Does Meeting Expectations Matter? Evidence from Analyst Forecast Revisions and Share Prices

Transcription

1 Does Meeting Expectations Matter? Evidence from Analyst Forecast Revisions and Share Prices Ron Kasznik Graduate School of Business Stanford University Stanford, CA (650) Fax: (650) Maureen F. McNichols Graduate School of Business Stanford University Stanford, CA (650) Fax: (650) August 1999 We appreciate the many helpful comments of workshop participants at the 1998 Stanford Accounting Summer Camp, University of Iowa, University of Michigan, MIT, 1998 Financial Economics and Accounting Conference at New York University, and 1999 American Accounting Association Annual Meetings. We also appreciate the financial support of the Stanford University Graduate School of Business Financial Research Initiative.

2 1. Introduction Recent research provides strong evidence that firms manage both earnings and expectations to report earnings that meet or exceed analysts earnings forecasts. 1 In addition, the Securities Exchange Commission has expressed concern that the pressure to meet expectations is eroding the quality of financial reporting (Levitt [1998]). Although many firms appear to devote resources to expectations management, its value is controversial, as typified by a recent article in CFO magazine: With Wall Street s earnings targets for 1998 higher than ever and investors skittish about the course of a long-running bull market, companies that miss targets, even by small margins, face unpleasant consequences in the stock market. No wonder strategies for nudging targets downward are about as legion as cold remedies, and seldom more reliable... A debate is brewing over how much, or even whether, companies should attempt to manage earnings expectations, and whether the strategy can really affect how the market reacts to earnings news. (McCafferty [1997]) This paper provides evidence relevant to this controversy by testing whether firms achieve greater share value, all else equal, by meeting analysts expectations. 2 We hypothesize that firms meeting expectations receive either of two kinds of market rewards, higher analysts earnings forecasts that lead to higher valuations or a market reward controlling for earnings forecasts. Our test for a forecast-based reward addresses two questions. First, to what extent do analysts forecasts differ for firms that meet expectations from those of firms that fail to meet expectations? Second, to what extent should analysts earnings forecasts differ for firms that meet expectations? We examine how the future earnings of firms that meet expectations differ from those of firms that do not, and compare the ex post difference to the difference in analysts earnings forecasts. The answers to these questions allow us to 1 See Burgstahler and Dichev [1997] and Burgstahler and Eames [1998]. 2 See also Ip [1997] and Norris [1997], who argue that Wall Street does reward companies for lowering expectations. 1

3 determine whether firms that meet expectations are rewarded with higher earnings expectations than are justified by future earnings. Our test for a market reward examines whether the market assigns a greater share price to firms that meet expectations, controlling for an estimate of the firm s fundamental value. Our estimate of fundamental value, following Ohlson [1995] and Frankel and Lee [1996], is based on book value of equity and analysts forecasts of future earnings. We also test, using the empirical methodology of Mishkin [1983] and Sloan [1996], whether any market premium earned is due to investors having higher earnings expectations than analysts forecast for firms meeting expectations. Our findings indicate that although analysts do not fully incorporate the information in negative earnings forecast errors, their forecasts for longer-term earnings fully reflect the information implicit in consistently meeting expectations. Firms that consistently meet expectations do not experience future abnormal earnings that exceed the amounts forecast by analysts. Further, we find that firms that meet expectations once have a positive but statistically insignificant market premium. Firms that meet expectations in two consecutive years have a significant market premium, and firms that meet expectations in three consecutive years have an even greater premium. Because we do not find evidence that firms that meet expectations experience higher ex post abnormal earnings than are forecast by analysts, we argue that the market premium we document for these firms is not a correction for errors in analysts forecasts. Our study contributes to the accounting literature in several ways. First, we provide evidence that suggests a rationale for the substantial emphasis placed by firms on meeting analysts expectations. A large literature examines management earnings forecasts to understand how these forecasts affect expectations and to understand firms incentives to disclose information. 3 Several recent studies examine firms incentives to 3 This literature includes Patell [1976], Penman [1980], Waymire [1984], McNichols [1989], King, Pownall, and Waymire [1990], Skinner [1994], Frankel, McNichols, and Wilson [1995], Kasznik and Lev [1995], and Kasznik [1999]. 2

4 manage expectations by studying firms that ex post met or failed to meet expectations. 4 We extend this literature by examining the valuation consequences of meeting expectations and by developing a design that allows us to determine the extent to which higher stock prices for meeting expectations are justified by higher future earnings. Kinney, Burgstahler, and Martin [1999] find that the market reaction to small unexpected earnings is symmetric, suggesting that there is no market reward for meeting expectations. However, their study does not consider the implications of meeting expectations in multiple periods. Barth, Elliott and Finn [1999] present strong evidence that firms that report increasing earnings receive a higher earnings multiple. Their study is silent however on whether the higher multiple is justified by higher future earnings. Our study also contributes to the literature on analysts forecasts. Prior research indicates that analysts under-react to publicly available information. For example, Abarbanell [1991] documents that analysts do not incorporate information available in price, in that firms with positive (negative) returns are more likely to experience positive (negative) forecast errors. Abarbanell and Bernard [1992] document that analysts underreact to earnings information, although the magnitude does not seem sufficient to explain the post-announcement drift. These studies do not examine whether meeting expectations affects analysts forecast revisions, all else equal, nor do they examine whether the magnitude of analysts forecast revisions associated with meeting expectations is consistent with its implications for future earnings. To address our research question, we develop a research design that compares how analysts weight information to its underlying predictive ability. This approach can be used for a broad class of informational variables to understand how analysts forecasts reflect relevant information, and where they deviate from the weights one would expect given an ex post distribution. 4 For example, Matsumoto [1998] uses analysts earnings forecasts errors to measure whether a firm consistently meets expectations. 3

