Corporate Disclosure,Analyst Forecast Dispersion,and Stock Returns

Abstract

This article examines whether a corporate disclosure practice is one of the reasons for the forecast dispersion anomaly—the negative relation between analyst forecast dispersion and future stock returns. Prior studies have shown that firms tend to delay the disclosure of bad news and that withholding of news leads to greater dispersion in analysts’ forecasts. Accordingly, we predict that firms with higher dispersion in analysts’ earnings forecasts are more likely to experience poor earnings in the subsequent quarter, and find evidence consistent with this prediction. After controlling for the relation between forecast dispersion and future earnings, we find that forecast dispersion is no longer significantly negatively related to future stock returns. These results suggest that temporary withholding of bad news by firms increases forecast dispersion among analysts and leads to low subsequent stock returns.

Keywords

corporate disclosure analyst forecast dispersion

Introduction

Diether, Malloy, and Scherbina (2002, hereafter DMS) find an inverse cross-sectional relation between the dispersion of analysts’ earnings forecasts and future stock returns. They show that stocks ranked in the highest quintile of forecast dispersion underperform those in the lowest quintile by 0.79% per month, during the sample period of 1983 to 2000. This pattern is at odds with the traditional view that analysts’ forecast dispersion reflects a stock’s fundamental risk, and thus should be positively correlated with expected returns.

The cause of this anomaly is being actively debated in the literature. DMS link it to E. Miller’s (1977) hypothesis that stock prices reflect optimism in the presence of short-sale constraints. They view analysts’ disagreement as a measure of investors’ difference of opinions. When investors’ opinions are more divergent, the optimistic bias in stock valuation induced by short-sale constraints is stronger, and future price reversal is more likely. The other main explanation is provided by Johnson (2004), who argues that the inverse relation between forecast dispersion and future stock returns is consistent with a rational leverage effect. In Johnson’s model, forecast dispersion proxies for the nonsystematic estimation risk on a firm’s cash flows, and stock is valued as an option on the asset of a levered firm. For a fixed terminal value of assets, higher estimation risk increases the current option value of equity and lowers future expected stock return.

Our article contributes to the understanding of this anomaly by highlighting a likely source of analyst forecast dispersion. We argue that the forecast dispersion anomaly is related to a firm’s disclosure behavior. Firms on average disclose good news about future earnings promptly and delay the disclosure of bad news. The evidence of this selective disclosure behavior has been provided by G. S. Miller (2002) and Kothari, Shu, and Wysocki (2009).¹ In addition, prior studies have shown that firms providing less public disclosure exhibit greater analyst forecast dispersion (e.g., Lang & Lundholm, 1996), presumably because market participants including analysts have to rely more on their private information, and this leads to greater divergence of opinion among them. Based on the above observations, we predict that firms with high forecast dispersion are likely to be the ones withholding bad news about future earnings. Furthermore, when the bad news is ultimately released to the public (e.g., via earnings announcements), these stocks tend to experience negative returns.

Our empirical evidence supports the proposed explanation. Using the Institutional Brokers’ Estimate System (I/B/E/S) analyst earnings forecast data, we confirm that forecast dispersion is negatively related to future returns. We then show that firms with higher forecast dispersion are more likely to report poor future earnings. These results suggest that when firms withhold information about their forthcoming unfavorable earnings, their forecast dispersion tends to be higher. Furthermore, as an example of managers’ incentives to withhold bad news, we show that external financing activities, a likely motivation for withholding bad news, are preceded by higher analyst forecast dispersion. More importantly, after controlling for its relation with future earnings, forecast dispersion is no longer negatively related to future returns. We also find a strong negative relation between forecast dispersion and stock returns during the future earnings announcement period. This relation is also explained by the negative relation between forecast dispersion and future earnings. These results suggest that withholding of bad news by firms leads to high forecast dispersion, and when the bad news reaches the market (e.g., via earnings announcements), stock prices decline.

Our study makes the following contribution to existing studies on the forecast dispersion anomaly. The explanation based on the short-sale constraints (e.g., DMS; Berkman, Dimitrov, Jain, Koch, & Tice, 2009) considers forecast dispersion as a proxy for difference of opinions among investors, and the explanation based on leverage (e.g., Johnson, 2004) regards forecast dispersion as a proxy for estimation risk; however, but none of them explore what causes difference of opinion or estimation risk in the first place. In particular, the existing explanations are silent on why forecast dispersion is related to future corporate earnings. Based on selective corporate disclosure behavior, we link analyst forecast dispersion to information about future corporate earnings, and show that the relation between forecast dispersion and future earnings is key to explain the relation between forecast dispersion and future stock returns.

By focusing on the potential source of analyst forecast dispersion, our study complements existing studies in explaining the forecast dispersion anomaly. For example, consider the effect of short-sale constraints contemplated by E. Miller (1977). It is plausible that withholding of bad news by some firms increases difference of opinions among investors, and due to short-sale constraints, the overvaluation of these firms is exacerbated. Similarly, consider the hypothesis of Johnson (2004). It is possible that withholding of bad news increases investors’ estimation risk, which, coupled with the leverage effect, increases the option value of stocks and correspondingly reduces the expected stock returns.²

The rest of the article is organized as follows: The “Variable Definitions and Stock Sample” section discusses definitions of key variables and the stock sample. The “Empirical Evidence” section presents our main empirical results. The “Further Analysis” section provides some further analyses, and the “Conclusion” section concludes.

Variable Definitions and Stock Sample

Analyst Forecast Dispersion

We obtain data on analyst earnings forecasts from the I/B/E/S-unadjusted Summary History file. The regular I/B/E/S Summary History file defines earnings per share (EPS) forecast on the current share basis, by adjusting all prior analyst earnings forecasts when there is a stock split. Payne and Thomas (2003) and DMS show that this stock split adjustment, coupled with the practice of rounding to the nearest cent, induces a downward bias in the analyst forecast dispersion measure even when it is scaled by similarly split-adjusted stock price or by absolute mean EPS forecast. The use of unadjusted Summary History file avoids this bias.

In each calendar quarter Q0, we measure analyst forecast dispersion (DISP) for a firm as the standard deviation of analysts’ EPS forecasts for the firm’s currently unreported fiscal year (referred to as “FY1” in I/B/E/S terminology) scaled by the absolute value of the mean EPS forecast for the same fiscal year. Both the standard deviation and mean EPS forecasts are measured during the last month of the calendar quarter Q0—to be exact, on the Thursday before the third Friday of that month according to the I/B/E/S practice. To ensure that the dispersion measure is meaningful, we follow DMS and require that DISP is calculated over earnings forecasts made by at least two analysts.

