Firm-Speifi Investor Sentiment David Aboody UCLA Anderson Graduate Shool of Management daboody@anderson.ula.edu Omri Even-Tov UCLA Anderson Graduate Shool of Management omri.even.tov.204@anderson.ula.edu Reuven Lehavy Ross Shool of Business University of Mihigan e-mail: rlehavy@umih.edu Brett Trueman UCLA Anderson Graduate Shool of Management brett.trueman@anderson.ula.edu Current draft: July 203 We thank Brad Barber, Avanidhar Subrahmanyam, and partiipants at the 203 London Business Shool Transatlanti Dotoral Conferene for their helpful omments. All remaining errors are our own.
Abstrat A signifiant body of researh is devoted to examining the effet of investor sentiment on the prie response to firm-level dislosures. Absent a firm-level measure of sentiment, these studies have used proxies of market-wide sentiment. In this paper we introdue a measure of firm-speifi investor sentiment and use it to gain insights into sentiment s effet on the prie response to earnings surprises. We develop a simple model whih predits that the greater the level of firm-speifi sentiment, the weaker the relation between announement return and earnings surprise and the more negative the market response to earnings that just meet expetations. Using our proxy for firm-speifi investor sentiment, we find strong empirial support for both of the model s key preditions. In addition, we show that our firm-speifi measure has greater explanatory power for announement returns than does the market-wide sentiment measure used in prior literature.
Firm-Speifi Investor Sentiment Introdution The effet of market sentiment on the ross-setional and time-series properties of stok returns is a topi of substantial researh interest. Among the proxies that have been used to measure market sentiment are NYSE share turnover, the losed-end mutual fund disount, the degree of underpriing in initial publi offerings, and the aggregate level of orporate investment. There is also a signifiant body of work studying how investor sentiment affets deision-making at the firm level and the prie response to firm-level dislosures. While this type of ross-setional researh naturally alls for the use of a measure of sentiment that is alulated at the firm level, there are no measures of this sort in the literature. As a onsequene, these studies utilize proxies of market-wide sentiment whih, although varying over time, are invariant in the ross-setion. In this paper we introdue a measure of firm-speifi investor sentiment and use it to gain new insights into sentiment s impat at the individual firm level. Baker and Wurgler (2006) disuss the potential weakness of using a market-wide measure of sentiment in firm-level studies. The main fous of their paper is on the onstrution of an index of market sentiment (hereafter the BW index), whih they use to study the timeseries properties of stok returns. However, in a final analysis, they test whether their marketwide sentiment index is useful for prediting the returns around individual firms earnings announements. While they find some evidene that it does, they aution against interpreting too muh into those results, aknowledging that measures of market-wide sentiment have limited power to detet sentiment at the firm level. They write that our [time-series return] results Papers inlude Arif and Lee (203), Baker and Wurgler (2006), Brown and Cliff (2004, 2005), Lee et al. (99), Lemmon and Portniaguina (2006), Ljungqvist et al. (2006), Neal and Wheatley (998), Qui and Welh (2004), Ritter (99), Stambaugh et al. (202), and Yu and Yuan (20).
are driven by the orrelated orretion of mispriing, but a firm s announement event return piks up the expetational orretions that our only to it alone, within its own announement window. Similar limitations are expressed by Brown et al. (202), who use the BW index to examine the effet of sentiment on a firm s deision whether to voluntarily release pro forma earnings statements. 2 Our measure of firm-speifi sentiment is based upon the notion that investor sentiment reflets either optimism or pessimism about a firm s value that is not justified by the available information. 3 It has as its underlying premise that investors who are influened by sentiment would be most likely to establish their speulative positions, and thereby influene stok pries, shortly before an earnings announement, sine that is when they expet their optimism or pessimism to be onfirmed. Establishing positions muh earlier would needlessly expose them to risks unrelated to the announement. Those investors who are optimisti would tend to purhase shares prior to the earnings announement, ausing pries to go up, while those who are pessimisti would sell shares, depressing pries. Refleting all this, our measure of firm-speifi investor sentiment is the market-adjusted return during the five trading days before the firm s earnings announement. We begin our analysis by developing a simple model of firm-level investor sentiment, whih generates theoretial preditions regarding the impat of sentiment on the stok reation to earnings surprises. Our model shows that the more positive (or less negative) the sentiment, the less sensitive will returns be to unexpeted earnings and the more negatively will the market 2 Other papers that examine firm-level issues using market-wide sentiment measures inlude Livnat and Petrovits (2009) and Mian and Sankaraguruswamy (202), who employ the BW index to study the effet of sentiment on the sensitivity of stok pries to firm-speifi earnings news, and Bergman and Royhowdhury (2008), who use the Mihigan Consumer Confidene Index to investigate the effet of investor sentiment on managerial foreasting behavior. Mikhail et al. (2009) employ the Mihigan Index of Consumer Expetations and Hribar and MInnis (202) use the BW Index to study the impat of sentiment on analysts foreast errors. 3 This is similar to the definition of Baker and Wurgler (2007), that sentiment is a belief about future ash flows and investment risks that is not justified by the fats at hand. 2
reat to an announement that earnings exatly met expetations. Intuitively, when investors have positive sentiment about a firm, the firm s pre-earnings announement stok prie will reflet an unjustifiably high earnings expetation. With expetations high, just meeting the prior earnings foreast will be looked upon negatively. Moreover, with the stok prie elevated, the marginal return to a one-ent hange in the earnings surprise will be muted. Conversely, when investors have negative sentiment, the pre-announement stok prie will reflet an earnings expetation that is unjustifiably low, and so meeting the prior foreast will be viewed positively. Also, with the stok prie depressed, the marginal return to a penny hange in the earnings surprise will be magnified. Our analyses are onduted on a sample of more than 400,000 quarterly earnings announements made over the years 973 through 200. We first rank eah quarter s announements aording to the five-day pre-announement return (umulated over days -6 through -2, where day 0 is the date of the earnings announement). We then partition the observations into deiles, with the highest (lowest) deile ontaining the stoks with the most positive (negative) pre-announement returns. Regressing the earnings announement return (umulated over days - through +) on unexpeted earnings, we find that both the estimated slope and interept deline with the sentiment deile, as our model predits. These results support our onjeture that the five-day pre-announement return serves as a measure of firmspeifi investor sentiment. We obtain similar results when we estimate the return-earnings surprise regression eah month over our sample period using a Fama-MaBeth approah. Results also remain substantially unhanged when we add size, book-to-market, and momentum as explanatory variables to ontrol for differenes in firm-speifi harateristis aross sentiment deiles, along with a dummy variable for firm-quarters with losses. 3
