SEARCH

SEARCH BY CITATION

Keywords:

  • Post-earnings-announcement drifts;
  • Time-series analysis
  • G14;
  • M41

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

In this paper, we generalize Bernard and Thomas’s [Journal of Accounting and Economics 13 (1990), 305]“delayed response” hypothesis as an explanation of post-earnings-announcement drifts. By applying a modified version of Beveridge and Nelson’s technique of decomposing a time-series process of earnings into permanent and temporary components, we estimate the relative weight to proxy for investor perception on the temporary component of earnings. We then provide evidence that our measure of investor misperception explains post-announcement drifts after controlling for firm size and investor sophistication. These findings reinforce Bernard and Thomas’s [Journal of Accounting and Economics 13 (1990), 305] conjecture that less weight is placed on temporary components of earnings than would be appropriate if earnings processes were well understood, although not zero as Bernard and Thomas implicitly assumed in their portfolio formation rule. Our results also complement Ball and Bartov’s [Journal of Accounting and Economics 21 (1996), 319] result that investors partially, but not fully, adjust for serial correlation in seasonal differences.


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

Several studies have documented a systematic pattern in abnormal returns related to corporate earnings announcements known as the post-earnings-announcement drift. Share prices tend to rise (fall) during the 9 months after the announcement of large positive (negative) quarterly earnings surprises. Fama (1998) describes the post-earnings-announcement drift as an anomaly “above suspicion.” In an article directed toward market professionals, Brennan (1991) describes the work of Bernard and Thomas (1989, 1990) on post-earnings-announcement drifts as “perhaps the most severe challenge to efficient market theorists.”1

Despite repeated attempts, prior research has failed to provide a satisfactory explanation for the drift. Of the three typical explanations for the drift, misestimation of expected returns, methodological shortcomings, and investor underreaction, we believe that investors’ misperception of the earnings process is most important. Building on previous studies, Bernard and Thomas (1990) exploit a simple idea in obtaining evidence that the stock market is inefficient. If investors price expected future earnings based on the perception that earnings follow a seasonal random walk when, in fact, seasonal differences are autocorrelated, then stock returns will display autocorrelations. Such autocorrelations in stock returns give rise to post-announcement drifts in returns on portfolios of firms displaying extreme unexpected earnings.

Ball and Bartov (1996) suggest that investors might be more sophisticated than Bernard and Thomas (1990) suggest by reporting evidence that the market does understand that earnings are seasonally correlated, but underestimates the degree of serial correlation by 50% on average. Soffer and Lys (1999) reconcile the apparent conflict between Bernard and Thomas (1990) and Ball and Bartov (1996) by showing that the degree of serial correlation in earnings impounded in market expectations increases from zero, following an earnings announcement, to 50% immediately before the next announcement. Jacob et al. (2000) also question the extreme form of investor naiveté implied by Bernard and Thomas’s analysis.

Our objectives in the present paper are to generalize previous characterizations of investor perceptions (e.g. Bernard and Thomas, 1990) by allowing for the possibility that investors give some weight to the non-seasonal random walk component of the earnings process, and to assess the extent to which cross-sectional variations in the weight are associated with variations in the absolute magnitude of post-announcement drifts. We begin by applying a modified version of Beveridge and Nelson’s (1981) model of decomposing time-series processes of earnings into permanent and temporary components (see Kormendi and Lipe, 1987; Lee, 1996a,b).2 The permanent component captures the seasonal random walk element of quarterly earnings, while the temporary component picks up the autocorrelation in seasonal differences.3

We measure investor perception on the temporary component of earnings with the relative weight on the temporary component. Therefore, the weight is decreasing in the level of investors’ misperception. Our model, which introduces a relative weight to be placed on the temporary component, is very general. Special cases include efficient pricing wherein investors fully perceive the temporary component in forming their expectations (a relative weight of one) and Bernard and Thomas’s (1990) characterization wherein investors completely overlook that component (a relative weight of zero).

We estimate this relative weight through a two-stage procedure. The first stage involves estimation of the earnings process. Residuals from the first-stage estimates of the earnings process and abnormal returns are then used to estimate both the relative weight placed on the temporary component and the earnings response coefficient. Having estimated relative weights for a sample of firms meeting our data requirements, we investigate whether post-announcement drifts are larger (smaller) in absolute magnitude for firms with smaller (larger) weights.

On average, we find that relative weights are between zero and one for the full sample, and closer to zero for small firms than for large firms. These results are consistent with investor misperceptions more likely being associated with small firms than with large firms. When we use relative weights to partition the sample, we find that post-announcement drifts are larger for firms with relative weights below the median than above, and that the difference is statistically significant for firms with positive earnings surprises. When the sample is first partitioned by firm size and repartitioned by relative weights, although post-announcement drifts are more pronounced for small firms as expected, drifts for large as well as small firms are generally larger for small relative weights than for large relative weights. A similar result is found when we partition our sample by the number of analysts following and institutional holdings, our proxies for investor sophistication, and relative weight. Overall, we are led to the conclusion that investor misperceptions of the earnings process contribute to post-announcement drifts, beyond their (low) coincidence with firm size and investor sophistication. As such, this paper advances the research on the post-earnings announcement drift by introducing a new measure of investor misperception.

The remainder of this paper is organized as follows. Section 2 provides a brief review of related studies, Section 3 characterizes investor perceptions of the earnings process, Section 4 discusses the estimation procedures, Section 5 reexamines post-announcement drifts in abnormal returns, and Section 6 briefly summarizes our main findings and provides conclusions.

2. Related Studies on Pricing Anomalies

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

Ball and Brown (1968) observed that drifts in cumulative abnormal returns continued beyond earnings announcements. Similar observations have been made in numerous other studies prior to Bernard and Thomas (1990). These include Jones and Litzenberger (1970), Brown and Kennelly (1972), Joy et al. (1977), Watts (1978), Grivoly and Lakonishok (1980), Latane and Jones (1979), Rendleman et al. (1982), Foster et al. (1984), Bernard and Thomas (1989), and Freeman and Tse (1989).

The academic literature offers three major explanations for the phenomenon: (i) shifts in risks of firms with extreme surprises, justifying higher expected return in equilibrium; (ii) methodological shortcomings in previous studies that document the phenomenon; and (iii) investors’ underreaction to the value-relevant earnings information during the announcement period, which is subsequently corrected due to new information (a type of “delayed response” to the previously released information).

Attempts to explain the drift as compensation for risk or as a result of flaws in research design have been mostly unsuccessful. Programming errors, chance variation, and potential research design flaws, including a failure to control fully for risk, have not provided fully satisfactory explanations for the drift. The drift cannot be traced entirely to transaction costs of liquidity and trading-mechanism effects (Mendenhall, 2004). There are many theories about the cause of the post-earnings-drift, and virtually all of the theories involve investors’ underreaction or overreaction to the announcement.

Rendleman et al. (1987), Freeman and Tse (1989), Mendenhall (1991), Wiggins (1991), and Bernard and Thomas (1989, 1990) consider the hypothesis that post-announcement drifts (or predictable future announcement period returns) arise because prices fail to reflect the information current earnings contain about future earnings (see also Bernard et al., 1993). Bartov (1992) further refines the investigation of the association between the predictability of earnings innovation and abnormal returns by separating firms into groups based on a logit estimate of the probability of future innovations. His evidence reinforces the conclusion that Bernard and Thomas’s (1990) pricing anomalies are driven by investor misperceptions of the earnings process in forming their expectations.4

There have been several attempts to explain investors’ underreaction. Barberis et al. (1998) provide a formal model to explain underreaction to earnings announcement. Their explanation of underreaction to earnings announcement is related to investors’ conservatism. Conservatism refers to the reluctance of individuals to update their beliefs upon receiving new information. Investors subject to conservatism might disregard the full information content of an earnings announcement because they tend to cling, at least partially, to their prior estimates of earnings rather than update their estimates based on the new information contained in the earnings announcement. Daniel et al. (1998) show that when investors overestimate the precision of private signals, they can generate an initial price reaction that is weaker than the fully rational one. In their model, investors do not react much to public signals when the content of the new information conflicts with private signals. Hong and Stein (1999) assume that firm-specific information diffuses gradually across the investing public, and investors cannot perform the rational-expectations trick of extracting information from prices. These assumptions generate underreaction and positive return autocorrelations.5

Similar to our paper, Ball and Bartov (1996) provide evidence that investors make some adjustment for serial correlation in seasonal differences, but only about half of the adjustment that should be made if they correctly perceived earnings processes. Soffer and Lys (1999) reconcile the apparent conflict between Bernard and Thomas (1990) and Ball and Bartov (1996) by showing gradual incorporation of earnings serial correlation in market expectations. This gradual recognition, however, is in response to earnings information released over the quarter that is itself predictable at the start of the quarter, so that the explanation still implicates investor irrationality. Jacob et al. (2000) also question the extreme form of investor misperception implied by the Bernard and Thomas (1990) analysis.

Consider a firm announcing $1 abnormal earnings. Its current stock price is $20. Investors would respond to this good news and purchase this undervalued stock. However, the challenge is determining what the new equilibrium price should be. If investors think that this $1 earnings surprise is temporary, they would buy the stock at $21. If they treat the $1 as permanent, they should trade it up to $30 (assuming a discount rate of 10%). The stock should be traded between $21 and $30, depending on investors’ weights on temporary and permanent components. The question is how to estimate the weight of the investors’ misperception. Our study provides an extension of Bernard and Thomas (1990) in that our model of investor misperception is more flexible, with a parameter that is hypothesized to capture cross-sectional variations in the degree of investor misperception. We investigate to what extent estimates of this parameter explain post-announcement drifts. Our findings are interesting in that investors behave less naively than is assumed in the design of earlier studies.

3. Investor Perceptions of the Earnings Process

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

The purpose of time-series decomposition is to additively separate the non-seasonal random walk component that Bernard and Thomas (1990) assume is ignored by investors. Additively separating the non-seasonal random walk or temporary component makes it possible to specify a proxy for investors’ expected earnings as a weighted sum of expected permanent and expected temporary components. Normalizing the weight on the permanent component at one reduces the parameters necessary for characterizing the degree of misperception to a single parameter, which we define as the relative weight, w. Given that assets are priced such that abnormal returns are proportional to unexpected earnings, wherein investor expectations reflect w, we have scope to estimate this parameter from earnings and price data. The following analysis demonstrates how this can be done.

