Earnings Information, Overvaluation, and the Forecast Dispersion Anomaly Soonho Kim* Pukyong National University Haejung Na† California State University, Los Angeles
This draft: January 2018
Abstract: This study investigates the effect of earnings components on the anomalous negative relation between analyst forecast dispersion and future stock returns. According to this study, higher discretionary earnings and lower nondiscretionary earnings amplify the analyst forecast dispersion anomaly. If firms are found with lower discretionary earnings or higher nondiscretionary earnings, the negative dispersion-return relation turns insignificant or almost disappears. The more prevalent dispersion anomaly for higher discretionary earnings stocks coincides with higher forecast error. It suggests that investors follow analysts’ overly optimistic forecasts for firms with higher discretionary earnings and overvalue them. Moreover, the relation between high discretionary or low nondiscretionary earnings and the forecast dispersion anomaly is more prevalent in subsamples with low institutional ownership which proxies for short-sale constraints. Such results also indicate that overvaluation in high dispersion stocks, as noted by Diether et al. (2002), can be associated with higher discretionary or lower nondiscretionary earnings. Overall, the results suggest that discretionary and nondiscretionary components of earnings play a significant role in the analyst forecast dispersion anomaly.
Keywords: analyst forecast dispersion, discretionary accruals, discretionary earnings, nondiscretionary earnings, overvaluation, overpricing, short-sale constraints
JEL classification: G12, G14
*
Pukyong National University Business School C25 605, Yongsorho 45, Namgu, Busan 48513, Korea. Phone: +82-51-629-5727, E-mail:
[email protected] † College of Business and Economics, California State University, Los Angeles, 5151 State University Drive, Los Angeles, CA 90032, E-mail:
[email protected]
Earnings Information, Overvaluation, and the Forecast Dispersion Anomaly
This draft: January 2018
Abstract This study investigates the effect of earnings components on the anomalous negative relation between analyst forecast dispersion and future stock returns. According to this study, higher discretionary earnings and lower nondiscretionary earnings amplify the analyst forecast dispersion anomaly. If firms are found with lower discretionary earnings or higher nondiscretionary earnings, the negative dispersion-return relation turns insignificant or almost disappears. The more prevalent dispersion anomaly for higher discretionary earnings stocks coincides with higher forecast error. It suggests that investors follow analysts’ overly optimistic forecasts for firms with higher discretionary earnings and overvalue them. Moreover, the relation between high discretionary or low nondiscretionary earnings and the forecast dispersion anomaly is more prevalent in subsamples with low institutional ownership which proxies for short-sale constraints. Such results also indicate that overvaluation in high dispersion stocks, as noted by Diether et al. (2002), can be associated with higher discretionary or lower nondiscretionary earnings. Overall, the results suggest that discretionary and nondiscretionary components of earnings play a significant role in the analyst forecast dispersion anomaly.
Keywords: analyst forecast dispersion, discretionary accruals, discretionary earnings, nondiscretionary earnings, overvaluation, overpricing, short-sale constraints
JEL classification: G12, G14
1
1. Introduction
Analyst forecast dispersion can be often regarded as a source of uncertainty and risk, and thought to yield higher returns in the stock market. However, investors and researchers witness an opposite anomalous pattern in the stock market that higher analyst forecast dispersions are followed by lower returns. With respect to this negative dispersion-stock return relation, Diether et al. (2002) relate it to difference of opinions and asymmetric arbitrage limits. According to Miller (1977), investors’ optimistic opinions are more likely to persist and prevail in the stock market as shortsale constraints make it harder for pessimistic viewers to take a short position. Thus, optimistic valuations inflate stock prices and are subsequently followed by a decrease in returns. Extending Diether et al.’s (2002) arguments, Boehme et al. (2006) also find that short-sale constraints are necessary conditions for the analyst forecast dispersion anomaly to hold. Kim et al. (2014) support the arguments by finding that the anomaly is stronger during high investor sentiment periods when overpricing becomes more severe due to short-sale impediments. In the related literature, many attempts to identify the source of the forecast dispersion anomaly can be found. For example, Johnson (2004) conjectures that analyst forecast dispersion is a proxy of firms’ idiosyncratic risk about cash flows and that the expected return should decrease with such risk. Sadka and Scherbina (2007) explain that high analyst dispersion and high trading costs coincide that the analyst dispersion anomaly is more prevalent for less liquid stocks. Meanwhile, according to Barron et al. (2009), the uncertainty component, but not the information asymmetry component, of forecast dispersion is attributable to the analyst forecast dispersion anomaly. Avramov et al. (2009) relate financial distress with analyst forecast dispersion and argue that behind the forecast dispersion puzzle stands the negative distress-return relation. 2
Among the previous efforts to explain the analyst dispersion anomaly, not many studies have linked it with components of earnings information: discretionary and nondiscretionary earnings. While Peng et al. (2016) examine the effect of accounting accruals on analysts’ opinions and stock returns, their research focus is on total accruals which contain both discretionary and nondiscretionary accruals. Discretionary accruals are often used as more direct proxies of managers’ discretionary or opportunistic behaviors (Healy (1985), DeAngelo (1988)), whereas nondiscretionary accruals have different characteristics in that they are affected by economic situations (Kaplan (1985), Jones (1991)). Thus, it can be misleading to use total accruals to measure the effect of manages’ discretions on analysts’ diverse opinions and the returns. As this study seeks to measure part of earnings affected by managers’ discretionary decisions, a more appropriate proxy should be discretionary accruals or discretionary earnings. On the contrary, nondiscretionary accruals share more similarities with operating cash flows, the remaining component of earnings, in that they are both affected by economic and fundamental factors. It is also worthwhile to examine how the two components, taken together as nondiscretionary earnings, affect the relation between analyst forecast dispersion and stock returns and investigate how differently discretionary earnings and nondiscretionary earnings affect the forecast dispersionreturn association. This study finds that the analyst dispersion anomaly is more prevalent among firms with higher discretionary or lower nondiscretionary earnings. For firms that report lower discretionary or higher nondiscretionary earnings, the anomalous association between forecast dispersion and stock returns turns insignificant or almost disappears. Moreover, the role of earnings information in the forecast dispersion anomaly is stronger in subsamples with low institutional ownership, but is weaker in those with high institutional ownership. As low institutional ownership proxies for 3
stronger short-sale constraints, the results indicate that high discretionary earnings (low nondiscretionary earnings) attribute to overpricing among high forecast dispersion stocks when faced with short-sale challenges. Subsequently, those stocks should go through a severe return decrease, and the analyst forecast dispersion anomaly becomes stronger. The remainder of this study proceeds as follows. Section 2 introduces the literature review and the three main hypotheses of the study. Section 3 describes the data and methodology to compute discretionary and nondiscretionary earnings, analyst forecast dispersion, and the shortsale constraints. Section 3 presents empirical evidence showing contrasting roles played by discretionary and nondiscretionary earnings in the analyst dispersion anomaly. Section 4 shows robustness checks, and Section 5 concludes the study.