Baker and Wurgler (2006), Hribar and MInnis (202), Mian and Sankaraguruswamy (202), and Seybart and Yang (202) all onjeture that sentiment will have a greater impat on the returns of firms that are harder to value. The evidene they present is supportive of this onjeture. If our measure aptures firm-speifi investor sentiment, then we would expet to find a similar result. Speifially, the impat of the pre-announement return on the slope and interept of the return-earnings surprise relation should be greater for harder-to-value firms. To test this predition, we employ four different proxies, widely used in the literature, to measure the degree of diffiulty in assessing firm value: firm size (smaller firms are likely harder to value), age (younger firms are expeted to be more diffiult to value), earnings volatility, and return volatility (higher volatility is expeted to be assoiated with harder-to-value firms). For all four proxies, the interept on the return-earnings surprise regression is more sensitive to hanges in investor sentiment for harder-to-value firms. For three of the four proxies, the regression slope is also more sensitive for the harder-to-value firms. These results provide additional support for our onjeture that the five-day pre-announement return aptures firmspeifi investor sentiment. Reognizing that a theory annot determine the preise length of the pre-announement period that should be used for umulating returns, we test the robustness of our results by extending the window over whih sentiment is measured to the two months prior to the earnings announement. In re-estimating our main regression using this longer window, we end the return aumulation period six days before earnings are released. We do so in order to ensure that none of our results are driven by the short-term return reversal pattern doumented by Lehmann (990). In his paper, Lehmann (990) finds that a portfolio long (short) in stoks with a positive (negative) prior five-day return earns positive hedge returns over the next five days. While that 4
pattern annot explain the inverse relation we doument between investor sentiment and the slope of the return-earnings surprise regression, it ould partially ontribute to the doumented negative relation between sentiment and the regression interept. Using the longer window, we find that the estimated regression slope and interept remain dereasing funtions of investor sentiment. This is evidene that our results are robust to the length of the return aumulation window and that they are not attributable to short-term return reversals. Berkman et al. (2009) show that a portfolio long (short) in stoks with high (low) dispersion of opinion generally earns a signifiantly positive hedge return prior to earnings announements and a signifiantly negative return afterwards. As with short-term return reversals, this result annot explain the negative relation we find between investor sentiment and the slope of the return-earnings surprise regression. However, it ould potentially be a ontributing fator to the negative relation between investor sentiment and the regression interept. To test whether differenes of opinion around earnings announements drives that result, we re-estimate our return-earnings surprise regression, adding a proxy for the magnitude of these differenes. Inluding this ontrol variable, we again find that the estimated slope and interept are dereasing funtions of the level of investor sentiment, onsistent with our model s predition. Using our theoretial model and data, we also re-examine whether the BW measure has value in explaining the returns around earnings announements. The evidene here is mixed. Taking this analysis one step further, we find that our measure of firm-speifi investor sentiment has signifiant inremental explanatory power for earnings announement returns over the BW index. These results reinfore the importane of using a firm-speifi measure to study the effet of sentiment at the individual firm level. 5
The plan of this paper is as follows. In Setion I we develop a model of firm-speifi investor sentiment. Our empirial methodology and hypothesis development appear in Setion II. This is followed in Setion III by a desription of our sample. Our full-sample empirial results are presented in Setion IV. An analysis of whether sentiment has the greatest effet in hard-to-value firms appears in Setion V. We present the results of robustness tests in Setion VI. Setion VII provides a summary and onlusions. I. Model development In this setion we develop a model whih will serve as the basis of our empirial analyses. We onsider a one-period, three-date eonomy and two types of risky firms one that is traded solely by investors who are influened by sentiment (as desribed below) and the other that is traded exlusively by investors unaffeted by sentiment. In all other respets the two types of firms are indistinguishable. All investors in the eonomy are risk neutral and the oneperiod disount rate is assumed equal to zero. The fration of firms in the eonomy that are traded by investors influened by sentiment is denoted by p. Eah firm s earnings for the period, 2 X, are normally distributed with a mean of EX ( ) > 0 and a standard deviation of σ X, and are assumed equal to the period s ash flows. At the beginning of the period, date 0, the prie of eah firm, P 0, is equal to the risk neutral investors prior expetation of earnings, EX ( ). At the end of the period, date 2, eah firm s earnings are formally announed and all firms are liquidated. Firm prie at that time, P 2, is therefore equal to X. With probability q the value of the realized earnings leaks out at date, just before the formal announement. In that ase, the date prie of the firm, P, would also equal X. The return at date, R, would be given by: 6
X EX ( ) R = () P 0 and the date 2 return, R 2, would equal zero. The returns at dates and 2 would be unrelated when there is information leakage. Note that the right hand side of () is just the firm s unexpeted earnings (normalized by prie). This quantity will sometimes be denoted by UE in our analysis below. With probability q there is no leakage of information at date. In that ase, investors who are influened by sentiment reeive in its plae a ompletely noisy signal of earnings, denoted by B, whih they inorretly believe provides information about the realized value of X. (The investors who are not influened by sentiment reognize that this signal is worthless.) These investors onjeture that B= X + ε B, where ε B is normally distributed with a mean of zero 2 and a finite standard deviation of σ B. Given their belief, the date prie of a firm traded by these investors equals we( X ) + ( w) B, where firm is equal to: σ w = σ 2 B 2 2 X + σb. The return at date for suh a we( X ) + ( w) B R = (2) EX ( ) The realized value of the signal, B, determines whether the investors are optimisti or pessimisti. If B is greater than the initial earnings expetation, EX ( ), then investors are optimisti and the return at date will be positive. If B is less than EX ( ), then they are pessimisti and the return will be negative. Of ourse, B is not observable to the researher. However, the date return, R, is observable and fully aptures the information in B. It is our measure of firm-speifi sentiment. Values of R greater than zero imply investor optimism. 7