5 The paper is organized as follows. Section 2 develops the hypotheses. Section 3 describes the research design, estimation equations and variables we use to test the hypotheses. Section 4 describes the data and sample firms and section 5 presents the findings. Section 6 describes additional analyses, and section 7 provides a summary and concluding remarks. 2. Hypotheses We hypothesize that firms that meet expectations may receive either of two kinds of rewards, higher expectations for subsequent earnings or a market premium controlling for earnings expectations. Our first hypothesis addresses the extent to which analysts reward firms meeting expectations with more favorable earnings forecasts, all else equal. If analysts expect, all else equal, that firms that fail to meet earnings targets will have lower future earnings, then negative share price consequences associated with failure to meet current year earnings expectations may reflect lower expected earnings. Similarly, if analysts expect, all else equal, that firms that meet earnings targets will have higher future earnings, then positive share price consequences associated with meeting current year earnings expectations may reflect higher expected earnings. Our first hypothesis focuses on understanding how analysts forecasts of a firm s subsequent earnings are affected by whether the firm meets analysts earnings expectations for the current year. Stated formally, in alternative form: Hypothesis 1: Analysts forecasts of subsequent earnings are higher (lower) for firms that meet (fail to meet) earnings expectations, controlling for the current year earnings news. Analysts may under-react or over-react to the earnings information of firms that meet or fail to meet analysts expectations. To assess this, we examine the ex post 4

6 distribution of subsequent earnings, conditional on whether firms met analysts expectations. Meeting expectations may be informative about future earnings incremental to the year s earnings information because firms choices may determine whether they meet expectations and these choices may be correlated with information about future earnings. By estimating the relation between subsequent earnings and information available to analysts, including whether the firm met expectations, we estimate a benchmark for how analysts should weight their information. Our second hypothesis tests for an association between meeting expectations and subsequent earnings. Hypothesis 2: Subsequent earnings are higher (lower) for firms that meet (fail to meet) earnings expectations, controlling for the current year earnings news. Our third hypothesis examines whether analysts post-announcement forecasts reflect information implicit in whether firms meet expectations. Essentially, we test whether the weights on informational variables implicit in analysts forecasts are consistent with the predictive ability of these variables for future earnings. In other words, we test whether analysts weight their information as they should, given its predictive ability. Hypothesis 3: The association between analysts forecasts of subsequent earnings and whether firms meet expectations is consistent with the predictive ability of meeting expectations for subsequent earnings. 5

7 This analysis casts light on how analysts respond to earnings surprises, and on how their response may provide incentives for management to meet expectations. Our fourth hypothesis examines whether share prices are higher for firms that meet expectations, controlling for expectations of subsequent earnings. We test whether meeting expectations is value-relevant, and if so, whether its association with share prices can be explained by its predictive ability for future earnings. Hypothesis 4: The earnings expectations in stock prices reflect the predictive ability of meeting expectations for future earnings. 3. Research Design Our study focuses on the implications of meeting expectations. Our research design requires a measure of the information in the current year s earnings realization, and a measure of whether expectations were met. Figure 1 describes the timeline that underlies the measurement of our forecast and earnings variables. We measure the information in year t s earnings as the forecast error for year t, AFE t. Specifically, AFE t equals realized earnings per share for year t, ACTEPS0, less the mean of analysts year t earnings per share forecasts at the beginning of year t, AF0 beg. The forecast error is comprised of two components: REVISION t, the revision from the mean beginning of year forecast, AF0 beg, to the mean pre-announcement forecast at t pre, AF0 pre, and SURPRISE t, the earnings surprise measured relative to the mean pre-announcement forecast. We define MEET, an indicator for firms that meet or exceed earnings expectations, where MEET=1 if SURPRISE t is non-negative and MEET=0 if SURPRISE t is negative. Our notion of meeting expectations therefore relates to the firm s annual earnings relative to expectations shortly before its announcement. We expect that 6

8 analysts and investors learn about a firm s ability to meet expectations over time. We therefore examine whether firms meet expectations in the current year (MEET1=1), the current and past year (MEET2=1), and the current and two prior years (MEET3=1). A key element of our study is its analysis of the future earnings implications of meeting expectations. Specifically, we test whether analysts revisions of subsequent year earnings forecasts are associated with whether firms met expectations in the current or prior years. To do so, we measure AF1 beg, the mean analysts forecast of earnings per share for year t+1 issued at the beginning of year t, and examine its association with AF1 post, the mean analysts forecast of year t+1 earnings per share after year t earnings are announced. Our first hypothesis concerns the relation between meeting consensus expectations in the current or past years and analysts expectations of subsequent-year earnings. We examine this with the following estimation equation: + AF1 = α + α AF + α AFE + α AFE + α MEETi + ε (1) post 0 1 1beg This equation specifies that analysts forecasts of subsequent year earnings are associated with their beginning of year forecast, AF1 beg, year t s earnings forecast error, AFE t, partitioned by sign, and with whether firms met expectations in the i most recent years, MEETi. The model assumes analysts posterior forecasts condition on their prior forecast for year t+1 earnings, and on the information in year t s earnings relative to the start of year expectation. We allow separate coefficients on the year t earnings forecast error given prior evidence that analysts are less likely to reflect negative earnings news than positive earnings news in their subsequent year earnings forecasts. Our second hypothesis concerns whether subsequent earnings differ for firms that meet expectations from that for firms that fail to meet expectations. By estimating the relation between subsequent earnings and available information, we have a basis for characterizing analysts response to the same conditioning variables. Accordingly, we 7