The above definition of forecast dispersion, which scales standard deviation by absolute mean, follows the statistical concept of coefficient of variation. It is also the definition used in existing studies, for example, DMS. However, the variable is undefined when the mean forecast is exactly zero. Such observations are not included in our sample. In addition, a mean forecast close to zero will cause this variable to have undesirable distributional properties. Nonetheless, we measure forecast dispersion this way to be consistent with the literature. Also, to make our conclusions robust to distributional assumptions of the underlying variables, our empirical analysis employs both sorted portfolios and cross-sectional regressions using the cross-sectional rank of forecast dispersion.³

Earnings Measures

To examine the association between forecast dispersion and future earnings, we use earnings measures that correspond to earnings reported in calendar quarter Q1, which is the quarter after calendar quarter Q0, the forecast dispersion measurement quarter. First, return on equity (ROE) is quarterly net income reported in quarter Q1, divided by the book value of equity at the beginning of the fiscal quarter. Second, standardized unexpected earnings (SUE) is the seasonally differenced quarterly EPS divided by its standard deviation. Both ROE and SUR are constructed using the Compustat data. Third, earnings surprise (SUR) is the actual quarterly EPS announced in quarter Q1 in excess of the corresponding consensus EPS analyst forecast in the month immediately prior to the earnings announcement, divided by stock price on the measurement date of the consensus EPS forecast. Finally, analyst forecast revision (FRV) is the consensus EPS forecast for the currently unreported fiscal year (FY1) in the last month of quarter Q1 minus the consensus forecast for the same fiscal year in the last month of quarter Q0, divided by stock price on the measurement date of the consensus forecast in quarter Q0. Note that FRV incorporates information released in the earnings report announced in calendar quarter Q1. Both SUR and FRV are based on the unadjusted I/B/E/S data.⁴ Earnings announcement dates are obtained from I/B/E/S. It is worthwhile to point out that as the ROE and SUE measures are based on the Compustat data, they are susceptible to earnings restatements; however, the SUR and FRV measures are based on I/B/E/S data, and are thus not subject to the restatement concern.

Stock Sample

For each calendar quarter from 1985 to 2015, we start with all stocks with valid analyst forecast dispersion data. To avoid market microstructure issues in measuring returns, we require that stock price at the end of the calendar quarter must be no less than US$5. We also require that the stock must have valid market capitalization data from Center for Research in Security Prices (CRSP) at the end of Q0. Table 1 provides summary statistics for stocks in our sample at the end of each year, from 1985 to 2015. The number of firms increases from 1,365 in 1985 to 2,454 in 2015. There are a good number of observations for each of the sample years, allowing us to effectively carry out cross-sectional analysis. We also report the average market capitalization and book-to-market ratio of sample stocks for each year. Market capitalization is measured at the end of each year. Book value of equity used in computing the book-to-market ratio is from the most recently reported fiscal quarter before the calendar year-end. As analysts tend to follow large-cap stocks, the average market capitalization in our sample is greater than that for the CRSP universe. The average book-to-market ratio of sample stocks becomes dramatically low in the late 1990s, and is temporarily high around 2008, consistent with what we know about the Internet bubble period and the recent financial crisis.

Table 1.

Summary Statistics: Analyst Forecast Dispersion and Firm Characteristics.

Year	n	DISP		SIZE		BM
Year	n	Mean	Median	Mean	Median	Mean	Median
1980	1,365	0.10	0.05	0.76	0.25	1.03	0.95
1981	1,508	0.12	0.05	0.67	0.23	1.11	1.00
1982	1,616	0.21	0.06	0.75	0.25	0.92	0.83
1983	1,935	0.21	0.06	0.79	0.25	0.77	0.68
1984	2,040	0.14	0.05	0.75	0.21	0.85	0.76
1985	2,112	0.20	0.06	0.92	0.25	0.72	0.65
1986	2,151	0.24	0.07	1.02	0.25	0.72	0.64
1987	2,110	0.16	0.06	1.03	0.24	0.82	0.73
1988	2,197	0.14	0.05	1.08	0.24	0.75	0.68
1989	2,266	0.15	0.05	1.29	0.25	0.70	0.61
1990	2,012	0.14	0.05	1.31	0.24	0.89	0.75
1991	2,229	0.14	0.05	1.57	0.31	0.66	0.57
1992	2,544	0.12	0.04	1.54	0.31	0.58	0.50
1993	2,880	0.12	0.04	1.53	0.33	0.55	0.46
1994	3,054	0.09	0.03	1.44	0.29	0.62	0.54
1995	3,274	0.11	0.03	1.82	0.35	0.54	0.46
1996	3,727	0.11	0.03	1.98	0.36	0.50	0.43
1997	3,798	0.09	0.03	2.55	0.42	0.45	0.38
1998	3,500	0.11	0.03	3.44	0.45	0.56	0.45
1999	3,426	0.10	0.03	4.50	0.59	0.56	0.42
2000	2,786	0.07	0.02	5.03	0.72	0.60	0.45
2001	2,558	0.10	0.03	4.86	0.79	0.54	0.45
2002	2,269	0.07	0.02	4.19	0.70	0.67	0.55
2003	2,566	0.08	0.02	4.85	0.85	0.48	0.42
2004	2,678	0.06	0.02	5.20	0.96	0.45	0.39
2005	2,720	0.07	0.02	5.34	1.00	0.47	0.41
2006	2,803	0.08	0.02	5.77	1.07	0.46	0.41
2007	2,638	0.09	0.03	6.15	1.06	0.56	0.45
2008	2,003	0.09	0.03	4.80	0.92	0.88	0.69
2009	2,234	0.12	0.03	5.45	1.05	0.64	0.54
2010	2,348	0.10	0.03	5.97	1.24	0.57	0.49
2011	2,223	0.09	0.03	6.23	1.24	0.68	0.57
2012	2,301	0.08	0.02	6.78	1.28	0.62	0.54
2013	2,436	0.08	0.02	8.46	1.71	0.47	0.40
2014	2,504	0.09	0.02	8.96	1.69	0.50	0.41
2015	2,454	0.08	0.02	8.90	1.58	0.56	0.45

Note. This table reports number of sample stocks (n), cross-sectional mean and median for analyst forecast dispersion (DISP), market capitalization (SIZE) in trillions, and book-to-market ratio (BM) for sample stocks during the last calendar quarter of each year from 1985 to 2015. The sample includes all stocks with valid forecast dispersion observations and valid market capitalization observations, with a stock price no less than US$5 at the beginning of a quarter.

