Is Guidance a Macro Factor? The Nature and Information Content of Aggregate Earnings Guidance*

Transcription

1 Is Guidance a Macro Factor? The Nature and Information Content of Aggregate Earnings Guidance* Carol Anilowski University of Michigan Business School Mei Feng Katz School of Business, University of Pittsburgh Douglas J. Skinner University of Michigan Business School March 2004, revised October 2004 Abstract Although there is a great deal of research that documents the information content of management earnings forecasts at the firm level, there is almost no research on the informativeness of aggregate guidance. We argue that aggregate earnings guidance is informative at the market/economy level through its effects on expectations about market-level expected future cash flows and expected returns. Consistent with aggregate guidance capturing information about economy-wide cash flows, we find that aggregate guidance, especially relative levels of quarterly downward guidance, is associated with analyst- and time-series-based measures of aggregate earnings news. We also find that guidance affects market returns in those months each quarter when the most guidance is released, and that relative levels of downward guidance are especially informative. Overall, the evidence supports our contention that guidance affects market returns by aggregating news about market-level cash flow shocks and through its effects on market uncertainty. * We received useful comments from workshop participants at Carnegie Mellon University, The University of Chicago, and the University of Exeter (Xfi Conference) as well as from Ray Ball, Phil Berger, Alan Gregory, and Greg Miller. Skinner s work on this project was supported by the Neubauer Faculty Fellows program at the University of Chicago Graduate School of Business during the academic year. We also appreciate the financial support of the University of Michigan Business School.

2 1. Introduction Earnings guidance is now pervasive, with thousands of management earnings forecasts released every year by close to two thousand firms. As a result, stock market commentators sometimes make statements about the implications of trends in aggregate earnings guidance for market returns. For example, financial press discussions of market-moving economic and financial news sometimes feature statistics on the relative number of negative earnings preannouncements ( earnings warnings ) in a quarter. The implication is that earnings guidance is informative with respect to overall earnings trends in the economy, and perhaps even for trends in macroeconomic variables such as corporate investment and GDP growth, and so affects market returns. 1 Our principal research question is whether aggregate earnings guidance provides information to the market. Whether aggregate guidance affects market returns depends on the implications of guidance for market-wide expected future cash flows and expected returns. With respect to cash flows, the informativeness of guidance depends on the extent to which earnings guidance, when aggregated across firms, is pervasive, representative, and timely, as well as on interactions among these factors. Because managers have traditionally issued guidance when their firms are performing unusually well or unusually poorly, it is not obvious that aggregate guidance, even if pervasive, will be informative about economy-wide earnings; guidance may not be representative. On the other hand, guidance may be informative even if it is issued by a 1 See, for example, Storm Warnings Replace Profit Dreams, Wall Street Journal, Abreast of the Market, September 23, 2002, C1; Stocks Face Test as Quarter Ends, Wall Street Journal, Abreast of the Market, June 30, 2003, C1; Wall Street Expects Rosy Earnings Figures Might Show Strongest Profits Growth Since 2000 and Further.., Financial Times, p. 27, October 13, Many of these articles cite Chuck Hill, until recently director of research at Thomson First Call. Thomson First Call produces a weekly earnings report ( This Week in Earnings ) which reports, among other statistics, summary data on the ratio of negative to positive preannouncements. The report for 23 January 2004 indicated that The 1Q04 pre-announcements are now up to meaningful levels with 198 total to date. The ratio of negative to positive pre-announcements is at a very low 1.4, well below the 1.9 at the same date for 1Q03, and even further below the 2.3 average over the last nine years. 1

3 relatively small number of firms if these are firms, such as GE and Intel, whose earnings are very representative of overall earnings trends. Evidence suggests that variation in firm-level stock returns is driven primarily by cash flow shocks, while variation in market-level returns is driven primarily by shocks to expected returns. The argument is that firms earnings news is largely idiosyncratic and diversifies away upon aggregation, while expected returns are driven by macro factors that are not diversifiable (Campbell, 1991; Vuolteenaho, 2002). We see two ways in which aggregate guidance might affect market-level expected returns. First, guidance can affect levels of market uncertainty. For example, if the arrival of earnings news increases market uncertainty, the associated increase in return volatility will increase expected returns, reducing stock prices. This tendency might be worse for bad news if earnings warnings have a contagion effect, so that warnings from a small number of firms have a disproportionately large effect on market uncertainty. 2 Note that this result may hold in spite of the fact that, at the firm level, one of the principal reasons for issuing earnings guidance is to avoid surprising investors, especially on the down side (e.g., Graham et al., 2004; Skinner, 1994). Second, the negative relation between economic conditions and expected returns (Fama and French, 1989) means that good news about aggregate earnings may be bad news for stocks because it lowers future expected returns, and vice versa. Alternatively, aggregate cash flow shocks may be linked directly to changes in expected returns, perhaps because they affect expectations about macroeconomic variables such as inflation or corporate investment, which in turn affect interest rates and expected returns (e.g., Boyd et al., 2005; Kothari et al., 2003). 2 Anecdotally, a spate of earnings warnings from technology companies apparently had this effect early in the summer of 2004; see Investor Jitters Greet Earnings Season, Financial Times, July 12,