9 estimate the relation between ACTEPS1, actual earnings per share for year t+1, and the conditioning variables in equation (1): + ACTEPS1 = β 0 + β1af1beg + β2afe + β3afe + β4meeti + ε (2) Equation (2) examines the relation between subsequent earnings and meeting expectations in the current year, controlling for the current year earnings news. This estimation equation reflects the ex post predictive ability of the information analysts condition on to form their forecasts of ACTEPS1, including the information in whether the firm met expectations. Our third hypothesis concerns whether analysts respond to earnings information as one would expect given the relation between subsequent earnings and available information. Comparison of the coefficients for forecast equation (1) and earnings equation (2) permits us to assess the extent to which analysts under-react or over-react to earnings information, and its relation to whether firms meet expectations. We provide tests of differences between the coefficients in the earnings equation and the related forecast equation by jointly estimating (1) and (2) using seemingly unrelated regression. Although the coefficient estimates and standard errors are identical to those estimated using OLS, the seemingly unrelated regression estimation allows for correlation between the error terms in equations (1) and (2). It also estimates the full variance-covariance matrix of the estimators, allowing us to test hypotheses about the coefficient estimates across equations. We also present evidence on the extent to which analyst forecasts reflect information in meeting expectations with the following specifications. The dependent variables, EPS for years t+1, t+2 and t+3, are regressed on the mean analysts forecasts of EPS for years t+1, t+2 and t+3 issued after EPS t, which we refer to as AF1 post, AF2 post and AF3 post. The estimation equations are: 8

10 ACTEPS1 φ + φ AF post + φ MEETi + e = ACTEPS2 φ + φ AF post + φ MEETi + e (3) = ACTEPS3 φ + φ AF post + φ MEETi + e = If analyst forecasts fully reflect information about year t earnings, then φ 2 should equal zero. These estimation equations therefore provide information about the predictive ability of meeting expectations, controlling for analysts post-announcement forecasts. If future earnings are related to whether firms have met expectations in recent years, controlling for analysts post-announcement forecasts, then these forecasts do not fully reflect the information in year t earnings. 5 Our final hypothesis concerns whether there is a market premium for meeting expectations. To test this hypothesis, we need to control for firm value that is due to fundamentals. Our valuation analysis comprises two steps. The first is to construct an accounting-based measure of firm value motivated by Ohlson [1995] and Frankel and Lee [1996]. Specifically, we measure firm value as the sum of BVE t, the end-of-year t book value of equity per share, and PVINCOME, the present value of expected abnormal earnings per share: PVINCOME = 2 i= 1 E ( EPS t t+ i ) re t t (1 + r) ( BVE t+ i 1 ) E t ( EPS + t+ 3 ) re (1 + r) t 2 r ( BVE where E t (EPS t+i ) equals the time t mean analyst forecast of earnings per share for year t + i, for i = 1 and 2. Because most analysts do not forecast annual earnings beyond two years into the future, but often issue a three-to-five year earnings growth estimate, we set E t (EPS t+3 ) equal to the time t mean analyst long-term earnings growth forecast times t+ 2 ) 5 This specification allows us to examine whether information in meeting expectations is reflected in analyst forecasts in a single equation rather than using both equations (1) and (2). Because we will test this for three years of future earnings and for three measures of meeting expectations, its parsimony is useful. Equations (1) and (2) are not redundant, however, as they allow us to examine how analysts weight the information in meeting expectations in forming their posterior expectations. 9

11 E t (EPS t+2 ). Et ( BVE t + i ) equals BVE t + i 1 + Et ( EPSt + i ) Et ( DIVt + i ), where DIV t+i is dividends in year t+i. We set Et ( DIV t + i ) equal to E t ( EPS t + i ) PAYOUT, where PAYOUT is the firm s dividend payout ratio. 6 The firm s cost of equity capital, r, equals the greater of R + β R R ) or R f, where β is estimated using the capital asset pricing f ( m f model, R f is the 10-year Treasury Note rate, and (R m R f ) is 7.6%, following Ibbotson Associates [1996]. Our test for a market effect associated with meeting expectations is based on the following equation: PRICE = γ 0 + γ 1BVE + γ 2PVINCOME + γ 3MEETi + ε (4) where PRICE is price per share three months after fiscal yearend, and BVE and PVINCOME are deflated by yearend shares outstanding. PVINCOME is based on analysts forecasts issued after year t s earnings are announced. This specification allows us to test whether there is a market penalty or reward for meeting expectations, controlling for an estimate of firm value based on analysts earnings forecasts. Analysts forecasts may fail to fully reflect earnings information and the extent to which earnings information is reflected may be correlated with whether expectations were met. The second part of our valuation analysis determines whether this explains the coefficient on MEETi in (4). Specifically, we compare γ 3 to the coefficient on MEET obtained from a specification that incorporates an objective measure of future earnings. Specifically, we estimate ACTPVINC, an ex post measure of PVINCOME using the subsequent three years earnings, as follows: ACTPVINC = 2 i= 1 EPS t+ i re t (1 + r) ( BVE t t+ i 1 ) EPS + t+ 3 re (1 + r) ( BVE t 2 r t+ 2 ) 6 We define the dividend payout ratio as the average of the prior three years ratio of dividends-to-net income. Consistent with Frankel and Lee [1996], among others, if net income in a particular year is negative, for purposes of this calculation, we set it equal to 6% of total assets. 10