3.1. Decomposition of Earnings Process

In keeping with time-series models by Foster (1977) and Brown and Rozeff (1979), we begin by assuming that seasonally differenced earnings can be portrayed as an autoregressive, moving average (ARMA) process of orders p and q; that is, ARMA (p, q):6

  • image(1)

where Qt denotes earnings (at time t), et represents a random shock, L is the backward shift or lag operator, and a(L) and b(L) are polynomials in L of orders p and q, respectively; that is,

  • image

We can isolate seasonally differenced earnings by inverting a(L) to obtain:

  • image(2)

where

  • image

To gain the decomposition, we re-express C(L) as follows:

  • image(3)

where

  • image(4)

Substituting equation (3) into equation (2) and multiplying through by (1 − L4)−1, we arrive at:

  • image(5)

where

  • image(6)
  • image(7)

It is noted that inline image follows a random walk with drift g (i.e. permanent component) and inline image is a non-random walk (temporary) component of earnings.

3.2. Market Expectations and Returns

We now assume that in forming its expectations the market places relative weights one and w on the permanent and temporary components of earnings, respectively. That is,

  • image(8)

where w ≥ 0, E is the expectations operator and superscript M denotes the market so that EM denotes the market expectation. Note that a relative weight equal to zero (w = 0) amounts to Bernard and Thomas’s (1990) stark assumption that investors perceive the earnings process as a seasonal random walk, and a relative weight equal to one (w = 1) is consistent with the assumption that investors perceive the whole earnings process and, therefore, investors fully recognize the temporary component as well as the random walk (i.e. permanent) component.

Similar to Bernard and Thomas (1990), we assume that assets are priced such that abnormal returns, ARt, are proportional to unexpected earnings:

  • image(9)

where ARt denotes abnormal returns and λ is the earnings response coefficient (ERC). Substituting for Qt from equation (5) and inline image from equation (8) leads to:

  • image(10)

Alternatively, by rearranging terms we obtain:

  • image(11)

where ARF is the abnormal return when w = 1 (i.e. assets are priced efficiently) and ARBT is the abnormal return when w = 0 (i.e. assets are priced as Bernard and Thomas assumed).7

Therefore, the parameter w defines both the relative weight placed by investors on the temporary component of the earnings process, and a convex combination of abnormal returns when investors’ expectations correspond to the “true” earnings process (ARF) and abnormal returns when investors view earnings as a seasonal random walk (ARBT). This characterization allows us to tie in market inefficiency to misperceptions of the earnings process. The difference from Bernard and Thomas’s (1990)equation (2) is simply that our equation (11) allows for partial recognition of the temporary component by investors, whereas theirs does not.

4. Estimation Procedures

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

4.1. Sampling Procedure

We selected all NYSE/AMEX firms with quarterly data on COMPUSTAT from the first quarter of 1984 to the fourth quarter of 2003. To estimate relative weights requires a reasonable time series of observations for each firm. It involves a trade-off between efficient estimation of relative weights and sample size. To be included, we required that a minimum of 34 observations of pre-differenced quarterly earnings be available,8 and that a full set of returns be available from the Center for Research in Securities Prices.9 This sample selection procedure biases the final sample toward larger, more established corporations. One disadvantage of this is that it limits the extent of cross-sectional variation in firm characteristics. In all, 1420 firms encompassing 96 976 observations met these criteria. For the purpose of replicating Bernard and Thomas’s (1990) results, we classified firms by deciles of the distribution of market values of common stocks outstanding. We label firms falling in deciles 1–5 as “small” and those in deciles 9–10 as “large.”10 The extra decile for small firms reflects a skewness toward larger firms due to survivorship bias induced by our sample inclusion criteria.11

4.2. Estimating Relative Weights

For estimation purposes, we standardize seasonally differenced earnings using their historical standard deviations:12

  • image(12)

As in Bernard and Thomas (1990), we winsorize by replacing inline image by its mean plus or minus two standard deviations depending on the direction of inline image from the mean.13 By assuming invertibility of the moving average process, we approximate the right-hand-side of equation (12) by a finite autoregressive process of order r.

Regression estimates of the parameters defining the autoregressive processes are used to generate residuals for use in extracting estimates of the earnings response coefficient λ and relative weight w from abnormal returns. Specifically, the two-stage procedure involves regressions of the form:14

  • image(13)
  • image(14)

where inline image and inline image is a disturbance term. It follows from equations (11), (13), and (14) that

  • image(15)

which implies the following estimates:15

  • image(16)

4.3. Abnormal Returns

To maintain comparability with past studies, abnormal returns are calculated in a manner identical to that used by Bernard and Thomas (1989):16

  • image(17)

where Ri,t is the return for firm i in period t and Rp,t is the return on an equally weighted portfolio of NYSE/AMEX firms from the same firm size decile as firm i in the year to which t belongs. The benefit of this “companion portfolio” approach is that it mitigates the well-known firm size effect detected in previous studies. Abnormal returns for purposes of estimating w and λ are accumulated over a 3-day window beginning two trading days before the release of an earning announcement.

4.4. Descriptive Statistics

Table 1 presents descriptive statistics on relative weights and ERC for the full sample and subsamples for small firms (1–5) and large firms (9–10).

Table 1.   Descriptive statistics on estimated relative weights and earnings response coefficients (ERC) This table provides descriptive statistics on estimated relative weights,inline image, and ERC, inline image, for full sample and subsamples based on firm size, analyst coverage, and institutional ownership. The inline image and inline image are estimated based on the following equations: first stage regressions: inline image second stage regressions: inline image implied estimates: inline image where inline image denotes standardized unexpected earnings, defined as the ratio of the detrended seasonal difference in quarterly earnings to the standard deviation of the detrended seasonal difference in quarterly earnings over the trend estimation period. ARt denotes abnormal returns, defined as the difference between the return for firm i in period t and the return on an equally weighted portfolio of NYSE/AMEX firms from the same firm size decile as firm i in the year to which t belongs. Firm size is based on 1984, 1993, or 2003 market values, as depicted in the table. Small (large) firms are those with market values that fall in deciles 1–5 (9–10) of all NYSE/AMEX firms in those years. Low (high) analysts coverage firms are those with the number of analysts following below (above) the median. Low (high) institutional holding firms are those with institutional ownership below (above) the median. *, **, and *** denote significance levels of 10%, 5%, and 1%, respectively.
Panel A: The mean and median of inline image and inline image for small and large firms
ParameterFull sampleSmall firms as ofLarge firms as of
198419932003198419932003
inline imageMean0.77940.35730.65990.58110.77690.76800.7859
Median0.95120.40780.74720.70500.89960.97230.9760
inline imageMean0.01020.00770.02250.01440.00760.00860.0065
Median0.00790.00790.01830.01150.00630.00620.0060
Panel B: The t-statistic for the significance of the difference of relative weights, inline image, between small and large firms
 198419932003
MeanSmall firms0.35730.65990.5811
Large firms0.77690.76800.7859
t-statistic(3.06)*(2.41)**(2.24)**
Panel C: The t-statistic for the significance of the difference of relative weights, inline image, between low and high analyst coverage firms
 198419932003
MeanLow analysts coverage0.67550.69240.7741
High analysts coverage0.86060.86740.8701
t-statistic(2.01)**(2.54)*(1.09)
Panel D: The t-statistic for the significance of the difference of relative weights, inline image, between low and high institutional holding firms.
 198419932003
MeanLow institutional holding0.74860.73930.4849
High institutional holding0.86070.86620.8395
t-statistic(2.34)**(1.60)(3.17)*

The ordering of mean estimated relative weights in Panel A is consistent with Bernard and Thomas’s (1990) finding of greater post-announcement drifts for small firms if misperceptions of the earnings process drive those drifts. That is, relative weights are somewhat closer to zero for small firms and somewhat closer to one for large firms. A relative weight of zero corresponds to investors’ perceptions of earnings as the permanent component alone, whereas a relative weight of one corresponds to the permanent and temporary components combined. Estimates of relative weights for large firms are statistically distinct from one, which suggests that investor misperceptions are not entirely confined to small firms. Panel B shows that the difference of w between small and large firms is statistically significant. Later, we report further evidence to this effect. Panels C and D will be discussed later.

An intuitively appealing way to think about the relationship between firm size and relative weights is to conjecture that smaller firms have less history or larger information asymmetry than larger firms from which to estimate the underlying earnings process. If investors started from the prior belief that earnings follow a seasonal random walk, then as they observe earnings realizations that are in part an artifact of a temporary component, they would begin to revise their beliefs regarding the true process.

5. Reexamination of Post-announcement Drifts

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

5.1. Replication of Bernard and Thomas’s Results

We begin our reexamination of post-announcement drifts by replicating part of Bernard and Thomas’s (1989) analysis using our sample data. Results from this replication provide a benchmark for examining the consequences of generalizing Bernard and Thomas’s assumption regarding the market’s relative weighing of the temporary component of the earnings process. Following Bernard and Thomas (1990), standardized unexpected earnings (SUE) are defined as the ratio of the detrended seasonal difference in quarterly earnings to the standard deviation of the detrended seasonal difference in quarterly earnings over the trend estimation period. The earnings trend is estimated using a history of up to 15 prior quarterly earnings. We form 10 equal size portfolios based on SUE. Portfolios are reformulated in each calendar quarter based on the SUE for earnings announcements released that quarter. To address outliers and potential nonlinearities in the abnormal return–SUE relation, Bernard and Thomas (1990) and others transform SUE into decile scores based on their rank within each calendar quarter. We also partition the full sample into subsamples based on firm size deciles. Table 2 presents cumulative abnormal returns (CAR) for the portfolios containing the lowest (denoted Portfolio 1) and the highest (denoted Portfolio 10) ranking SUE, over 60, 120, 180, and 240 day accumulation periods beginning on the first trading day following earnings announcements for the full sample and for the two subsamples based on firm size.17