2. Literature Review & Hypotheses Development
Prior literature on analysts shows that the level of firms’ accruals affects individual analysts’ earnings forecast accuracy. According to Bradshaw et al. (2001), high accruals lead analysts to report overly optimistic forecasts, and are negatively associated with forecast errors, i.e., the difference between actual earnings and analysts’ forecasts. Teoh and Wang (2002) also document similar results, showing that newly issuing firms with high accruals are followed by overly optimistic analysts. Drake and Myers (2009) explain that accrual-related over-optimism is related to analyst characteristics. The previous studies argue that analysts in general are not able to accurately understand the characteristics of accruals or to identify that firm with high accruals go through earnings reversals. Due to lack of understanding and sophistication, analysts do not fully incorporate the 4
relation between accruals and subsequent earnings into their forecasts and tend to overestimate firms with high accruals (Bradshaw et al. (2001), Collins et al. (2003), Mashruwala et al. (2006), Drake and Myers (2009)). Moreover, Chan et al. (2001) document that if accruals are separated into discretionary and nondiscretionary accruals, higher discretionary accruals are more likely followed by a decline in earnings and other operating performance proxies than higher nondiscretionary accruals. If so, the component of accruals attributable to overly optimistic forecasts should be discretionary accruals, instead of the nondiscretionary part. Accordingly, it is possible to conjecture that higher discretionary accruals lead to more erroneous and inflated forecasts, thus should coincide with greater dispersion among analyst forecasts. Also suggested by prior studies is that higher accruals firms attract more analyst coverage. Lobo et al. (2012) report that poorer quality in accruals, which is measured by higher discretionary accruals, is interpreted as more opportunities for analysts to take advantage of the benefit created by private information gathering. Thus, more analysts tend to follow firms with high discretionary accruals, making it more likely that such firms experience diverse opinions. In sum, the previous arguments suggest that high discretionary accruals are followed by greater analyst coverage and more forecast errors. In turn, it is likely to lead to wider dispersion among analysts, and, according to Miller’s (1977) argument, also to more severe mispricing among investors. Thus, as such stocks’ subsequent period returns go down, the negative relation between analyst dispersion and return should hold strongly and significantly. On the other hand, according to Guay et al. (1996), earnings can be divided into two components: discretionary and nondiscretionary earnings. While discretionary earnings are equivalent to discretionary accruals, nondiscretionary earnings are the sum of cash flows and nondiscretionary accruals. They report that discretionary and nondiscretionary earnings are 5
associated with stock returns in the opposite direction and that discretionary (nondiscretionary) earnings are negatively (positively) correlated with stock returns. 1 On the contrary, as nondiscretionary earnings are also composed of cash flows, they are more closely linked to economic fundamental values of a firm. Thus, higher nondiscretionary earnings are likely to coincide with less analyst forecast errors and less analyst forecast dispersion. Accordingly, the analyst dispersion anomaly should be less severe for firms with more nondiscretionary earnings.
Hypothesis 1a. The analyst dispersion anomaly increases as the discretionary earnings information increases. Hypothesis 1b. The analyst dispersion anomaly decreases as the nondiscretionary earnings information increases.
In addition to the effect of earnings information on the dispersion anomaly, we are interested in discovering the channel between earnings information and the dispersion anomaly. Among many candidates, we seek to examine the dispersion, forecast error, and illiquidity as the variable to bridge between earnings information and the dispersion anomaly. First, as firms with more discretionary earnings are followed by greater dispersion of analyst forecasts, the amount of increase in dispersion may be greater for high discretionary earnings firms. If so, the dispersion anomaly can be more severe among high discretionary earnings firms. Another possible reason can be overly optimistic forecasts of analysts for firms with higher discretionary earnings (Bradshaw et al. (2001), Collins et al. (2003), Mashruwala et al. (2006), Drake and Myers (2009)).
1
Such results are based on the Jones and Modified Jones (1991) models to estimate discretionary accruals, which Guay et al. (1996) conclude to be the most consistent with prior research hypotheses. 6
If investors credulously follow analysts’ overly optimistic forecasts and overprice those stocks, the subsequent return reversal and the dispersion anomaly should be more sever among high discretionary earnings firms. Last, it is possible that higher discretionary earnings and higher dispersion coincide with higher illiquidity of the stock as the quality of information from both managers and analysts become worse (Sadka and Scherbina (2007)). In such case, it is likely that the dispersion anomaly will become more prevalent among high discretionary earnings stocks.
Hypothesis 2a. The analyst dispersion should be higher and the increase of dispersion should be more severe for stocks with higher discretionary earnings. Hypothesis 2b. The analyst forecast error should be more positive and inflated, and the dispersionforecast error link should be more severe for stocks with higher discretionary earnings. Hypothesis 2c. The illiquidity should deteriorate and the dispersion-illiquidity relation should be more severe for stocks with higher discretionary earnings.
If the role of discretionary and nondiscretionary earnings in the analyst forecast dispersion anomaly is related to causing or amplifying overpricing, then its role should remain or grow more significant among stocks that are harder to short. Miller (1977) argues that overpricing is more prevalent than underpricing if there is a diversity of opinions in valuation as pessimistic views are harder to be reflected due to short-sale impediments. Also, as short selling is known to constrain firms’ earnings management efforts (Fang et al. (2015)), a challenge to short selling may coincide with higher changes of earnings management or higher discretionary earnings. Thus, for firms that face short-sale impediments and that are found with high discretionary (low nondiscretionary) earnings, it is likely that the analyst dispersion anomaly will exacerbate. As short-sale impediments 7
are often measured as the sum of firms’ outstanding shares held by institutional owners, a stronger link between earnings items and analyst forecast dispersion anomaly is expected for firms with low institutional ownership.
Hypothesis 3a. The increasing effect of discretionary earnings on the analyst dispersion anomaly becomes greater for firms with lower institutional ownership. Hypothesis 3b. The decreasing effect of nondiscretionary earnings on the analyst dispersion anomaly becomes greater for firms with lower institutional ownership.
3. Data & Methodology 3.1. Earnings Information, Earnings Forecast, and Short-sale Impediments
The sample of this study consists of all NYSE, AMEX, and NASDAQ stocks available on the Institutional Brokers Estimate System (I/B/E/S), COMPUSTAT, and the CRSP. The sample period spans from 1983 to 2014. The dispersion in analyst earnings forecasts, DISP, is measured by standardizing the standard deviation of analysts’ current-fiscal-year EPS forecasts by the absolute value of mean forecast. As the standard deviation in analyst earnings forecasts from the adjusted file in I/B/E/S is known to be subject to the rounding error issue and the rounding problem becomes more severe for the summary file (Diether et al. (2002) and Payne and Thomas (2003)), we use the Unadjusted Detailed History File. At the end of every month, we compute standard deviation and average based on all the outstanding analyst forecasts for the same forecast period. The forecast dispersion is updated on a monthly basis. If there are more than one forecast from each brokerage firm for 8
the same firm and the same forecast period, only the latest estimate is used. If the forecast is voided by I/B/E/S with an “Excluded” or “Stopped” flag, then it is excluded.2 Also, stocks with less than three estimates for the given month and penny stocks, i.e. whose previous end-month price is less than five dollars, are dropped from the sample. Following Jones (1991), Dechow et al. (1995), and Guay et al. (1996), we compute discretionary and nondiscretionary earnings based on the below regression model, T𝐶𝐴𝑖𝑗𝑡 = 𝛽𝑗𝑡 (𝐴𝑇
1 𝑖𝑗𝑡−1
) + 𝛾𝑗𝑡 (∆𝑆𝑎𝑙𝑒𝑖𝑗𝑡 − ∆𝑅𝑒𝑐𝑖𝑗𝑡 ) + 𝛿𝑗𝑡 𝑃𝑃𝐸𝑖𝑗𝑡 + 𝜀𝑖𝑗𝑡
where 𝑇𝐶𝐴𝑖𝑗𝑡 indicates the total current accruals of Firm we of Industry j, in calendar year t, 𝐴𝑇𝑖𝑗𝑡 total assets, and ∆𝑆𝑎𝑙𝑒𝑖𝑗𝑡 − ∆𝑅𝑒𝑐𝑖𝑗𝑡 the difference between sales and account receivables, while 𝑃𝑃𝐸𝑖𝑗𝑡 stands for property, plant, and equipment. Total current accruals are calculated as follows: T𝐶𝐴𝑖𝑡 = ∆𝐶𝐴𝑖𝑡 − ∆𝐶𝐿𝑖𝑡 − ∆𝐶𝐴𝑆𝐻𝑖𝑡 + ∆𝑆𝑇𝐷𝐸𝐵𝑇𝑖𝑡 where ∆𝐶𝐴𝑖𝑡 = change in current assets; ∆𝐶𝐿𝑖𝑡 = change in current liabilities; ∆𝐶𝐴𝑆𝐻𝑖𝑡 = change in cash and cash equivalents; and ∆𝑆𝑇𝐷𝐸𝐵𝑇𝑖𝑡 = change in debt included in current liabilities. All items are deflated by the previous-year total assets. Discretionary earnings (DE) are proxied by the residual obtained from the model, while nondiscretionary earnings (NDE) refer to the sum of the predictive value from the model and operating cash flows deflated by total assets. The previous literature suggests that short-sale constraints should be harsher for stocks with lower institutional ownership (D’Avolio (2002), Nagel (2005), Hirshleifer et al. (2011), Stambaugh et al. (2015)). In particular, following Nagel (2005), we measure the degree of shortsale constraints a firm faces by residual institutional ownership, which is the size-adjusted
2
We closely follow the procedure introduced in the I/B/E/S Manual provided by Wharton Research Data Services (WRDS), “A Note on Recreating Summary Statistics from Detail History”. 9
institutional ownership. Residual institutional ownership (RIO) denotes the residual from the following regression: 𝑙𝑜𝑔𝑖𝑡(𝐼𝑂𝑖t ) = 𝛾𝑖0 + 𝛾𝑖1 log(𝑠𝑖𝑧𝑒𝑖𝑡 ) + 𝛾𝑖2 log(𝑠𝑖𝑧𝑒𝑖𝑡 )2 + ε𝑖𝑡 , where 𝐼𝑂𝑖𝑡 is the fraction of shares outstanding held by institutional investors, and log(𝑠𝑖𝑧𝑒𝑖𝑡 ) refers to the log of market capitalization.