12 We then estimate the relation between ACTPVINC, our estimate of the ex post present value of abnormal earnings, and analysts predictions reflected in PVINCOME. We test whether ACTPVINC is greater for firms that meet expectations than for firms that do not, controlling for analysts predictions of the present value of future abnormal income, with the following specification: ACTPVINC λ + λ PVINCOME + MEETi + e (5) = 0 1 λ2 We compare the coefficients on MEETi in (4), the valuation equation, and in (5), the ex post abnormal earnings equation, to test whether the valuation difference associated with meeting expectations is justified by its implications for future abnormal earnings. Specifically, using the methodology of Mishkin [1983], we jointly estimate the system of equations comprised of (4) and (5) using iterative weighted non-linear least squares, and test for a difference in coefficients across the equations with a likelihood ratio statistic Data The initial sample includes all firms in the 1996 COMPUSTAT Merged Annual Industrials, Full Coverage and Primary-Supplementary-Tertiary files. Sample firms are required to have forecasts of EPS one and two years ahead, and forecasts of long-term earnings growth from I/B/E/S. The sample period begins in 1986 to ensure a reasonable number of observations with data available on variables required for our research design. The sample period ends in 1993 to allow for tests using data for three subsequent years. To permit comparisons of the effects of meeting expectations in one year to multiple years, we present our results for the sample of firms with data for at least three years. Our sample period is therefore , and includes 3,373 firm-year observations. 7 We thank Richard Sloan for providing programming code used to conduct these tests. 11

13 Realized earnings per share, ACTEPS0, ACTEPS1, ACTEPS2, and ACTEPS3, are obtained from I/B/E/S to ensure appropriate stock split and dividend adjustments, and to avoid inconsistencies in the definitions of earnings between the COMPUSTAT and I/B/E/S databases. We measure analysts forecasts of EPS t and EPS t+1 at t beg, the beginning of the year, and analysts forecasts of EPS t at t pre, prior to the earnings announcement. We measure analysts forecasts of EPS t+1, EPS t+2, and EPS t+3 at t post, following the year t earnings announcement. The mean analyst forecast at t beg (t post ) is calculated using the forecast issued by each analyst closest to but within 90 days following the year t-1 (t) earnings announcement. The mean analyst forecast at t pre is calculated using the forecast issued by each analyst closest to but within the 90 days preceding the year t earnings announcement. Table 1 presents descriptive statistics on the variables used in our research design. The mean analysts forecast for EPS t issued at the beginning of year t, AF0 beg, is $1.60, with a median of $1.45. The mean analysts earnings forecast for EPS t issued prior to the announcement of EPS t, AF0 pre, is $1.44, with a median of $1.32. The mean forecast for EPS t+1 at the beginning of year t, AF1 beg, is $1.84 with a median of $1.70. The mean of AF1 post, analysts earnings forecasts for EPS t+1 after the announcement of EPS t declines to $1.63. The means of AF2 post and AF3 post, the post-announcement forecasts of EPS t+2 and EPS t+3, are $1.91 and $2.14 respectively. The means of EPS for years t to t+3, ACTEPS0, ACTEPS1, ACTEPS2, and ACTEPS3, are well below the means for the respective forecast variables, at $1.43, $1.46, $1.54 and $1.68, consistent with the optimism in analysts forecasts documented in prior research (Brown Foster and Noreen [1985], O Brien [1988]). The mean earnings forecast error relative to the mean beginning of year forecast is -$0.17, with a mean revision over the year of -$0.15, and a mean earnings surprise of -$0.01. The mean for MEET1 is 0.59, indicating that 59% of sample firms met expectations in the current year. The means for MEET2 and MEET3 indicate 12