Table 2.   Replication of Bernard and Thomas’s post-announcement drifts This table reports post-announcement drifts by replicating part of Bernard and Thomas’s (1989) analysis using our sample data. Standardized unexpected earnings (SUE), defined as the ratio of the detrended seasonal difference in quarterly earnings to the standard deviation of the detrended seasonal difference in quarterly earnings over the trend estimation period, are used for Panels A–C. In Panel D, SUE are defined as actual earnings minus the mean analyst forecast of earnings, scaled by the dispersion of analyst forecasts. The consensus forecast is defined as the most recent monthly median forecast before earnings announcement when there are at least four earnings forecasts for the firm. Portfolio 10(1) corresponds to the decile subset of firms with the highest (lowest) SUE. ARt denotes abnormal returns defined as the difference between the return for firm i in period t and the return on an equally weighted portfolio of NYSE/AMEX firms from the same firm size decile as firm i in the year to which t belongs. CAR denotes cumulative abnormal returns over 60, 120, 180, and 240 day accumulation periods beginning on the first trading day following earnings announcements. Small (large) firms are those with market values that fall in deciles 1–5 (9–10) of all NYSE/AMEX firms in those years. Low (high) analysts coverage firms are those with the number of analysts following below (above) the median. Low (high) institutional holding firms are those with institutional ownership below (above) the median. t-Statistics are provided in parentheses.
Panel A: The mean of post-announcement drifts when SUE = (Earningst − Earningst−4)/standard deviation of unexpected earnings
 All firmsSmall firmsLarge firmsLow analyst coverageHigh analyst coverageLow institutional holdingHigh institutional holding
Portfolio (1)
 Observations8163113032321637233016901940
 SUE−2.24−2.32−2.28−1.7447−1.8041−1.5909−1.7429
 CAR (1–60)−0.0036 (−0.34)−0.0020 (−0.21)−0.0044 (−0.91)−0.0022 (−0.47)0.0067 (1.70)0.0008 (0.17)−0.0033 (−0.78)
 CAR (1–120)0.0019 (0.98)0.0097 (0.79)0.0036 (0.52)0.0072 (1.14)0.0176 (3.18)0.0078 (1.23)0.0037 (0.65)
 CAR (1–180)0.0056 (1.32)0.0181 (1.20)−0.0044 (−0.48)0.0127 (1.64)0.0187 (2.72)0.0207 (2.69)0.0078 (1.09)
 CAR (1–240)0.0158 (2.41)0.0389 (2.34)0.0065 (0.59)0.0257 (2.91)0.0276 (3.54)0.0259 (3.08)0.0148 (1.81)
Portfolio (10)
 Observations8182107732931439275816032028
 SUE1.481.491.501.90972.02241.89452.0171
 CAR (1–60)0.0408 (11.80)0.0623 (6.94)0.0207 (5.28)0.0250 (5.81)0.0076 (2.84)0.0192 (5.94)0.0063 (1.92)
 CAR (1–120)0.0517 (10.72)0.0823 (6.74)0.0277 (4.91)0.0374 (6.17)0.0109 (2.96)0.0230 (4.82)0.0094 (2.12)
 CAR (1–180)0.0621 (11.19)0.0919 (6.60)0.0355 (5.18)0.0434 (6.02)0.0162 (3.61)0.0348 (6.11)0.0109 (2.01)
 CAR (1–240)0.0734 (11.21)0.1121 (6.78)0.0416 (5.65)0.0490 (5.81)0.0204 (4.08)0.0426 (6.54)0.0110 (1.80)
Panel B: The median of post-announcement drifts when SUE = (Earningst − Earningst−4)/standard deviation of unexpected earnings
 All firmsSmall firmsLarge firmsSmall analysts coverageLarge analysts coverageLow institutional holdingHigh institutional holding
Portfolio (1)
 Observations8163113032321637233016901940
 SUE−1.4671−1.5241−1.4575−1.4771−1.5351−1.324−1.4885
 CAR (1–60)−0.0021 (−0.20)−0.0082 (−0.86)−0.0016 (−0.33)−0.0055 (−1.18)0.0052 (1.32)−0.0033 (−0.70)−0.0028 (−0.66)
 CAR (1–120)0.0052 (2.68)0.0010 (0.08)0.0067 (0.97)0.0043 (0.68)0.0155 (2.80)0.0045 (0.71)0.0045 (0.79)
 CAR (1–180)0.0115 (2.71)0.0130 (0.86)0.0065 (0.71)0.0182 (2.35)0.0180 (2.62)0.0165 (2.15)0.0080 (1.12)
 CAR (1–240)0.0172 (2.62)0.0210 (1.26)0.0123 (1.12)0.0272 (3.08)0.0204 (2.62)0.0205 (2.44)0.0069 (0.84)
Portfolio (10)
 Observations8182107732931439275816032028
 SUE1.72181.70621.73891.69181.81421.71041.8185
 CAR (1–60)0.0154 (4.45)0.0316 (3.52)0.0129 (3.29)0.0187 (4.35)0.0127 (4.75)0.0120 (3.71)0.0133 (4.05)
 CAR (1–120)0.0211 (4.38)0.0399 (3.27)0.0130 (2.30)0.0331 (5.46)0.0143 (3.88)0.0198 (4.15)0.0165 (3.72)
 CAR (1–180)0.0286 (5.15)0.0505 (3.63)0.0235 (3.43)0.0390 (5.41)0.0239 (5.33)0.0361 (6.34)0.0183 (3.37)
 CAR (1–240)0.0259 (3.96)0.0509 (3.08)0.0205 (2.78)0.0336 (3.98)0.0229 (4.58)0.0325 (4.99)0.0176 (2.88)
Panel C: The mean of post-announcement drifts reported on a yearly base when SUE = (Earningst − Earningst−4)/standard deviation of unexpected earnings
 1985198619871988198919901991199219931994
Portfolio (1)
 CAR (1–60)−0.0069 (−0.62)0.0099 (1.02)0.0009 (0.13)−0.0183 (−1.75)−0.0116 (−1.20)0.0055 (0.48)−0.0049 (−0.52)0.0097 (1.04)0.0067 (0.70) 0.0073 (0.69)
 CAR (1–120)−0.0378 (−3.20)0.0429 (3.98)−0.0006 (−0.08)−0.0041 (−0.37)−0.0334 (−3.28)0.0322 (3.10)−0.0042 (−0.38)0.0113 (1.20)0.0107 (1.03) 0.0120 (1.32)
 CAR (1–180)−0.0447 (−4.92)0.0587 (6.01)−0.0121 (−1.10)−0.0070 (−0.63)−0.0438 (−4.27)0.0451 (4.39)0.0194 (1.97)0.0026 (0.30)0.0235 (2.40) 0.0124 (1.35)
 CAR (1–240)−0.0383 (−3.72)0.0795 (7.90)0.0047 (0.50)−0.0032 (−0.28)−0.0594 (−6.03)0.0671 (6.82)0.0343 (3.27)0.0204 (2.11)0.0377 (3.80) 0.0084 (0.78)
Portfolio (10)
 CAR (1–60)0.0158 (1.60)0.0306 (3.00)0.0319 (3.08)−0.0023 (−0.20)0.0047 (0.43)0.0280 (2.68)0.0420 (4.07)0.0157 (1.38)0.0065 (0.45) 0.0135 (1.29)
 CAR (1–120)0.0065 (0.73)0.0296 (2.70)0.0479 (4.58)0.0010 (0.08)0.0167 (1.54)0.0432 (4.28)0.0660 (6.35)0.0097 (0.86)0.0060 (0.43) 0.0049 (0.37)
 CAR (1–180)0.0367 (3.70)0.0292 (2.67)0.0682 (6.23)0.0050 (0.43)0.0282 (2.79)0.0640 (6.23)0.0552 (5.45)0.0210 (2.03)0.0032 (0.28) 0.0003 (0.00)
 CAR (1–240)0.0363 (3.67)0.0349 (3.34)0.0683 (6.23)0.0083 (0.78)0.0310 (3.05)0.0669 (6.58)0.0743 (7.28)0.0294 (2.81)−0.0016 (−0.11)−0.010 (−0.07)
 199519961997199819992000200120022003
Portfolio (1)
 CAR (1–60)−0.0230 (−2.17)−0.0245 (−2.32)−0.0080 (−0.73)−0.0110 (−1.01)0.0121 (1.10)0.0217 (2.01)0.0289 (2.58)−0.0080 (−0.63)0.0720 (7.00)
 CAR (1–120)−0.0300 (−2.88)−0.0330 (−3.18)−0.0079 (−0.68)−0.0088 (−0.73)0.0112 (1.02)0.0461 (4.23)0.0330 (3.12)0.0155 (1.43)0.0850 (8.33)
 CAR (1–180)−0.0431 (−4.17)−0.0371 (−3.92)−0.0146 (−1.32)0.0012 (0.08)0.0170 (1.52)0.1000 (10.71)0.0110 (1.01)0.0585 (5.35)0.0907 (9.17)
 CAR (1–240)−0.0463 (−4.72)−0.0376 (−3.53)−0.0273 (−2.42)0.0023 (0.18)−0.0005 (−0.01)0.1400 (13.32)0.0304 (2.89)0.0746 (7.23)0.1000 (10.32)
 CAR (1–60)−0.0067 (−0.53)0.0159 (2.01)0.0181 (1.70)0.0148 (1.52)−0.0068 (−0.53)0.0286 (2.93)0.0210 (2.30)0.0361 (4.01)−0.0026 (−0.12)
 CAR (1–120)−0.0263 (−2.38)0.0221 (2.83)0.0205 (1.98)0.0031 (0.27)−0.0057 (−0.42)0.0528 (5.40)0.0419 (4.89)0.0352 (3.87)0.0178 (1.58)
 CAR (1−180)−0.0356 (−3.07)0.0255 (3.12)0.0183 (1.70)−0.0031 (−0.27)−0.0012 (−0.08)0.0665 (7.01)0.0670 (8.03)0.0316 (3.20)0.0242 (2.70)
 CAR (1–240)−0.0440 (−4.70)0.0388 (4.70)0.0190 (1.83)−0.0197 (−2.03)0.0029 (0.18)0.0978 (11.32)0.0844 (10.70)0.0246 (2.84)0.0163 (1.43)
Panel D: The mean of post-announcement drifts when SUE = (actual earnings − forecast consensus)/standard deviation of unexpected earnings
 All firmsSmall firmsLarge firmsLow analyst coverageHigh analyst coverageLow institutional holdingHigh institutional holding
Portfolio (1)
 Observations816311303232103261021592018
 SUE−2.1387−2.2445−2.1247−2.1774−2.2548−1.9286 (−76.13)−2.0483 (−70.40)
 CAR (1–60)−0.0003 (−0.08)−0.0010 (−0.10)0.0003 (0.06)−0.0033 (−0.60)0.0015 (0.23)−0.0014 (−0.35)−0.0068 (−1.71)
 CAR (1–120)0.0096 (1.76)0.0168 (1.32)0.0096 (1.39)0.0055 (0.73)0.0103 (1.10)0.0077 (1.35)0.0062 (1.16)
 CAR (1–180)0.0163 (2.39)0.0294 (1.85)0.0045 (0.51)0.0112 (1.20)0.0083 (0.714)0.0093 (1.32)0.0111 (1.73)
 CAR (1–240)0.0293 (3.66)0.0518 (2.91)0.0183 (1.70)0.0239 (2.26)0.0149 (1.09)0.0125 (1.6)0.0191 (2.62)
Portfolio (10)
 Observations81821077329393657819642225
 SUE1.47051.44561.52971.51081.51491.4411 (71.48)1.5515 (83.18)
 CAR (1–60)0.0329 (9.90)0.0459 (5.96)0.0160 (3.41)0.0411 (8.44)0.0349 (6.20)0.0302 (8.68)0.0257 (8.07)
 CAR (1–120)0.0395 (8.38)0.0646 (5.82)0.0169 (2.97)0.0477 (7.25)0.0485 (6.04)0.0392 (7.92)0.0235 (5.22)
 CAR (1–180)0.0455 (8.26)0.0687 (5.29)0.0200 (2.91)0.0543 (6.93)0.0566 (5.98)0.0525 (8.85)0.0259 (4.69)
 CAR (1–240)0.0476 (7.50)0.0753 (5.09)0.0196 (2.54)0.0664 (7.18)0.0696 (6.28)0.0625 (9.33)0.0313 (4.86)

