns on the characteristics portfolios:
R Xit = High, t RX it = Low, t = c + dSENTIMENTt1 + (e + fSENTIMENTt 1 )RMRF t + it .

(8)

The systematic risk hypothesis predicts that the composite coefficient f , reported in Table 7, will be of the same sign as the estimates of d in Table 6. A quick comparison of these tables shows, however, that when the coefficient f is statistically significant, it is usually the wrong sign. We have repeated this analysis with RMRF replaced by aggregate consumption growth. Similar results obtain, i.e. when the composite coefficient is significant, it is typically of the wrong sign for a systematic risk interpretation. A table is available upon request. E. Predictive regressions for earnings announcement returns Our last exercise looks for conditional characteristics effects in returns around earnings announcements. La Porta, Lakonishok, Shleifer, and Vishny (1997) find that low book-to-market stocks have lower earnings announcement returns than high book-to-market stocks, consistent with systematic errors in earnings expectations. Likewise, if errors in earnings expectations account for some of the results, we might expect that the average earnings announcement return on small, young, volatile, unprofitable, nonpaying, extreme growth and/or extreme distress firms would tend to be negatively related to sentiment.

27

This methodology has only limited power to detect how expectational errors affect our results, however, because the essential feature of our results is the correlated correction of mispricing. A firms announcement event returns, on the other hand, pick up the expectational corrections that occur only to it alone and within its own announcement window. A quote from Malkiel (1999) illustrates the problem: The music slowed drastically for the conglomerates on January 19, 1968. On that day, the granddaddy of the conglomerates, Litton Industries, announced that earnings for the second quarter of that year would be substantially less than forecast. the announcement was greeted with disbelief and shock. In the selling wave that followed, conglomerate stocks declined by roughly 40 percent (p. 67). And so while an analysis of announcement event returns will capture the idiosyncratic corrective effect of Litton Industries announcement on its own stock, it will capture none of its broader effects, which are important to our main results. This limitation notwithstanding, we gather quarterly earnings announcement dates from the merged CRSP-Compustat file, which are available beginning in 1971. For each firm-calendar quarter observation, we compute the cumulative abnormal return over the value-weighted market index across trading days [-1, +1] around the report date. We then construct a quarterly series of average announcement effects for each characteristic decile, and attempt to predict it with the composite sentiment index:
CARX it = Decile, t = c + dSENTIMENTt 1 + it .

(9)

Table 8 reports the coefficient estimates for each characteristic decile. The patterns in Table 8 should be compared to those in Table 3b. In Table 8, 12 of the 82 coefficients are significant at the 5% level. In Table 3b, on the other hand, sentiment is associated with an especially large conditional difference in monthly returns (more than 1.50

28

percentage points) in 10 of the 82 corresponding portfolios. The intersection of these strong results is 6 cells, and the signs agree in all cases. Among the 5 cells that are strong in Table 3b but insignificant in Table 8, 4 have matching signs. Likewise, among the 6 significant results in Table 8 that are not associated with especially large conditional differences in Table 3b, 5 are of the same sign. The clear anomaly is the significantly negative coefficient on the oldest firms; perhaps one such anomaly is a chance outcome of so many comparisons. Overall, these results suggest that at least a portion of our results appears to reflect the correction of conditional errors in earnings expectations.

V.

Summary and implications We develop and test a conditional characteristics model of the cross-section of expected

returns. In a nutshell, it is an empirical framework that captures the corrections of mispricing that arise in a world in which investors view stocks as bundles of salient investment characteristics, are prone to time-varying sentiment, and trade stocks that vary cross-sectionally in the difficulty of valuation and arbitrage. This approach ties together several important strands in the recent asset pricing literature. We motivate it further with descriptions of historical bubbles and crashes, as well as economic and psychological considerations. Empirical tests suggest that the conditional characteristics model is a useful empirical framework. Conditional on optimistic investor sentiment, measured by proxies that have been orthogonalized with respect to macroeconomic conditions, we find that expected returns are low on small, young, high return volatility, unprofitable, non-dividend paying, and extreme growth or distressed stocks. Conditional on pessimistic sentiment, on the other hand, these same stocks are generally associated with higher expected returns. These conditional effects appear to be largely

29

distinct from well-known cross-sectional patterns. The approach thus helps us to identify several new conditional patterns in the cross-section that leave no unconditional trace, because their sign depends on sentiment conditions. Several aspects of the results indicate that they do not reflect compensation for bearing systematic risk. For instance, we directly consider, and reject, a connection between the conditional patterns in expected returns and conditional patterns in covariance with market returns or aggregate consumption growth. In addition, we find some conditional patterns in earnings announcement returns, which suggests that at least a component of the basic results can be directly attributed to expectational errors. If the conditional characteristics model and our interpretation are correct, the implications for finance are significant. For corporate finance, one implication is the relevance of dividends. Black and Scholes (1974) view the unstable cross-sectional effect of dividends as evidence that dividends are irrelevant to market value, as in the efficient markets setup of Miller and Modigliani (1961). Our results suggest that this conclusion is incorrect. While the cross-sectional effect of dividends is indeed unstable, its sign is predictable from prevailing sentiment. This suggests that dividends are in fact quite relevant, but in a time-varying direction, as in Baker and Wurgler (2003). More generally, the conditional characteristics framework may be useful for examining whether other corporate behaviors that involve the adoption of salient characteristics should be viewed as catering responses to mispricings, for instance earnings manipulation as a function of sentiment. For asset pricing, the implications appear quite significant. The results call into question whether the cross-section of expected returns is usefully viewed as an equilibrium outcome of cross-sectional differences in systematic risk. Indeed, some proxies for investor sentiment, after

30

being orthogonalized to macroeconomic conditions, can predict the factor portfolios of the Fama and French (1993) three-factor model. The results also suggest that stock returns are difficult to properly interpret without some appreciation of historical stock market bubbles and crashes. More research will help to determine whether the systematic risk model should, as an empirical matter, be replaced by the conditional characteristics model.