14 that 37% of sample firms met expectations in two consecutive years, and 24% met expectations in three consecutive years. The average share price is $22.79, average book value of equity is $11.77 per share, and the average present value of expected abnormal earnings is $5.15 per share. Firms in our sample are followed by analysts, on average. The mean of DCHEPS indicates that 68% of sample firms experienced an increase in earnings relative to the prior year. 5. Findings Table 2 presents the estimation results for equations (1) and (2). The left side of Table 2 documents that AF1 post is significantly associated with AF1 beg, AFE + and AFE -. The coefficient on AFE +, 1.26, is greater than the coefficient on AFE -, 0.75, indicating that analysts weight positive earnings news more heavily than negative news in forming their conditional expectations of future earnings. The coefficient on MEET1 is 0.04 and is significant at less than 0.01, indicating that, surprisingly, analysts forecasts are 4 cents per share lower, on average, for firms that meet or exceed expectations than for firms that do not. The right side of Table 2 documents that ACTEPS1 is positively associated with AF1 beg, AFE +, AFE -, and MEET1. The coefficient on AF1 beg is 0.84, which is significantly less than 0.91, the respective coefficient in the forecast equation. This indicates that analysts overweight their prior expectations relative to their ex post predictive ability. The coefficient on AFE + is 1.26, the same as the estimated coefficient in (1). This suggests analysts weight on AFE + is consistent with its ex post predictive weight. In contrast, the coefficient on AFE - exceeds the respective coefficient in (1), indicating that analysts underweight the information in negative current year earnings news. Finally, the coefficient on MEET1 is positive but insignificant. In contrast, the estimated coefficient in (1) indicates that analysts forecasts for the subsequent year are on average lower for these firms. These findings indicate that analysts do not fully reflect 13

15 the implications of meeting expectations in their subsequent year earnings forecasts, and that they do not reward firms that meet expectations with higher earnings forecasts, all else equal. Panel B of Table 2 presents the estimation results when firms meeting expectations have done so for the current and prior year. The estimated coefficients are similar, though the coefficient on MEET2 is less negative, suggesting that analysts forecasts are slightly higher for firms that meet expectations in two years rather than in one. This trend continues in Panel C, where we find the coefficient on MEET3 is 0.02, and is no longer statistically significant. However, comparison to the ex post distribution on the right side of Table 2 indicates that this information remains underweighted. Firms meeting expectations in the three most recent years have significantly higher ex post earnings in the subsequent year than firms that do not, as indicated by the significant positive coefficient on MEET3. Table 3 presents the estimation results for equation (3), in which ACTEPS1, ACTEPS2 and ACTEPS3 are regressed on the post-announcement forecast of earnings for the respective year and MEETi. As discussed earlier, this estimation equation provides a parsimonious way to examine whether forecasts reflect the information in meeting expectations. The findings in Panel A are consistent with those in Table 2 and indicate that analysts post-announcement forecasts do not fully incorporate information in MEET1 for future earnings. In particular, subsequent earnings are higher in years t+1, t+2 and t+3 for firms that meet or exceed expectations in year t, controlling for analysts post-announcement forecasts, though to a lesser extent in year t+3. Panel B presents the estimation results for the model comparing future earnings for firms that meet expectations in the current and past year to those that did not meet expectations in both years. Earnings in the following two years are significantly greater than what analysts forecast for firms that met expectations in the current and prior year, but longer-term earnings, ACTEPS3, are not significantly higher. This trend is also 14

16 observed in Panel C, where the coefficient on MEET3 in the ACTEPS3 regression decreases to 0.03, and is insignificantly positive. These findings suggest that analysts forecasts reflect more information about firms longer-term earnings when they consistently meet expectations. Table 4 presents the estimation results for equations (4) and (5). The coefficients indicate that share prices are significantly positively associated with BVE and PVINCOME, as documented by prior research. Furthermore, controlling for BVE and PVINCOME, the coefficient on MEET1, 0.48, is positive though not significantly so (t=1.52). The right side of table 4 presents the estimation results for equation (5), relating the ex post present value of abnormal earnings to analysts estimates and to MEET1. It documents that ACTPVINC is significantly positively associated with PVINCOME, our estimate of the present value of abnormal earnings based on analysts forecasts. The intercept is significantly negative, consistent with the over-optimism in analysts earnings forecasts for years t+1, t+2 and t+3. The coefficient on MEET1 is significantly positive, indicating that firms that meet expectations experience a present value of abnormal earnings that is $0.58 per share greater than firms that do not, controlling for PVINCOME. These findings indicate, consistent with tables 2 and 3, that meeting expectations in a given year is associated with higher earnings than analysts forecast for subsequent years. Finally, we examine whether the magnitude of the coefficient on MEETi in the valuation equation is consistent with the future earnings associated with meeting expectations, incremental to analysts future earnings forecasts. We test this using the methodology developed by Mishkin [1983] and Sloan [1996]. For firms meeting expectations once, we cannot reject the hypothesis that the coefficients on MEET1 in (4) and (5) are equal; the p-value of the likelihood ratio test is Panel B indicates that firms meeting expectations in both the current and prior year receive a value that is $1.06 greater per share than firms that fail to meet expectations. Furthermore, the estimation 15

17 results for equation (5) indicate that the present value of ex post abnormal earnings is $0.33 per share greater for these firms. We reject the null hypothesis that γ 3 and λ 2 are equal at the level. This finding suggests that at least part of the market premium is not due to higher abnormal earnings that could rationally be expected for these firms. The findings in Panel C support this to an even greater degree. The coefficient on MEET3 is 1.92, indicating the market premium is greater still for firms that meet expectations in three years relative to that for firms meeting expectations in two years or less. Furthermore, the estimation results for equation (5) indicate that the market premium we identify in (4) is unlikely to reflect expectations by investors of higher future abnormal earnings than analysts forecast. The coefficient on MEET3 in equation (5) is only 0.18, and is insignificantly different from zero. The likelihood ratio statistic that γ 3 equals λ 2 is which is significantly different from zero at the level. We therefore reject the null hypothesis that the coefficients are equal. Taken as a whole, the findings in Table 4 indicate that firms that meet expectations receive a market premium that increases as they continue to meet expectations. The equation (5) estimation results also indicate that analysts forecasts better reflect the implications of their earnings for future earnings for firms that consistently meet expectations. Firms that met expectations in only the current year experience higher future abnormal earnings than analysts forecast, but firms that have met expectations in three consecutive years do not. 6. Additional analyses In Table 5, we examine how the implications of meeting expectations vary across firms that are more widely or less widely followed by analysts. We estimate equations (4) and (5) for firm-year observations with following greater than or equal to and less than the sample median for the year. The findings document a generally similar pattern for both subsamples. For the more widely-followed firms, the coefficients in the valuation equation for MEET1, MEET2 and MEET3 are 0.10, 0.31, and 1.26 respectively. Similar 16