In Panel A, the mean earnings SUE for portfolio 1 is negative, which is similar to previous studies. We observe a significant positive post-announcement drift for the extreme positive SUE portfolio (i.e. portfolio 10). The CAR for portfolio 10 over 240 days is 7.34%. However, we do not find a negative post-announcement drift for the extreme negative SUE portfolio (i.e. portfolio 1). Although the CAR for portfolio 1 over 60 days is −0.36%, over 240 days it is +1.58%, not negative. Our sample period differs somewhat from prior studies. It has some effect on the negative drift. However, our results are consistent with what is found by Johnson and Schwartz (2001). They investigate the relation between earnings surprises and post-announcement stock returns for 1991–1997, and show that the profit opportunities previously associated with simple trading strategies designed to exploit the drift phenomenon have now been substantially eliminated. The other potential explanation would be that the disappearing drift for extreme negative earnings is due to a survivorship bias. To check the robustness of the disappearing negative drift, we report the post-earning drifts for both lowest and highest SUE portfolios on a yearly basis in Panel C. It shows that the negative post-announcement drift following the negative earnings surprises seems to fluctuate over time. The post-announcement drifts is significantly positive in 1990–1993 and 2000–2003 for portfolio 1.

Given this discrepancy from previous findings for negative surprises, we conduct separate analyses for positive and negative surprises. A combined zero-investment with a long position in firms with extreme positive earnings surprises and a short position in firms with extreme negative surprises generates an abnormal return of 5.76%. The magnitude of this post-announcement price movement is both economically and statistically significant.

Partitioning the sample based on firm size decile shows that drifts are most pronounced for small firms. The facts that the positive post-announcement drift tends to be concentrated in small firms and the relative weights are lower for small firms than for large firms support our relative weights measuring misperception of the earnings process. A hedge portfolio with a long position in the highest SUE decile and a short position in the lowest SUE decile generates an average abnormal return of 7.32% for the small firms, while the same strategy earns an average abnormal return of 3.51% for the large firms. It supports the view that investor misperceptions are a factor in explaining the positive drift, although not necessarily the only factor. Moreover, for companies of all sizes, the estimated post-announcement abnormal returns are nearly half as large as the returns that occur during the quarter leading up to and including the earnings announcement. To eliminate the impact of outliers, we also report the median abnormal returns in Panel B. The results are somewhat weaker compared with those of mean abnormal returns reported in Panel A.

Some researchers argue that the post-announcement drift occurs because a flawed proxy for the expected earnings is used. Using a better earnings expectation model would presumably mitigate the drift. Modeling earnings expectations is a challenging task. In other published studies, earnings expectations are based on a seasonal random walk with a trend. This gives a noisy proxy of expectations. Recent studies use the mean analyst forecast as a proxy for the market’s expectation of earnings and show that drift is related to analyst forecasts (Mendenhall, 1991; Abarbanell and Bernard, 1992; Shane and Brous, 2001).

Analysts’ forecasts are taken from the I/B/E/S analyst forecast database, where the surprise is based on actual earnings minus the mean analyst forecast of earnings, scaled by the dispersion of analyst forecasts. The consensus forecast is defined as the most recent monthly median forecast before earnings announcement when there are at least four earnings forecasts for the firm. The main advantage of this approach is that it is based on actual earnings as reported by the firm originally, not including any subsequent restatements of the original data, and adjusted for special items. Note that this approach does not require a long history of earnings and is suitable for young firms as well. The main problem with this approach is that it is limited to firms that are followed by analysts, introducing a potentially significant sample-selection bias.

We use earnings surprises based on analyst forecasts as a robustness test. The results are reported in Panel D of Table 2. The firms with extreme positive earnings surprises earn excess returns of 4.76%. However, a combined zero-investment with a long position in firms with extreme positive earnings surprises and a short position in firms with extreme negative surprises generates an abnormal return of 1.83%. The post-announcement drift is approximately 70% smaller when earning surprises are measured relative to analysts’ forecasts rather than the Bernard and Thomas model. The results suggest that the post-announcement drift in Bernard and Thomas (1990) might have been overestimated because a simple proxy (i.e. seasonal random walk) for the expected earnings is used.

5.2. Autocorrelations in Earnings

Prior accounting research shows that SUE in quarter t is positively related to the SUE in quarters t − 1 through t − 3 in a descending order and negatively related to the SUE in quarter t − 4. Although an efficient stock market would immediately reflect the implications of current earnings for future earnings, the post-earnings announcement drift literature finds that a portion of the future SUE that can be predicted based on the current quarter’s SUE is gradually reflected in stock prices only after the current quarter’s earnings announcement. Having seasonally differenced earnings, we estimate autocorrelation functions up to eight lags for each firm from data over the full sample period, 1984–2003. Means and quartiles of estimated autocorrelation coefficient distributions for all firms, small firms and large firms are reported in Panels A–C of Table 3, respectively.

Table 3.   Distributions of seasonally differenced earnings autocorrelation coefficients This table reports means and quartiles of estimated autocorrelation coefficient distributions for all firms, small firms, and large firms. Having seasonally differenced earnings, we estimate autocorrelation functions up to eight lags for each firm from data over the full sample period, 1984–2003. Small (large) firms are those with market values that fall in deciles 1–5 (9–10) of all NYSE/AMEX firms in those years
Quartile meanLag 1Lag 2Lag 3Lag 4Lag 5Lag 6Lag 7Lag 8
Panel A: All firms
25%0.0251−0.0964−0.0923−0.1219−0.1197−0.1402−0.1063−0.1461
50%0.16960.02030.0181−0.0033−0.0184−0.0345−0.01429−0.0253
75%0.32670.13740.13290.11130.07430.05710.08550.0654
Mean0.18200.03750.0306−0.0136−0.0186−0.04080.0185−0.0442
Panel B: Small firms
25%0.0275−0.1065−0.1207−0.1609−0.0971−0.1657−0.1200−0.1935
50%0.17830.00780.0021−0.0241−0.0109−0.0303−0.0142−0.0475
75%0.34520.13820.13980.08810.09340.05080.10030.0565
Mean0.18700.01330.0135−0.0049−0.0021−0.04170.0101−0.0429
Panel C: Large firms
25%0.0281−0.0820−0.0936−0.1126−0.1178−0.1271−0.1022−0.1135
50%0.17940.02310.0133−0.0005−0.224−0.0387−0.0142−0.0225
75%0.31410.12310.11470.09840.06200.04250.08190.0662
Mean0.18530.02540.01480.0036−0.0217−0.04720.0037−0.0204

For all firms and small firms, we find a pattern of declining positive autocorrelations over lags 1, 2, and 3, negative autocorrelations at lag 4, and small autocorrelations beyond the fourth lag. This pattern is similar to that found by Bernard and Thomas (1990) and others. However, the negative autocorrelation occurs at lag 5 for large firms. The autocorrelation pattern is not consistent across different quartiles.

5.3. Drifts and Investor Expectations

Our next task is to determine whether post-announcement drifts are more likely for firms with small relative weights than for firms with large relative weights ignoring, for the moment, the effect of firm size. Again, the idea is that if post-announcement drifts are a manifestation of investor misperceptions of the earnings process and those weights measure the degree of investor misperception, then drifts should be more evident when the temporary component is ignored in forming expectations.18 We use two approaches in forming portfolios. First, we form portfolios based on SUE computed the same way as before; that is, where earnings expectations are determined consistent with a seasonal random walk. Our second approach is to use the same partition based on inline image, but to compute SUE by using earnings expectations that are also based on inline image; that is, expectations defined by equation (8) above. The results using the two approaches are reported in Table 4 (Panel A, conventional SUE; Panel B, inline image-based SUE).19

Table 4.   Relative weights and post-announcement drifts using two portfolio formation rules This table shows the post-announcement drifts using two portfolio formation rules. We use the following two approaches in forming portfolios. First, we form portfolios based on SUE, defined as the ratio of the detrended seasonal difference in quarterly earnings to the standard deviation of the detrended seasonal difference in quarterly earnings over the trend estimation period. The results are reported in Panel A. Our second approach is to employ the same partition based on estimated relative weights, inline image, but to compute SUE by using earnings expectations, which are also based on inline image; that is, expectations defined by equation (8). The results are shown in Panel B. ARt denotes abnormal returns defined as the difference between the return for firm i in period t and the return on an equally weighted portfolio of NYSE/AMEX firms from the same firm size decile as firm i in the year to which t belongs. CAR denotes cumulative abnormal returns over 60, 120, 180, and 240 day accumulation periods beginning on the first trading day following earnings announcements. Portfolio 10(1) corresponds to the decile subset of firms with the highest (lowest) SUE.
 All firms (%)Small inline image (%)Large inline image (%)
Panel A: SUE based on seasonal random walk (w = 0)
Portfolio 1
 CAR (1–60)−0.53−0.010.32
 CAR (1–120)0.37−0.611.04
 CAR (1–180)0.50−1.011.42
 CAR (1–240)1.18−0.881.88
Portfolio 10
 CAR (1–60)2.952.722.85
 CAR (1–120)3.363.142.78
 CAR (1–180)4.083.863.19
 CAR (1–240)4.704.453.46
Panel B: SUE based on relative weights (inline image)
Portfolio 1
 CAR (1–60)−0.29−1.630.66
 CAR (1–120)0.12−2.461.75
 CAR (1–180)0.36−2.962.80
 CAR (1–240)1.15−2.383.68
Portfolio 10
 CAR (1–60)0.250.370.35
 CAR (1–120)0.410.720.86
 CAR (1–180)1.251.920.48
 CAR (1–240)1.582.680.54

From Panel A, we observe that post-announcement drifts are generally larger for small relative weights than for large relative weights for portfolio 10. We find a negative post-announcement drift for the extreme negative SUE portfolio (i.e. portfolio 1) when relative weights are small. A zero investment strategy based on long and short stocks in the top and bottom deciles of standardized unexpected earnings yields mean cumulative abnormal returns of 5.33% and 1.58% for small, inline image, and large, inline image, portfolios, respectively. The differences in CAR are statistically significant for Portfolio 10 over accumulation periods of 120 or 240 days. Although the differences in CAR are systematically in keeping with our prediction of more pronounced drifts when relative weights are small, it is evident by comparison with the effects of firm size reported in Table 2 that investor misperceptions might not be the only factor contributing to those drifts. We also observe that post-announcement drifts are positive for Portfolio 1 (extreme negative earnings) when the relative weights are large. This result is consistent with what we find in Table 2.