31

References Amihud, Yakov, and Haim Mendelson, 1986, Asset pricing and the bid-ask spread, Journal of Financial Economics, 17, 223-49. Baker, Malcolm, and Jeremy Stein, 2002, Market liquidity as a sentiment indicator, NBER Working Paper Series, No. 8816. Baker, Malcolm, and Jeffrey Wurgler, 2000, The equity share in new issues and aggregate stock returns, Journal of Finance 55, 2219-57. Baker, Malcolm, and Jeffrey Wurgler, 2002, A catering theory of dividends, Journal of Finance, forthcoming. Banz, Rolf, 1981, The relationship between return and market value of common stocks, Journal of Financial Economics 9, 3-18. Barberis, Nicholas, and Andrei Shleifer, 2003, Style investing, Journal of Financial Economics, forthcoming. Barberis, Nicholas, Andrei Shleifer, and Jeffrey Wurgler, 2003, Comovement, University of Chicago working paper. Basu, Sanjoy, 1983, The relationship between earnings yield, market value, and return for NYSE common stocks: Further evidence, Journal of Financial Economics 12, 129-156. Benveniste, Lawrence M., Alexander P. Ljungqvist, William J. Wilhelm, Jr., and Xiaoyun Yu, 2003, Evidence of information spillovers in the production of investment banking services, Journal of Finance 58, 577-608. Black, Fischer, and Myron S. Scholes, 1974, The effects of dividend yield and dividend policy on common stock prices and returns, Journal of Financial Economics 1, 1-22. Blume, Marshall, and Robert Stambaugh, 1983, Biases in computed returns: An application to the size effect, Journal of Financial Economics 12, 387-404. Breeden, Douglas T., 1979, An intertemporal asset pricing model with stochastic consumption and investment opportunities, Journal of Financial Economics 7, 265-296. Brennan, Michael J., 1970, Taxes, market valuation and corporate financial policy, National Tax Journal 23: 417-427. Brown, John Dennis, 1991, 101 Years on Wall Street , Prentice Hall. Choe, Hyuk, Ronald Masulis, and Vikram Nanda, 1993, Common stock offerings across the business cycle: Theory and evidence, Journal of Empirical Finance 1, 3-31. Christie, William, 1990, Dividend yield and expected returns: The zero-dividend puzzle, Journal of Financial Economics 14, 451-471. Christie, William, and Roger Huang, 1994, The changing functional relation between stock returns and dividend yields, Journal of Empirical Finance 1, 161-191. DAvolio, 2002, The market for borrowing stock, Journal of Financial Economics, forthcoming. Daniel, Kent, and Sheridan Titman, 1997, Evidence on the characteristics of cross-sectional variation in stock returns, Journal of Finance 46, 1739-1764. De Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert Waldmann, 1990, Noise trader risk in financial markets, Journal of Political Economy 98, 703-738. Dreman, 1979, Contrarian Investment Strategy, Dreman Contrarian Group. Fama, Eugene F., and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, 427-465. Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3-56.

32

Fama, Eugene F., and Kenneth R. French, 2001, Disappearing dividends: Changing firm characteristics or lower propensity to pay?, Journal of Financial Economics 60, 3-44. Greenwood, Robin, and Nathan Sosner, 2002, Trade and the comovement of stock returns: Evidence from Japan, Harvard University working paper. Ibbotson, Roger, and Jeffrey F. Jaffe, 1975, Hot issue markets, Journal of Finance 30, 1027-42. Ibbotson, Roger, Jody Sindelar, and Jay Ritter, 1994, The market's problems with the pricing of initial public offerings, Journal of Applied Corporate Finance 7, 66-74. Jones, Charles, 2001, A century of stock market liquidity and trading costs, Working paper, Columbia University. Keim, Donald, 1983, Size-related anomalies and stock return seasonality: Further empirical evidence, Journal of Financial Economics 12, 13-32. Keim, Donald, 1985, Dividend yields and stock returns: Implications of abnormal January returns, Journal of Financial Economics 14, 473-489. Kindleberger, Charles, 2001, Manias, Panics, and Crashes, Wiley. La Porta, Rafael, Josef Lakonishok, Andrei Shleifer, and Robert Vishny, 1997, Good news for value stocks: Further evidence on market efficiency, Journal of Finance 52:859-74. Lee, Charles, Andrei Shleifer, and Richard Thaler, Investor sentiment and the closed-end fund puzzle, Journal of Finance 46, 75-109. Lettau, Martin, and Sydney Ludvigson, 2001, Consumption, aggregate wealth, and expected stock returns, Journal of Finance 56, 815-49. Lettau, Martin, and Sydney Ludvigson, 2001, Resurrecting the (C)CAPM: A cross-sectional test when risk premia are time-varying, Journal of Political Economy 109, 1238-87. Lewellen, Jonathan, and Stefan Nagel, 2003, The conditional CAPM does not explain assetpricing anomalies, MIT working paper. Litzenberger, Robert H., and Krishna Ramaswamy, 1979, The effect of personal taxes and dividends on capital asset prices, Journal of Financial Economics 7, 163-195. Ljunqvist, Alexander and William Wilhelm, 2003, IPO pricing in the dot-com bubble, Journal of Finance, forthcoming. Loughran, Tim and Jay Ritter, 1995, The new issues puzzle, Journal of Finance 50, 23-51. Lowry, Michelle, 2003, Why does IPO volume fluctuate so much? Journal of Financial Economics 67, 3-40. Lowry, Michelle and G. William Schwert, 2002, IPO market cycles: Bubbles or sequential learning?, Journal of Finance 57, 1171 - 1200. Malkiel, Burton, 1990, A Random Walk Down Wall Street , W.W. Norton. Malkiel, Burton, 1999, A Random Walk Down Wall Street , W.W. Norton. Markowitz, Harry, 1959, Portfolio Selection: Efficient Diversification of Investments, John Wiley, New York. Merton, Robert C., 1973, An intertemporal capital asset pricing model, Econometrica 41, 867887. Miller, Merton H., and Franco Modigliani, 1961, Dividend policy, growth and the valuation of shares, Journal of Business 34, 411-433. Miller, Edward M., 1977, Risk, uncertainty, and divergence of opinion, Journal of Finance 32, 1151-1168. Neal, Robert, with Simon Wheatley, 1998, Do measures of investor sentiment predict stock returns, Journal of Financial and Quantitative Analysis 34, 523-547.