18 to the findings for the sample as a whole, the valuation effect of meeting expectations is greater the more frequently expectations are met. In contrast, the predictive ability of MEET2 and MEET3 for ACTPVINC is smaller than that for MEET1, suggesting analysts better reflect underlying information for firms that consistently meet expectations. However, for all three specifications, we cannot reject the hypothesis that the coefficients on MEETi in (4) and (5) are equal; p-values of the likelihood ratio test for MEET1, MEET2, and MEET3 are 0.225, 0.920, and 0.137, respectively. Panel B of Table 5 presents the estimation results for firms with analyst following less than the sample median. In contrast to the more widely-followed firms, for less widely-followed firms there is evidence of a market reward for meeting expectations even in a single period. The coefficient on MEET1 is 1.47 and is significantly greater than the coefficient on MEET1 in equation (5). For firms meeting expectations twice, the valuation coefficient on MEET2 climbs to 2.03 while the equation (5) coefficient on MEET2 falls to For firms meeting expectations three times, the valuation coefficient on MEET3 climbs further still, to 2.54, and the coefficient on MEET3 in the ACTPVINC equation falls further, to The hypothesis that the coefficients on MEETi in (4) and (5) are equal is rejected in all three specifications; p-values of the likelihood ratio test for MEET1, MEET2, and MEET3 are 0.061, 0.001, and 0.001, respectively. These findings indicate that firms with less analyst following experience the greatest rewards associated with consistently meeting expectations. Finally, in Table 6, we compare the magnitude of the market reward for meeting expectations to the market reward for increasing earnings documented by Barth, Elliott and Finn 1999]. Our findings indicate that controlling for this effect, firms meeting expectations in one year achieve no incremental premium. The findings indicate a premium of $2.52 per share, however, for firms with increasing earnings. For firms meeting expectations in two consecutive years, our estimation results indicate an incremental market reward of $0.63 per share, and a market reward of $2.34 for 17

19 increasing earnings. For firms meeting expectations in three consecutive years, our estimation results indicate that there is an incremental market reward of $1.57 per share associated with meeting expectations consistently, and a market reward of $2.26 for increasing earnings. Furthermore, the equation (5) results indicate that MEET2, MEET3, and DCHEPS are not significantly associated with higher future abnormal earnings controlling for PVINCOME. Although we cannot reject the hypothesis that the coefficients in (4) and (5) are equal for either MEET1 or MEET2, we reject the hypothesis that the coefficients on MEET3 are equal across the two equations at the level. This indicates that the premia associated with increasing earnings and with consistently meeting expectations are not due to their predictive ability for future earnings, as future abnormal earnings are not significantly higher than analysts expectations for either set of firms. Our findings indicate that the market rewards we document reflect distinct premia rather than correction for misspecification in analysts earnings forecasts. 7. Summary and Conclusions This study examines whether firms achieve greater market value by meeting analysts expectations. We hypothesize that such firms may be rewarded with higher earnings forecasts that lead to higher share values, or with higher share prices controlling for analysts forecasts. We find that analysts forecasts are not higher for firms that meet expectations relative to those that do not, controlling for the level of the current year s earnings information. We also find that one or two years ahead earnings are higher for firms that meet expectations relative to those that do not, controlling for the level of analysts postannouncement earnings forecasts. Earnings three years ahead, however, are not generally greater for firms that consistently meet expectations, controlling for analysts postannouncement earnings forecasts. These findings indicate firms meeting expectations are not rewarded by analysts with higher earnings forecasts than are warranted. 18

20 Consistent with the hypothesis that the market rewards firms for meeting expectations, we find that share prices are higher for firms that meet expectations in more than one year than for those that do not. These tests control for an estimate of fundamental value based on the book value of equity and analysts estimates of the present value of abnormal earnings. For firms meeting expectations in one year, the magnitude of the share price difference can be explained by the higher present value of future abnormal earnings realized by these firms. For firms meeting expectations in two years, the market reward is greater and the incremental future abnormal earnings realized by these firms are of insufficient magnitude to explain the market reward. For firms meeting expectations in three consecutive years, the market reward is greater still, and the incremental future abnormal earnings of these firms are insignificantly positive. Our findings therefore support a market reward for firms that consistently meet expectations. This market reward could reflect a lower cost of capital for these firms or a distinct market premium for firms that meet expectations. In addition to documenting a market reward for meeting expectations, our paper contributes to the literature on financial analysts processing of accounting information. To address whether analysts forecasts fully reflect the information inherent in a firm s meeting expectations, we develop a methodology to compare how analysts weight information with the ex post predictive ability of that information. This approach can be applied to a broad class of questions related to how analysts process information, including questions about how analysts process information generated from alternative accounting systems. Finally, our paper has implications for the literature on accounting-based valuation. Our evidence indicates that analysts forecasts do not fully reflect available information, and that failure to adjust for this in valuation analyses can lead to misleading inferences. Our paper contributes to this literature by developing a methodology for 19