The more positive is R, the greater the level of optimism and the greater will be the positive deviation of prie from underlying firm value. Values of R less than zero imply investor pessimism. More negative values reflet greater levels of pessimism and a greater negative deviation of prie from underlying firm value. The date 2 return for a firm traded by investors influened by sentiment is: R 2 = X [ we( X ) + ( w) B] [ we( X ) + ( w) B] (3) From equations (2) and (3), we an express R 2 as a funtion of R and UE as follows: R 2 = UE + R + + R (4) The final possible senario in this eonomy is one in whih there is no leakage of information and the investors trading in the firm are unaffeted by sentiment. In this ase, the prie of the firm at date is simply equal to EX ( ) and the date return is zero. The date 2 return is equal to unexpeted earnings, UE. As in the ase of information leakage, the returns at dates and 2 would be unrelated. Calulating the expeted return at date 2, ER ( 2), over all possible senarios in this eonomy yields: UE E( R2 ) = q p + + ( p) UE + ( q) * 0, + R + R (5) whih an be rewritten as: UE E( R2 ) = qp + q( p) UE + qp + qp, + Sent + Sent (6) 8
where we use Sent in plae of R in (6) to emphasize that this is our measure of firm-speifi sentiment. II. Empirial methodology and hypothesis development The empirial analog to equation (6) is: R = α + β UE + β + β UE + ε, 2 3 + D ( Sent ) + D ( Sent ) (7) where: R = the umulative market-adjusted return for firm j (omputed using the CRSP valueweighted market index) in the three days surrounding the announement of quarter t earnings (trading days -, 0, and, where day 0 denotes the date of the earnings announement); 4 UE X X 4 =, where P 4 X is earnings before extraordinary items for firm j in quarter t and P is the share prie of firm j at the end of quarter t-4; 4 Sent = the umulative market-adjusted return for firm j over the period from six trading days before to two trading days before the announement of quarter t earnings (sometimes referred to as the pre-announement return); D ( Sent ) = the deile rank of Sent, where the rank is determined relative to the values of Sent it for all firms, i, that announe in the same alendar quarter (higher deiles reflet greater optimism/less pessimism); and ε = the regression residual for firm j and quarter t. 4 Our results are robust to the use of raw returns, rather than market-adjusted returns. 9
We use a three-day earnings announement window in order to mitigate the effet of possible database errors with respet to the exat date of the earnings announement as well as to aount for unertainty over whether earnings were announed after hours. The earnings announement date used for our analysis is that reported in Compustat, unless the announement date also appears in the I/B/E/S database and the I/B/E/S date is the earlier of the two. In that ase, the I/B/E/S date is used instead. If I/B/E/S also speifies the exat time of the announement and it is after trading hours, then the earnings announement date is advaned by one trading day. We use a firm s four-quarters ago earnings, rather than the onsensus analyst foreast, as our expetation of urrent-quarter earnings in order to be able to retain in our sample those firms without analyst overage. Reognizing that the ross-setional distribution of our sentiment measure hanges over time, we use the sentiment deile in our regression, rather than the value of the measure, itself. This allows us to maintain omparability aross our sample period and is onsistent with Baker and Wurgler (2006), Hribar and MInnis (202), and Seybart and Yang (202). 5 We use a relatively short period of five trading days prior to the earnings announement window to measure sentiment beause individuals who believe they have private information about upoming earnings are more likely to trade on their information lose in time to the earnings announement. Establishing positions earlier would subjet the traders to a greater level of risk unrelated to the forthoming announement. 6 Our formal hypothesis follows diretly from expression (7): 5 Using the atual value of sentiment, rather than the sentiment deile, does not qualitatively affet our main results. 6 Using similar reasoning, Berkman et al. (2009) hypothesize that firms subjet to high dispersion of opinion should experiene a share prie run-up in the few days before their earnings announements. Their empirial findings support this onjeture. As we report later, our main results are qualitatively unhanged when we ontrol for dispersion of opinion. 0
Hypothesis: In a regression of earnings announement return on earnings surprise, the interept and slope will be dereasing funtions of firm-speifi investor sentiment. That is, the oeffiients β 2 and β 3 in expression (7) will be positive. 7 III. Data seletion and desriptive statistis Our initial sample is the set of quarterly earnings announements ontained in the merged CRSP-Compustat quarterly file for the period from January 973 through Deember 200. We end our sample period in Deember 200 beause some of our tests involve the BW index, for whih data is provided only through the end of that year. 8 From this sample we remove those announements for whih either the announement date is not available or is more than 90 days after quarter-end. We also delete any announement of quarter t earnings by firm j if one or more of the following variables are missing from the CRSP-Compustat dataset: () the earnings of firm j for quarter t-4, (2) the stok prie of firm j at the end of quarter t-4, (3) firm j s market apitalization at the end of quarter t, (4) monthly returns for firm j over the period from 2 months before the end of quarter t through two months before the end of that quarter (this data is used to alulate return volatility, whih is defined as the standard deviation of monthly stok returns over those months, and is also used to alulate stok prie momentum), (5) firm j s book value of equity at the end of quarter t, and (6) any day s return during the period from six trading days before to one trading day after firm j s announement of quarter t s earnings. We also remove announements for whih (a) the announement date of quarter t earnings reorded in Compustat and the date reorded in I/B/E/S differ by more than five trading days, (b) firm j s prie per share at the end of quarter t is less than $, or () firm j s market value 7 Sine our measure of sentiment appears in the denominator of the independent variables in (7), an inverse relation between sentiment and the slope and interept of the return-earnings surprise relation requires that the two regression oeffiients be positive. 8 Our main results are robust to extending our sample period to the end of 202.