33

Ofek, Eli and Matthew Richardson, 2002a, DotCom Mania: The rise and fall of internet stocks, Journal of Finance, forthcoming. Ofek, Eli, and Matthew Richardson, 2002b, The valuation and market rationality of internet stock prices, NYU working paper. Reinganum, Marc R., 1983, The anomalous stock market behavior of small firms in January: Empirical tests for tax-loss selling effects, Journal of Financial Economics 12, 89-104. Ritter, Jay, 1984, The 'hot issue' market of 1980, The Journal of Business 57, 215-40. Ritter, Jay, 1991, The long-run performance of initial public offerings, Journal of Finance 46, 327. Rosch, Eleanor, 1978, Principles of categorization, in Eleanor Rosch and Barbara B. Lloyd, eds.: Cognition and Categorization (Lawrence Erlbaum Associates, Hillsdale, NJ). Ross, Stephen A., 1976, The arbitrage theory of capital asset pricing, Journal of Economic Theory 13, 341-360. Sharpe, William F., 1964, Capital asset prices: A theory of market equilibrium under conditions of risk, Journal of Finance 19, 425-442. Shiller, Robert, 2000, Irrational Exuberance, Princeton University Press. Shleifer, Andrei, 2000, Inefficient Markets, Oxford University press. Shleifer, Andrei, and Robert W. Vishny, 1997, The limits of arbitrage, Journal of Finance 52, 35-55. Siegel, 1998, Stocks for the Long Run, McGraw-Hill. Speiss, D. Katherine, and John Affleck-Graves, 1995, Underperformance in long-run stock returns following seasoned equity offerings, Journal of Financial Economics 38, 243-267. Spiess, Katherine, and John Affleck-Graves, 1999, The long-run performance of stock returns following debt offerings, Journal of Financial Economics 54, 45-73. Stambaugh, Robert F., 1999, Predictive regressions, Journal of Financial Economics 54, 375421. Stigler, George J., 1964, Public regulation of the securities markets, Journal of Business 37, 117142. Wurgler, Jeffrey, and Katia Zhuravskaya, 2002, Does arbitrage flatten demand curves for stocks?, Journal of Business 75, 583-608. Zweig, Martin E., 1973, An investor expectations stock price predictive model using closed-end fund premiums, Journal of Finance 28, 67-87.

34

Figure 1. Investor Sentiment. The first panel shows the year-end, value-weighted average discount on closed-end mutual funds. The data on prices and net asset values (NAVs) come from Neal and Wheatley (1998) for 1962 through 1993; CDA/Wiesenberger for 1994 through 1998; and turn-of-the-year issues of the Wall Street Journal for 1999 and 2000. The second panel shows detrended log turnover. Turnover is the ratio of reported share volume to average shares listed from the NYSE Fact Book. We detrend using the past five-year average. The third panel shows the annual number of initial public offerings. The fourth panel shows the average annual first-day returns of initial public offerings. Both series come from Jay Ritter, updating data analyzed in Ibbotson, Sindelar, and Ritter (1994). The fifth panel shows gross annual equity issuance divided by gross annual equity plus debt issuance from Baker and Wurgler (2000). We adjust the series for the one-time change in the share from 1983 to 1984. The sixth panel shows the year-end log ratio of the value-weighted average market-to-book ratios of payers and nonpayers from Baker and Wurgler (2003). The dashed line is raw data. We regress each measure on the growth in industrial production, the growth in durable, nondurable and services consumption, the growth in employment and a flag for NBER recessions. The solid line is the residuals from this regression. The final panel presents a first principal component index of the six measures. In the index, turnover, the average annual first-day return, and the dividend premium are lagged one year relative to the other three measures. Panel A. Closed-end fund discount %
30 25 20 15 10 5 0 -5 -10 -15 1962 1967 1972 1977 1982 1987 1992 1997 15 10 5 0 -5 -10 -15 -20 25 20 15 10 5 0 -5 -10 -15 1962 1967 1972 1977 1982 1987 1992 1997

Panel B. Turnover %
15 10 5 0 -5 -10 -15

Panel C. Number of IPOs


1200 1000 800 600 400 200 0 1962 1967 1972 1977 1982 1987 1992 1997 600 400 200 0 -200 -400 -600 80 70 60 50 40 30 20 10 0 -10

Panel D. Average first-day return


60 50 40 30 20 10 0 -10 -20 -30 1962 1967 1972 1977 1982 1987 1992 1997

Panel E. Equity share in new issues


60 50 40 30 20 10 0 1962 1967 1972 1977 1982 1987 1992 1997 30 20 10 0 -10 -20 -30
40 30 20 10 0 -10 -20 -30 -40 1962

Panel F. Dividend premium


50 40 30 20 10 0 -10 -20 -30 -40 -50 1967 1972 1977 1982 1987 1992 1997

Panel E. Sentiment index


2.5 1.5 0.5 -0.5 -1.5 -2.5 1962 1967 1972 1977 1982 1987 1992 1997

Figure 2. Average Annual Coefficients. Average annual coefficients from monthly univariate regressions of returns on firm characteristics X.

Rit = a t + bt X it 1 + eit
The firm characteristics are size, age, total risk, indicator variables for profitable firms and dividend payers, book-to-market ratio, external finance over assets, and sales growth decile. Size is the log of market equity. Market equity (ME) is price times shares outstanding from CRSP. Age is the number of years since the firms first appearance on CRSP. Total risk is the annual standard deviation in monthly returns from CRSP. Earnings (E) is defined as income before extraordinary items (Item 18) plus income statement deferred taxes (Item 50) minus preferred dividends (Item 19). The book-to-market ratio is the log of the ratio of book equity to market equity. Book equity (BE) is defined as shareholders equity (Item 60) plus balance sheet deferred taxes (Item 35). External finance (EF) is equal to the change in assets (Item 6) less the change in retained earnings (Item 36). When the change in retained earnings is not available we use net income (Item 172) less common dividends (Item 21) instead. Sales growth decile is formed using NYSE breakpoints for sales growth. Sales growth is the percentage change in net sales (Item 12). For the last three characteristics, we analyze the top (solid) and bottom (dashed) seven deciles separately. Panel A. log(ME)
1.0 0.5 0.0 -0.5 -1.0 -1.5 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 0.08 0.06 0.04 0.02 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

Panel B. Age
50 40 30 20 10 0 -10 -20

Panel C. Total Risk

Panel D. E>0
3.0 2.0 1.0 0.0 -1.0 -2.0 -3.0 -4.0 -5.0 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 3.0 2.0 1.0 0.0 -1.0 -2.0 -3.0 -4.0 -5.0

Panel E. D>0
2.5 2.0 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

Panel G. EF/A
10.0 5.0 0.0 -5.0 -10.0 -15.0 -20.0 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 6.0 4.0 2.0 0.0 -2.0 -4.0 -6.0 -8.0