21 testing whether the value-relevance of a given factor derives from its implications for future abnormal earnings. 20

22 References Abarbanell, J.S Do Analysts Earnings Forecasts Incorporate Information in Prior Stock Price Changes? Journal of Accounting and Economics 14 (2): Abarbanell, J.S., and V.L. Bernard Tests of Analysts Overreaction/underreaction to Earnings Information as an Explanation for Anomalous Stock Price Behavior. Journal of Finance 47 (3): Barth, M., J. Elliott, and M. Finn Market Rewards for Increasing Earnings Patterns. Forthcoming, Journal of Accounting Research. Brown, P., G. Foster, and E. Noreen Security Analyst Multi-year Earnings Forecasts and the Capital Market. Studies in Accounting Research No. 21. American Accounting Association, Sarasota FL. Burgstahler, D., and I. Dichev Earnings Management to Avoid Earnings Decreases and Losses. Journal of Accounting & Economics 24: Burgstahler, D., and M. Eames Management of Earnings and Analyst Forecasts. Working Paper, University of Washington, Seattle, WA. Frankel, R., and C. Lee Accounting Valuation, Market Expectation, and the Bookto-Market Effect. Journal of Accounting & Economics 25: Frankel, R., M. McNichols, and G.P. Wilson Discretionary Disclosure and External Financing. The Accounting Review 70 (1): Ibbotson Associates. Stocks Bonds Bills and Inflation. Ibbotson Associates, Ip, Greg. Rise in Profit Guidance Dilutes Positive Surprises. The Wall Street Journal, June 23, 1997, C1. Kasznik, R On the Association between Voluntary Disclosure and Earnings Management. Journal of Accounting Research 37 (Spring): Kasznik, R., and B. Lev To Warn or not to Warn: Management Disclosures in the Face of an Earnings Surprise. The Accounting Review 70 (1): King, R.G., G. Pownall, and G. Waymire Expectations Adjustment via Timely Management Forecasts: Review, Synthesis, and Suggestions for Future Research. Journal of Accounting Literature 9: Kinney, W., D. Burgstahler, and R. Martin The Materiality of Earnings Surprises. Working Paper, University of Washington, Seattle, WA. 21

23 Matsumoto, D.A Management s Incentives to Influence Analysts Forecasts. Working Paper, University of Washington, Seattle, WA. McCafferty, J Why Managing Expectations often doesn t Work. CFO Magazine (October): McNichols, M Evidence of Informational Asymmetries from Management Earning Forecasts and Stock Returns. The Accounting Review 64 (January): Mishkin, F.S A Rational Approach to Macroecononometrics Testing Policy Ineffectiveness and Efficient-Markets Models, Chicago, IL: The University of Chicago Press. Norris, F A C is Great, if an F was Expected. The New York Times April 27, 1997, C1. O Brien, P Analysts Forecasts as Earnings Expectations. Journal of Accounting and Economics 10 (1): Ohlson, J Earnings, Book Values and Dividends in Security Valuation. Contemporary Accounting Research 12: Patell, J.M Corporate Forecasts of Earnings per Share and Stock Price Behavior: Empirical Tests. Journal of Accounting Research 14 (2): Penman, S. H An Empirical Investigation of the Voluntary Disclosure of Corporate Earnings Forecasts. Journal of Accounting Research 18 (1): Skinner, D Why Firms Voluntarily Disclose Bad News. Journal of Accounting Research 32 (Spring): Sloan, R.G Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? The Accounting Review 71 (3): Waymire, G. Additional Evidence on the Information Content of Management Forecasts. Journal of Accounting Research 22 (Autumn):

24 Figure 1 Timeline for Measurement of Forecasts EPSt-1 announced Analysts beginning of year forecast of EPS t,eps t+1 Analysts forecast of EPS t prior to announcement of EPS t EPS t is announced Analysts forecast of EPS t+1 after announcement of EPS t EPS t+1 is announced t-1 t beg t pre t t post t+1 AF1 beg= forecast[epst+1] AF1 post= forecast[epst+1] ACTEPS1=EPS t+1 AF0 beg= forecast[epst] AF0 pre= forecast[epst] ACTEPS0=EPS t REVISION SURPRISE AFE

25 Table 1 Distribution statistics for variables used in regressions, pooled over time. Sample of 3,373 firm-year observations from Variable Mean Median Std dev AF0 beg AF0 pre AF1 beg AF1 post AF2 post AF3 post ACTEPS ACTEPS ACTEPS ACTEPS AFE AFE AFE REVISION SURPRISE MEET MEET MEET PRICE BVE PVINCOME #ANALYSTS DCHEPS Variable definitions: AF0 beg = mean analysts forecast of EPS t issued at the beginning of year t. AF0 pre = mean analysts forecast of EPS t issued shortly before announcement of EPS t. AF1 beg = mean analysts forecast of EPS t+1 issued at the beginning of year t. AF1 post = mean analysts forecast of EPS t+1 issued after the announcement of EPS t. AF2 post = mean analysts forecast of EPS t+2 issued after the announcement of EPS t. AF3 post = mean analysts forecast of EPS t+3 issued after the announcement of EPS t, measured as AF2 post times analysts earnings growth forecast. 23