Empirical Evidence

Forecast Dispersion and Stock Returns

We first confirm the forecast dispersion-stock return anomaly using sorted portfolios. For each calendar quarter from 1985 to the third calendar quarter of 2015 (2015Q3), we sort stocks into equal-weighted quintile portfolios based on analysts’ forecast dispersion, DISP. Our portfolio ranking period ends in 2015Q3 because our next-quarter return data end in 2015Q4. The portfolios are held without rebalancing for one quarter. We calculate the average return and the Carhart (1997) four-factor alpha for each quintile portfolio. Stock return data are obtained from CRSP. Delisting returns are included when computing buy-and-hold returns for each stock. When CRSP delisting return is missing, we follow Shumway (1997) and replace missing delisting return by −30% if delisting is performance related, and by zero otherwise. The Carhart four-factor alpha is estimated using the following regression:

R_{i τ} = α + b_{1} (R_{m t} - R_{f t}) + b_{2} S M B_{t} + b_{3} H M L_{t} + b_{4} U M D_{t} + e_{t},

where $R_{it}$ is the quarterly portfolio return in excess of the risk-free rate for the dispersion-sorted quintile portfolios, or the quarterly return spread between the top and bottom dispersion quintile portfolios. $R_{f t}$ , the risk-free rate, is proxied by the 3-month T-bill yield, and $R_{m t}$ is the quarterly return of the CRSP value-weighted index. $S M B_{t}$ , $H M L_{t}$ , and $U M D_{t}$ are the Fama–French size, book-to-market, and momentum factors. Monthly data on these factors are obtained from Ken French’s website.⁵ We compound these monthly factors, which are in essence portfolio returns, into quarterly factors for our regression analysis.

The results are reported in Table 2 and are consistent with those reported by DMS. Portfolio returns and four-factor alphas monotonically decrease with DISP quintile ranks. The difference in average returns between the highest and lowest dispersion quintiles is −1.51% per quarter (t = −2.15), and the corresponding difference in four-factor alpha is −1.67% (t = −3.85).⁶ This finding suggests the existence of the forecast dispersion anomaly in our sample.

Table 2.

Performance of Stock Portfolios Sorted on Forecast Dispersion.

DISP quintile	Portfolio return	Four-factor alpha
1 (Low)	3.72	.67
2	3.46	.45
3	3.30	.29
4	2.85	−.23
5 (High)	2.21	−1.00
High − Low	−1.51 (−2.15)	−1.67 (−3.85)

Note. This table reports the performance of stock portfolios sorted by forecast dispersion. In each calendar quarter Q0, we sort stocks on analyst forecast dispersion (DISP) into equal-weighted quintile portfolios. For each portfolio, we report the time-series averages of quarterly portfolio returns and the Carhart (1997) four-factor alphas for calendar quarter Q1. Returns and alphas are expressed in percentage points. The Newey–West t statistics are computed with a one-quarter lag and are reported in parentheses. The sample period is from 1985 to 2015.

Forecast Dispersion and Future Earnings

Portfolio analysis

We compute the averages of our four future earnings measures, ROE, SUE, SUR, and FRV, for each quintile portfolio sorted on forecast dispersion. To alleviate the influence of outliers on statistical inference, in each quarter we cross-sectionally winsorize each variable at the top and bottom 1%.⁷ The results, reported in Panel A of Table 3, show a strong pattern that stocks with higher analyst forecast dispersion have lower values of future earnings measures. For example, the average ROE for the next quarter is −1.39% for stocks in the highest dispersion quintile and 4.29% for stocks in the lowest dispersion quintile. The ROE difference between the two quintiles is highly significant (t = 27.07). The corresponding differences for SUE, SUR, and FRV are also significant. These results suggest that stocks with higher dispersion are associated with poorer earnings performance in the subsequent quarter.⁸

Table 3.

Forecast Dispersion and Future Earnings.

Panel A: Future Earnings.
DISP quintile	ROE (%)	SUE	SUR (%)	FRV (%)
1 (Low)	4.29	0.45	0.00	−0.28
2	3.72	0.33	0.00	−0.29
3	2.85	0.20	−0.03	−0.36
4	1.37	0.06	−0.09	−0.53
5 (High)	−1.40	−0.09	−0.37	−1.08
High − Low	−5.69 (−27.08)	−0.54 (−18.61)	−0.36 (−4.23)	−0.80 (−8.94)
Panel B: Characteristic- and Industry-Adjusted Future Earnings.
DISP quintile	ROE (%)	SUE	SUR (%)	FRV (%)
1 (Low)	1.02	0.13	0.03	0.08
2	0.86	0.07	0.04	0.08
3	0.51	0.01	0.04	0.08
4	−0.30	−0.06	0.02	0.01
5 (High)	−2.03	−0.13	−0.14	−0.29
High − Low	−3.04 (−21.11)	−0.25 (−17.35)	−0.17 (−4.51)	−0.37 (−9.26)

Note. For each calendar quarter Q0, we sort stocks on analyst forecast dispersion (DISP) into quintile portfolios. For each DISP quintile portfolio, we report the time-series averages of quarterly earnings measures for earnings announced in the next calendar quarter Q1. Earnings measures include return on equity (ROE), standardized unexpected earnings (SUE), earnings surprises relative to consensus forecasts (SUR), and analyst forecast revisions from before to after the earnings announcement in calendar quarter Q1 (FRV). Panel A reports unadjusted future earnings measures, and Panel B reports the characteristic- and industry-adjusted earnings measures. The Newey–West t statistics are computed with a four-quarter lag and are reported in parentheses.

In Panel B of Table 3, we report future earnings measures that are adjusted for firm characteristics as well as industry membership. We use a two-step procedure to estimate characteristic- and industry-adjusted future earnings measures for each firm. The first step follows Daniel, Grinblatt, Titman, and Wermers (1997). In each quarter Q0, we sort firms sequentially into 125 groups by size, book-to-market ratio, and past 12-month returns (five groups in each dimension). Firm size (SIZE), book-to-market ratio (BM), and price momentum (MOM) are based on information available at the end of Q0. Definitions of SIZE, BM, and MOM are provided in the appendix. Firms with any of the three characteristics missing are put into an additional group. Therefore, there are altogether 126 groups. For each group, we compute the average future earnings measure, and then for each firm, we calculate a characteristic-adjusted future earnings measure by subtracting the corresponding group’s average future earnings measure from the firm’s raw earnings measure. In the second step, we classify firms into 12 industries using the industry classification from Ken French’s website, and compute the average characteristic-adjusted future earnings measure (obtained in the first step) for each of the 12 industries in each quarter. Then, for each firm, we subtract the corresponding industry’s average characteristic-adjusted earnings measure from the firm’s characteristic-adjusted earnings measure to obtain a characteristic- and industry-adjusted future earnings measure.