Panel H. GS/10

1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

Panel F. log(BE/ME)

Table 1. Summary Statistics, 1963-2001. Panel A summarizes the returns variables. Returns are measured monthly. Momentum (MOM) is defined as the cumulative return for the eleven-month period between 12 and two months prior to t. Panel B summarizes the size, age, and risk characteristics. Size is the log of market equity. Market equity (ME) is price times shares outstanding from CRSP in the June prior to t. Age is the number of years between the firms first appearance on CRSP and t. Total risk () is the annual standard deviation in monthly returns from CRSP for the 12 months ending in the June prior to t. Panel C summarizes profitability variables. The earnings-price ratio is defined for firms with positive earnings. Earnings (E) is defined as income before extraordinary items (Item 18) plus income statement deferred taxes (Item 50) minus preferred dividends (Item 19). Book equity (BE) is defined as shareholders equity (Item 60) plus balance sheet deferred taxes (Item 35). Panel D reports dividend variables. Dividends (D) are equal to dividends per share at the ex date (Item 26) times shares outstanding (Item 25). Panel E reports variables used as proxies for growth opportunities and distress. The book-to-market ratio is the log of the ratio of book equity to market equity. External finance (EF) is equal to the change in assets (Item 6) less the change in retained earnings (Item 36). When the change in retained earnings is not available we use net income (Item 172) less common dividends (Item 21) instead. Sales growth decile is formed using NYSE breakpoints for sales growth. Sales growth is the percentage change in net sales (Item 12). In Panels C through E, accounting data from the fiscal year ending in t-1 are matched to monthly returns from July of year t through June of year t+1. All variables are Winsorized at 99.5 and 0.5 percent. Full Sample N Rt (%) MOM t -1 (%) MEt -1 ($M) Age t (Years) t-1 (%) E+/BEt -1 (%) E>0t -1 D/BEt -1 (%) D>0t -1 BE/MEt -1 EF/At -1 (%) GSt -1 (Decile) 1,398,432 1,398,432 1,398,432 1,398,432 1,396,198 1,398,432 1,398,432 1,398,432 1,398,432 1,398,432 1,365,115 1,349,463 Mean SD Min Max 1960s Panel A. Returns 1.38 13.44 576 13.68 14.39 10.62 0.76 1.96 0.45 0.93 11.70 5.94 18.70 60.32 2,152 13.89 9.07 10.32 0.42 2.96 0.50 0.86 24.99 3.19 -97.37 -86.00 1 0.03 0.00 0.00 0.00 0.00 0.00 0.02 -69.00 1.00 2400.00 359.40 21,623 69.17 64.00 67.48 1.00 17.61 1.00 5.78 131.18 10.00 1.06 21.45 392 15.91 9.20 12.14 0.96 4.42 0.78 0.70 7.09 5.66 1.59 12.51 238 12.97 12.62 12.10 0.91 2.72 0.65 1.35 6.34 5.66 1.24 14.85 371 13.80 13.69 11.29 0.77 1.98 0.48 0.92 10.72 5.97 1.45 12.11 805 13.65 15.28 9.28 0.68 1.36 0.30 0.72 14.77 6.12 1.19 11.95 1,411 14.06 21.53 9.08 0.63 1.05 0.24 0.81 19.16 5.91 Subsample Means 1970s 1980s 1990s 2000-1

Panel B. Size and Age

Panel C. Profitability

Panel D. Dividend Policy

Panel E. Growth Opportunities and Distress

Table 2. Investor Sentiment. Means, standard deviations, and correlations for measures of investor sentiment. The first measure (CEFD) is the year-end, valueweighted average discount on closed-end mutual funds. The data on prices and net asset values (NAVs) come from Neal and Wheatley (1998) for 1962 through 1993; CDA/Wiesenberger for 1994 through 1998; and turn-of-the-year issues of the Wall Street Journal for 1999 and 2000. The second measure (TURN) is detrended log turnover. Turnover is the ratio of reported share volume to average shares listed from the NYSE Fact Book. We detrend using the past five-year average. The third measure (NIPO) is the annual number of initial public offerings. The fourth measure (RIPO) is the average annual first-day returns of initial public offerings. Both IPO series come from Jay Ritter, updating data analyzed in Ibbotson, Sindelar, and Ritter (1994). The fifth measure (S) is gross annual equity issuance divided by gross annual equity plus debt issuance from Baker and Wurgler (2000). We adjust the series for the one-time change in the share from 1983 to 1984. The sixth measure (PD-ND ) is the year-end log ratio of the value-weighted average market-to-book ratios of payers and nonpayers from Baker and Wurgler (2003). Turnover, the average annual first-day return, and the dividend premium are lagged one year relative to the other three measures. The index is the first principal component of the six measures. In the first panel, we present raw data. In the second panel, we regress each measure on the growth in industrial production, the growth in durable, nondurable and services consumption, the growth in employment and a flag for NBER recessions. The adjusted measures are the residuals from these regressions. The sentiment index is the first principal component of the six measures. a, b, and c denote statistical significance at 1%, 5%, and 10%. Correlations Mean CEFDt TURNt -1 NIPOt RIPOt -1 St Pt -1
D-ND

SD 8.12 7.94 262.76 14.93 10.08 18.67 6.14 6.73 226.30 14.31 9.47 16.89

Min -10.41 -11.60 9.00 -1.67 11.15 -33.17 -18.31 -11.31 -435.98 -23.55 -22.17 -43.20

Max 23.70 18.66 953.00 69.53 50.03 36.06 9.27 11.45 484.15 46.54 18.78 35.96

Index -0.51a 0.62 0.74 0.78 -0.85


a a

CEFD 1.00 -0.30c -0.57 -0.31 0.53


a

TURN

NIPO

RIPO

PD-ND

Panel A. Raw data 8.87 5.21 358.41 16.94 28.97 0.20 0.00 0.00 0.00 0.00 0.00
D-ND

1.00 0.38b 0.50a 0.31 -0.50


c a

1.00 0.35b 0.77 -0.56


a a

0.73a
a a

-0.42a
c a

1.00 0.33b -0.58


a

1.00 -0.67a 1.00

Panel B. Controlling for macroeconomic conditions CEFDt TURNt -1 NIPOt RIPOt -1 St Pt -1 -0.51a 0.65 0.82 0.76 0.80 -0.85
a a a a a