3.2. Sample Characteristics
Panel A and B of Table 1 present summary statistics and correlation coefficients of the variables used in the analysis, respectively. In addition to DISP, DE, and NDE, income before extraordinary items standardized by the previous-year assets (EARNINGS), percentage of the sum of outstanding shares held by institutional owners adjusted for firm size (RIO), market beta (𝛽), natural log of firms’ size (SIZE), and book-to-market ratio are analyzed. All items are winsorized at the 1% level. As suggested by the previous literature (Chan et al. (2001)), the correlation between DE and NDE is as high as -0.58 and highly significant. Also, both discretionary and nondiscretionary earnings are significantly correlated with earnings, with the correlation coefficient standing at 0.045 and 0.543, respectively. It is also noteworthy in that DISP’s correlations with DE and NDE are in the opposite direction. DISP is positively correlated with DE, with the correlation coefficient standing at 0.004, but is negatively correlated with NDE (correlation coefficient=-0.02). The preliminary results are consistent with the previous literature that states the level of accruals increases analysts’ forecast errors (Teoh and Wong (2001), Bradshaw et al. (2001), Drake and Myers (2011)). The significantly negative correlation between analyst forecast dispersion and
10
institutional ownership, with a correlation coefficient of -0.021, also closely matches prior findings (Nagel (2005)).
4. Empirical Analysis 4.1. Discretionary Earnings and the Analyst Forecast Dispersion Anomaly
In this section, we investigate how earnings information affects the relation between the analyst forecast dispersion and stock returns. Table 2 presents average monthly returns of the 25 portfolios, excess returns in Panel A and benchmark-adjusted returns in Panel B, formed by the following approach. At the end of each month, 5x5 portfolios are sorted independently according to DISP and DE and held for a month. The DISP breakpoints are decided by the most recent monthly dispersion available at the formation period. On the other hand, according to Sloan (1996), starting from four months after the previous fiscal yearend t, DE of year t are used to decide the breakpoints for a year, until four months following the fiscal yearend t+1. Consistent with Guay et al. (1996) and Chan et al. (2001), both excess and benchmarkadjusted returns of DE portfolios decrease as the degree of DE increases. Also, overall, portfolios with higher analyst forecast dispersion are found with smaller returns in the subsequent months. The difference between the highest and lowest analyst dispersion portfolios’ average excess returns is -0.43% (t-statistic=-2.08) and the difference measured with benchmark-adjusted returns is -0.54% (t-statistic=-3.74). The result lends support to the previous literature on the analyst forecast dispersion anomaly, as noted by Diether et al. (2002), Johnson (2004), and Barron et al. (2009).
11
However, the degree of dispersion anomaly varies along the amount of discretionary earnings. In the portfolios at the highest DE quintile, the negative relation between analyst dispersion and stock returns is prevalent and monotonic. Also, the difference between the highest and lowest analyst dispersion portfolios’ average excess returns is -0.61% (t-statistic=-2.27). On the contrary, the significant negative relation is not maintained at every DE quintile. At the lower four quintiles, the excess return differences between the highest and lowest dispersion portfolios are lower than that measured with all stocks in the sample, and mostly insignificant or significant only at the 10%. At the lowest DE quintile, the negative relation is neither monotonic nor significant, yielding an excess return difference of -0.20% (t-statistic=-0.80) between the highest and lowest dispersion portfolios. While the benchmark-adjusted return differences are mostly significant at every quintile, the magnitude of the return difference is still highest at the highest DE quintile and smallest at the lowest DE quintile. The results are in support of our first hypothesis that the analyst dispersion anomaly increases as the discretionary earnings information increases. The different patterns of the analyst forecast dispersion anomaly along discretionary earnings suggest that the information content in earnings affects the relation between analyst forecast dispersion and stock returns. According to Bradshaw et al. (2001) and Drake and Myers (2009), as more non-cash flows component or managers’ discretionary items are included in the earnings information, there is a higher chance that analysts make more inaccurate and overoptimistic forecasts. Thus, it becomes more likely higher discretionary earnings coincide with more optimistic views among analysts and investors, leading to greater overpricing and becoming a source of the analyst dispersion puzzle.
4.2. Nondiscretionary Earnings and the Analyst Forecast Dispersion Anomaly 12
As a subsequent analysis, we examine the effect of nondiscretionary earnings on the analyst forecast dispersion-stock returns relation. Table 3 displays average monthly returns of the 25 portfolios, excess returns in Panel A and benchmark-adjusted returns in Panel B, constructed according to the following method. Similar to the portfolio construction in the previous section, at the end of each month, 5x5 portfolios are sorted independently according to DISP and NDE and held for a month. The DISP and NDE breakpoints are decided by the same approach as above. Also consistent with Guay et al. (1996) and Chan et al. (2001), both excess and benchmark-adjusted returns of NDE portfolios increase as the degree of NDE increases, which is an opposite pattern from Table 2. The negative relation between analyst forecast dispersion and subsequent-period return also holds, with the difference between the highest and lowest analyst dispersion portfolios’ average excess returns and benchmark-adjusted returns marking -0.43% (tstatistic=-2.00) and -0.51% (t-statistic=-3.38), respectively. As high discretionary items in earnings are found to lead to overly optimistic analyst forecasts and lower nondiscretionary coincide with higher discretionary earnings, a stronger analyst forecast dispersion anomaly is expected in the lower nondiscretionary earnings portfolios. As expected, the analyst dispersion-return relation is not same at every NDE quintile, but Table 3 shows a different pattern than what is found in Table 2. In the portfolios at the lower NDE quintiles, the negative relation between analyst dispersion and stock returns is prevalent and almost monotonic. In Panel A, At the lowest and 2nd lowest quintiles, the differences between the highest and lowest analyst dispersion portfolios’ average excess returns are -0.44% (t-statistic=-1.65) and -0.50% (t-statistic=-2.01), respectively. Also, the 3rd lowest quintile shows a monotonically decreasing dispersion-return pattern, and the excess return difference is as high as -0.45% (t13
statistic=-2.19). However, the significant negative relation is not maintained at the lower two NDE quintiles. The excess return differences between the highest and lowest dispersion portfolios are as low as -0.09% and statistically insignificant. In Panel B, the relations between analyst forecast dispersion and the benchmark-adjusted return differences show a pattern similar to what is found with excess returns. The lower three NDE quintiles show a significant negative difference in benchmark-adjusted returns between the highest and lowest analyst dispersion portfolios, while the higher two NDE quintiles do not show such a pattern. Thus, the results altogether support Hypothesis 1b that the analyst dispersion anomaly decreases as the nondiscretionary earnings information increases. The different patterns of the analyst forecast dispersion anomaly along nondiscretionary earnings can be another example that shows the role of earnings information on the relation between analyst forecast dispersion and stock returns. As less cash flows or fundamental component, there is a higher chance of less dispersion in analyst forecasts, which is also supported by the correlation coefficient in Table 1. Thus, it becomes more likely lower nondiscretionary earnings coincide the greater analyst dispersion anomaly.