The results reported in Panel B are qualitatively similar to those in Panel A: post-announcement drifts are generally larger for small relative weights than for large relative weights for portfolio 10, and the post-announcement drift for the extreme negative SUE portfolio (i.e. portfolio 1) is negative when relative weights are small. However, the magnitudes of the drifts are generally lower. The smaller drifts might be attributable to the generalization of the relative weights, inline image, to non-zero values. Accordingly, we use both rules in further analysis involving partitions based on firm size.

5.4. Interaction of Firm Size and Relative Weights

Correlations between the relative weights and firm size, where size is measured both in the first year and the last year of our data, are less than 9% using either the Pearson product moment or the Spearman rank order coefficients. This implies that investor misperceptions and firm size are two distinct dimensions, notwithstanding the association implied by the smaller (larger) relative weights for small (larger) firms in the analysis reported earlier. Accordingly, to control for firm size we now partition the data first based on firm size and then based on inline image. Panel A of Table 5 reports average firm sizes for Portfolios 1 and 10 after partitioning based on both firm size and relative weights. It confirms the low correlations between firm size and the relative weight: average firm sizes for small inline image and large inline image firms in each firm size group are very similar.

Table 5.   Average firm size, analysts coverage and institutional holdings for small and large relative weights Panel A reports average firm sizes for Portfolios 1 and 10 after partitioning based on firm size and relative weights. Portfolio 10(1) corresponds to the decile subset of firms with the highest (lowest) SUE. Panel B shows average number of analyst following for Portfolios 1 and 10 after partitioning based on analyst coverage and relative weights. Panel C exhibits average percentage of institutional holding for Portfolios 1 and 10 after partitioning based on institutional holding and relative weights. Small (large) firms are those with market values that fall in deciles 1–5 (9–10) of all NYSE/AMEX firms in those years. Low (high) analysts coverage firms are those with the number of analysts following below (above) the median. Low (high) institutional holding firms are those with institutional ownership below (above) the median. The number of observations is provided in the parentheses.
Panel A
 Small firm size (1067 observations)Large firm size (4072 observations)
Small inline imageLarge inline imageSmall inline imageLarge inline image
Portfolio 14.244.369.449.50
Portfolio 104.434.499.519.55
Panel B
 Low analyst coverage (2339 observations)High analyst coverage (2543 observations)
Small inline imageLarge inline imageSmall inline imageLarge inline image
Portfolio12.412.6911.7312.09
Portfolio 102.282.5812.0413.22
Panel C
 Low institutional holding (2325 observations)High institutional holding (1917 observations)
Small inline imageLarge inline imageSmall inline imageLarge inline image
Portfolio133.5336.6764.9965.78
Portfolio 1033.2139.0767.9066.69

Given the similarity of average firm sizes across small and large inline image, we surmise that the magnitude of relative weights does not proxy for firm size within the subsamples to be compared. Our findings for each of the subsamples appear in Table 6. Comparing the results reported in Panels A and B of Table 6, we find that post-announcement drifts are, once again, somewhat more pronounced for subsamples of smaller firms than for subsamples of larger firms. We also observe that the magnitude of post-announcement drifts is dependent on the weighting of the temporary component of earnings; that is, smaller weights are accompanied by larger absolute drifts. The differences in CAR are statistically significant for Portfolio 10, specifically over accumulation periods of 180 and 240 days for small and large firms, respectively.20 The other interesting finding worth mentioning is that CAR over an accumulation period of 240 days for Portfolio 1 of small size with small inline image is −4.98%, which is statistically and economically significant. In contrast, the same CAR for Portfolio 1 of small size with large inline image is 0.71%. This result is consistent with investor misperceptions of the earnings process.

Table 6.   Post-announcement drifts for subsamples based on firm size, analyst coverage, institutional ownership and relative weight This table presents post-announcement drifts for subsamples based firm size, analyst coverage, institutional ownership and relative weight. ARt denotes abnormal returns defined as the difference between the return for firm i in period t and the return on an equally weighted portfolio of NYSE/AMEX firms from the same firm size decile as firm i in the year to which t belongs. CAR denote cumulative abnormal returns over 60, 120, 180, and 240 day accumulation periods beginning on the first trading day following earnings announcements. Small (large) firms are those with market values that fall in deciles 1–5 (9–10) of all NYSE/AMEX firms in those years. Low (high) analysts coverage firms are those with the number of analysts following below (above) the median. Low (high) institutional holding firms are those with institutional ownership below (above) the median. Portfolio 10(1) corresponds to the decile subset of firms with the highest (lowest) SUE. The number of observations is provided in the parentheses.
Accumulation periodSmall inline image (%)Large inline image (%)Accumulation periodSmall inline image (%)Large inline image (%)
Panel A: Small firm size
Portfolio 1(655 observations)(476 observations)Portfolio 10(412 observations)(327 observations)
CAR (1–60)−2.050.60CAR (1–60)5.725.81
CAR (1–120)−2.81−0.24CAR (1–120)5.375.31
CAR (1–180)−3.80−1.42CAR (1–180)6.005.45
CAR (1–240)−4.980.71CAR (1–240)5.241.80
Panel B: Large firm size
Portfolio 1(1915 observations)(1600 observations)Portfolio 10(2157 observations)(1868 observations)
CAR (1–60)−0.91−0.26CAR (1–60)2.122.33
CAR (1–120)0.340.90CAR (1–120)2.641.96
CAR (1–180)−0.101.01CAR (1–180)3.812.87
CAR (1–240)0.261.65CAR (1–240)4.413.80
Panel C: Low analyst coverage
Portfolio 1(1187 observations)(1001 observations)Portfolio 10(952 observations)(838 observations)
CAR (1–60)−1.01–0.87CAR (1–60)1.862.25
CAR (1–120)−1.440.69CAR (1–120)2.762.36
CAR (1–180)−1.922.04CAR (1–180)3.022.66
CAR (1–240)−1.542.77CAR (1–240)2.872.84
Panel D: High analyst coverage
Portfolio 1(924 observations)(1005 observations)Portfolio 10(1168 observations)(1264 observations)
CAR (1–60)0.251.69CAR (1–60)1.350.83
CAR (1–120)0.832.63CAR (1–120)2.330.73
CAR (1–180)1.152.99CAR (1–180)3.151.69
CAR (1–240)2.444.13CAR (1–240)3.512.13
Panel E: Low institutional holdings
Portfolio 1(836 observations)(853 observations)Portfolio 10(810 observations)(794 observations)
1–60−0.320.46CAR (1–60)0.462.09
1–1200.001.56CAR (1–120)3.061.52
1–1801.182.94CAR (1–180)4.342.59
1–2401.693.49CAR (1–240)5.043.44
Panel F: High institutional holdings
Portfolio 1(976 observations)(961 observations)Portfolio 10(998 observations)(1032 observations)
CAR (1–60)−0.810.13CAR (1–60)0.660.61
CAR (1–120)−0.951.65CAR (1–120)1.580.45
CAR (1–180)−1.232.75CAR (1–180)1.131.11
CAR (1–240)−0.363.30CAR (1–240)0.421.80

If we apply an alternative portfolio formation rule based on SUE using estimated inline image in computing earnings expectations, the results, which are not tabulated to save space, remain qualitatively similar in terms of spreads between drifts for small inline image versus large inline image. This robust finding further suggests that investor misperceptions of the earnings process are a relevant factor in explaining post-announcement drifts beyond a coincidence with firm size.

5.5. Future Announcement Period Returns

Although the analysis to this point has focused on post-announcement drifts, the findings remain about the same in a qualitative sense when we examine abnormal returns corresponding to short windows centered on future earnings announcements. Bernard and Thomas (1990) show that a significant portion of post-earnings announcement drift is concentrated in a 3-day window around the announcement of the next four quarterly earnings. Their results imply that investors, on average, do not understand the implications of current quarterly earnings for future earnings and instead rely on a simple random walk model with a drift to form earnings expectations. Similar to Bernard and Thomas (1990), we calculate 3-day abnormal returns corresponding to quarterly earnings announcements for each of the four quarters following an earnings announcement from which SUE are determined.

Table 7 presents these returns for Portfolios 1 and 10 using subsamples based on either small (large) firm size or small (large) relative weights. To ensure comparability of subsample sizes, we redefined small and large firm sizes as below and above the median, respectively. The results reported in Panels A and B of Table 7 suggest that our relative weights play a role comparable in importance to that of firm size in explaining the predictability of market reactions to future earnings announcements. Abnormal returns for Portfolio 10 at future announcements are lower for small inline image than for small size, implying that there is no clear ordering across the two subsamples in this respect. These results combined with the low correlations between firm size and relative weights indicate that their respective roles in explaining post-announcement drifts are similar in importance but distinct.