at the end of quarter t is less than $ million. Finally, for eah alendar quarter we trunate those observations that are at the top and bottom one perent of the unexpeted earnings distribution, in order to lessen the impat of outliers. After removing all announements that meet any of these riteria, we are left with a final sample of 425,70 observations. A few of our regressions inlude earnings volatility as an independent variable. We define earnings volatility as the standard deviation of the ratio of quarterly operating inome before depreiation to average total assets, alulated over the prior 20 quarters. For these tests we drop any earnings announement for whih this operating inome-to-asset ratio is not available for at least eight of the prior 20 quarters. This redues the sample size in these tests to 332,033 observations. Desriptive statistis for our variables are presented in Table I, panel A. The average umulative market-adjusted return over the three-day earnings announement window (days -, 0, and ) is 0.22 perent, while the average value of our firm sentiment measure, the umulative market-adjusted return over the pre-announement period (days -6 through -2), is 0.28 perent. These positive average returns likely reflet the earnings announement premium, whih has been doumented by Ball and Kothari (99), Cohen et al. (2007), Frazzini and Lamont (2007), and Barber et al. (202), among others. The mean (median) market value of the firms in our earnings announement sample is $2. billion ($24 million), while the mean (median) bookto-market ratio is 0.99 (0.63). 9 These numbers ompare to a mean (median) end-of-quarter market value for all New York Stok Exhange firms over our sample period of $3.3 billion ($490 million) and a mean (median) book-to-market ratio of 2.06 (0.68). 0 The average age of the firms in our sample (where age is defined as the number of years a firm s shares have been 9 For firms with negative book value, we set the book-to-market ratio equal to zero. 0 As we demonstrate by adding ontrols for firm harateristis to our regression, our results are not driven by the speifi size and book-to-market attributes of our sample. 2
publily trading) is just under 7 years. Unexpeted earnings has a mean of 0.34 perent of stok prie, refleting growth in earnings, on average, over our sample period. Return volatility has a mean of 2.4 perent, while earnings volatility averages 3.8 perent. The pairwise Pearson orrelation oeffiients for the variables of interest are presented in Table I, panel B. As expeted, age and firm size are strongly positively orrelated (orrelation = 0.225). Earnings volatility and return volatility are also signifiantly positively orrelated, as expeted; however, the magnitude of the orrelation, 0.005, is low. The orrelation between the market-adjusted return during the earnings announement window and the orresponding return during the pre-announement period is a signifiantly negative 0.087. The negative orrelation is onsistent with the notion that the announement of a firm s earnings serves to partially orret the misvaluations of investors who are influened by sentiment. IV. Full-sample results We begin our empirial analysis by estimating equation (7). Results are presented in olumn of Table II. Consistent with our hypothesis, and the onjeture that pre-earnings announement returns are a measure of firm-speifi investor sentiment, the oeffiients on both + D ( Sent ) and UE + D ( Sent ) in (7) are positive and signifiant. Alternatively stated, the sensitivity of announement returns to earnings surprises is dereasing in investor sentiment as is the expeted stok prie reation to an announement that earnings just met expetations. To ensure that our results are not driven by differenes in firm harateristis aross sentiment deiles, we add ln(size), ln(book-to-market ratio), momentum, and a dummy variable Cohen et al (2007) show that unexpeted delays in earnings announements are aompanied by negative returns in the pre-announement period. To determine whether this phenomenon has an impat on our results, we repeat our analysis, exluding any firm-quarter observation for whih the earnings announement date falls more than seven days after the announement date in the orresponding quarter of the previous year. Untabulated results are similar to those presented in Table II. 3
equal to one for firm-quarters with losses as explanatory variables to (7). 2 As reported in olumn 2 of Table II, the oeffiients on our independent variables, + D ( Sent ) and UE, remain signifiantly greater than zero and are of similar magnitudes to the + D ( Sent ) values reported in olumn of the table. From this we onlude that our findings are not due to differenes in firm harateristis aross the sentiment deiles. In order to redue the potential impat of influential observations of we re-estimate (7) using the rank of UE, in plae of UE : UE on our results, R = α + β Rank( UE ) + β + β Rank( UE ) + ε. 2 3 + D ( Sent ) + D ( Sent ) (8) We do not replae the variable + D ( Sent ) by the rank of that variable sine we are already using the deile rank of Sent, rather than Sent, itself, in the regression. The results of estimating this rank regression appear in olumn 3 of Table II. The oeffiients on + D ( Sent ) and Rank( UE ) + D ( Sent ) are signifiantly positive, onfirming that influential observations are not driving our results. We also employ a Fama-MaBeth approah and estimate regression (8) for eah month of our sample period. 3 The averages of the monthly oeffiients on + D ( Sent ) and Rank( UE ) + D ( Sent ) are reported in olumn 4 of Table II. Both are reliably positive. 2 For firms with negative book value, we set ln(book-to-market ratio) equal to zero. 3 We exlude from our regressions the 2 months at the beginning of our sample period that have fewer than 00 observations. 4