1.00 -0.28c -0.46 -0.46 0.28


a a

1.00 0.39b 0.53 0.35 -0.60


a b a

1.00 0.44a 0.77 -0.46


a a

1.00 0.37b -0.68


a

-0.18
c

1.00 -0.61a 1.00

0.00

Table 3a. Two-way Sorts: Closed-End Fund Discount and Firm Characteristics. For each month, we form ten portfolios according to the NYSE breakpoints of firm size (ME), age, total risk, earnings-book ratio for profitable firms (E/BE), dividend-book ratio for dividend payers (D/BE), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth (GS). We also calculate portfolio returns for unprofitable firms and nonpayers. We then report average portfolio returns over months where the closed-end fund discount is positive, negative, and the difference between the two averages. ClosedEnd Fund Discount 0 ME Positive Negative Difference Age Positive Negative Difference Positive Negative Difference E/BE Positive Negative Difference D/BE Positive Negative Difference BE/ME Positive Negative Difference EF/A Positive Negative Difference GS Positive Negative Difference 2.48 0.25 2.23 2.35 0.12 2.23 1 2.40 0.46 1.94 1.97 -0.23 2.20 1.20 0.57 0.64 2.21 0.53 1.68 2.08 0.60 1.48 1.47 -0.27 1.74 2.52 0.76 1.76 2.37 0.66 1.71 2 1.93 0.21 1.73 1.99 0.50 1.49 1.39 0.50 0.89 2.08 0.65 1.43 1.90 0.51 1.39 1.72 0.09 1.63 2.18 0.75 1.43 2.04 0.61 1.43 3 1.81 0.36 1.46 2.05 0.67 1.38 1.43 0.54 0.90 2.11 0.89 1.22 1.92 0.73 1.19 1.78 0.35 1.43 2.08 0.84 1.25 1.97 0.58 1.39 4 1.79 0.24 1.56 1.94 0.47 1.46 1.52 0.50 1.02 1.92 0.53 1.38 1.77 0.47 1.30 1.86 0.38 1.48 1.98 0.77 1.22 1.79 0.65 1.15

Decile 5 1.83 0.53 1.29 2.02 0.62 1.40 1.55 0.50 1.05 1.79 0.46 1.32 1.71 0.58 1.13 1.87 0.56 1.31 1.82 0.75 1.08 1.82 0.62 1.20 6 1.63 0.33 1.30 2.01 0.64 1.37 1.68 0.48 1.20 1.95 0.49 1.46 1.59 0.68 0.91 1.93 0.67 1.26 1.89 0.63 1.27 1.88 0.67 1.21 7 1.62 0.45 1.17 1.57 0.66 0.91 1.72 0.48 1.24 1.90 0.38 1.52 1.53 0.76 0.77 2.12 0.67 1.45 1.88 0.38 1.50 1.98 0.68 1.31 8 1.49 0.55 0.94 1.67 0.65 1.01 1.80 0.26 1.54 1.88 0.50 1.38 1.47 0.70 0.77 2.12 0.77 1.35 1.80 0.38 1.42 2.08 0.56 1.52 9 1.38 0.46 0.92 1.63 0.58 1.05 1.98 0.13 1.85 1.77 0.48 1.29 1.44 0.70 0.74 2.33 0.89 1.45 1.92 0.24 1.67 1.90 0.34 1.57 10 1.31 0.38 0.93 1.48 0.72 0.76 2.86 0.60 2.26 1.87 0.36 1.51 1.38 0.81 0.56 2.58 0.94 1.64 1.66 -0.46 2.12 1.81 -0.41 2.22 10-1 -1.08 -0.08 -1.00 -0.49 0.95 -1.44 1.66 0.03 1.62 -0.34 -0.17 -0.18 -0.70 0.21 -0.91 1.11 1.21 -0.10 -0.87 -1.23 0.36 -0.56 -1.07 0.51

Overall 10-5 5-1 >00

-0.61 0.23 -0.84 -0.63 0.52 -1.16 0.71 0.38 0.33 -0.17 -1.21 1.05 -0.01 -1.03 1.02 0.40 0.83 -0.43 -0.70 -0.02 -0.68 -0.54 -0.04 -0.51

Table 3b. Two-way Sorts: Sentiment Index and Firm Characteristics. For each month, we form ten portfolios according to the NYSE breakpoints of firm size (ME), age, total risk, earnings-book ratio for profitable firms (E/BE), dividend-book ratio for dividend payers (D/BE), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth (GS). We also calculate portfolio returns for unprofitable firms and nonpayers. We then report average portfolio returns over months where the sentiment index is positive, negative, and the difference between the two averages. Sentiment Index 0 ME Positive Negative Difference Age Positive Negative Difference Positive Negative Difference E/BE Positive Negative Difference D/BE Positive Negative Difference BE/ME Positive Negative Difference EF/A Positive Negative Difference GS Positive Negative Difference 0.60 2.89 -2.29 0.63 2.55 -1.92 1 0.89 2.59 -1.70 0.34 2.01 -1.67 0.97 0.92 0.04 0.78 2.54 -1.77 1.02 2.11 -1.09 0.13 1.63 -1.51 1.29 2.51 -1.22 1.00 2.59 -1.59 2 0.74 1.91 -1.17 0.89 2.06 -1.16 0.99 1.09 -0.10 0.99 2.18 -1.19 0.98 1.83 -0.85 0.69 1.56 -0.88 1.19 2.15 -0.96 1.21 1.79 -0.58 3 0.80 1.80 -0.99 1.03 2.11 -1.08 0.95 1.23 -0.28 0.92 2.54 -1.62 1.15 1.82 -0.66 0.88 1.63 -0.75 1.37 1.86 -0.49 1.20 1.69 -0.49 4 0.75 1.73 -0.98 0.97 1.82 -0.85 0.88 1.42 -0.55 0.94 1.92 -0.99 0.92 1.68 -0.76 1.00 1.62 -0.62 1.28 1.78 -0.50 1.13 1.60 -0.48