The results reported in Panel B of Table 3 for the characteristic- and industry-adjusted future earnings are qualitatively similar to those reported in Panel A, suggesting that the negative association between forecast dispersion and future earnings measures cannot be explained away by firm size, book-to-market ratio, momentum, or industry membership. Furthermore, all the characteristic- and industry-adjusted future earnings measures are negative and have the largest magnitude for the highest dispersion quintile, suggesting that the future earnings news is the largest for firms with highest forecast dispersion.

In addition, we note that the future earnings measures reported above require that firms survive the future quarter Q1. To check whether this requirement biases our results, we compute, for each dispersion quintile, the proportion of sample firms with forecast dispersion data in quarter Q0 but without any valid future earnings measures in Q1. By the end of Q1, the fraction of nonsurviving firms for both the lowest dispersion quintile and the highest dispersion quintile is less than 0.5%. The small magnitudes of these fractions suggest that the impact of the survival bias on our inference is not material. Furthermore, as poor operating performance tends to be the dominant reason for nonsurvival, the slightly higher nonsurvival rate for the highest dispersion quintile firms means that it is unlikely that the observed negative relation between dispersion and future earnings is overstated due to the survival bias.

Regression analysis

It is possible that the negative association between forecast dispersion and future earnings documented in Table 3 and discussed in the “Portfolio analysis” section is driven by explanations other than the one we propose. We therefore use cross-sectional regressions with control variables related to likely alternative explanations. We consider five alternative explanations.

First, firms experiencing high growth tend to be risky, and analyst forecasts are likely to be highly dispersed for such firms. Due to the mean-reverting nature of sales growth and earnings, high growth firms also tend to disappoint investors in terms of future operating performance (Chan, Karceski, & Lakonishok, 2003; Lakonishok, Shleifer, & Vishny, 1994). Therefore, the negative correlation between forecast dispersion and future earnings might be due to this growth-firm effect. To control for this effect, we use two variables of firm growth: the four-quarter sales growth rate (SG) and analysts’ consensus long-term growth forecast (LTG; following La Porta, 1996), measured during the last month of quarter Q0.

Second, when firms engage in large investments, their business fundamentals often change dramatically, increasing uncertainty about future earnings. Therefore, there may be a positive association between analyst forecast dispersion and firms’ capital expenditure. Existing studies such as Titman, Wei, and Xie (2004) also document that firms with high capital expenditure tend to have low operating performance in the future, possibly due to the overinvestment tendency of empire-building managers. Thus, capital investments may induce a negative relation between forecast dispersion and future earnings. Similarly, large R&D spending increases uncertainty of future earnings, and hence may increase forecast dispersion. R&D expenditure is typically expensed rather than capitalized, and therefore depresses current earnings. If firms’ R&D spending is persistent, future earnings are likely to remain depressed, causing a negative relationship between forecast dispersion and future earnings. To control for the corporate investment effects, we follow Jegadeesh, Kim, Krische, and Lee (2004) and Chan, Lakonishok, and Sougiannis (2001) to construct two variables: annual capital expenditure (CAPEX) and annual R&D intensity (RD). We use annual capital expenditure and R&D measures instead of the quarterly measures because the Compustat quarterly file has frequent missing observations for these items.

Third, the relation between analyst forecast dispersion and future earnings may reflect the effect of accruals quality and idiosyncratic stock return volatility (see, for example, Ang, Hodrick, Xing, & Zhang, 2006; Francis, Lafond, Olsson, & Schipper, 2005; Jiang, Xu, & Yao, 2009). To control for this possibility, we include these two variables as controls. The accruals quality measure (AQ) follows Francis et al. (2005), except that we use quarterly data instead of annual data, and we require eight quarterly observations to measure AQ. Idiosyncratic volatility (IVOL) is the standard deviation of estimated residuals from regressing daily stock returns onto the contemporaneous daily market return as well as three lags of market returns. The regression is performed using daily returns of quarter Q0. In addition, we include a related firm characteristic, earnings variability (EV), measured as the standard deviation of return on equity (ROE) of quarters Q7 to Q0. We require a minimum of four quarterly ROE observations to measure EV.

Finally, analysts may respond asymmetrically to public news. Specifically, when bad news is disclosed analysts’ opinions about future earnings may become more dispersed than when good news is disclosed. Zhang (2006) shows that this causes stock mispricing—investors tend to have a higher degree of underreaction to bad news when there is more uncertainty. Also, our future earnings measures (ROE, SUE, SUR, and FRV) have been shown in the literature to be positively autocorrelated (e.g., Ali, Klein, & Rosenfeld, 1992; Bernard & Thomas, 1989, 1990; Freeman & Tse, 1989; Mendenhall, 1991). These two effects together predict a negative relation between forecast dispersion and future earnings. To control for this factor, we use lagged earnings measures (LAGEM) as a control variable. For example, when the future earnings measure is SUE, LAGEM is measured as SUE of quarter Q0.

We estimate quarterly cross-sectional regressions. The dependent variable is one of our future earnings measures. The main explanatory variable is the cross-sectional percentile rank of forecast dispersion (DISPRANK), which takes value from 0 to 1. In addition to the control variables discussed above, we include the inverse of the absolute mean EPS forecast (1/ABSFEPS, that is, the denominator of DISP), log market capitalization (Ln(SIZE)), log book-to-market ratio (Ln(BM)), stock returns during past 12 months (MOM), and 12 industry dummy variables (based on Fama–French 12-industry classification) as control variables.⁹SIZE, BM, and MOM are measured at the end of quarter Q0. The inclusion of 1/ABSFEPS is to control for potential artificial correlation between DISP and the absolute value of EPS forecast due to the lack of variation in the standard deviation of analyst forecast, as noted by Cheong and Thomas (2011) and discussed in the “Analyst Forecast Dispersion” section. The regression results are reported in Table 4 and are consistent with the portfolio analysis results reported in Table 3. The coefficient for DISPRANK remains negative and significant in the presence of the control variables. These results suggest that the negative association between forecast dispersion and future earnings is not explained away by control variables representing alternative explanations to the one we propose.