Figure I, panel A (panel B) plots the estimated value of β 2 ( β 3 ) for eah month of our sample period. As is lear from the figure, the months with positive oeffiients are not onentrated within any narrow time frame. Rather, they are spread out over our entire sample period. The oeffiient, β 2, is positive in 9. perent of the months, while β 3 is positive for 53.7 perent of the months. 4 We alternatively alulate unexpeted earnings using as our expetation the onsensus analyst foreast, omputed as the average of all valid individual foreasts in I/B/E/S that are outstanding seven days before the earnings announement. 5 Unexpeted earnings in this ase is equal to the I/B/E/S reported earnings minus the onsensus analyst foreast. (We normalize eah earnings surprise by the stok prie at the end of the prior quarter.) As before, we trunate the highest and lowest one perent of the unexpeted earnings observations. The requirement that I/B/E/S onsensus foreasts be available, along with the trunation of the top and bottom one perent of the unexpeted earnings observations, redues our sample to slightly more than 220,000 announements. The results of re-estimating regression (8) using this alternative alulation of unexpeted earnings are presented in panel A of Table III. Both the oeffiient on + D ( Sent ) and the oeffiient on Rank( UE ) + D ( Sent ) remain signifiantly greater than zero. 4 When estimating panel regressions it is ommon pratie to adjust the standard errors for orrelation over time and aross firms (usually referred as double lustering). This proedure is used beause market-wide shoks may indue orrelation between firms at a given point in time, and persistent firm-speifi shoks may indue orrelation aross time for a given firm. In addition, researhers use the White orretion to aount for potential rosssetional heteroskedastiity (see Thompson, 20). Our researh design, though, eliminates the need to adjust our standard errors for auto-orrelation and heteroskedastiity. Our firm-speifi sentiment measure is the pre-earnings announement return; as suh, the errors should not be orrelated over time. Moreover, we use the rank of the sentiment measure; hene, market-wide shoks, while affeting all firms, should not affet their deile ranking. This allays onerns over the orrelation between firms at a given point in time and heteroskedastiity. Nevertheless, we repliated our analysis using both double lustering (over time and aross firms) and employing the White orretion and obtained qualitatively similar results. 5 A valid foreast is one that is issued no more than 90 days prior to the earnings announement. 5
The I/B/E/S sample also gives us a onvenient alternative means of testing the hypothesis that the greater the level of firm-speifi sentiment, the less positive (or more negative) the stok prie reation to an announement of earnings that just meets expetations (UE=0). Using the subsample of observations for whih realized earnings equal the onsensus foreast, we regress the market-adjusted announement return on the announement s sentiment deile. As we report in Table III, panel B, the oeffiient on D ( Sent ) in this regression is negative and signifiant, onsistent with our onjeture. While not the main fous of their paper, Baker and Wurgler (2006) examine whether their market-wide sentiment index is useful for prediting the returns around individual firms earnings announements. Although they find some evidene that it does, they aknowledge that their methodology has limited power in this setting sine their index is not expliitly designed to detet sentiment at the firm level. We re-examine the effetiveness of their measure at the firm level using our theoretial model and our data, by substituting the BW index for our measure of sentiment in expression (7): R = α + β UE + β + β UE + ε, 2 3 + BWSentt + BWSentt (9) where BWSent t is the value of the BW index during the last month of quarter t. As reported in Table IV, olumn, the regression results are mixed. Consistent with the hypothesis, the oeffiient, β 2, on + BWSentt is signifiantly positive. However, ontrary to the hypothesis, the oeffiient, β 3, on UE + BWSent t is insignifiantly different from zero. We take this analysis one step further and test whether our measure of firm-speifi investor sentiment has inremental explanatory power for earnings announement returns over 6
the BW index. We do so by adding the two independent variables, + BWSentt and UE + BWSent t, to regression (7): R = α + βue + β + β UE ( ) ( ) 2 3 + D Sent + D Sent + β + β UE + ε. + + BWSent 2 3 BWSentt t (0) Coeffiient estimates from this regression are reported in Table IV, olumn 2. The oeffiients on + D ( Sent ) and UE + D ( Sent ) remain positive and signifiant, providing evidene that our firm-speifi sentiment measure has inremental explanatory power for announement window returns over the BW index. As before, the oeffiient β 2 is signifiantly positive, while the oeffiient β 3 is insignifiantly different from zero. These results reinfore the importane of using a firm-speifi measure to study the effet of sentiment at the individual firm level. V. The impat of sentiment on hard-to-value firms Baker and Wurgler (2006), Hribar and MInnis (202), Mian and Sankaraguruswamy (202), and Seybart and Yang (202) all onjeture that sentiment will have a greater effet on returns for subsets of firms that are harder to objetively value. The empirial evidene they present is onsistent with this onjeture. If our measure aptures firm-speifi investor sentiment, then we should find a similar result. Speifially, the impat of the pre-announement return on the slope and interept of the return-earnings surprise relation should be greater for harder-to-value firms. 7