4.3. Between Discretionary Earnings and the Analyst Forecast Dispersion Anomaly
In this section, we investigate the relation between earnings information and the analyst forecast dispersion anomaly more thoroughly. In fact, there can be many possible channels between earnings information and the forecast dispersion anomaly. Possibly, firms with more discretionary earnings are found with greater dispersion of analyst forecasts due to more inaccurate and overly optimistic forecasts (Bradshaw et al. (2001); Drake and Myers (2009)), and for such 14
firms greater reduction in the subsequent returns can be inevitable. Another possible reason is that overly optimistic forecasts driven by higher discretionary earnings were followed by investors’ overvaluation and reflected in prices. In this case, overvaluation should be prevalent only among those with high discretionary earnings. On the other hand, it is also likely that higher discretionary earnings and higher dispersion deteriorated liquidity of the stock as information from both managers and analysts became less credible (Sadka and Scherbina (2007)), which led to a greater anomaly among those stocks. To examine these hypotheses, we focus on the level of forecast dispersion, forecast error, and liquidity of the portfolios sorted by discretionary earnings and dispersion. Specifically, we sort the stocks first based on discretionary earnings, and then on forecast dispersion so we may be able to observe the pattern of each proxy along the dispersion portfolio, and compare the patterns along the discretionary earnings. The forecast dispersion is defined the same as DISP, and the signed forecast error is defined as signed |mean forecast EPS – actual EPS/actual EPS|. If mean forecast EPS > actual EPS, the proxy is positive; otherwise, the proxy is negative. Bid-ask spread is the monthly average of the bid-ask spreads at the end of the day. Table 4 presents average monthly forecast dispersions, signed forecast errors, and bid-ask spread of portfolios sorted by discretionary earnings (DE) and analysts’ earnings forecast dispersions (DISP) in Panel A, B, and C, respectively. In Panel A, we find that the increase of dispersion along the dispersion portfolio is universal across all discretionary earnings portfolios. However, the amount of increase is higher among those with lower discretionary earnings than higher discretionary earnings. Also, the average dispersion is higher in the lowest discretionary earnings portfolio than the highest portfolio. This is contrary to Hypothesis 2a that the average dispersion and the increase of dispersion are higher among the stocks with higher discretionary 15
earnings. In other words, although higher forecast dispersion and higher discretionary earnings can be coincidental, the greater dispersion anomaly for high discretionary earning stocks is not likely to be caused by the sks direct relation between discretionary earnings by management and dispersion among analysts. On the other hand, Panel B shows not only that the increase of analyst dispersion and analyst forecast error are coincidental across all discretionary earnings portfolios but also that the high dispersion-high forecast error link becomes more severe as discretionary earnings goes higher. Moreover, the average forecast error is higher for stocks with higher discretionary earnings. Such evidence indicates that managers’ increase of discretionary earnings lead to overly optimistic forecasts by analysts and also investors’ overvaluation of such stocks. As the high dispersion-high forecast error link is more prevalent among high DE stocks, subsequent return reversal should be more prevalent among them as well. In Panel C, however, the dispersion-illiquidity relation remains unchanged along discretionary earnings. It is also contrary to Hypothesis 2c that the average dispersion and the increase of dispersion are higher among the stocks with higher discretionary earnings. In other words, the greater dispersion anomaly for high dispersion earning stocks is unlikely to be driven by higher illiquidity of high discretionary earnings stocks. As our empirical results support Hypothesis 2b, not 2a and 2c, we may say that investors overprice stocks with overly optimistic forecasts, which should be more severe for firms have higher discretionary earnings.
4.4. Discretionary Earnings, Short-sale Impediments, and the Analyst Forecast Dispersion Anomaly
16
The objective of this section is to further find the source of the different pattern in the analyst forecast dispersion anomaly along discretionary earnings. As suggested by Bradshaw et al. (2001) and Drake and Myers (2011) and the former section, higher discretionary items in earnings make analysts announce more overly optimistic forecasts, and lead to overly optimistic opinions among investors and to overpricing. Then, the significant forecast dispersion anomaly among the highest discretionary earnings portfolios should hold or amplify among stocks that face short-sale impediments. According to Miller (1977) and Stambaugh et al. (2015), there is an arbitrage asymmetry in mispriced stocks, as investors find it easier to take a long position in than to take a short position due to short-sale limits. If so, pessimistic valuation is less likely to be arbitraged and overpricing will prevail. Thus, if high discretionary earnings lead to high analyst dispersion in a group of stocks and those stocks face short-sale impediments, overpriced stocks will see a decrease in returns in the following months, which suggests a possible explanation to the analyst forecast anomaly. Table 5 presents average monthly excess returns of portfolios are double-sorted by analysts’ earnings forecast dispersions (DISP) and discretionary earnings (DE) within the subsample of low and high institutional ownership. Panel A and B display low and high institutional ownership subsamples, respectively, and Panel C shows the difference between low and high residual institutional ownership (RIO) subsamples. It is noteworthy that the overall analyst forecast dispersion anomaly is only prevalent among low institutional ownership stocks, but not among high institutional ownership stocks, which is consistent with the findings by Nagel (2005) and Boehme et al. (2006). As predicted, the analyst dispersion anomaly at the high DE quintile is more significant and stronger in the low institutional ownership subsample. In Panel A, at the highest DE quintile, 17
the average return difference between the highest and the lowest analyst dispersion portfolios is 0.89% (t-statistic=-2.76). On the contrary, the significant negative relation is not maintained at every DE quintile. At the lower four quintiles, the excess return differences between the highest and lowest dispersion portfolios are lower than that measured with all stocks in the sample. At the lowest DE quintile, the average return difference is as low as -0.48% (t-statistic=-1.62). Compared to the result reported with all sample stocks in Table 2, it is noteworthy that firms with high dispersion and high discretionary earnings are more strongly negatively related to stock returns, when faced with short-sale impediments. Thus, it is possible to say that the overpricing created by managers’ stronger discretion in earnings and analysts’ overly optimistic forecasts persists and amplifies via short-sale constraints, which becomes the source of the analyst forecast dispersion anomaly. On the other hand, in the high institutional ownership subsample, the significant negative relation between analyst forecast dispersion and stock returns is found at none of the DE quintiles. In Panel B, even at the highest DE quintile, the average return difference between the highest and lowest DISP portfolios is -0.10 (t-statistics=-0.41). Surprisingly, it is smaller than the degree of return difference reported in most of the lower DE quintiles. Hence, it can be noted that overpricing in the higher DE and higher DISP does not tend to persist but is more easily resolved if shorting is less of a problem. The above results are more clearly shown in Panel C, in which the anomaly difference at the highest DE quintile between low and high RIO samples is as high as -0.79 (t-statistic=-2.66). The anomaly difference at the lowest DE quintile between low and high RIO samples, on the other hand, is insignificant. In sum, the results are in support of Hypothesis 3a that the increasing effect
18
of discretionary earnings on the analyst dispersion anomaly becomes greater with short-sale constraints.
4.5. Nondiscretionary Earnings, Short-sale Impediments, and the Analyst Forecast Dispersion Anomaly
This section is devoted to finding the role of short-sale constraints on the different patterns in the analyst forecast dispersion anomaly along nondiscretionary earnings. Considering the negative relation between discretionary and nondiscretionary earnings, an opposite pattern from Table 5 is expected along nondiscretionary earnings portfolios. In other words, the significant forecast dispersion anomaly among the lowest nondiscretionary earnings portfolios should remain intact or amplify among stocks that face short-sale impediments. As low nondiscretionary earnings are correlated with high analyst dispersion in a group of stocks, if those stocks face short-sale impediments, their returns should decrease in the following months, which constitutes another possible explanation to the analyst forecast anomaly. Table 6 presents average monthly excess returns of portfolios double-sorted by analysts’ earnings forecast dispersions (DISP) and nondiscretionary earnings (NDE) within the subsample of low and high institutional ownership. Panel A and B display low and high institutional ownership subsamples, respectively, and Panel C shows the difference between low and high institutional ownership subsamples. Again, consistent with the previous literature (Nagel (2005)), the overall analyst forecast dispersion anomaly is only prevalent among low institutional ownership stocks, but not among high institutional ownership stocks.