4 There is little research on the relation between firm-level earnings news and market-wide returns. Penman (1987) finds systematic variation in aggregate earnings news during calendar years and links this to variation in aggregate returns. Kothari, Lewellen and Warner (2003) investigate a number of properties of aggregate earnings news and its relation to market returns and find, among other things, that market returns are negatively related to aggregate earnings growth. Our study is also similar in spirit to Seyhun (1988), who investigates the information content of aggregate insider trading. We use an extensive database of management-issued earnings guidance from Thomson First Call to investigate our predictions. First, we find that there has been a substantial increase in guidance over the 1994 to 2002 sample period. Consistent with previous research, we find that the majority of earnings guidance is quarterly rather than annual (a 60/40 ratio), and that over half (56%) of all quarterly guidance is downward guidance, with the remainder equally divided between upward and neutral guidance. We confirm previous evidence that quarterly guidance, especially downward quarterly guidance, is more informative than other types of guidance. All of this is consistent with the view that the relative extent of quarterly downward guidance (also known as earnings warnings or preannouncements) is potentially informative about market level returns. Our evidence indicates that guidance is increasingly pervasive and representative. In addition to a consistent increase in the proportion of firms issuing guidance, from around 10% in the mid 1990s to around 30% in 2001 and 2002, we find that firms issuing guidance now represent over one-half of the total market capitalization of Compustat and nearly one-half of 3

5 Compustat total assets, up from 10-20% in the mid-1990s. 3 We also find a shift over time in the variables that explain which firms issue guidance. In the 1990s, measures of firm performance such as ROA, losses, and earnings surprises were most important in explaining when managers issued guidance, consistent with previous research. In recent years, however, firm characteristics such as size, book-to-market, and growth are relatively more important. This suggests that firms guidance policies are now more consistent than in the past, and that guidance is now more representative, in the sense that it is now less likely to be driven by unusual/surprising earnings performance in a particular period. There are clear patterns in the timing of guidance. In relative terms, managers tend to issue neutral guidance early in the quarter, downward guidance toward the end of the quarter, and upward guidance after the end of the quarter. In addition, the information content of guidance increases during the quarter. This evidence supports conventional wisdom that, in each quarter, it is the relative extent of downward guidance available the end of the fiscal quarter that is informative about overall earnings trends. We also find that aggregate quarterly guidance measures, and especially the relative extent of downward guidance (as measured by the ratio of downward to upward guidance), vary a good deal from one quarter to the next, and vary more than corresponding aggregate annual guidance measures. This reinforces the idea that variation in aggregate downward guidance has the potential to capture variation in overall earnings news. Finally, we assess whether aggregate measures of quarterly earnings guidance capture overall earnings news and drive market returns. Using quarterly data, we find that the guidance measures, and especially the relative extent of downward guidance, are strongly associated with 3 These trends may be overstated if the completeness of First Call s coverage of earnings guidance has improved over time. Nevertheless, it is clear that the value-weighted proportion of firms issuing guidance has increased over time to levels that are now very substantial. 4