Table 7.   Future announcement period abnormal returns The table reports future announcement period abnormal returns for subsamples based on firm size, analyst coverage institutional ownership and relative weights. ARt denotes abnormal returns defined as the difference between the return for firm i in period t and the return on an equally weighted portfolio of NYSE/AMEX firms from the same firm size decile as firm i in the year to which t belongs. CAR denote 3-day abnormal returns corresponding to quarterly earnings announcements for each of the four quarters (T + 1, T + 2, T + 3, and T + 4) following an earnings announcement (T) from which SUE are determined. Small (large) firms are those with market values that fall in deciles 1–5 (9–10) of all NYSE/AMEX firms in those years. Low (high) analysts coverage firms are those with the number of analysts following below (above) the median. Low (high) institutional holding firms are those with institutional ownership below (above) the median. Portfolio 10(1) corresponds to the decile subset of firms with the highest (lowest) SUE.
 Small size (%)Small inline image (%) Small size (%)Small inline image (%)
Panel A: Small firm size
Portfolio 1Portfolio 10
CAR in quarter T−1.73−1.02CAR in quarter T3.181.95
CAR in quarter T + 10.15−0.09CAR in quarter T + 10.890.70
CAR in quarter T + 20.590.001CAR in quarter T + 20.840.63
CAR in quarter T + 30.530.14CAR in quarter T + 30.550.42
CAR in quarter T + 40.590.23CAR in quarter T + 40.700.27
Panel B: Large firm size
 Large size (%)Large inline image (%) Large size (%)Large inline image (%)
Portfolio 1Portfolio 10
CAR in quarter T−0.88−1.05CAR in quarter T1.621.77
CAR in quarter T + 10.270.52CAR in quarter T + 10.140.14
CAR in quarter T + 20.130.42CAR in quarter T + 20.280.15
CAR in quarter T + 30.320.29CAR in quarter T + 30.140.05
CAR in quarter T + 40.430.25CAR in quarter T + 40.110.23
Panel C: Low analyst coverage
 Low analysts coverage (%)Small inline image (%) Low analysts coverage (%)Small inline image (%)
Portfolio 1Portfolio 10
CAR in quarter T−0.9−0.76CAR in quarter T1.31.25
CAR in quarter T + 1−0.07−0.31CAR in quarter T + 10.30.65
CAR in quarter T + 2−0.05−0.14CAR in quarter T + 20.240.54
CAR in quarter T + 30.30.08CAR in quarter T + 30.340.56
CAR in quarter T + 40.380.42CAR in quarter T + 40.0040.16
Panel D: High analyst coverage
 High analysts coverage (%)Large inline image (%) High analysts coverage (%)Large inline image (%)
Portfolio 1Portfolio 10
CAR in quarter T−0.170.006CAR in quarter T0.600.45
CAR in quarter T + 10.40.51CAR in quarter T + 10.13−0.025
CAR in quarter T + 20.380.41CAR in quarter T + 20.060.005
CAR in quarter T + 30.360.46CAR in quarter T + 30.220.11
CAR in quarter T + 40.50.42CAR in quarter T + 40.07−0.034
Panel E: Low institutional ownership
 Low institutional holding (%)Small inline image (%) Low institutional holding (%)Small inline image (%)
Portfolio 1Portfolio 10
CAR in quarter T−0.62−0.57CAR in quarter T1.521.58
CAR in quarter T + 10.07−0.38CAR in quarter T + 10.561.21
CAR in quarter T + 20.25−0.16CAR in quarter T + 20.170.16
CAR in quarter T + 30.380.23CAR in quarter T + 30.540.95
CAR in quarter T + 40.540.36CAR in quarter T + 4−0.100.18
Panel F: High institutional ownership
 High institutional holding (%)Large inline image (%) High institutional holding (%)Large inline image (%)
Portfolio 1Portfolio 10
CAR in quarter T−0.16−0.11CAR in quarter T0.750.51
CAR in quarter T + 10.330.58CAR in quarter T + 10.01−0.19
CAR in quarter T + 20.150.48CAR in quarter T + 20.21−0.02
CAR in quarter T + 30.400.43CAR in quarter T + 30.280.08
CAR in quarter T + 40.580.32CAR in quarter T + 40.130.06

5.6. Regression Analysis

Thus far, our analysis has been based on portfolio-level data. Although, compared with firm-level analyses, portfolio-level analyses potentially reduce noise in regressions and do not suffer from cross-correlation problems, Kothari et al. (2006) show that results from analyses of portfolio-level SUE differ substantially from those based on firm-level SUE. We examine the sensitivity of our prior conclusions to estimating regressions using firm-level data. Such tests also allow us to better differentiate the interactive role played by both size and relative weight. To assess how relative weights and size might affect the magnitude of the drift, we use interaction variables to allow the slope of the SUE-return relation to vary with each firm. The results on regression of CAR over an accumulation period of 180 days are shown in Panel A of Table 8, while those over 240 days are reported in Panel B. Both interaction coefficients of size and relative weight with SUE are significantly negative. This finding implies that both size and relative weight significantly reduce the ability of SUE to predict future returns. This reinforces our argument that investor misperception or underestimation of the earnings process is an explanation of post-earning drifts.

Table 8.   Regression analysis This table reports the results of regression analysis based on the following model: Drifts = α + β0Inst + β1Size + β2W + β3Coverage + β4SUE + β5SUE × Size + β6SUE × W + β7SUE × Coverage + β8SUE × Inst, where Inst is the institutional ownership, Size is firm size deciles, W is the relative weight estimated in Table 1, Coverage is the number of analysts following and SUE = (Earningst − Earningst−4)/standard deviation of unexpected earnings. The results on regression of cumulative abnormal returns (CAR) over accumulation periods of 180 and 240 days are shown in Panels A and B, respectively. t-Statistics are provided in the parentheses. *, **, and *** denote significance levels of 10%, 5%, and 1%, respectively.
Modelsαβ0β1β2β3β4β5β6β7β8R2
Panel A: Drifts = CAR 180
10.0261* (17.81)−0.0068* (−4.61)   −0.0002 (−0.14)   −0.0006 (−0.47)0.06
2−0.0067 (−1.49) 0.0036* (6.34)  0.0142* (3.18)−0.0016* (−2.92)   0.07
30.0184* (13.60)  0.0042* (3.92) 0.0037* (2.93) −0.0024** (−2.38)  0.03
40.0081* (4.05)   0.0029 (7.96)0.0064* (3.28)  −0.0010* (−2.97) 0.10
5−0.0086*** (−1.85) 0.0034* (6.10)0.0037* (3.51) 0.0153* (3.42)−0.0015* (−2.78)−0.0022** (−2.21)  0.09
60.0230* (12.51)−0.0068* (−4.63) 0.0040* (2.83) 0.0009 (0.49) −0.0014 (−1.02) −0.0006 (−0.44)0.09
70.0059* (2.76)  0.0036* (3.37)0.0028* (7.68)0.0077* (3.79) −0.0022** (−2.16)−0.0009* (−2.79) 0.12
80.0028 (0.54) 0.0005 (0.63)0.0036* (3.35)0.0026* (4.70)0.0135* (2.68)−0.0010 (−1.26)−0.0022** (−2.15)−0.0005 (−0.91) 0.13
Modelsαβ0β1β2β3β4β5β6β7β8R2
Panel B: Drifts = CAR240
10.0354* (20.85)−0.0092* (−5.39)   −0.0019 (−1.17)   −0.0015 (−0.93)0.10
2−0.0057 (−1.09) 0.0045* (6.89)  0.0117** (2.27)−0.0015** (−2.41)   0.07
30.0257* (16.49)  0.0055* (4.45) 0.0021 (1.43) −0.0026** (−2.27)  0.03
40.0116* (5.01)   0.0041* (9.35)0.0052** (2.31)  −0.0012* (−2.99) 0.13
5−0.0082 (−1.52) 0.0044* (6.61)0.0049* (4.00) 0.0129** (2.50)−0.0014** (−2.28)−0.0025** (−2.13)  0.10
60.0307* (14.45)−0.0093* (−5.49) 0.0059* (3.62) −0.0014 (−0.68) −0.0007 (−0.046) −0.0015 (−0.93)0.13
70.0086* (3.52)  0.0046* (3.80)0.0039* (9.07)0.0067* (2.83) −0.0023** (−2.05)−0.0011* (−2.82) 0.15
80.0093 (1.52) −0.0001 (−0.10)0.0047* (3.81)0.0039* (6.20)0.0096 (1.65)−0.0005 (−0.54)−0.0023** (−2.04)−0.0009 (−1.47) 0.16

5.7. Robustness Tests

Ball and Bartov (1996) show that the post-earnings-announcement drift is attributable to investors’ underestimating the magnitude of serial correlation in seasonally differenced quarterly earnings. Bartov et al. (2000) show that post-earnings-announcement drift is negatively correlated with the percentage of institutional ownership proxying for investor sophistication. This suggests that stock prices are more efficient in firms with a larger proportion of sophisticated investors. The underreaction hypothesis of the drift may be interpreted as suggesting that some (naive) investors underestimate the implications of current earnings innovations for future earnings levels whereas other (sophisticated) investors do not.

One variable that might be related to investor sophistication is the number of analysts who follow the stock (Bhushan, 1994). Financial analysts provide investors with interpretations of financial reports and commentary on firms’ financial performance. Investors consider financial analysts’ forecasts and recommendations to be an important information source. Barber et al. (2001) document that an investment strategy based on consensus recommendations of financial analysts earns positive returns. Brennan and Subrahmanyam (2000) find that firms with a high analysts’ coverage have lower trading costs because of the reduction in information asymmetry. Hong et al. (2000) suggest that firms with low analyst coverage should be ones where firm-specific information moves more slowly across the investing firms after controlling for other factors. Hence, evidence is consistent with the perception that information will reach investors faster when analysts’ coverage is higher.

The other proxy for the sophistication of the firm’s ownership is institutional investor holding. Lev (1988) suggests that institutional investors are better informed than individual investors because of the lower marginal costs of collecting information. Potter (1992) attributes the higher level of information advantage of institutional investors to substantial information research resources. El-Gazzar (1998) shows that institutional investors develop private information to satisfy their fiduciary responsibilities and to improve portfolio performance. Walther (1997) finds that market participants place more weight on the analyst forecast relative to the time-series model forecast as institutional ownership increases. This supports the idea that institutional ownership captures the degree of sophistication of the marginal investor.

To control for the investor sophistication, we redo our tests adding two new variables: the amount of analyst coverage, which is the number of analysts reporting quarterly forecasts to I/B/E/S in the 90 days prior to the earnings announcement, and institutional ownership for each firm-quarter. We obtain data on institutional holdings from the CDA Spectrum database, which contains the total number of common shares held by institutions at the end of each calendar quarter based on Section 13(f) filings. We require that institutional ownership be available for the calendar quarter immediately preceding the quarterly earnings announcement based on Bartov et al. (2000).

Panel C of Table 1 shows that relative weights, inline image, are smaller for firms with low analyst coverage and closer to one for firms with high analyst coverage. This is consistent with Bernard and Thomas’s finding of greater post-earning drifts for small firms if it is the misperceptions of the earnings process that are driving those drifts. We find a similar pattern in Panel D of Table 1 using institutional holdings as a proxy for investors’ sophistication: relative weights, inline image, are smaller for firms with low institutional holdings and closer to one for firms with high institutional holdings. Panel A of Table 2 documents that a hedge portfolio with a long position in the highest SUE decile and a short position in the lowest SUE decile earnings generates an average abnormal return of 2.33% for the firms with low analyst coverage, whereas the same strategy earns an average abnormal return of −0.72% for the firms with large analyst following. Similarly, the return on a zero-investment portfolio is 1.67% for the firms with low institutional holdings, whereas the return is −0.38% for the firms with high institutional holdings.