19
Consistent with the prediction, the analyst dispersion anomaly at the lower NDE quintile is more significant and stronger in the low institutional ownership subsample. In Panel A, at the lowest NDE quintile, the average return difference between the highest and the lowest analyst dispersion portfolios is -0.95% (t-statistic=-3.22). On the contrary, at the two highest NDE quintiles, the average return differences are as low as -0.53% (t-statistic=-2.45) and -0.28 (tstatistic=-1.13), respectively. It is also noteworthy that the pattern found in Table 3, the higher dispersion anomaly at the lower NDE portfolios, becomes more severe with short-sale impediments. Hence, it is also possible to say that the short-sale constraints add to overpricing created by less cash-flow or fundamental-related items in earnings and analysts’ overly optimistic forecasts and constitute a negative relation with subsequent stock returns. It can also be counted as one of the sources of the analyst forecast dispersion anomaly. Meanwhile, in the high institutional ownership subsample, the significant negative relation between analyst forecast dispersion and stock returns becomes less significant or even disappears. In Panel B, at the lowest NDE quintile, the average return difference between the highest and lowest DISP portfolios is -0.45 (t-statistics=-1.83). More surprisingly, at the higher NDE quintiles, the average return differences turn positive, which means that when there is less severe arbitrage asymmetry and stronger cash flows or fundamental-related information in earnings, the analyst forecast dispersion anomaly is resolved. In sum, the results are also in support of Hypothesis 3b that the decreasing effect of nondiscretionary earnings on the analyst dispersion anomaly becomes greater with short-sale constraints.
4.6. Robustness Check: Cross-sectional Regressions
20
The focus of this section is to check whether a robust relation holds between each earnings component and the analyst dispersion anomaly. Following Fama and MacBeth (1973), we run a two-stage regression and compute time-series averages (𝜆̂̅𝑡 ) of monthly cross-sectional regression coefficient estimates from the following regression model: 𝑟𝑖𝑡 − 𝑟𝑓𝑡 = 𝜆0𝑡 + 𝜆1𝑡 DISP𝑖𝑡 + 𝜆2𝑡 earnings info𝑖𝑡 + 𝜆3𝑡 (DISP𝑖𝑡 × earnings info𝑖𝑡 ) + 𝜆4𝑡 (𝐶𝑜𝑛𝑡𝑟𝑜𝑙) + 𝜀𝑖𝑡 , where 𝑟𝑖𝑡 − 𝑟𝑓𝑡 is the excess return, DISP𝑖𝑡 is Analyst Forecast Dispersion, and earnings info𝑖𝑡 refers to either DE or NDE. As control variables, market beta (𝛽), log of market capitalization (LN(SIZE)), and book-to-market ratio (BM) are used. ̅̅̅̅̅ Adj 𝑅 2 is the time-series average of monthby-month cross-sectional regression’s adjusted 𝑅 2 . According to Table 7, the interaction between discretionary earnings and analyst forecast dispersion is negatively associated with stock returns, with the coefficient scoring -0.91% (tstatistic=-2.07). Even if control variables are added to the model, the coefficient of the interactive term remains roust at -0.88% (t-statistic=-2.12). Consistent with the previous analysis, both DISP and DE are significantly negative at least at the 10% level. While the interactive term between DISP and NDE are insignificant, both DISP and DE’s relation with stock returns remain intact.
5. Conclusion This study investigates the effect of earnings components on the anomalous negative relation between analyst forecast dispersion and future stock returns. Based on Guay’s (1996) decomposition of earnings and previously found links between discretionary accruals and analyst dispersion, we seek to find contrasting effects of discretionary and nondiscretionary earnings on 21
the forecast dispersion anomaly. According to the empirical analysis, we find that higher discretionary or lower nondiscretionary earnings are associated with and amplify the analyst forecast dispersion anomaly. If firms are found with lower discretionary earnings or higher nondiscretionary earnings, the negative dispersion-return relation turns insignificant or almost disappears. The more prevalent dispersion anomaly among higher discretionary earnings stocks coincides with higher forecast error, but not with higher dispersion and higher illiquidity. It indicates that investors credulously follow analysts’ overly optimistic forecasts and it is more severe for firms have higher discretionary earnings. Moreover, the relation between discretionary (nondiscretionary) earnings and the forecast dispersion anomaly is more prevalent in subsamples with low institutional ownership than those with high institutional ownership, indicating that overpricing in high dispersion-high discretionary (low nondiscretionary) earnings stocks becomes more serious with short-sale constraints. Such results indicate that discretionary and nondiscretionary earnings stand behind the analyst forecast dispersion anomaly, but plays an opposite role.
22
References
Avramov, D., T. Chordia, G. Jostova, and A. Philipov, 2009, “Dispersion in analysts' earnings forecasts and credit rating,” Journal of Financial Economics 91, 83-101. Barron, O. E., M. H.s Stanford, and Y. Yu, 2009, “Further evidence on the relation between analysts' forecast dispersion and stock returns,” Contemporary Accounting Research 26, 329-357. Boehme, R. D., Danielsen, B. R., & Sorescu, S. M. 2006, "Short-sale constraints, differences of opinion, and overvaluation," Journal of Financial and Quantitative Analysis, 41(02), 455-487 Bradshaw, M. T., Richardson, S. A., & Sloan, R. G., 2001, "Do analysts and auditors use information in accruals?" Journal of Accounting Research, 39, 45–74. Carhart, M. M., 1997, “On persistence in mutual fund performance,” Journal of Finance, 52(1), 57-82. Chan, K., Chan, L. K., Jegadeesh, N., & Lakonishok, J., 2001, "Earnings quality and stock returns," (No. w8308). National bureau of economic research. Collins, D. W., Gong, G., & Hribar, P., 2003, "Investor sophistication and the mispricing of accruals," Review of Accounting Studies, 8, 251–276. D’Avolio, G., 2002, "The market for borrowing stock," Journal of Financial Economics, 66, 271–306. DeAngelo, L.E., 1988, “Managerial competition, information costs, and corporate governance: The use of accounting performance measures in proxy contests,” Journal of Accounting and Economics, 10(1), pp.3-36. Dechow, P. M., Sloan, R. G., & Sweeney, A. P., 1995, "Detecting earnings management," Accounting review, 193-225. Diether, K. B., C. J. Malloy, and A. Scherbina, 2002, “Differences of opinion and the crosssection of stock returns,” Journal of Finance 57, 2113-2141 Drake, M. S., & Myers, L. A., 2011, "Analysts’ accrual-related over-optimism: do analyst characteristics play a role?" Review of Accounting Studies, 16(1), 59-88. Fama, E. F., and K. R. French, 1993, “Common risk factors in the returns on stocks and bonds,” Journal of Financial Economics 33, 3-56. Fama, E., and J. D. MacBeth, 1973, “Risk, return and equilibrium: Empirical tests,” Journal of Political Economy, 607-636. Fang, V.W., Huang, A.H. and Karpoff, J.M., 2016, “Short selling and earnings management: A controlled experiment,” Journal of Finance, 71(3), 1251-1294.
23
Guay, W. R., Kothari, S. P., & Watts, R. L., 1996, "A market-based evaluation of discretionary accrual models," Journal of Accounting Research, 83-105. Healy, P. M., 1985, "The Effect of Bonus Schemes on Accounting Decisions," Journal of Accounting and Economics, 85-107. Hirshleifer, D., Teoh, S. H., & Yu, J. J., 2011, "Short arbitrage, return asymmetry, and the accrual anomaly," Review of Financial Studies, 24(7), 2429-2461. Johnson, T. C., 2004, “Forecast dispersion and the cross section of expected returns,” Journal of Finance 59, 1957-1978. Jones, J. J., 1991, "Earnings management during import relief investigations," Journal of Accounting Research, 193-228. Kim, J. S., Ryu, D., & Seo, S. W., 2014, "Investor sentiment and return predictability of disagreement," Journal of Banking & Finance, 42, 166-178. Lobo, G. J., Song, M., & Stanford, M., 2012, "Accruals quality and analyst coverage," Journal of Banking & Finance, 36(2), 497-508. Mashruwala, C., Rajgopal, S., & Shevlin, T., 2006, "Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and transaction costs," Journal of Accounting & Economics, 42, 3–33. Miller, E. M., 1977. “Risk, uncertainty, and divergence of opinion,” Journal of Finance 32, 1151-1168. Nagel, S., 2005, "Short sales, institutional investors and the cross-section of stock returns," Journal of Financial Economics 78, 277–309. Payne, J. L., and W. B. Thomas, 2003, “The implications of using stock-split adjusted I/B/E/S data in empirical research,” Accounting Review 78, 1049-1067. Peng, E.Y., Yan, A. and Yan, M., 2016, “Accounting accruals, heterogeneous investor beliefs, and stock returns,” Journal of Financial Stability, 24, pp.88-103. Sadka, R. and A. Scherbina, 2007, “Analyst Disagreement, Mispricing, and Liquidity,” Journal of Finance 62, 2367–2403. Sloan, R. G., 1996. “Do stock prices fully reflect information in accruals and cash flows about future earnings?” Accounting Review 71, 289-315. Stambaugh, R. F., Yu, J., & Yuan, Y., 2015, "Arbitrage asymmetry and the idiosyncratic volatility puzzle," The Journal of Finance. Teoh, S. H., & Wong, T. J., 2002, "Why new issues and high-accrual firms underperform: The role of analysts’ credulity," Review of Financial Studies, 15, 869–900.