6 other measures of aggregate earnings news such as analyst forecast-based surprise measures and more conventional time series-based measures of earnings news. These results hold for both overall guidance and for S&P 500 guidance. The results are significant given that the guidance measures are available earlier in the quarter than the other measures, which require earnings realizations. There is no evidence of a relation between these guidance measures and quarterly stock market returns. Using monthly data, however, we find evidence that aggregate guidance is related to market returns. In particular, we find that the relative extent of downward guidance in the last month of each quarter is negatively related to contemporaneous market returns. There is a similar, but weaker, result for neutral guidance in this month, suggesting that neutral guidance sometimes conveys bad news. These results are stronger when we look at the most recent time period ( ), as expected given the increases in pervasiveness and representativeness that occur over our sample period. In addition, when we divide quarters into good news and bad news quarters, we find different patterns of intra-quarter market returns. While positive market returns in good news quarters tend to accrue smoothly and steadily over the entire period, negative market returns in bad news quarters are concentrated in the last 2-3 weeks of the fiscal period, so that returns are essentially flat in the announcement month. These patterns appear to reflect systematic differences in the way that guidance is released during the quarter: bad news quarters are characterized by sharp increases in the relative extent of downward guidance in the last 2-3 weeks of the fiscal period; no such patterns exist in good news quarters. We also find that bad news quarters are characterized by higher levels of market volatility, consistent with greater market uncertainty in these quarters. Finally, we find that earnings guidance issued by the largest firms ( bellwethers ) is associated with market returns in short windows around its 5

7 release. This is evidence that firm-level earnings guidance is informative at the market level, at least for large and influential firms. Section 2 discusses our predictions in more detail. Section 3 describes sample selection and provides descriptive information. Section 4 reports evidence on the informativeness of guidance at the firm level while Section 5 reports on the pervasiveness and representativeness of aggregate guidance. Section 6 presents evidence on the intra-quarter timing of aggregate guidance. Section 7 presents evidence on the relation between aggregate guidance, aggregate earnings news, and market returns. Section 8 concludes and discusses implications. 2. Development of Empirical Predictions Our principal research question is whether earnings guidance is informative for returns at the market level. We motivate this question with the observation that there are three sources of variation in stock returns (e.g., Fama, 1990): (1) shocks to expected cash flows, (2) predictable variation in expected returns, (3) shocks to expected returns. We discuss how earnings guidance is likely to affect market returns though its effects on cash flows in Section 2.1 and expected returns in Section Earnings guidance and expected future cash flows To the extent that earnings guidance provides information about firms expected future cash flows, it seems likely that aggregating guidance would provide information about the expected future cash flows of firms in general, and thus for the market as a whole. This is similar to the intuition behind papers that investigate intra-industry information transfer, the idea that earnings news for one firm in an industry will affect the returns of other firms in the same industry (e.g., see Clinch and Sinclair, 1987; Pownall and Waymire, 1989; Lang and 6

8 Lundholm, 1996). We extend this argument to the economy level to the extent that macroeconomic factors affect firm-level cash flows, we expect that the earnings news of firms that provide earnings guidance will have implications for the cash flows of firms in general. This argument implies that firm-level earnings news is informative about market level macro variables such as real GDP, industrial production, and investment, an argument often made in the financial press. The more firms that issue earnings guidance, the more likely it is to have implications for the market as a whole. If all firms in the economy issue guidance, aggregating that guidance should provide us with a complete picture of economy-wide earnings. Thus, we expect that the extent to which guidance is informative for the market as a whole will increase with the pervasiveness of the guidance. We measure the pervasiveness of guidance using both equal-weighted and valueweighted measures. Recent evidence shows that the cross-sectional distribution of earnings has become more and more concentrated an increasingly small number of firms is responsible for an increasing fraction of total earnings implying that value-weighting will be important. DeAngelo, DeAngelo and Skinner (2004) show that the 25 firms paying the largest dividends in 2000 account for over one half of aggregate Compustat earnings. This means that even if guidance is issued by a relatively small number of firms, their size may make the guidance informative for market-wide earnings. 4 4 General Electric s (GE) size and diversification across both industries and geographical regions makes its earnings news informative about overall economic trends. For example, see GE buoys market with growth optimism (Financial Times, July 9, 2004): General Electric a closely watched economic bellwether, on Friday forecast a strengthening global economy as it reported better-than-expected second-quarter earnings. Its optimism lifted financial markets spirits The US conglomerate sounded an upbeat tone for key economic and business indicators...(emph. added). 7