To test this we estimate the following regression: R = β0 + βue + β2 + β3ue + D ( Sent ) + D ( Sent ) () + γ D ( HTV ) + γ D ( HTV ) + γ D ( HTV ) UE + ε, 2 3 + D ( Sent ) + D ( Sent ) where: HTV = the value of the proxy (see below) used to measure the diffiulty of valuing firm j in quarter t and D ( HTV ) = the deile rank of HTV, where the rank is determined relative to the values of HTV it for all firms, i, that announe in the same alendar quarter. 6 If the market-adjusted return over days -6 through -2 is, indeed, a measure of firm-level investor sentiment, then the oeffiients γ 2 and γ 3 in expression () should be positive. We proxy for the extent to whih a firm is diffiult to value by four different variables: firm size (smaller firms are likely harder to value), age (it is expeted that younger firms will be more diffiult to value), earnings volatility, and return volatility (higher volatility is likely assoiated with harder-to-value firms). 7 We expet that firm-speifi investor sentiment will have the greatest impat on the return-earnings surprise relation for the smallest and youngest firms, and for those with the greatest earnings volatility and return volatility. Table V reports the results of estimating () over our entire sample period for eah hardto-value proxy. The oeffiient γ 2 is signifiantly positive for all the proxies age (olumn ), size (olumn 2), earnings volatility (olumn 3), and return volatility (olumn 4). The oeffiient γ 3 is positive and signifiant for three of the four proxies (all but return volatility, where it is 6 The deiles are onstruted suh that firms in higher deiles are more diffiult to value. 7 These proxies are also used by Baker and Wurgler (2006), Hribar and MInnis (202), Mian and Sankaraguruswamy (202), and Seybert and Yang (202). 8
negative). These findings provide support for the hypothesis that the interept and slope of the return-earnings surprise relation are more sensitive to firm-speifi investor sentiment for firms that are harder to value. As suh, the results are also supportive of our onjeture that the fiveday pre-announement return aptures firm-speifi investor sentiment. VI. Robustness tests Reognizing that theory annot determine the preise length of the pre-announement period that should be used to measure sentiment, we test the robustness of our results by extending the return measurement window to the two months prior to the earnings announement. In re-estimating our main regression we exlude from this return aumulation period the returns over days -6 through -2. We do so in order to ensure that none of our results are driven by the short-term return reversal pattern doumented by Lehmann (990). In his paper, Lehmann (990) finds that stoks whose pries inrease (derease) over the previous five days tend to deline (rise) over the next five. While this pattern annot aount for the inverse relation we find between investor sentiment and the sensitivity of announement returns to earnings surprises, it ould potentially be a ontributing fator to the negative relation between investor sentiment and the interept of the return-earnings surprise regression. We estimate the following regression using the longer return aumulation window: R = α + βue + β + β UE + β Sent + ε, 2 3 4 + D ( Sent ) + D ( Sent ) (2) where Sent is the umulative market-adjusted return for firm j over the period from 60 trading days before to seven trading days before the announement of quarter t earnings and D ( Sent ) is the deile rank of Sent, where the rank is determined relative to the values of 9 Sent it for all firms, i, that announe in the same alendar quarter. As a ontrol variable, we also inlude in
the regression the umulative market-adjusted return over days -6 through -2. Results of estimating (2) are reported in Table VI, panel A. Although smaller than their ounterparts in Table II, olumn, both β 2 and β 3 are reliably greater than zero. This is evidene that our results are robust to the length of the pre-announement return aumulation window and that they are not attributable to short-term return reversals. That the umulative market-adjusted return over days -60 through -7 reflets firmspeifi investor sentiment invites the question of whether our five-day sentiment measure provides inremental explanatory power over this longer-window measure. We expet that it will have additional explanatory power, given that investors who are influened by sentiment are arguably more likely to establish their speulative positions shortly before an earnings announement. To test this onjeture, we add our measure as a main effet to regression (2) and also interat it with the longer-term sentiment measure: R = α + βue + β + β UE + β ( ) ( ) ( ) 2 3 4 + D Sent + D Sent + D Sent + β x + β UE x + ε. ( ) ( ) ( ) ( ) 5 6 + D Sent + D Sent + D Sent + D Sent (3) Results of estimating (3) are presented in Table VI, panel B. As expeted, the oeffiient on the five-day measure of firm-speifi investor sentiment, β 4, is signifiantly greater than zero, as are the oeffiients on the interation terms between the short-window and longer-window measures, β 5 and β 6. Moreover, the oeffiient on + D ( Sent ), β 2, is no longer positive, onsistent with the market-adjusted return over days -6 through -2 being the more powerful of the two sentiment measures. 20