Decile 5 0.96 1.76 -0.80 1.23 1.75 -0.52 0.85 1.49 -0.63 0.94 1.68 -0.74 1.06 1.53 -0.47 1.11 1.65 -0.54 1.21 1.62 -0.41 1.20 1.53 -0.33 6 0.86 1.45 -0.59 1.32 1.64 -0.32 0.88 1.61 -0.73 0.94 1.93 -0.98 1.02 1.49 -0.46 1.15 1.78 -0.63 1.19 1.64 -0.45 1.18 1.67 -0.49 7 1.05 1.29 -0.24 1.04 1.41 -0.37 0.82 1.75 -0.93 0.92 1.78 -0.86 1.09 1.39 -0.30 1.18 2.02 -0.84 1.07 1.56 -0.49 1.19 1.82 -0.63 8 1.00 1.25 -0.25 1.13 1.43 -0.30 0.63 1.91 -1.28 1.04 1.70 -0.66 1.09 1.26 -0.17 1.19 2.08 -0.89 0.89 1.67 -0.78 1.09 1.98 -0.89 9 1.01 1.02 -0.01 1.05 1.43 -0.38 0.59 2.09 -1.50 1.03 1.54 -0.51 1.11 1.20 -0.09 1.33 2.32 -1.00 0.77 1.88 -1.11 0.89 1.78 -0.88 10 1.01 0.84 0.17 1.07 1.32 -0.26 1.13 3.04 -1.91 0.89 1.76 -0.87 1.01 1.34 -0.32 1.46 2.53 -1.08 0.06 1.79 -1.73 0.18 1.89 -1.71 10-1 0.12 -1.75 1.87 0.73 -0.69 1.42 0.17 2.12 -1.95 0.12 -0.78 0.90 -0.01 -0.77 0.77 1.33 0.90 0.43 -1.23 -0.72 -0.51 -0.81 -0.70 -0.12

Overall 10-5 5-1 >00

0.33 -1.08 1.42 0.42 -0.95 1.36 0.34 0.89 -0.54 -1.15 0.17 -1.32 -1.02 0.37 -1.38 0.99 0.02 0.97 -0.08 -0.89 0.80 0.20 -1.06 1.26

Table 4. Baseline Regressions, Monthly Returns. Average coefficients and t-statistics from monthly regressions of returns on firm characteristics (X), size (log(ME)), book-to-market (log(BE/ME)), and momentum (MOM).

Rit = at + bt X it 1 + st log (ME )it 1 + ht log ( BE ME )it 1 + mt MOM it 1 + eit

We only report the average of b t . The firm characteristics are firm size (log(ME)), age (log(Age)), total risk (), an indicator variable for profitable firms (E>0) and dividend payers (D>0), book-to-market ratio (log(BE/ME)), external finance over assets (EF/A), and sales growth decile (GS). The first panel shows univariate results. The second panel includes book-to-market, size, and momentum as control variables. Standard errors are equal to the time series standard deviation of b t divided by the number of months. ME s -0.17 -0.18 0.02 -0.16 -0.16 -0.14 -0.16 -0.15 t(s) [-3.3] [-3.3] [0.4] [-3.3] [-3.5] [-2.4] [-3.0] [-2.8] Age b 0.03 0.08 t(b) [0.4] [2.3] 9.24 [5.3] -0.05 [-0.4] 0.05 [0.4] b 8.28 t(b) [4.5] E>0 b -0.26 t(b) [-1.3] D>0 b -0.17 t(b) [-0.8] BE/ME h 0.47 0.32 0.51 0.35 0.34 0.51 0.26 0.32 t(h) [6.5] [4.5] [8.8] [4.9] [5.1] [6.7] [3.9] [4.4] -1.16 [-6.8] -0.26 [-2.7] EF/A b -1.63 t(b) [-6.8] GS/10 b -0.59 t(b) [-4.6]

Panel A. Univariate Panel B. Controlling for Book-to-Market, Size, and Momentum

Table 5. Time Series Regressions, Monthly Returns Coefficients. Two-stage regression. In the first stage, we compute coefficients from monthly regressions of returns on firm characteristics (X), size (log(ME)), book-to-market (log(BE/ME)), and momentum (MOM).

Rit = at + bt X it 1 + st log (ME )it 1 + ht log ( BE ME )it 1 + mt MOM it 1 + eit


The firm characteristics are firm size (log(ME)), age (log(Age)), total risk (), an indicator variable for profitable firms (E>0) and dividend payers (D>0), bookto-market ratio (log(BE/ME)), external finance over assets (EF/A), and sales growth decile (GS). When we analyze growth opportunities, we exclude the top (bottom) three deciles for book-to-market ratio (external finance, sales growth). When we analyze distress, we exclude the bottom (top) three deciles for book-tomarket ratio (external finance, sales growth). In the second stage, we regress the monthly coefficients b t on measures of investor sentiment, each standardized to have unit variance.

= c + dSENTIMENT + u b t t 1 t
Coefficients are matched to the closed-end fund discount (CEFD), the number of IPOs (NIPO), and the equity share (S) for the calendar year one year before t. Coefficeints are matched to detrended log turnover (TURN), the average annual first-day return (RIPO), and the dividend premium (PD-ND ) for the calendar year two years before t. The first seven columns show univariate results. The last column includes size, book-to-market, and momentum as control variables in the first-stage regression and uses the sentiment index in the second stage. Bootstrap p-values are in braces. Controlling for Size, B/M, and MOM d 0.2 0.1 -3.9 -0.5 0.5 0.2 -0.4 -0.2 0.3 -0.9 -0.1 0.0 1.1 0.0 p(d) [.00] [.15] [.04] [.00] [.00] [.02] [.08] [.22] [.01] [.01] [.00] [.64] [.01] [.09]

CEFDt -1 d ME Age E>0 D>0 BE/ME EF/A GS BE/ME (1-7) EF/A (4-10) GS (4-10) BE/ME (4-10) EF/A (1-7) GS (1-7) -0.2 -0.2 6.4 0.7 -0.8 -0.0 0.6 0.2 -0.1 1.8 0.1 0.1 -2.1 -0.1 p(d) [.00] [.02] [.00] [.01] [.00] [.68] [.02] [.16] [.31] [.00] [.00] [.30] [.00] [.02]

TURN t -2 d 0.1 0.1 -3.9 -0.5 0.5 0.1 -0.2 -0.1 0.1 -0.8 -0.1 -0.1 1.1 0.1 p(d) [.06] [.11] [.06] [.03] [.03] [.48] [.52] [.40] [.15] [.11] [.01] [.17] [.11] [.03]