9 Whether earnings guidance is a useful indicator of market-wide earnings also depends on the representativeness of firms that provide guidance. If firms that provide guidance do so for particular economic reasons or otherwise have unusual characteristics, their results may not be representative, making it less obvious that we can draw implications from their results for the economy as a whole. Previous evidence shows that firms that issue annual earnings guidance tend to be larger and more profitable than other firms (e.g., Patell, 1976; Lev and Penman, 1990) while firms that provide guidance often do so when they have adverse earnings news (e.g., Skinner, 1994; Kasznik and Lev, 1995). Thus, we investigate the characteristics and circumstances of firms that provide earnings guidance. Earnings guidance need not be pervasive if it is representative. That is, it may be that there are a small number of firms whose results are very representative of those for the market as a whole and which therefore have a disproportionate effect on market-wide earnings expectations. 5 For example, the press sometimes features discussions of the earnings guidance of firms such as Intel or Dell, which are seen as bellwethers for the entire tech sector, if not for the market as a whole. Our empirical tests attempt to isolate some of these firms, to see whether their guidance is, in fact, informative at the market level. The informativeness of earnings guidance for market-wide returns also depends on the timeliness of the guidance the earlier in the quarter guidance becomes available, the more likely it is that it will drive market-wide earnings expectations. However, it may be that there is a relation between the timing of the news and its representativeness. For example, if adverse earnings news tends to be disclosed on a more timely basis than good news, 6 the market will 5 In the extreme, one could imagine a single firm whose earnings represented a sufficient statistic for the earnings of the economy, in the same way that the voting results of a single small town are sometimes seen as highly predictive of national election results. 6 See, for example, Skinner (1994, 1997), Soffer et al. (2000), Graham et al. (2004). 8

10 observe systematically more adverse earnings news early in the quarter and will condition its response accordingly, and only respond when guidance is more or less negative than expected. 7 Thus, we also provide evidence on the timeliness of earnings guidance and on whether and how the timing of disclosure varies according to the type of the guidance that firms provide. The timing of guidance may also affect market uncertainty, as discussed next. 2.2 Earnings guidance and expected returns Recent evidence indicates that news about expected cash flows is relatively more important in explaining firm-level returns than it is in explaining market-level returns (Vuolteenaho, 2002). In fact, the evidence suggests that information about expected returns explains most of the variation in market returns (e.g., Campbell, 1991) while cash flow news explains most of the variation in firm-level returns. These papers argue that because firm-level earnings news is largely idiosyncratic, its effects will be diversified away when aggregated. In contrast, variation in expected returns is predominantly driven by systematic macro factors, which are not diversifiable. The implication is that earnings guidance will not be all that informative about market-level expected future cash flows, but may be informative with respect to expected returns. There are at least two ways in which earnings guidance potentially affects expected market returns. First, we might expect a relation between the aggregate news conveyed by earnings guidance and expected returns at the market level. For example, it is sometimes argued that good cash flow news suggests strengthening aggregate demand which increases inflationary pressures in the macroeconomy, which in turn leads to an increase in discount rates and to lower 7 Consistent with this, market commentators sometimes compute the ratio of negative to positive guidance for a quarter and discuss the fact that this ratio typically exceeds one. See, e.g., Investor Jitters Greet Earnings Season, Financial Times, July 12, 2004: John Butter, research analyst at Thompson, said the ratio of 1.5 negative second quarter pre-announcements to every positive one was better than the typical ratio of

11 stock prices (the reverse would also follow). Fama and French (1989) provide evidence that expected returns are lower when economic conditions are strong and higher when economic conditions are weak. This argument implies a negative relation between economy-wide cash flow news and market-level stock returns. 8 Second, earnings guidance potentially affects the level of market uncertainty, and hence market volatility and expected returns. This argument is based on the idea that increased market uncertainty increases the volatility of market returns, that volatility shocks are persistent, and that this, in turn, increases expected returns and drives down stock returns (e.g., French, Schwert and Stambaugh, 1987; Schwert, 1989). This argument can help explain the asymmetric reaction to earnings news, as in Skinner and Sloan (2002) and Conrad, Cornell and Landsman (2002). Campbell and Hentschel (1992) discuss the volatility feedback effect, under which an information shock increases volatility by increasing uncertainty. If the news is good, it will naturally increase stock prices, but this effect will be offset by an accompanying increase in uncertainty, volatility and expected returns. If the news is bad, on the other hand, stock prices will naturally decline, and this effect will be reinforced by the accompanying increase in uncertainty, volatility and expected returns. Thus, volatility feedback will cause an asymmetric response to earnings news. 9 This effect will be amplified if there is something inherently different about the nature of good and bad news such that bad news naturally creates more uncertainty than good news. For example, there may be a contagion effect for earnings warnings: if investors see an unusually large number of earnings warnings in a quarter, it is likely to increase their uncertainty about earnings news overall, 8 Kothari et al. (2003) use this type of argument to explain their finding that aggregate quarterly earnings news is negatively related to market returns. 9 The argument is related to that of Veronesi (1999), who predicts that the market overreacts to bad news in good times (when the unexpected nature of the news increases uncertainty) and underreacts to good news in bad times (for the same reason). 10