Table 4.

Forecast Dispersion and Future Earnings: Fama–MacBeth Regressions.

	ROE		SUE		SUR		FRV
DISPRANK	−1.8637	(−15.59)	−0.2929	(−16.56)	−0.1251	(−4.40)	−0.3531	(−8.97)
1/ABSFEPS	−1.4370	(−2.46)	0.1586	(3.65)	−0.1202	(−1.00)	0.4077	(2.91)
Ln(SIZE)	3.0217	(1.16)	−0.4240	(−0.91)	0.6892	(2.42)	0.2649	(0.47)
Ln(BM)	−1.4064	(−10.70)	−0.1226	(−13.85)	−0.0496	(−3.58)	−0.1123	(−8.88)
MOM	0.4627	(6.08)	0.2373	(12.10)	0.0760	(3.29)	0.1389	(6.01)
SG	−0.3233	(−4.65)	0.0709	(6.42)	0.0054	(0.64)	−0.0311	(−1.82)
LTG	−0.0115	(−3.34)	0.0051	(7.21)	0.0005	(0.80)	−0.0012	(−1.80)
CAPEX	−0.6731	(−2.51)	−0.3412	(−5.47)	−0.0441	(−1.33)	−0.1151	(−2.09)
R&D	−12.1819	(−9.60)	−0.3634	(−4.99)	0.0947	(2.07)	0.1071	(1.37)
AQ	0.6184	(1.07)	−0.0277	(−0.33)	−0.5816	(−1.24)	−0.3027	(−1.63)
IVOL	−24.9970	(−8.38)	−3.7809	(−7.02)	−2.3506	(−2.94)	−3.0792	(−4.42)
EV	42.2022	(1.71)	−7.8565	(−2.20)	−3.8261	(−0.95)	−11.6559	(−3.34)
LAGEM	0.3533	(27.76)	0.2250	(31.70)	0.1727	(17.90)	0.1301	(23.33)
Industry dummies	Yes		Yes		Yes		Yes
Adjusted R²	.33		.16		.06		.07

Note. This table reports the results of Fama–MacBeth regressions of future earnings measures on cross-sectional percentile ranking of analyst forecast dispersion (DISPRANK) along with various control variables. Control variables include the inverse of absolute EPS forecast (1/ABSFEPS), log market cap (Ln(SIZE)), log book-to-market ratio (Ln(BM)), past 12-month return (MOM), sales growth (SG), analyst long-term growth forecast (LTG), capital investments (CAPEX), R&D Intensity (RD), accruals quality (AQ), idiosyncratic return volatility (IVOL), earnings volatility (EV), lagged earnings measures (LAGEM), and 12 industry dummy variables. Estimated intercepts and coefficients on industry dummies are not reported. Adjusted R² is the time-series average of adjusted R-squares from cross-sectional regressions. The Newey–West t statistics are computed with a four-quarter lag and are reported in parentheses.

Forecast Dispersion and Stock Returns: Controlling for Future Earnings

Next, we examine whether the negative relation between forecast dispersion and future earnings drives the negative relation between forecast dispersion and future stock returns. We perform quarterly cross-sectional regressions, with the next quarter’s stock returns as the dependent variable and percentile rank of forecast dispersion (DISPRANK) as the main explanatory variable. To control for the relation between forecast dispersion and future earnings, we include future earnings measures (EM) in the model:

R_{i t + 1} = a_{0} + a_{1} D I S P R A N K_{i t} + a_{2} E M_{i t + 1} + C O N T R O L S_{i t} + e_{i t + 1} .

The control variables (CONTROLS) used in the model include the inverse of absolute mean EPS forecast, log of market capitalization, log of book-to-market ratio, and past 12-month returns. In addition, we include the control variables that we use in the regression analysis of relation between forecast dispersion and future earnings—sales growth (SG), analyst long-term growth forecast (LTG), capital expenditure (CAPEX), R&D intensity (RD), accruals quality (AQ), idiosyncratic return volatility (IVOL), earnings variability (EV), and a lagged earnings measure, LAGSUE.¹⁰ Definitions of these variables are provided in the appendix. If the negative association between DISPRANK and future stock returns is due to the relation between DISPRANK and future earnings, then after controlling for future earnings, the coefficient for DISPRANK in the above regression should no longer be significantly negative.

Table 5 reports the regression results. The baseline regression reported in column (1) shows that the negative relation between DISPRANK and future stock returns are not explained by a basic set of firm characteristics that include Ln(SIZE), Ln(BM), MOM, and 1/ABSFEPS. In the regression reported in column (2), we add firm characteristics to control for the alternative explanations discussed in the “Regression Analysis” section. Relative to the baseline regression in column (1), the magnitude of coefficients for DISPRANK and the corresponding t statistics in these regressions become smaller. However, the t statistics for the coefficients of DISPRANK remain significant at the 10% level.

Table 5.

Forecast Dispersion and Future Stock Returns After Controlling for Future Earnings: Fama–MacBeth Regressions.

	(1)		(2)		(3)		(4)		(5)		(6)
DISPRANK	−1.7660	(−2.30)	−0.3811	(−1.82)	0.0702	(0.15)	0.3201	(0.70)	−0.0060	(−0.01)	0.5836	(1.32)
1/ABSFEPS	−0.0061	(−0.51)	0.0075	(0.59)	0.0145	(1.05)	0.0035	(0.27)	0.0118	(0.92)	0.0000	(0.00)
Ln(SIZE)	−0.0456	(−0.45)	−0.3144	(−2.75)	−0.3074	(−2.72)	−0.3045	(−2.73)	−0.3372	(−2.98)	−0.3247	(−2.82)
Ln(BM)	0.4660	(1.24)	0.1194	(0.53)	0.4855	(1.98)	0.4258	(1.85)	0.2360	(1.05)	0.4399	(2.00)
MOM	1.1130	(1.59)	0.7850	(1.18)	0.6664	(1.02)	0.1584	(0.23)	0.4919	(0.77)	0.2906	(0.45)
SG			−0.9519	(−4.43)	−0.9265	(−4.25)	−1.1305	(−5.15)	−0.9616	(−4.60)	−0.8914	(−4.25)
LTG			0.0796	(0.05)	0.3223	(0.19)	−1.3113	(−0.78)	0.3096	(0.18)	0.5261	(0.30)
CAPEX			−0.0269	(−1.46)	−0.0260	(−1.41)	−0.0172	(−0.93)	−0.0258	(−1.38)	−0.0266	(−1.48)
RD			0.0398	(1.69)	0.0656	(2.83)	0.0484	(2.06)	0.0325	(1.36)	0.0323	(1.36)
AQ			−0.0054	(−2.17)	−0.0056	(−2.25)	−0.0053	(−2.13)	−0.0047	(−1.90)	−0.0044	(−1.66)
IVOL			−0.6864	(−3.80)	−0.6305	(−3.49)	−0.6030	(−3.40)	−0.6238	(−3.59)	−0.4934	(−2.82)
EV			−5.1313	(−0.89)	−5.6282	(−1.00)	−2.9829	(−0.52)	−4.6629	(−0.82)	−1.4629	(−0.25)
LAGSUE			5.2223	(5.78)
ROE					0.1633	(6.46)
SUE							2.4200	(22.09)
SUR									4.7984	(12.26)
FRV											2.7161	(19.74)
Adjusted R²	.04		.08		.08		.09		.10		.11