Panel B of Table 5 reports the average number of analysts following for Portfolios 1 and 10 after partitioning based on both analyst coverage and relative weights. It shows low correlations between analyst coverage and the relative weight: average numbers of analysts following small inline image and large inline image firms in each firm analyst coverage group are very similar. We find a similar pattern between institutional holdings and relative weights reported in Panel C of Table 5. These findings suggest that investor misperceptions represented by the relative weights are quite independent of investor sophistication, represented by either analysts following or institutional holdings within the subsamples to be compared, although both investor misperceptions and investor sophistication are related to post-announcement drifts.

In Panel C of Table 6, a hedge portfolio with a long position in the highest SUE decile and a short position in the lowest SUE decile earnings generates an average abnormal return of 4.41% for the firms with low analysts coverage when inline image is low, whereas the same strategy earns an average abnormal return of 0.07% for the firms with low analysts following when inline image is high. In Panel E of Table 6, we find a similar pattern for firms with low institutional holdings. The results for an alternative portfolio formation rule based on SUE using estimated inline image in computing earnings expectations remain qualitatively similar in terms of spreads between drifts for small inline image versus large inline image. This robust finding further suggests that investor misperceptions of the earnings process are a relevant factor in explaining post-announcement drifts beyond a coincidence with analyst coverage and institutional ownership. Panels E and F of Table 6 reveal that post-announcement drifts are somewhat more pronounced for subsamples of firms with low institutional holdings than for subsamples of firms with high institutional holdings.

Table 7 reports that 3-day abnormal returns, corresponding to quarterly earnings announcements for each of the four quarters following an earnings announcement from which SUE are determined. We find that there is no clear ordering across the subsamples of analyst coverage and relative weights and the subsamples of institutional ownership and relative weights in explaining the predictability of market reactions to future earnings announcements.

In Table 8, we include the interaction terms SUE × Coverage and SUE × Inst to capture the model’s implication that the number of analysts following and institutional holdings matter only when the information content of the earnings signal is salient. We find that both interaction coefficients of analyst following and relative weight with SUE are significantly negative in Model 7. However, when we include the three interaction terms of size, analyst coverage and relative weight with SUE, we observe both interaction coefficients of size and analyst following, with SUE become insignificant, while the interaction coefficient of relative weight with SUE is still significant in Model 8. This result suggests that investor underestimation of the impact of current earnings on future earnings growth plays an important role in the post-earnings-announcement drift. The respective roles of firm size, investor sophistication and relative weight in explaining post-announcement drifts are quite distinct. Our measure of investor misperception seems relatively more important.

However, both interaction coefficients of institutional holding and relative weight with SUE are not significant in Model 6. The “disappointing” results might be due to the invalid proxy for investor sophistication. When Bartov et al. (2000) evaluate the validity of the institutional holdings variable as proxy for investor sophistication, in terms of incorporating earnings information into stock prices in a timely fashion, they find mixed results. They suggest the possibility that the institutional-holdings variable is not a valid proxy for investor sophistication and, therefore, call for caution in the interpretation of their findings. Grffin et al. (2003) find that many institutional investors adopt strategies, such as return momentum, that do not yield significant future abnormal returns. Ke and Ramalingegowda (2005) show that not all institutional investors, but transient institutional investors, trade to exploit the post-earnings announcement drift.21

6. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References

The predictable drift in stock returns following earnings announcements is one of the longest-standing anomalies in the accounting and finance literature. Kothari (2001) argues that the drift provides a serious challenge to the efficient markets hypothesis because it has survived rigorous testing for over 30 years. We observe a significant positive post-announcement drift for the extreme positive SUE portfolio. However, the negative post-announcement drift for the extreme negative SUE portfolio seems to disappear. Having decomposed the earnings process into permanent and temporary components, we estimated the relative weights implicitly assigned to the temporary component by investors in forming their expectations. A relative weight of zero implies that investors misperceive the earnings process in a manner suggested by Bernard and Thomas (1990), while a relative weight of one is consistent with market efficiency. Accordingly, we predict that more pronounced post-announcement drifts will be associated with firms for which the relative weight on the temporary component is closer to zero than with firms for which the relative weight is closer to one.

To test these predictions, we estimate relative weights for 1420 firms, employ these weights to partition this sample, and estimate cumulative abnormal returns over periods following earnings announcements. We find that estimates of post-announcement drifts are generally larger in absolute value for firms with small relative weights, and that the differences in estimates between subsamples based on the magnitude of relative weights are no less prominent for large firms than for small firms. Recent studies show that the post-earnings-announcement drift is negatively correlated with proxies of investor sophistication. We find that number of analyst following and institutional ownership reduce the ability of SUE to predict future returns. However, the roles of firm size and sophistication are mitigated when we include the relative weight in the analysis. Taken together, these findings suggest that firm size and investor sophistication do not fully explain post-announcement drifts and that investors’ oversight of the autoregressive structure in seasonally differenced earnings might also play a role.

In summary, we provide empirical evidence that investors’ misperceptions contribute to post-announcement drifts beyond the contribution of firm size and investor sophistication. These findings reinforce Bernard and Thomas’s conjecture that less weight is placed on temporary components of earnings than would be appropriate if earnings processes were well understood, although not zero as Bernard and Thomas implicitly assume in their portfolio formation rule. The findings also complement Ball and Bartov’s (1996) result that investors partially, but not fully, adjust for serial correlation in seasonal differences. If one accepts the conclusion that factors other than firm size and investor sophistication contribute to investor misperceptions of earnings processes, then the next task would be to develop theory and evidence regarding what those factors might be.22

Footnotes
  • 1

    This quote also applies to work by Ou and Penman (1989).

  • 2

    Kormendi and Lipe (1987) examine the relation between the time-series process for earnings and the responsiveness of stock prices to earnings changes. They find that the greater the response coefficient, the more permanent or persistent the changes in earnings. Ali and Zarowin (1992) adopt a similar perspective. Operating from the assumption that annual earnings follow an IMA (1,1) process, they consider the effects of misspecifying unexpected earnings in estimating earnings response coefficients. Our approach estimates the earnings process from the data, and, hence, is more general. Lee (1996a) decomposes earnings into permanent and temporary components based on a bivariate model and finds support for the hypothesis that dividend decisions are primarily influenced by a measure of permanent earnings. For a comovement of earnings, dividends, and stock prices, see Lee (1996b). For the usefulness of earnings data in predicting stock prices, see Campbell and Shiller (1988) and Lee (1996b).

  • 3

    Quah (1992) criticizes that the random walk trend implicit in the Beveridge and Nelson’s (1981) decomposition maximizes the importance of the permanent component. Another potential criticism is that Beveridge and Nelson’s (1981) decomposition assumes the permanent and temporary components are not orthogonal. For this purpose, we may need to use a structural vector autoregression based decomposition (e.g. Blanchard and Quah, 1989). This should always be recognized when interpreting the results of the Bernard and Thomas (1990) decomposition.

  • 4

    Michaely et al. (1995) provide evidence of the post-dividend initiation/omission price drift, which is distinct from the post-earnings-announcement drift.

  • 5

    Apart from studies linking stock returns to earnings announcements, there is considerable evidence of autocorrelations in the former. Studies providing this evidence include Keim and Stambaugh (1986), French et al. (1987), Fama and French (1988), Lo and MacKinlay (1988), and Poterba and Summers (1988). Lo and MacKinlay (1988) report positive autocorrelations in weekly and monthly returns over the period 1962–1985. Fama and French (1988), using data from 1926 to 1985 report negative autocorrelations for return horizons longer than a year. They explain the correlation by introducing a temporary component into the model of stock price (see also Lee, 1995). Both Lo and MacKinlay (1988) and Fama and French (1988) find greater evidence of autocorrelations for portfolios of small firms than for portfolios of large firms.

  • 6

    We suppress subscripts denoting firm for economy of notation.

  • 7

    One might conjecture that the market would use different ERC for the two components of abnormal returns in equation (10). The difficulty with modifying the analysis to accommodate this conjecture is that both components are a function of the same innovation, et. As a result, we can identify only one ERC from the later regression, or two parameters, λ and w, in all.

  • 8

    This provides a minimum of 30 sample observations after seasonal differencing.

  • 9

    These requirements are costly in terms of firms. However, we need to ensure sufficient time-series observations to estimate parameters firm by firm.

  • 10

    Firm size is redetermined each year.

  • 11

    The numbers of firms in the 10 size deciles are 713 (size 1), 2507, 4015, 6276, 8266, 10 685, 12 708, 14 921, 17 500, 17 500, and 19 385 (size 10), respectively. We also label firms falling in deciles 1–4 as “small” and those in deciles 9–10 as “large.”

  • 12

    Recent studies use the mean analyst forecast as proxy for the market’s expectation of earnings and show that drift is related to analyst forecasts (Mendenhall, 1991; Abarbanell and Bernard, 1992; Shane and Brous, 2001).

  • 13

    This procedure affected 2.6% of our data points.

  • 14

    In deriving equation (14), we use inline image by assuming that inline image Specifically,

    • image
  • 15

    Standard errors of inline image are computed by a delta method, which is standard procedure for computing asymptotic variances.

  • 16

    Subscripts denoting firm are included to avoid confusion.

  • 17

    The alternative compound abnormal return is computed as the buy-and-hold return on the stock minus the buy-and-hold return on the CRSP equally weighted size-decile of which the stock is a member at the beginning of the calendar year. The similar qualitative results emerge when we adopt this measure. For brevity, we do not report the results using the buy-and-hold return.

  • 18

    We need to be careful here in that autocorrelations in abnormal returns might not be monotone in w. However, we can contrast autocorrelations when w = 1 and w = 0.

  • 19

    Notwithstanding that the inline image are estimated from all available data, these approaches correspond to Bernard and Thomas’s (1990) Hypotheses 1 and 2.

  • 20

    The t-values are 1.67 and 1.77, respectively.

  • 21

    Bushee (2001) classifies institutional investors into transient and dedicated investors based on their expected investment horizons. Transient institutional investors are characterized as having high portfolio turnover and highly diversified portfolio holdings, whereas dedicated institutional investors are characterized by large average investment in portfolio firms and extremely low turnover.