NIPO t -1 d 0.2 0.3 -7.2 -0.9 1.0 0.1 -0.6 -0.3 0.2 -1.9 -0.1 -0.1 2.4 0.1 p(d) [.00] [.00] [.00] [.00] [.00] [.19] [.02] [.06] [.01] [.00] [.00] [.21] [.00] [.01]

RIPO t -2 d 0.2 0.2 -5.2 -0.7 0.7 0.1 -0.5 -0.3 0.1 -1.6 -0.1 -0.1 1.9 0.1 p(d) [.01] [.01] [.01] [.00] [.00] [.55] [.06] [.04] [.12] [.01] [.00] [.31] [.01] [.05]

S t -1 d 0.1 0.2 -5.3 -0.7 0.8 0.1 -0.7 -0.4 0.3 -1.5 -0.1 -0.1 1.7 0.1 p(d) [.03] [.00] [.03] [.01] [.00] [.13] [.02] [.03] [.02] [.01] [.00] [.24] [.03] [.09]

P t -2 D-ND d -0.1 -0.1 4.0 0.6 -0.5 -0.0 0.2 0.2 -0.1 1.0 0.1 0.2 -1.6 -0.1 p(d) [.13] [.21] [.09] [.03] [.09] [.89] [.52] [.34] [.27] [.11] [.04] [.07] [.04] [.09]

Sentiment Index d 0.2 0.2 -7.0 -0.9 0.9 0.1 -0.6 -0.3 0.2 -1.8 -0.1 -0.2 2.4 0.1 p(d) [.01] [.00] [.00] [.00] [.00] [.28] [.05] [.05] [.03] [.01] [.00] [.10] [.00] [.01]

Table 6. Time Series Regressions, Portfolios. Regressions of long-short portfolio returns on measures of sentiment (S), each standardized to have unit variance, the market risk premium (RMRF), the Fama-French factors (HML and SMB), and a momentum factor (UMD).

R X it = high,t R Xit =low,t = c + dSENTIMENTt 1 + RMRF t + sSMB t + hHML t + mUMDt + u t


The long-short portfolios are formed based on firm characteristics (X): firm size (ME), age, total risk (), profitability (E), dividends (D), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth decile (GS). High is defined as a firm in the top three NYSE deciles; low is defined as a firm in the bottom three NYSE deciles; medium is defined as a firm in the middle four NYSE deciles. Monthly returns are matched to the closed-end fund discount (CEFD), the number of IPOs (NIPO), and the equity share (S) for the calendar year one year before t. Monthly returns are matched to detrended log turnover (TURN), the average annual first-day return (RIPO), and the dividend premium (PD-ND ) for the calendar year two years before t. The first seven columns show univariate results. The last column includes RMRF, SMB, HML, and UMD as control variables. Bootstrap p-values are in braces. Controlling for RMRF, SMB, HML, and UMD d -0.4 0.3 -0.3 -0.5 0.4 0.1 -0.2 -0.2 0.2 -0.3 -0.3 -0.1 0.2 0.1 p(d) [.02] [.01] [.09] [.01] [.00] [.75] [.06] [.16] [.14] [.00] [.00] [.45] [.00] [.17]

CEFDt -1 d ME Age E D BE/ME EF/A GS BE/ME EF/A GS BE/ME EF/A GS SMB High-Low High-Low >0 <0 >0 =0 HML High-Low High-Low Low-Medium High-Medium High-Medium High-Medium Low-Medium Low-Medium 0.4 -0.6 0.8 0.7 -0.8 -0.2 0.2 0.1 -0.1 0.4 0.4 0.1 -0.2 -0.3 p(d) [.02] [.01] [.00] [.00] [.00] [.41] [.12] [.27] [.23] [.01] [.00] [.41] [.00] [.01]

TURN t -2 d -0.2 0.4 -0.6 -0.5 0.5 0.1 -0.1 -0.1 0.2 -0.3 -0.3 -0.1 0.2 0.2 p(d) [.26] [.09] [.04] [.03] [.03] [.58] [.43] [.59] [.20] [.03] [.02] [.43] [.01] [.03]

NIPO t -1 d -0.6 0.9 -1.0 -0.9 1.0 0.2 -0.3 -0.2 0.3 -0.5 -0.5 -0.1 0.2 0.3 p(d) [.00] [.00] [.00] [.00] [.00] [.18] [.01] [.09] [.01] [.00] [.00] [.26] [.00] [.00]

RIPO t -2 d -0.4 0.6 -0.6 -0.7 0.7 0.2 -0.2 -0.2 0.2 -0.4 -0.4 -0.1 0.2 0.2 p(d) [.05] [.00] [.02] [.01] [.00] [.15] [.06] [.05] [.17] [.00] [.00] [.43] [.02] [.05]

S t -1 d -0.4 0.7 -0.8 -0.7 0.8 0.2 -0.4 -0.2 0.3 -0.5 -0.4 -0.1 0.2 0.2 p(d) [.05] [.00] [.01] [.01] [.00] [.22] [.00] [.05] [.01] [.00] [.00] [.36] [.07] [.09]

P t -2 D-ND d 0.2 -0.3 0.6 0.6 -0.5 0.1 0.2 0.1 -0.1 0.4 0.3 0.1 -0.2 -0.2 p(d) [.21] [.20] [.08] [.03] [.09] [.60] [.15] [.39] [.34] [.01] [.03] [.09] [.02] [.08]

Sentiment Index d -0.5 0.8 -1.0 -0.9 0.9 0.2 -0.3 -0.2 0.3 -0.5 -0.5 -0.1 0.3 0.3 p(d) [.01] [.00] [.00] [.00] [.00] [.38] [.02] [.10] [.06] [.00] [.00] [.18] [.01] [.01]

Table 7. Conditional Market Betas. Regressions of long-short portfolio returns on the market risk premium (RMRF) and the market risk premium interacted with measures of sentiment (S) , each standardized to have unit variance.