We next test whether differenes of opinion ould be driving any of our results. In the presene of differenes of opinion and short-sales onstraints, Miller (977) onjetures that stok pries will be positively biased; investors with favorable information will bid pries up, while those with unfavorable information will be onstrained from fully realigning pries to fundamental value. Berkman et al. (2009) posit that this bias will be stronger just before earnings announements (when investors trade on their private information), but will dissipate one earnings are announed and differenes of opinion are redued. Consistent with their onjeture, they find that a hedge portfolio long in stoks with high dispersion of opinion and short in stoks with low dispersion generally earns a signifiantly positive return during the two weeks before the earnings announement and a signifiantly negative return during the subsequent two weeks. As with short-term return reversals, this phenomenon annot aount for the negative relation we find between investor sentiment and the sensitivity of returns to earnings surprises. However, it, too, ould partially ontribute to the doumented negative relation between sentiment and the interept of the return-earnings surprise regression. To test whether differenes of opinion is driving the interept result, we re-estimate (7), adding an explanatory variable to apture the degree to whih there are differenes of opinion prior to an earnings announement: R = α + β UE + β + β UE + β D ( Diff ) + ε, 2 3 4 + D ( Sent ) + D ( Sent ) (4) where Diff is our proxy (see below) for differenes of opinion prior to firm j s announement of quarter t earnings and D ( Diff ) is the deile rank of Diff, where the rank is determined relative to the values of Diff it, for all firms, i, that announe earnings in alendar quarter (higher deiles reflet greater differenes of opinion). We employ two separate proxies, 2
earnings volatility and return volatility. Regression results are reported in Table VI, panel C. Consistent with Berkman et al. (2009), the oeffiient on the earnings volatility variable (olumn ) and on the return volatility variable (olumn 2) are signifiantly negative. More importantly for our purposes, the oeffiients, β 2 and β 3, remain signifiantly greater than zero and are of a magnitude similar to those estimated for regression (7), when no differene of opinion proxy is inluded. We infer from this that our results are not driven by differenes of opinion. VII. Summary and onlusions This paper tests the onjeture that firm-speifi investor sentiment an be measured by a firm s pre-earnings announement return. To guide our empirial analysis, we develop a simple theoretial model to show how firm-level investor sentiment is expeted to impat the relation between earnings announement returns and unexpeted earnings. The model leads to the preditions that the return-earnings surprise relation will be weaker the greater the level of investor sentiment and that the announement return in ases where earnings just meet expetations will be more negative the greater the level of investor sentiment. These preditions are a onsequene of the role that earnings play in mitigating investors misvaluation of the firms in whih they invest. Our empirial analysis provides support for eah of these preditions. We also onjeture that the impat of investor sentiment on the return-earnings surprise relation will be greater for harder-to-value firms than for those that are easier to value. Our empirial results are supportive of this predition as well. Market-wide measures of sentiment have been used in the past to examine ross-setional and time-series properties of stok returns. Absent a firm-speifi measure of sentiment, researhers have also employed these market-wide measures to study the effet of sentiment on 22
deisions and pries at the individual firm level. By developing a firm-level measure of sentiment, we provide a tool that an be used in future researh on this topi. 23
Referenes Arif, Salman, and Charles Lee, 203, Does aggregate investment reflet investor sentiment?, Indiana University working paper. Baker, Malolm, and Jeffrey Wurgler, 2006, Investor sentiment and the ross-setion of stok returns, Journal of Finane, 6(4), 645 680. Baker, Malolm, and Jeffrey Wurgler, 2007, Investor sentiment in the stok market, Journal of Eonomi Perspetives, 2(2), 29 5. Ball, Ray, and S.P. Kothari, 99, Seurity returns around earnings announements, The Aounting Review, 66(4), 78 738. Barber, Brad, Emmanuel De George, Reuven Lehavy, and Brett Trueman, 203, The earnings announement premium around the globe, Journal of Finanial Eonomis, 08(), 8-38. Bergman, Nittai, and Sugata Royhowdhury, 2008, Investor sentiment and orporate dislosure, Journal of Aounting Researh, 46(5), 057 083. Berkman, Henk, Valentin Dimitrov, Prem Jain, Paul Koh, and Sheri Tie, 2009, Sell on the news: Differenes of opinion, short-sales onstraints, and returns around earnings announements, Journal of Finanial Eonomis, 92(3), 376-399. Brown, Gregory, and Mihael Cliff, 2004, Investor sentiment and the near-term stok market, Journal of Empirial Finane, (), -27. Brown, Gregory, and Mihael Cliff, 2005, Investor sentiment and asset valuation, Journal of Business, 78(2), 405-440. Brown, Nerissa, Theodore Christensen, W. Brooke Elliott, and Rihard Mergenthaler, 202, Investor sentiment and pro forma earnings dislosures, Journal of Aounting Researh, 50(), -40. Cohen, Daniel, Aiyesha Dey, Thomas Lys, and Shyam Sunder, 2007, Earnings announement premia and the limits to arbitrage. Journal of Aounting and Eonomis, 43(2-3), 53-80. Frazzini, Andrea, and Owen Lamont, 2007, The earnings announement premium and trading volume, National Bureau of Eonomi Researh working paper. Hribar, Paul, and John MInnis, 202, Investor sentiment and analysts earnings foreast errors, Management Siene, 58(2), 293 307. Lee, Charles, Andrei Shleifer, and Rihard H. Thaler, 99, Investor sentiment and the losedend fund puzzle, Journal of Finane 46(), 75 09. 24