  • 22

    For example, using annual data, Sloan (1996) finds indirect evidence that investor misperceptions are related to the accrual component of earnings. Liang (2003) examines the post-earnings announcement drift in the context of theories that consider investors’ non-Bayesian behaviors. Hirshleifer et al. (2003) examine whether individual investors are the source of post-earnings announcement drift.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Related Studies on Pricing Anomalies
  5. 3. Investor Perceptions of the Earnings Process
  6. 4. Estimation Procedures
  7. 5. Reexamination of Post-announcement Drifts
  8. 6. Conclusion
  9. References
  • Abarbanell, J., and V. Bernard, 1992, Tests of analysts’ overreaction/underreaction to earnings information as an explanation for anomalous stock price behavior, Journal of Finance 47, pp. 11811207.
  • Ali, A., and P. Zarowin, 1992, Permanent versus transitory components of annual earnings and estimation error in earnings response coefficients, Journal of Accounting and Economics 15, pp. 249264.
  • Ball, R., and E. Bartov, 1996, How naive is the stock market’s use of earnings information? Journal of Accounting and Economics 21, pp. 319337.
  • Ball, R., and P. Brown, 1968, An empirical evaluation of accounting income, Journal of Accounting Research 6, pp. 159178.
  • Barber, B., R. Lehavy, M. McNichols, and B. Trueman, 2001, Can investors profit from the prophets? Security analyst recommendations and stock returns, Journal of Finance 56, pp. 531563.
  • Barberis, N., A. Shleifer, and R. Vishney, 1998, A model of investor sentiment, Journal of Financial Economics 49, pp. 307343.
  • Bartov, E., 1992, Patterns in unexpected earnings as an explanation for post-announcement drift, The Accounting Review 67, pp. 610622.
  • Bartov, E., S. Radhakrishnan, and I. Krinsky, 2000, Investor sophistication and patterns in stock returns after earnings announcements, The Accounting Review 75, pp. 4363.
  • Bernard, V., and J. Thomas, 1989, Post-earnings announcement drift: Delayed price response or risk premium? Journal of Accounting Research 27, pp. 136.
  • Bernard, V., and J. Thomas, 1990, Evidence that stock prices do not fully reflect the implications of current earnings for future earnings, Journal of Accounting and Economics 13, pp. 305340.
  • Bernard, V. L., J. K. Thomas, and J. S. Abarbanell, 1993, How sophisticated is the market in interpreting earnings news? Journal of Applied Corporate Finance 6, pp. 5463.
  • Beveridge, S., and C. R. Nelson, 1981, A new approach to decomposition of economic time series into permanent and transitory components with particular attention to measurement of the ‘business cycle’, Journal of Monetary Economic 7, pp. 151174.
  • Bhushan, R., 1994, An information efficiency perspective on the post-earnings announcement drift, Journal of Accounting and Economics 18, pp. 4565.
  • Blanchard, Olivier. J., and D. Quah, 1989, The dynamic effects of aggregate demand and supply disturbances, American Economic Review 79, pp. 655673.
  • Brennan, M., 1991, A perspective on accounting and stock prices, Accounting Review 66, pp. 6779.
  • Brennan, M., and A. Subrahmanyam, 2000, Investment analysis and price formation in securities markets, Journal of Financial Economics 38, pp. 361381.
  • Brown, D., and J. W. Kennelly, 1972, The information content of quarterly earnings: An extension and some further evidence, Journal of Business 45, pp. 403415.
  • Brown, L., and M. Rozeff, 1979, Univariate time-series models of quarterly accounting earnings per share: A proposed model, Journal of Accounting Research 17, pp. 179189.
  • Bushee, B., 2001, Do institutional investors prefer near-term earnings over long-run value? Contemporary Accounting Research 18, pp. 207246.
  • Campbell, J. Y., and R. J. Shiller, 1988, Stock prices, earnings and expected dividends, Journal of Finance 43, pp. 661676.
  • Daniel, K., D. Hirshleifer, and A. Subrahmanyam, 1998, Investor psychology and security market under- and overreaction, Journal of Finance 53, pp. 18391885.
  • El-Gazzar, S. M., 1998, Predisclosure information and institutional ownership: A cross-sectional examination of market revaluations during earnings announcement periods, Accounting Review 73, pp. 119129.
  • Fama, E. F., 1998, Market efficiency, long-term returns, and behavioral finance, Journal of Financial Economics 49, pp. 283306.
  • Fama, E. F., and K. French, 1988, Permanent and temporary components of stock prices, Journal of Political Economy 96, pp. 246273.
  • Foster, G., 1977, Quarterly accounting data: Time-series properties and predictive-ability results, The Accounting Review 52, pp. 121.
  • Foster, G., C. Olsen, and T. Shevlin, 1984, Earnings release, anomalies, and the behavior of security returns, The Accounting Review 59, pp. 574603.
  • Freeman, R., and S. Tse, 1989, The multi-period information content of accounting earnings: Confirmations and contradictions of previous earnings reports, Journal of Accounting Research 27, pp. 4979.
  • French, K., G. W. Schwert, and R. Stambaugh, 1987, Expected stock returns and volatility, Journal of Financial Economics 19, pp. 329.
  • Grffin, J., J. Harris, and S. Topaloglu, 2003, The dynamics of institutional and individual trading, Journal of Finance 58, pp. 22852320.
  • Grivoly, D., and J. Lakonishok, 1980, Financial analysts’ forecast of earnings: Their value to investors, Journal of Banking and Finance 4, pp. 221233.
  • Hirshleifer, D., J. Myers, L. Myers, and S. H. Teoh, 2003,Do individual investors drive post-earnings announcement drift? Direct evidence from personal traders, Working Paper, Ohio State University, Columbus, Ohio.
  • Hong, H., and J. C. Stein, 1999, A unified theory of underreaction, momentum trading and overreaction in asset markets, Journal of Finance 54, pp. 21432184.
  • Hong, H., T. Lim, and J. Stein, 2000, Bad news travels slowly: Size, analyst coverage, and the profitability of momentum strategies, Journal of Finance 55, pp. 265295.
  • Jacob, J., T. Lys, and J. Sabino, 2000, Autocorrelation structure of forecast errors from time-series models: Alternative assessments of the causes of post-earnings announcement drift, Journal of Accounting and Economics 28, pp. 329358.
  • Johnson, W., and W. Schwartz Jr, 2001, Evidence that capital markets learn from academic research: Earnings surprises and the persistence of post-announcement drift, Working paper, University of Iowa, Iowa City, Iowa.
  • Jones, C., and R. Litzenberger, 1970, Quarterly earnings reports and intermediate stock price trends, Journal of Finance 25, pp. 143148.
  • Joy, O., R. Litzenberger, and R. McEnally, 1977, The adjustment of stock prices to announcements of unanticipated changes in quarterly earnings, Journal of Accounting Research 15, pp. 207225.
  • Ke, B., and S. Ramalingegowda, 2005, Do institutional investors exploit the post-earnings announcement drift? Journal of Accounting and Economics 39, pp. 2553.
  • Keim, D., and R. Stambaugh, 1986, Predicting returns in the stock and bond markets, Journal of Financial Economics 17, pp. 357390.
  • Kormendi, R., and R. Lipe, 1987, Earnings innovations, earnings persistence, and stock returns, Journal of Finance 60, pp. 323346.
  • Kothari, S. P., 2001, Capital markets research in accounting, Journal of Accounting and Economics 31, pp. 105231.
  • Kothari, S. P., J. Lewellen, and J. B. Warner, 2006, Stock returns, aggregate earnings surprises, and behavioral finance, Journal of Financial Economics 79, pp. 537568.
  • Latane, H., and C. Jones, 1979, Standardized unexpected earnings – 1971–77, Journal of Finance 34, pp. 717724.
  • Lee, B. S., 1995, The response of stock prices to permanent and temporary shocks to dividends, Journal of Financial and Quantitative Analysis 30, pp. 122.
  • Lee, B. S., 1996a, Time-series implications of aggregate dividend behavior, Review of Financial Studies 9, pp. 589618.
  • Lee, B. S., 1996b, Comovements of earnings, dividends, and stock prices, Journal of Empirical Finance 3, pp. 327346.
  • Lev, B., 1988, Toward a theory of equitable and efficient accounting Policy, Accounting Review 63, pp. 121.
  • Liang, L. H., 2003, Post-earnings announcement drift and market participants’ information processing biases, Review of Accounting Studies 8, pp. 321345.
  • Lo, A., and A. MacKinlay, 1988, Stock market prices do not follow random walks: Evidence from a simple specification test, Review of Financial Studies 1, pp. 4166.
  • Mendenhall, R., 1991, Evidence of the possible underweighting of earnings-related information, Journal of Accounting Research 29, pp. 170179.
  • Mendenhall, R., 2004, Arbitrage risk and post-earnings-announcement drift, Journal of Business 77, pp. 875894.
  • Michaely, R., R. H. Thaler, and K. L. Womack, 1995, Price reactions to dividend initiations and omissions: Overreaction or drift? Journal of Finance 50, pp. 573608.
  • Ou, I., and S. Penman, 1989, Financial statement analysis and the prediction of stock returns, Journal of Accounting and Economics 11, pp. 295330.
  • Poterba, J., and L. Summers, 1988, Mean reversion in stock prices: Evidence and implications, Journal of Financial Economics 22, pp. 2759.
  • Potter, G., 1992, Accounting earnings announcements, institutional investor concentration, and common stock returns, Journal of Accounting Research 30, pp. 146155.
  • Quah, D., 1992, The relative importance of permanent and transitory components: Identification and some theoretical bounds, Econometrica 60, pp. 107118.
  • Rendleman, R., C. Jones, and H. Latane, 1982, Empirical anomalies based on unexpected earnings and the importance of risk adjustments, Journal of Financial Economics 10, pp. 269287.
  • Rendleman, R., C. Jones, and H. Latane, 1987, Further insight into the standardized unexpected earnings anomaly: Size and serial correlation effects, The Financial Review 22, pp. 131144.
  • Shane, P., and R. Brous, 2001, Investor and (Value Line) analyst underreaction to information about future earnings: The corrective role of non-earnings-surprise information, Journal of Accounting Research 39, pp. 387404.
  • Sloan, R. G., 1996, Do stock prices fully reflect information in accruals and cash flows about future earnings? The Accounting Review 71, pp. 289315.
  • Soffer, L. C., and T. Lys, 1999, Post-earnings announcement drift and the dissemination of predictable information, Contemporary Accounting Research 16, pp. 305331.
  • Walther, B. R., 1997, Investor sophistication and market earnings expectations, Journal of Accounting Research 35, pp. 157179.
  • Watts, R., 1978, Systematic abnormal returns after quarterly earnings announcements, Journal of Financial Economics 6, pp. 127150.
  • Wiggins, J., 1991, Do misspecifications about the earnings process contribute to post-announcement drift? Working Paper, Cornell University, Ithaca, New York.