R X it = high,t R Xit =low,t = c + dSENTIMENTt 1 + (e + fSENTIMENTt 1 )RMRFt + u t


The long-short portfolios are formed based on firm characteristics (X): firm size (ME), age, total risk (), profitability (E), dividends (D), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth decile (GS). High is defined as a firm in the top three NYSE deciles; low is defined as a firm in the bottom three NYSE deciles; medium is defined as a firm in the middle four NYSE deciles. Monthly returns are matched to the closed-end fund discount (CEFD), the number of IPOs (NIPO), and the equity share (S) for the calendar year one year before t. Monthly returns are matched to detrended log turnover (TURN), the average annual first-day return (RIPO), and the dividend premium (PD-ND ) for the calendar year two years before t. T-statistics are heteroskedasticity robust. CEFDt -1 f ME Age E D BE/ME EF/A GS BE/ME EF/A GS BE/ME EF/A GS
a b

TURN t -2 f -0.02 -0.13


b

NIPO t -1 f -0.03 0.01 -0.03 -0.06 0.05 -0.02 -0.06 -0.00 -0.19 -0.01 -0.02 -0.01 -0.05
a

RIPO t -2 f 0.02 -0.13 0.11


b b

S t -1 f -0.10
a

P t -2 D-ND t( f) [-2.6] [-0.3] [-0.7] [-0.8] [1.1] [0.8] [-4.1] [2.3] [0.7] [-3.0] [0.4] [0.3] [-3.3] [2.5] f 0.05 0.15
b

Sentiment Index f -0.04 -0.10


b

t( f) [-0.2] [-0.5] [0.2] [1.7] [-0.5] [-0.4] [0.9] [0.1] [-0.4] [0.2] [0.6] [-0.2] [1.9] [-0.8]

t( f) [-0.6] [-2.9] [1.6] [1.9] [-2.3] [2.9] [-3.0] [4.1] [1.3] [-2.5] [3.1] [3.1] [-3.3] [1.2]

t( f) [-0.8] [0.3] [-0.7] [-1.6] [1.2] [-0.5] [-1.7] [0.0] [-1.0] [-0.4] [-0.9] [-0.6] [-2.6] [1.3]

t( f) [0.5] [-2.7] [2.1] [1.0] [-1.8] [3.4] [-2.2] [4.5] [3.4] [-2.9] [3.2] [3.6] [-3.1] [1.3]

t( f) [1.4] [3.0] [-1.4] [-0.7] [1.4] [-3.8] [3.6] [-6.8] [-3.9] [3.4] [-3.5] [-4.1] [4.4] [-3.4]

t( f) [-1.1] [-2.1] [0.9] [-0.1] [-0.6] [2.8] [-3.7] [4.2] [1.9] [-3.0] [2.2] [2.6] [-4.3] [2.4]

SMB High-Low High-Low >0-<0 >0-<0 HML High-Low High-Low Low-Medium High-Medium High-Medium High-Medium Low-Medium Low-Medium

-0.01 -0.02 0.01 0.08 -0.02 -0.03 0.03 0.00 -0.07 0.01 0.01 -0.00 0.04 -0.01

-0.01 -0.03 -0.04 0.05 0.04 -0.16 0.05 -0.09


b b

0.08 0.09 -0.10b 0.12 -0.12 0.09 -0.07 0.07 0.07


b b b

-0.07 -0.04 0.06 -0.17 -0.14 -0.85 0.12 -0.09 -0.10


b a

0.05 -0.00 -0.03 0.14 -0.15 0.09 -0.10 0.06 0.06


b b b

0.05 -0.08 0.16 -0.08 0.10 0.55 -0.09 0.08 0.08


a b b b b b b a

0.16

b b b b b a

0.30
b b b a

0.15
b

0.44
b b b a

0.01 0.01 -0.07


a

-0.06

-0.04

0.10

-0.09

0.02

0.02

0.02

0.04a

-0.05a

0.04a

Statistically significant f that matches the sign of the return predictability from Tables 5 and 6. Statistically significant f that does not match the sign of the return predictability from Tables 5 and 6.

Table 8. Announcement Effects. For each calendar quarter, we form ten portfolios according to the NYSE breakpoints of firm size (ME), age, total risk, earnings-book ratio for profitable firms (E/BE), dividend-book ratio for dividend payers (D/BE), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth (GS). We also calculate average announcement effects for unprofitable firms and nonpayers. We then regress the average quarterly earnings announcement effects for each portfolio on lagged values of the sentiment index.

Rt = a + bSENTIMENTt 1 + u t
We report b . Quarterly average announcement effects are matched to the sentiment index for the calendar year one year before t. T-statistics are heteroskedasticity robust. Decile 0 ME Age E/BE D/BE BE/ME EF/A GS -0.38 [-3.26] -0.18 [-2.11] 1 -0.20 [-2.21] -0.06 [-0.49] -0.04 [-0.71] -0.24 [-2.22] -0.02 [-0.30] -0.08 [-1.21] -0.08 [-0.75] -0.29 [-2.68] 2 -0.06 [-0.97] -0.02 [-0.14] -0.04 [-0.88] 0.06 [0.66] -0.10 [-1.36] 0.01 [0.19] -0.06 [-0.73] -0.06 [-0.66] 3 -0.03 [-0.42] -0.09 [-1.22] -0.03 [-0.58] 0.14 [1.16] -0.06 [-0.77] 0.04 [0.61] 0.00 [-0.05] -0.12 [-1.51] 4 -0.08 [-0.92] -0.12 [-1.43] 0.00 [0.06] -0.17 [-1.73] 0.00 [0.04] -0.03 [-0.50] -0.06 [-0.84] -0.02 [-0.27] 5 -0.06 [-0.95] -0.17 [-2.12] -0.03 [-0.59] -0.17 [-1.88] -0.10 [-1.59] -0.03 [-0.43] 0.01 [0.15] -0.02 [-0.39] 6 -0.05 [-0.72] 0.01 [0.09] -0.01 [-0.18] -0.06 [-0.66] 0.00 [0.05] 0.02 [0.28] -0.07 [-1.02] 0.04 [0.58] 7 -0.01 [-0.13] 0.03 [0.56] -0.04 [-0.56] -0.03 [-0.35] -0.09 [-1.53] -0.13 [-1.74] -0.13 [-1.96] -0.05 [-0.79] 8 -0.05 [-0.79] -0.06 [-0.90] -0.01 [-0.20] 0.02 [0.33] -0.12 [-2.11] -0.01 [-0.07] -0.03 [-0.39] 0.01 [0.14] 9 0.00 [0.08] -0.03 [-0.48] -0.08 [-0.91] -0.02 [-0.31] 0.03 [0.45] -0.14 [-1.64] -0.14 [-1.99] -0.04 [-0.65] 10 -0.02 [-0.41] -0.16 [-3.10] -0.27 [-2.49] 0.13 [1.99] -0.05 [-0.65] -0.21 [-1.77] -0.04 [-0.48] -0.04 [-0.46]