24
Table 1 Summary Statistics of Portfolios Sorted by Discretionary and Non-discretionary Earnings Information This table presents summary statistics (Panel A) and correlation coefficients (Panel B) of variables used in the analysis. The dispersion in analyst earnings forecasts, DISP, is measured by standardizing the standard deviation of analysts’ currentfiscal-year EPS forecasts by the absolute value of mean forecast. At the end of every month, we compute standard deviation and average based on all the outstanding analyst forecasts for the same forecast period. The forecast dispersion is updated on a monthly basis. Discretionary earnings (DE) and nondiscretionary earnings (NDE) are based on the following regression model, T𝐶𝐴𝑖𝑗𝑡 = 𝛽𝑗𝑡 (
1
𝐴𝑇𝑖𝑗𝑡−1
) + 𝛾𝑗𝑡 (∆𝑆𝑎𝑙𝑒𝑖𝑗𝑡 − ∆𝑅𝑒𝑐𝑖𝑗𝑡 ) + 𝛿𝑗𝑡 𝑃𝑃𝐸𝑖𝑗𝑡 + 𝜀𝑖𝑗𝑡 ,
where 𝑇𝐶𝐴𝑖𝑗𝑡 indicates the total current accruals of Firm i of Industry j, in calendar year t, 𝐴𝑇𝑖𝑗𝑡 total assets, and ∆𝑆𝑎𝑙𝑒𝑖𝑗𝑡 − ∆𝑅𝑒𝑐𝑖𝑗𝑡 the difference between sales and account receivables, while 𝑃𝑃𝐸𝑖𝑗𝑡 stands for property, plant, and equipment. All items are scaled by the previous-year total assets. Discretionary earnings (DE) is proxied by the residual from the model, while nondiscretionary earnings (NDE) is the sum of the predictive value from the model and operating cash flow. Residual institutional ownership (RIO) denotes the residual from the following regression: 𝑙𝑜𝑔𝑖𝑡(𝐼𝑂𝑖t ) = 𝛾𝑖0 + 𝛾𝑖1 log(𝑠𝑖𝑧𝑒𝑖𝑡 ) + 𝛾𝑖2 log(𝑠𝑖𝑧𝑒𝑖𝑡 )2 + ε𝑖𝑡 , where 𝐼𝑂𝑖𝑡 is the fraction of shares outstanding held by institutional investors, and log(𝑠𝑖𝑧𝑒𝑖𝑡 ) refers to the log of market capitalization. Also presented are income before extraordinary items standardized by the previous-year assets (EARNINGS), percentage of the sum of outstanding shares held by institutional owners adjusted for firm size (RIO), market beta (𝛽), natural log of firms’ size (SIZE), and book-to-market ratio (BM). All items are winsorized at the 1% level. The sample period is January 1983 to December 2014. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.
Min Mean Median Std. Dev. Max DISP DE NDE Earnings RIO beta ln(size) BM
DISP DE NDE Panel A. Summary Statistics 0.000 -2.292 -2.219 0.201 0.011 0.041 0.06 -0.001 0.064 0.474 0.273 0.32 5.849 2.008 1.798 Panel B. Correlation Coefficients 1 0.004*** 1 [0.001] -0.02*** -0.581*** 1 [0.000] [0.000] -0.021*** 0.045*** 0.543*** [0.000] [0.000] [0.000] -0.021*** 0.002 0.018*** [0.000] [0.118] [0.000] 0.106*** -0.011*** -0.058*** [0.000] [0.000] [0.000] -0.2*** -0.018*** 0.055*** [0.000] [0.000] [0.000] 0.054*** 0.006*** -0.002** [0.000] [0.000] [0.04]
Earnings
RIO
beta
ln(size)
BM
-1.305 0.024 0.054 0.202 0.425
-8.246 1.328 2.507 4.104 11.246
0.139 1.221 1.143 0.571 3.162
16.046 20.248 20.171 1.74 24.658
0.000 0.551 0.458 0.422 2.459
1 0.025*** [0.000] -0.092*** [0.000] 0.054*** [0.000] 0.009*** [0.000]
1 0.034*** [0.000] -0.093*** [0.000] -0.02*** [0.000]
1 -0.122*** [0.000] -0.075*** [0.000]
1 -0.093*** [0.000]
1 -
25
Table 2 Average Monthly Returns of Portfolios Sorted by Analysts’ Earnings Forecast Dispersions and Discretionary Earnings This table presents average monthly returns of portfolios sorted by discretionary earnings (DE) and analysts’ earnings forecast dispersions (DISP). Panel A presents excess returns, and Panel B displays the Carhart (1997) 4 factor-adjusted returns. At the end of each month, 5x5 portfolios are sorted independently according to DISP and DE and held for a month. The DISP breakpoints are decided by the most recent monthly dispersion available at the formation period. On the other hand, according to Sloan (1996), starting from four months after the previous fiscal yearend t, DE of year t are used to decide the breakpoints for a year, until four months following the fiscal yearend t+1. The sample period is January 1983 to December 2014. Numbers in parentheses indicate autocorrelation-adjusted 𝑡-statistics. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.
DE1 DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
1.06*** (3.65) 1.08*** (3.19) 0.8** (2.32) 0.75** (2.00) 0.86** (2.00) -0.20 (-0.80) 0.88*** (2.51) 0.35*** (2.55) 0.31*** (2.51) -0.01 (-0.05) -0.04 (-0.25) 0.00 (0.03) -0.34* (-1.68) 0.09 (0.95)
DE2
DE3 DE4 DE5 Panel A. Excess Return 1.03*** 0.89*** 0.92*** 0.78*** (4.13) (3.76) (3.84) (2.61) *** *** *** 0.94 0.90 0.85 0.56* (3.31) (3.49) (3.22) (1.71) 0.85*** 0.79*** 0.84*** 0.51 (2.73) (2.78) (2.81) (1.37) 0.96*** 0.85*** 0.7** 0.4 (2.74) (2.58) (2.13) (0.99) 0.63 0.57 0.61* 0.17 (1.60) (1.54) (1.65) (0.4) -0.41* -0.32 -0.31 -0.61** (-1.74) (-1.41) (-1.37) (-2.27) *** *** *** 0.87 0.82 0.83 0.48 (2.87) (2.98) (2.87) (1.39) Panel B. Benchmark-adjusted Return 0.27** 0.16 0.20* 0.04 (2.32) (1.4) (1.84) (0.24) 0.15 0.16* 0.14 -0.13 (1.29) (1.7) (1.38) (-1.11) 0.06 0.04 0.13 -0.21* (0.52) (0.45) (1.24) (-1.73) 0.13 0.05 -0.10 -0.31*** (1.24) (0.4) (-0.82) (-2.64) -0.25* -0.28** -0.20 -0.61*** (-1.82) (-2.27) (-1.55) (-4.29) -0.54*** -0.44*** -0.41*** -0.65*** (-2.97) (-2.58) (-2.41) (-2.80) 0.05 0.06 0.07 -0.25*** (0.70) (0.76) (0.86) (-3.23)
26
P5-P1
OVERALL
-0.28** (-2.2) -0.45*** (-3.27) -0.29** (-2) -0.34** (-2.12) -0.69*** (-3.5)
0.94*** (3.75) 0.88*** (3.16) 0.78*** (2.52) 0.73** (2.13) 0.51 (1.33) -0.43** (-2.08)
-0.4*** (-4.05) -0.31*** (-2.47) -0.41*** (-2.85) -0.20 (-1.48) -0.28* (-1.71) -0.62*** (-3.25)
-0.34*** (-3.43)
0.20** (2.07) 0.14* (1.88) 0.01 (0.10) -0.05 (-0.71) -0.34*** (-3.84) -0.54*** (-3.74)
Table 3 Average Monthly Returns of Portfolios Sorted by Analysts’ Earnings Forecast Dispersions and Nondiscretionary Earnings This table presents average monthly returns of portfolios sorted by nondiscretionary earnings (NDE) and analysts’ earnings forecast dispersions (DISP). Panel A presents excess returns, and Panel B displays the Carhart (1997) 4 factor-adjusted returns. At the end of each month, 5x5 portfolios are sorted independently according to DISP and NDE and held for a month. The DISP breakpoints are decided by the most recent monthly dispersion available at the formation period. On the other hand, according to Sloan (1996), starting from four months after the previous fiscal yearend t, NDE of year t are used to decide the breakpoints for a year, until four months following the fiscal yearend t+1. The sample period is January 1983 to December 2014. Numbers in parentheses indicate autocorrelation-adjusted 𝑡-statistics. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.