Note. This table reports the results of Fama–MacBeth regressions of stock returns over the next quarter on cross-sectional percentile ranking of forecast dispersion (DISPRANK). Control variables include the inverse of absolute EPS forecast (1/ABSFEPS), log market cap (Ln(SIZE)), log book-to-market ratio (Ln(BM)), past 12-month return (MOM), sales growth (SG), analyst long-term growth forecast (LTG), capital investments (CAPEX), R&D Intensity (RD), accruals quality (AQ), idiosyncratic return volatility (IVOL), earnings volatility (EV), standardized unexpected earnings (LAGSUE, only in the second regression specification), and one of the four future earnings measures, ROE, SUE, SUR, and FRV. The dependent variable, stock return, is expressed in percentage points. Estimated intercepts are not reported. Adjusted R² is the time-series average of adjusted R-squares from cross-sectional regressions. The Newey–West t statistics (in parentheses) are computed with a four-quarter lag.

Models in columns (3) to (6) of Table 5 add a future earnings measure (ROE, SUE, SUR, or FRV) as an explanatory variable. The coefficient on DISPRANK is no longer significantly negative in these models. This result suggests that the negative association between forecast dispersion and future stock returns is at least in part due to the relation between forecast dispersion and future earnings.

Note that analyst forecast dispersion is influenced by the amount of information selectively disclosed by firms and the fundamental uncertainty of a firm’s business. Once we control for future earnings, the relation between forecast dispersion and stock return may be interpreted as the effect of the fundamental uncertainty. If such uncertainty represents a form of systematic risk that commands risk premium, then we should expect a positive relation between forecast dispersion and stock returns after controlling for future earnings (i.e., controlling for selective disclosure). On the contrary, if such uncertainty is nonsystematic and not priced, then after controlling for future earnings there should be no significant relation between forecast dispersion and stock returns. To what extent the latter scenario is true is an empirical question. The results in columns (3) to (6) of Table 5 show that after controlling for future earnings, the relation between forecast dispersion and future stock returns is not significant. We reestimate the models in columns (3) to (6) by keeping forecast dispersion and future earnings measures but removing all other explanatory variables. We find that the coefficients on forecast dispersion are again not significant. Thus, our analysis does not provide conclusive evidence for whether forecast dispersion has a component that commands risk premium.

Further Analysis

Firms’ Short-Term Incentives to Withhold Bad News

In this section, we provide some evidence in support of the notion that firms may knowingly withhold bad news. We conjecture that around the time of raising financing from external sources, firms are more likely to withhold bad news, so that they can obtain better financing terms. That is, we predict that analyst forecast dispersion would be higher around firms’ external financing activities. Following Bradshaw, Richardson, and Sloan (2006), we construct a variable on external financing, XFIN, which summarizes the net equity and debt financing of a firm (with details provided in the appendix).

In each quarter Q0 (i.e., the ranking quarter), we measure XFIN during the rolling four quarters up until Q0 and refer to this as the XFIN window. We rank stocks by XFIN into quintiles and compute the average DISP during this XFIN window for each quintile. As a comparison, we also compute the average DISP during the four quarters prior to the XFIN window, for each XFIN quintile, and compute the difference in the average DISP between during and before the window. The results are reported in Table 6. The results suggest that there is a positive contemporaneous relation between analyst forecast dispersion and the level of external financing activities during the XFIN window. The average DISP for firms in the top XFIN quintile is 0.1510 versus 0.0976 for firms in the bottom XFIN quintile, and the difference is statistically significant. Furthermore, for firms in the top XFIN quintile, there is a significant increase in the average DISP from the four quarters prior to the XFIN window to during the XFIN window. By contrast, for firms in the bottom XFIN quintile, the average DISP decreases. These results are consistent with the notion that firms may withhold bad news around the external financing activities to obtain better financing terms.

Table 6.

Analyst Forecast Dispersion and Corporate External Financing Activities.

XFIN quintile	Average DISP during XFIN window	Average DISP during four quarters prior to XFIN window	Difference in average DISP
1 (Low)	0.0976	0.1030	−0.0053	(−2.65)
2	0.1093	0.1077	0.0016	(0.98)
3	0.1319	0.1280	0.0039	(1.73)
4	0.1399	0.1364	0.0035	(2.07)
5 (High)	0.1510	0.1461	0.0049	(2.54)
High − Low	0.0534 (11.71)	0.0452 (7.94)	0.0102	(3.05)

Note. This table reports the relation between analyst forecast dispersion and corporate external financing activities. In each quarter, we sort firms into quintiles based on XFIN, a measure of corporate external financing over a four-quarter window, including the current quarter. We calculate the average analyst forecast dispersion DISP during the XFIN window and during the four quarters prior to the XFIN window, as well as their differences. The Newey–West t statistics (in parentheses) are computed with an eight-quarter lag.

Forecast Dispersion and Future Earnings Announcement Returns

If the negative relation between forecast dispersion and subsequent stock returns is due to the withholding of bad news, this relation should be observed for stock returns during the earnings announcement window as well, because that is when large amount of information about the firm’s performance is released. We define a 3-day earnings announcement window from 1 trading day before to 1 trading day after a quarterly earnings announcement date. Announcements made on a nontrading day are considered effective on the following trading day. Earnings announcement dates are obtained from I/B/E/S. We calculate earnings announcement return (EAR) as the 3-day earnings announcement window buy-and-hold returns for a stock in excess of the CRSP value-weighted index return during the same 3-day window. In Table 7, we sort firms into quintiles based on forecast dispersion in Q0 and report the average EAR for each quintile for earnings announcements made during quarter Q1. The average EAR for stocks in the lowest DISP quintile is 0.40% versus −0.25% for stocks in the top DISP quintile. The difference in EAR between the top and bottom quintiles is significantly negative.