26 ACTEPS0 = realized EPS t. ACTEPS1 = realized EPS t+1. ACTEPS2 = realized EPS t+2. ACTEPS3 = realized EPS t+3. AFE = analysts forecast error, measured as ACTEPS0 less AF0 beg. AFE + = equals AFE when the analysts forecast error is positive, and zero otherwise. AFE = equals AFE when the analysts forecast error is negative, and zero otherwise. REVISION = analysts forecast revision, measured as AF0 pre less AF0 beg. SURPRISE = earnings surprise, measured as ACTEPS0 less AF0 pre. MEET1 = indicator equal to one if SURPRISE for year t is non-negative, and zero otherwise. MEET2 = indicator equal to one if SURPRISE for years t and t 1 are both non-negative, and zero otherwise. MEET3 = indicator equal to one if SURPRISE for years t, t 1, and t 2 are all nonnegative, and zero otherwise. PRICE = price per share three months after the fiscal yearend. BVE = end-of-year book value of equity. PVINCOME = the present value of expected abnormal earnings per share, calculated as: PVINCOME = 2 i= 1 E ( EPS t t+ i ) re (1 + r) t t ( BVE t+ i 1 ) E t ( EPS + t+ 3 ) re (1 + r) t 2 r ( BVE where E t (EPS t+i ) is time t mean analyst forecast of earnings per share for t+i for i = 1 and 2, and E t (EPS t+3 ) is time t mean analyst long-term earnings growth forecast times E t (EPS t+2 ). Et ( BVE t + i ) equals BVE t + i 1 + Et ( EPSt + i ) Et ( DIVt + i ), where DIV t+i is dividends in year t+i. Et ( DIV t + i ) is set equal to E t ( EPS t + i ) PAYOUT, where PAYOUT is the firm s dividend payout ratio. The firm s cost of equity capital, r, equals the greater of R f + β ( Rm R f ) or R f, where β is estimated using the capital asset pricing model, R f is the 10-year Treasury Note rate, and (R m R f ) is 7.6%, following Ibbotson Associates [1996]. #ANALYSTS = number of analysts following the firm. DCHEPS = indicator equal to one if realized EPS t is greater than, or equal to, realized EPS t 1, and zero otherwise. All variables, except for indicator variables and #ANALYSTS, are per-share. t+ 2 ) 24

27 Table 2 Estimation results from regressions of AF1 post, mean analysts forecast of subsequent year EPS and ACTEPS1, realized subsequent year EPS, on AF1 beg, AFE +, AFE, and an indicator for meeting analyst expectations. Analysis is performed separately using MEET1, MEET2, and MEET3, reflecting whether the company has met analysts forecasts in the most recent year, most recent two years, and most recent three years, respectively. Sample of 3,373 firm year observations from AF1 α α α + α post = 0 + 1AF1beg + 2 AFE + 3AFE + α 4 MEETi + ε (1) ACTEPS1 = β β β + + β 0 + 1AF1beg + 2 AFE 3AFE + β 4 MEETi + ε (2) Panel A: Meeting Expectations in Most Recent Year AF1 post ACTEPS1 p-value a Coefficient t-statistic Coefficient t-statistic Intercept AF1 beg AFE AFE MEET Adjusted R Panel B: Meeting Expectations in Recent Two Years AF1 post ACTEPS1 p-value a Coefficient t-statistic Coefficient t-statistic Intercept AF1 beg AFE AFE MEET Adjusted R Panel C: Meeting Expectations in Recent Three Years AF1 post ACTEPS1 p-value a Coefficient t-statistic Coefficient t-statistic Intercept AF1 beg AFE AFE MEET Adjusted R See Table 1 for variable definitions a We test whether the coefficients on each independent variable are equal in the regressions with AF1 post and ACTEPS1 as dependent variables using seemingly unrelated regressions. Two-tailed probability values are reported. 25

28 Table 3 Estimation results from regressions of ACTEPS1, ACTEPS2, and ACTEPS3, realized EPS one, two and three years following year t, on mean analysts forecast of subsequent years EPS, AF1 post,, AF2 post, and AF3 post, respectively, and an indicator for meeting analyst expectations. Analysis is performed separately using MEET1, MEET2, and MEET3, reflecting whether the company has met analysts forecasts in the most recent year, most recent two years, and most recent three years, respectively. Sample of 3,373 firm-year observations from Panel A: Meeting Expectations in Most Recent Year ACTEPS1 ACTEPS2 ACTEPS3 Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Intercept AF1 post AF2 post AF3 post MEET Adjusted R Panel B: Meeting Expectations in Recent Two Years ACTEPS1 ACTEPS2 ACTEPS3 Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Intercept AF1 post AF2 post AF3 post MEET Adjusted R Panel C: Meeting Expectations in Recent Three Years ACTEPS1 ACTEPS2 ACTEPS3 Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Intercept AF1 post AF2 post AF3 post MEET Adjusted R See Table 1 for variable definitions 26