NDE1 DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
0.53* (1.74) 0.5 (1.39) 0.51 (1.26) 0.43 (1.01) 0.10 (0.21) -0.44* (-1.65) 0.34 (0.86) -0.20 (-1.53) -0.23* (-1.8) -0.17 (-1.21) -0.29** (-2.04) -0.67*** (-4.62) -0.47** (-2.15) -0.38*** (-3.91)
NDE2
NDE3 NDE4 NDE5 Panel A. Excess Return 0.92*** 0.93*** 0.92*** 0.99*** (3.90) (4.00) (3.68) (3.49) 0.82*** 0.81*** 0.92*** 0.97*** (3.22) (3.14) (3.21) (2.95) 0.82*** 0.75*** 0.74*** 0.83*** (2.81) (2.7) (2.38) (2.4) * *** *** 0.63 0.76 0.91 0.79** (1.88) (2.44) (2.68) (2.14) 0.43 0.48 0.82** 0.91** (1.11) (1.36) (2.31) (2.24) ** ** -0.5 -0.45 -0.09 -0.09 (-2.01) (-2.19) (-0.48) (-0.42) 0.71*** 0.78*** 0.87*** 0.9*** (2.41) (2.86) (2.99) (2.69) Panel B. Benchmark-adjusted Return 0.20* 0.20* 0.24** 0.33*** (1.79) (1.78) (2.09) (2.72) 0.12 0.07 0.18* 0.3** (1.23) (0.68) (1.72) (2.31) 0.08 -0.01 -0.03 0.08 (0.94) (-0.11) (-0.32) (0.6) -0.16 -0.02 0.17 0.08 (-1.61) (-0.23) (1.45) (0.62) -0.38*** -0.34*** 0.04 0.17 (-3.03) (-2.94) (0.32) (1.13) -0.57*** -0.54*** -0.19 -0.16 (-3.18) (-3.39) (-1.2) (-0.89) -0.05 0.02 0.14* 0.20** (-0.81) (0.22) (1.65) (2.09)
27
P5-P1
OVERALL
0.46*** (3.48) 0.48*** (2.49) 0.32 (1.49) 0.36* (1.82) 0.81*** (3.91)
0.89*** (3.6) 0.83*** (3) 0.72*** (2.34) 0.68** (1.98) 0.46 (1.2) -0.43** (-2)
0.56*** (3.31) 0.54*** (4.19) 0.53*** (2.81) 0.24 (1.26) 0.37* (1.94) 0.84*** (4.20)
0.58*** (3.94)
0.19* (1.95) 0.12* (1.76) -0.01 (-0.21) -0.06 (-0.83) -0.32*** (-3.52) -0.51*** (-3.38)
Table 4. Average Monthly Forecast Dispersion, Signed Forecast Error, and Bid-ask Spread of Portfolios Sorted by Forecast Dispersions and Discretionary Earnings This table presents average monthly forecast dispersions, signed forecast errors, and bid-ask spread of portfolios sorted by discretionary earnings (DE) and analysts’ earnings forecast dispersions (DISP) in Panel A, B, and C, respectively. The signed forecast error is defined as signed |mean forecast EPS – actual EPS/actual EPS|. If Mean forecast EPS > actual EPS, the proxy is positive; otherwise, the proxy is negative. Bid-ask spread is the monthly average of the bid-ask spreads at the end of the day. At the end of each month, 5x5 portfolios are sorted first based on DISP and then based on NDE and held for a month. The DISP breakpoints are decided by the most recent monthly dispersion available at the formation period. On the other hand, according to Sloan (1996), starting from four months after the previous fiscal yearend t, DE of year t are used to decide the breakpoints for a year, until four months following the fiscal yearend t+1. The sample period is January 1983 to December 2014. Numbers in parentheses indicate autocorrelation-adjusted 𝑡-statistics. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.
DE1 DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
0.02*** (21.51) 0.04*** (22.92) 0.08*** (24.09) 0.16*** (30.81) 0.84*** (44.75) 0.82*** (44.85) 0.23*** (41.79) 0.49*** (8.89) 0.52*** (9.98) 0.71*** (10.19) 0.73*** (9.90) 0.96*** (9.54) 0.46*** (3.83) 0.68*** (16.67)
DE2
DE3 DE4 DE5 Panel A. Forecast Dispersion 0.02*** 0.01*** 0.01*** 0.02*** (22.86) (23.14) (24.01) (19.28) *** *** *** 0.03 0.03 0.03 0.04*** (23.36) (25.4) (25.2) (18.94) 0.06*** 0.05*** 0.06*** 0.07*** (23.3) (24.69) (23.53) (19.95) 0.13*** 0.11*** 0.12*** 0.15*** (27.05) (26.55) (27.17) (24.76) *** *** *** 0.70 0.62 0.65 0.78*** (35.00) (36.64) (41.06) (30.24) 0.68*** 0.60*** 0.63*** 0.76*** (35.03) (36.75) (40.82) (30.34) *** *** *** 0.19 0.16 0.17 0.21*** (33.36) (35.10) (39.11) (28.69) Panel B. Signed Forecast Error 0.44*** 0.82*** 0.39*** 0.71*** (4.51) (3.01) (11.06) (6.55) 0.40*** 0.37*** 0.41*** 0.66*** (7.95) (8.27) (11.87) (10.73) 0.62*** 0.55*** 0.56*** 0.82*** (10.99) (10.61) (10.38) (11.63) 0.86*** 0.91*** 0.82*** 1.28*** (15.91) (12.02) (15.78) (7.18) 1.28*** 1.35*** 1.08*** 1.39*** (7.68) (12.49) (15.76) (8.39) 0.84*** 0.53* 0.69*** 0.67*** (5.39) (1.80) (8.77) (3.24) 0.72*** 0.81*** 0.65*** 0.97*** (12.32) (10.33) (20.55) (14.64)
28
P5-P1
OVERALL
-0.00*** (-7.23) -0.00*** (-5.45) -0.01*** (-6.29) -0.02*** (-6.02) -0.06** (-2.59)
0.02*** (22.74) 0.03*** (23.59) 0.06*** (23.33) 0.13*** (28.09) 0.72*** (46.56) 0.70*** (47.16)
-0.02*** (-3.45) 0.22** (2.08) 0.14* (1.73) 0.10 (1.15) 0.55*** (3.00) 0.43** (2.42)
0.29*** (3.89)
0.58*** (6.49) 0.46*** (18.47) 0.65*** (16.96) 0.97*** (18.33) 1.21*** (15.33) 0.63*** (5.39)
***
DISP1 DISP2 DISP3 DISP4 DISP5 P5-P1 Overall
0.27 (14.90) 0.24*** (15.41) 0.23*** (15.23) 0.21*** (15.30) 0.19*** (15.18) -0.07*** (-10.41) 0.23*** (15.39)
***
0.27 (15.12) 0.25*** (15.44) 0.24*** (14.77) 0.23*** (15.33) 0.20*** (15.18) -0.07*** (-11.03) 0.24*** (15.48)
Panel C. Bid-ask Spread 0.26*** 0.26*** (14.96) (15.26) 0.25*** 0.25*** (15.16) (14.98) *** 0.24 0.24*** (15.31) (14.91) 0.23*** 0.23*** (15.17) (14.62) 0.21*** 0.20*** (15.11) (14.74) -0.05*** -0.06*** (-8.16) (-9.12) 0.24*** 0.24*** (15.52) (15.2)
29
0.26*** (15.34) 0.24*** (15.06) 0.23*** (14.56) 0.21*** (14.55) 0.19*** (14.81) -0.07*** (-12.05) 0.23*** (15.27)
-0.00 (-0.32) 0.01 (1.30) 0.01 (0.88) 0.00 (0.61) 0.00 (0.11)
0.00 (0.82)
0.27*** (15.54) 0.25*** (15.58) 0.23*** (15.53) 0.22*** (15.17) 0.20*** (15.28) -0.06*** (-13.34)
Table 5 Average Returns of Portfolios Sorted by Analysts’ Earnings Forecast Dispersions and Discretionary Earnings in Subsamples of High vs. Low Institutional Ownership This table presents average monthly excess returns of portfolios sorted by analysts’ earnings forecast dispersions (DISP) and institutional ownership (RIO) within each quintile of portfolios constructed on discretionary earnings (DE). Panel A and B display low and high IO subsamples, respectively, and Panel C show the difference between low and high IO subsamples. The sample period is January 1983 to December 2014. Numbers in parentheses indicate autocorrelation-adjusted 𝑡-statistics. *** ** , , and * indicate significance at the 1%, 5%, and 10% level, respectively.