Table 7.

Future Earnings Announcements Returns for Portfolios Sorted on Forecast Dispersion.

DISP quintile	EAR% (t stat)
1 (Low)	0.40	(3.64)
2	0.39	(2.69)
3	0.06	(0.47)
4	0.17	(1.20)
5 (High)	−0.25	(−1.66)
High − Low	0.65	(3.19)

Note. This table reports the averages EAR for each quintile portfolio sorted by forecast dispersion in quarter Q0. EAR is the stock return during the 3-day earnings announcement window in the subsequent quarter Q1, in excess of the CRSP value-weighted index return during the same window. EARs reported in this table are in percentage points. The Newey–West t statistics (in parentheses) are computed with a one-quarter lag.

Next, we repeat the analysis in Table 5 using earnings announcement returns instead of long window returns to further examine whether the dispersion anomaly is driven by the negative relation between forecast dispersion and future earnings. The benefit of using short window returns is that omission of any risk factors in the pricing equation is unlikely to bias the coefficients on the included variables by much, because expected returns associated with the omitted risk factors are likely to be small over short windows (Bernard & Thomas, 1990).

Table 8 reports the results from regressions that are similar to those in Table 5, except that the dependent variable is EAR. When a future earnings measure is not included as an explanatory variable, EAR is negatively associated with past forecast dispersion (see columns (1) and (2)). However, after controlling for future earnings, the coefficient on past forecast dispersion becomes insignificant. This result suggests that the negative association between forecast dispersion and future returns around earnings announcements is driven by the relation between forecast dispersion and future earnings.

Table 8.

Forecast Dispersion and Future Earnings Announcement Returns After Controlling for Future Earnings: Fama–MacBeth Regressions.

	(1)		(2)		(3)		(4)		(5)		(6)
DISPRANK	−0.7132	(−2.82)	−0.7257	(−2.54)	−0.4088	(−1.53)	−0.3843	(−1.42)	−0.3551	(−1.37)	−0.3594	(−1.37)
1/ABSFEPS	0.0010	(0.06)	0.0049	(0.23)	−0.0031	(−0.13)	−0.0046	(−0.18)	−0.0050	(−0.20)	−0.0062	(−0.25)
Ln(SIZE)	−0.0428	(−0.73)	−0.1455	(−2.71)	−0.0834	(−1.96)	−0.0999	(−2.32)	−0.0758	(−1.91)	−0.0770	(−1.90)
Ln(BM)	0.1198	(1.08)	−0.0496	(−0.39)	−0.0711	(−0.76)	−0.0714	(−0.70)	−0.0787	(−0.84)	−0.0941	(−0.97)
MOM	0.4454	(2.32)	0.3876	(1.76)	0.2718	(2.00)	0.2305	(1.71)	0.2667	(2.03)	0.2756	(2.01)
SG			0.3848	(1.23)	−0.3488	(−1.47)	−0.3808	(−1.60)	−0.3641	(−1.55)	−0.3543	(−1.48)
LTG			−0.9826	(−0.78)	−1.5285	(−1.54)	−1.6795	(−1.70)	−1.4859	(−1.52)	−1.4029	(−1.42)
CAPEX			−0.0161	(−1.80)	−0.0048	(−0.67)	−0.0053	(−0.73)	−0.0048	(−0.67)	−0.0050	(−0.66)
RD			−0.0017	(−0.16)	−0.0082	(−0.92)	−0.0095	(−1.05)	−0.0109	(−1.20)	−0.0116	(−1.34)
AQ			−0.3539	(−0.17)	2.6078	(1.07)	2.1112	(0.90)	3.8827	(1.13)	2.8926	(1.24)
IVOL			−0.1596	(−2.20)	−0.1448	(−2.72)	−0.1544	(−3.00)	−0.1461	(−2.70)	−0.1520	(−2.92)
EV			−1.6481	(−0.26)	−0.5619	(−0.12)	0.0591	(0.01)	0.4449	(0.09)	0.3798	(0.08)
LAGSUE			0.0112	(1.62)
ROE					1.0439	(2.75)
SUE							0.2192	(3.95)
SUR									0.4986	(4.18)
FRV											0.1955	(3.34)
Adjusted R²	.0104		.0232		.0154		.0183		.0187		.0167

Note. This table reports the results of Fama–MacBeth regressions of EAR in the subsequent quarter on cross-sectional percentile rank of forecast dispersion (DISPRANK). Control variables include the inverse of absolute EPS forecast (1/ABSFEPS), log market cap (Ln(SIZE)), log book-to-market ratio (Ln(BM)), past 12-month return (MOM), sales growth (SG), long-term growth forecast (LTG), capital investments (CAPEX), R&D Intensity (RD), accruals quality (AQ), idiosyncratic volatility (IVOL), earnings volatility (EV), and one of the four future earnings measures, ROE, SUE, SUR, and FRV. The dependent variable EAR is reported in percentage points. Estimated intercepts are not reported. Adjusted R² is the time-series average of adjusted R-squares of cross-sectional regressions. The Newey–West t statistics (in parentheses) are computed with a four-quarter lag.

Controlling for Stale Forecasts

Analysts may not timely update their forecasts when they receive negative information about the firm. If their stale forecasts are included in calculating the mean and standard deviation of analysts’ forecasts, both the consensus forecast and forecast dispersion will be biased upward. These arguments suggest that dispersion will be negatively related to future earnings as well as to future stock returns, if investors rely on the consensus forecast to value stocks and are not aware of the potential bias.

To control for the effect of stale forecasts, we construct a new analyst forecast dispersion measure, which uses only those forecasts that are made within the previous 30 days of the measurement date of consensus forecast and forecast dispersion. We obtain data for the new forecast dispersion measure from the I/B/E/S-Unadjusted Detailed History file. This alternative forecast dispersion measure should be largely free from the influence of stale forecasts. We find that all the key findings are robust to the use of this alternative dispersion measure. For brevity, we do not tabulate these results in the article.