Lehmann, Brue, 990, Fads, martingales, and market effiieny, Quarterly Journal of Eonomis, 05(), -28. Lemmon, Mihael, and Evgenia Portniaguina, 2006, Consumer onfidene and asset pries: some empirial evidene, Review of Finanial Studies, 9(4), 499-529. Livnat, Joshua, and Christine Petrovits, 2009, Investor sentiment, post-earnings announement drift, and aruals, New York University working paper. Ljungqvist, Alexander, Vikram Nanda, and Rajdeep Singh, 2006, Hot markets, investor sentiment, and IPO priing, Journal of Business, 79(4), 667-702. Mian, G. Muaba, and Srinivasan Sankaraguruswamy, 202, Investor sentiment and stok market response to earnings news, The Aounting Review, 87(4), 357-384. Mikhail, Mihael, Beverly Walther, and Rihard Willis, 2009, Does investor sentiment affet sell-side analysts foreast auray?, Arizona State University working paper. Miller, Edward, 977, Risk, unertainty, and divergene of opinion, Journal of Finane, 32(4), 5 68. Neal, Robert, and Simon Wheatley, 998, Do measures of sentiment predit returns?, Journal of Finanial and Quantitative Analysis, 33(4), 523-535. Qiu, Lily, and Ivo Welh, 2004, Investor sentiment measures, NBER Working Paper No. 0794. Ritter, Jay, 99, The long-run performane of initial publi offerings, Journal of Finane, 46(), 3-27. Seybart, Niholas, and Holly Yang, 202, The Party s Over: The role of earnings guidane in resolving sentiment-driven overvaluation, Management Siene, 58(2), 308-39. Stambaugh, Robert, Jianfeng Yu, and Yu Yuan, 202, The short of it: investor sentiment and anomalies, Journal of Finanial Eonomis, 04(2), 288-302. Thompson, Samuel, 20, Simple formulas for standard errors that luster by both firm and time, Journal of Finanial Eonomis, 99(), -0. Yu, Jianfeng, and Yu Yuan, 20, Investor sentiment and the mean-variane relation, Journal of Finanial Eonomis, 00(2), 367-38. 25
Figure I Monthly oeffiients from regressions of earnings announement return on firm-speifi sentiment This figure reports the β (panel A) and β (panel B) oeffiients for monthly regressions of 2 3 R = α + βue + β + β UE + ε, where 2 3 + D ( Sent ) + D ( Sent ) Panel A: Monthly oeffiients on 0.3 0.25 0.2 0.5 0. 0.05 0-0.05-0. De-73 De-74 De-75 De-76 De-77 De-78 De-79 De-80 De-8 De-82 De-83 De-84 De-85 De-86 De-87 De-88 De-89 De-90 De-9 De-92 De-93 De-94 De-95 De-96 De-97 De-98 De-99 De-00 De-0 De-02 De-03 De-04 De-05 De-06 De-07 De-08 De-09 De-0 Panel B: Monthly oeffiients on 6 + D( Sent) R is the umulative marketadjusted return (omputed using the CRSP value-weighted market index) over the three-day window (days -, 0, and ) surrounding the announement of quarter t earnings for firm j; UE is the differene between firm j's earnings before extraordinary items for quarter t (Compustat item EPSPXQ) and for quarter t-4, normalized by the share prie of firm j at the end of quarter t-4, and then multiplied by 00 (UE is trunated at the top and bottom one perent of its distribution); Sent is the umulative market-adjusted return for firm j over the period from six trading days before to two trading days before the announement of quarter t earnings; D ( Sent ) is the deile rank of Sent, where the rank is determined relative to the values of Sent for all firms, i, that announe in alendar quarter. Our sample period spans the years it 973 through 200. UE + D( Sent) 4 2 0-2 -4-6 De-73 De-74 De-75 De-76 De-77 De-78 De-79 De-80 De-8 De-82 De-83 De-84 De-85 De-86 De-87 De-88 De-89 De-90 De-9 De-92 De-93 De-94 De-95 De-96 De-97 De-98 De-99 De-00 De-0 De-02 De-03 De-04 De-05 De-06 De-07 De-08 De-09 De-0
Table I Desriptive statistis and orrelation oeffiients This table provides sample desriptive statistis for our variables (panel A) and the average of the quarterly pairwise Pearson orrelation oeffiients (panel B). For a given quarter, t, earnings announement return is equal to the umulative market-adjusted return (using the CRSP value-weighted market index) over the three-day window (days -, 0, and ) surrounding the announement of quarter t earnings. Sentiment is defined as the umulative market-adjusted return over the pre-announement period (days -6 through -2), prior to the announement of quarter t earnings. Market value is the end-of-quarter t market value of equity. Book-to-market is the end-ofquarter t book value of equity divided by the end-of-quarter t market value of equity (we set this variable to zero for firms with negative book value). Age is defined as the number of years a firm s shares have been publily trading, as of the end of quarter t. Unexpeted earnings is omputed as the differene between earnings before extraordinary items for quarter t (Compustat item EPSPXQ) and for quarter t-4, normalized by the share prie at the end of quarter t-4, and then multiplied by 00. Unexpeted earnings is trunated at the top and bottom one perent of its distribution. Return volatility is omputed as the standard deviation of monthly stok returns over months t-2 through t-2. Earnings volatility is defined as the standard deviation of the ratio of quarterly operating inome before depreiation to average total assets, alulated over the 20 quarters prior to quarter t (with a minimum of 8 out of 20 quarters required for omputation). Our sample period spans the years 973 through 200. In Panel B, numbers in bold are signifiant at the 0% level or better. Panel A: Desriptive statistis Variable No. of observations Mean Median Std. Dev. 25th perentile 75th perentile Earnings announement return 425,70 0.22% 0.000 0.082-0.032 0.033 Sentiment 425,70 0.28% -0.00 0.070-0.029 0.028 Market value 425,70 2,05 24,238 55 903 Book-to-market 425,70 0.987 0.632 7.057 0.37.02 Age 425,70 6.832 2.66 4.98 6.333 2.750 Unexpeted earnings 425,70 0.335 0.86 3.97-0.577 0.923 Return volatility 425,70 0.24 0.05 0.086 0.072 0.5 Earnings volatility 332,033 0.038 0.04 2.788 0.007 0.025 Panel B: Pairwise Pearson orrelation oeffiients Earnings announement return Sentiment Market value Book-tomarket Age Unexpeted earnings Return volatility Sentiment -0.0868 Market value -0.0023-0.0042 Book-to-market 0.0028 0.0030-0.043 Age -0.0004-0.053 0.225-0.0075 Unexpeted earnings 0.397 0.0494-0.0049-0.0036-0.06 Return volatility -0.024 0.045-0.0792-0.0042-0.270 0.0720 Earnings volatility 0.0006-0.003-0.007-0.0009-0.0072 0.007 0.0048