DE1 DISP1 DISP5 P5-P1
DISP1 DISP5 P5-P1
DISP1 DISP5 P5-P1
0.83*** (2.66) 0.35 (0.77) -0.48 (-1.62) 1.12*** (3.8) 0.83** (2.04) -0.29 (-1.07) -0.29 (-1.6) -0.48** (-2.06) -0.19 (-0.66)
DE2
DE3 DE4 Panel A. Low RIO 0.96*** 0.82*** 0.82*** (3.98) (3.68) (3.63) 0.42 0.2 0.23 (1.04) (0.53) (0.55) -0.54* -0.62** -0.59** (-1.91) (-2.11) (-2.04) Panel B. High RIO 0.98*** 0.83*** 0.96*** (3.6) (3.16) (3.73) 0.78* 0.75** 0.67* (1.93) (2.05) (1.86) -0.2 -0.08 -0.29 (-0.76) (-0.36) (-1.3) Panel C. Low - High RIO -0.02 -0.01 -0.15 (-0.11) (-0.05) (-1.06) -0.36 -0.55*** -0.45* (-1.57) (-2.38) (-1.96) -0.34 -0.54** -0.3 (-1.38) (-1.99) (-1.19)
30
DE5
Overall
0.72*** (2.35) -0.17 (-0.37) -0.89*** (-2.76)
0.83*** (3.56) 0.2 (0.49) -0.63*** (-2.52)
0.54* (1.78) 0.44 (1.11) -0.1 (-0.41)
0.87*** (3.33) 0.69* (1.9) -0.18 (-0.96)
0.18 (0.99) -0.61*** (-2.5) -0.79*** (-2.66)
-0.05 (-0.5) -0.5*** (-3.55) -0.45*** (-2.95)
Table 6 Average Returns of Portfolios Sorted by Analysts’ Earnings Forecast Dispersions and Nondiscretionary Earnings in Subsamples of High vs. Low Institutional Ownership This table presents average monthly excess returns of portfolios sorted by analysts’ earnings forecast dispersions (DISP) and institutional ownership (RIO) within each quintile of portfolios constructed on nondiscretionary earnings (NDE). Panel A and B display low and high IO subsamples, respectively, and Panel C show the difference between low and high IO subsamples. The sample period is January 1983 to December 2014. Numbers in parentheses indicate autocorrelationadjusted 𝑡-statistics. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.
NDE1 DISP1 DISP5 P5-P1
DISP1 DISP5 P5-P1
DISP1 DISP5 P5-P1
0.42 (1.16) -0.53 (-1.14) -0.95*** (-3.22) 0.66** (2.18) 0.21 (0.54) -0.45* (-1.83) -0.24 (-0.88) -0.74*** (-3.43) -0.51* (-1.72)
NDE2 NDE3 NDE4 Panel A. Low RIO 0.76*** 0.91*** 0.87*** (3.08) (4.15) (3.51) 0.06 0.01 0.59 (0.13) (0.04) (1.6) -0.71*** -0.89*** -0.28 (-2.42) (-3.29) (-1.13) Panel B. High RIO 0.95*** 0.85*** 0.85*** (3.27) (3.27) (3.27) 0.45 0.73** 0.92*** (1.22) (1.97) (2.41) *** -0.5 -0.13 0.07 (-2.42) (-0.58) (0.36) Panel C. Low – High RIO -0.19 0.05 0.02 (-1.08) (0.39) (0.17) -0.4* -0.71*** -0.33 (-1.75) (-3.45) (-1.49) -0.21 -0.77*** -0.35 (-0.78) (-3.29) (-1.48)
31
NDE5
Overall
0.93*** (3.08) 0.4 (0.98) -0.53*** (-2.45)
0.82*** (3.44) 0.03 (0.07) -0.79*** (-3.54)
1.03*** (3.64) 1.07*** (2.63) 0.03 (0.14)
0.88*** (3.37) 0.57 (1.58) -0.21 (-1.11)
-0.11 (-0.71) -0.67*** (-3.09) -0.56*** (-2.41)
-0.11 (-0.71) -0.67*** (-3.09) -0.45*** (-2.96)
Table 7 Time-series Averages of Monthly Cross-sectional Regression Coefficient Estimates This table presents time-series averages ( 𝜆̂̅𝑡 ) of monthly cross-sectional regression coefficient estimates from the following regression model, 𝑟𝑖𝑡 − 𝑟𝑓𝑡 = 𝜆0𝑡 + 𝜆1𝑡 DISP𝑖𝑡 + 𝜆2𝑡 earnings info𝑖𝑡 + 𝜆3𝑡 (DISP𝑖𝑡 × earnings info𝑖𝑡 ) + 𝜆4𝑡 (𝐶𝑜𝑛𝑡𝑟𝑜𝑙) + 𝜀𝑖𝑡 , where 𝑟𝑖𝑡 − 𝑟𝑓𝑡 is the excess return, DISP𝑖𝑡 is Analyst Forecast Dispersion, and earnings info𝑖𝑡 refers to either Discretionary Earnings (DE) or Nondiscretionary Earnings (NDE). As control variables, market beta (𝛽), log of market capitalization (LN(SIZE)), and book-to-market ratio (BM) are used. ̅̅̅̅̅ Adj 𝑅2 is the time-series 2 average of month-by-month cross-sectional regression’s adjusted 𝑅 . The sample period is January 1983 to December 2014. Numbers in parentheses indicate autocorrelation-adjusted 𝑡-statistics. ***, **, and * indicate significance at the 1%, 5%, and 10% level, respectively.
DISP DE
(1) -0.06** (-2.05) -0.44*** (-3.86)
DISP*DE
(2) -0.11* (-1.81) -0.57*** (-3.56) -0.91** (-2.07)
(3) -0.08*** (-3.13) -0.43*** (-4.35)
NDE
Model (4) (5) -0.14*** -0.06*** (-2.75) (-3.74) *** -0.46 (-3.25) -0.88** (-2.12) 0.46*** (4.01)
DISP*NDE 𝜷 LN(SIZE) BM ̅̅̅̅̅ Adj 𝑅 2
0.25%
0.44%
-0.02 (-0.31) 0.18** (2.13) 0.22 (0.82)
-0.01 (-0.33) 0.11 (1.17) 0.22 (0.93)
4.92%
5.21%
32
0.36%
(6) -0.07*** (-3.64)
(7) -0.06*** (-4.36)
(8) -0.07*** (-4.6)
0.46*** (3.78) 0.09 (0.71)
0.49*** (5.14)
-0.01 (-0.22) 0.21*** (2.59) 0.21 (0.79)
0.51*** (5.13) 0.03 (0.39) -0.01 (-0.25) 0.21*** (2.57) 0.22 (0.79)
5.00%
5.05%
0.42%