Conclusion

We examine whether the corporate disclosure practice of disclosing good news promptly and withholding bad news is one of the reasons for the forecast dispersion anomaly—the negative relation between analyst forecast dispersion and future stock returns. We argue that withholding bad news increases dispersion in analyst forecasts of earnings, because analysts have to rely more on their private and idiosyncratic sources of information, which in turn leads to greater forecast dispersion. Furthermore, when the bad news is eventually released to the public (e.g., via earnings announcements), these stocks tend to experience negative returns. Consequently, a negative association between forecast dispersion and future returns is observed.

The empirical results support our explanation for the forecast dispersion anomaly. We show that firms with higher forecast dispersion are more likely to report poor future earnings. These results suggest that withholding information about their forthcoming unfavorable earnings causes firms’ forecast dispersion to be higher. In addition, after controlling for its relation with future earnings, forecast dispersion is no longer significantly negatively related to future returns.

Overall, our results suggest that the strategic corporate disclosure behavior of asymmetric disclosure of good versus bad news, combined with its effect on analyst earnings forecasts, has important implications on stock returns. In particular, such behavior plays an important role in explaining the negative association between analyst forecast dispersion and future stock returns.

Footnotes

Appendix

Acknowledgements

We appreciate the helpful comments from Anup Agrawal, Jon Garfinkal, Doug Hanna (associate editor), Clifton Green, George Jiang, Kathy Kahle, Frank Moers, Alexei Ovtchinnikov, Hong Yan, an anonymous referee, and seminar participants at the University of Arizona, National University of Singapore, annual meetings of American Accounting Association, Western Finance Association, China International Conference in Finance, and European Finance Association.

Authors’ Note

Data on analyst forecasts are provided by Institutional Brokers’ Estimate System (I/B/E/S) under a program to encourage academic research.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Ali

Klein

Rosenfeld

(1992). Analysts’ use of information about permanent and transitory earnings components in forecasting annual EPS. The Accounting Review, 67, 183-198.

Ang

Hodrick

Xing

Zhang

(2006). The cross-section of volatility and expected returns. Journal of Finance, 51, 259-299.

Avramov

Chordia

Jostova

Philipov

(2009). Dispersion in analysts’ earnings forecasts and credit rating. Journal of Financial Economics, 91, 83-101.

Berkman

Dimitrov

Jain

Koch

Tice

(2009). Sell on the news: Differences of opinion, short-sales constraints, and returns around earnings announcements. Journal of Financial Economics, 92, 372-399.

Bernard

Thomas

(1989). Post-earnings-announcement drift: Delayed price response or risk premium. Journal of Accounting Research, 27, 1-36.

Bernard

Thomas

(1990). Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting & Economics, 13, 305-340.

Bradshaw

Richardson

Sloan

(2006). The relation between corporate financing activities, analysts’ forecasts and stock returns. Journal of Accounting & Economics, 42, 53-85.

Carhart

(1997). On persistence in mutual fund performance. Journal of Finance, 52, 57-82.

Chan

K. C.

Karceski

Lakonishok

(2003). The level and persistence of growth rates. Journal of Finance, 58, 643-684.

10.

Chan

K. C.

Lakonishok

Sougiannis

(2001). The stock market valuation of research and development expenditures. Journal of Finance, 56, 2431-2456.

11.

Cheong

F. S.

Thomas

(2011). Why do EPS forecast error and dispersion not vary with scale? Implications for analyst and managerial behavior. Journal of Accounting Research, 49, 359-401.

12.

Cohen

D. A.

Hann

Ogneva

(2007). Another look at GAAP versus the street: An empirical assessment of measurement error bias. Review of Accounting Studies, 12, 271-303.

13.

Daniel

Grinblatt

Titman

Wermers

(1997). Measuring mutual fund performance with characteristic-based benchmarks. Journal of Finance, 52, 1035-1058.

14.

Diether

Malloy

Scherbina

(2002). Differences of opinion and the cross section of stock returns. Journal of Finance, 57, 2113-2141.

15.

Francis

Lafond

Olsson

Schipper

(2005). The market pricing of accruals quality. Journal of Accounting & Economics, 39, 295-327.

16.

Frankel

McNichols

Wilson

G. P.

(1995). Discretionary disclosure and external financing. Accounting Review, 70, 135-150.

17.

Freeman

R. N.

Tse

S. Y.

(1989). The multiperiod information content of accounting earnings: Confirmations and contradictions of previous earnings reports. Journal of Accounting Research, 27, 49-79.

18.

Jegadeesh

Kim

Krische

Lee

(2004). Analyzing the analysts: When do recommendations add value?Journal of Finance, 59, 1083-1124.

19.

Jiang

Yao

(2009). The information content of idiosyncratic volatility. Journal of Financial and Quantitative Analysis, 46, 1-28.

20.

Johnson

(2004). Forecast dispersion and the cross section of expected returns. Journal of Finance, 59, 1957-1978.

21.

Kothari

S. P.

Shu

Wysocki

(2009). Do managers withhold bad news?Journal of Accounting Research, 47, 241-276.

22.

La Porta

(1996). Expectations and the cross-section of stock returns. Journal of Finance, 56, 1985-2010.

23.

Lakonishok

Shleifer

Vishny

(1994). Contrarian investment, extrapolation, and risk. Journal of Finance, 49, 1541-1578.

24.

Lang

Lundholm

(1996). Corporate disclosure policy and analyst behavior. The Accounting Review, 71, 467-492.

25.

Mendenhall

(1991). Evidence on the possible underweighting of earnings information. Journal of Accounting Research, 29, 170-179.

26.

Miller

(1977). Risk, uncertainty, and divergence of opinion. Journal of Finance, 32, 1151-1168.

27.

Miller

G. S.

(2002). Earnings performance and discretionary disclosures. Journal of Accounting Research, 40, 173-204.

28.

Payne

Thomas

(2003). The implications of using stock-split adjusted I/B/E/S data in empirical research. The Accounting Review, 78, 1049-1067.

29.

Sadka

Scherbina

(2007). Analyst disagreement, mispricing, and liquidity. Journal of Finance, 52, 2367-2403.

30.

Shumway

(1997). The delisting bias in CRSP data. Journal of Finance, 52, 3274-3340.

31.

Skinner

(1994). Why firms voluntarily disclose bad news. Journal of Accounting Research, 32, 38-60.

32.

Titman

Wei

Xie

(2004). Capital investment and stock returns. Journal of Financial and Quantitative Analysis, 39, 677-700.

33.

Yermack

(1997). Good timing: CEO stock option awards and company news announcements. Journal of Finance, 52, 449-476.

34.

Zhang

(2006). Information uncertainty and stock returns. Journal of Finance, 61, 105-137.