12 driving up expected returns at the market level and compounding the negative effect of the news itself. An alternative argument is that guidance reduces levels of market uncertainty. By reducing the surprise element of earnings announcements and/or by helping to explain the news in advance, guidance potentially reduces the extent to which earnings news creates uncertainty, reducing volatility feedback. 10 Whether these second moment effects of guidance are relatively more important for market returns than for the returns of individual firms, where mean effects dominate, is an empirical question on which we hope to shed some light. 3. Sample Selection and Descriptive Statistics on Forecast Pervasiveness Table 1 summarizes sample construction. We begin with all company issued guidance available on First Call from , a total of 48,239 firm/period observations. We include forecasts of both annual and quarterly EPS and drop all forecasts from because coverage in those years is sporadic. We remove observations that fail to satisfy various selection criteria, most notably forecasts that do not provide information about EPS, which reduces the sample to 45,516 observations. We next remove observations that do not have accompanying analysts consensus earnings estimates from First Call because we need to measure the earnings surprise (loss of 9,890 observations), those for which there is no earnings announcement date available from Compustat (loss of 4,201 observations), and those for which the management forecast is made outside of a reasonable window relative to the fiscal period being forecast (loss of 253 observations). 11 The final sample contains 31,172 management earnings forecasts. 10 The argument is similar to that for why managers carefully manage their firms earnings expectations, as discussed in Skinner (1994), Soffer et al. (2000), and Matsumoto (2002). 11 On the last point, we require that a quarterly management forecast is made before the earnings announcement date of the fiscal quarter being forecast and no more than 90 days before the end of that fiscal quarter. For annual forecasts we require that the forecast is made before the earnings announcement date of the fiscal year being forecast 11

13 The advantage of First Call is that the sample is larger and covers many more firms than would be possible if the management forecasts were collected using databases that search newspapers and wire services, the approach typically used in the accounting literature. Having a comprehensive dataset is important for assessing the aggregate effects of disclosure. The disadvantage is that we have to rely on First Call, whose sampling criteria may be different from what we might want. For example, they may prefer to focus on larger firms of more interest to their clients. Table 2, Panel A reports the number of forecasts in each year, in total and categorized by reporting period (annual versus quarterly), along with the number of firms releasing forecasts in each year. As is well known (e.g., Pownall et al., 1993) a majority of forecasts are shorter horizon quarterly forecasts: 59% of forecasts are of quarterly earnings while 41% are of annual earnings. The number of firms with forecasts on First Call increases through time, implying increasing coverage. The number of firms increases fairly significantly from 1994 to 1998, is fairly stable between 1998 and 2000, increases in 2001, and declines in This implies that, at a minimum, coverage should be fairly complete from 1998 to 2002 and consequently we restrict some analyses to data drawn from this subperiod. There is a strong overall increase in the number of forecasts issued over time, which is unlikely to be fully explained by changes in coverage by First Call. This suggests that the propensity of firms to issue earnings guidance has increased over our sample period, as we might expect given other evidence that firms are and no more than 730 days (two years) before the end of that fiscal year. This means that annual forecasts can be made no earlier than the beginning of the fiscal year that precedes that being forecast. With regard to the corresponding analysts forecasts, we require that a consensus analysts earnings forecast is available at least 5 and no more than 90 days before the corresponding quarterly management forecast, and at most 120 days before the fiscal quarter end. For annual forecasts, we require that: (1) if the management forecast is released in the same fiscal year, the consensus analyst forecast be available at least 5 and no more than 120 days before the management earnings forecasts, or (2) if the management forecast is released in the fiscal year prior to that being forecast, there is no restriction. With regard to the Compustat earnings announcement date requirement, we also drop a small number of observations for which the earnings announcement date is clearly in error given the fiscal period for which the forecast is being made. 12

14 increasing their levels of disclosure (e.g., through conference calls; Frankel et al., 1999; Bushee et al., 2003). Table 2, Panel B reports on the number of forecasts issued by firms in each year. Consistent with an increasing propensity to forecast, the average number of forecasts per firm/year increases steadily through time, from 1.18 in 1994 to 3.95 in 2002, while the median increases from 1 in to 2 in and then to 3 in 2001 and This trend holds for both annual and quarterly forecasts (not reported in tables). Table 2, Panel C reports on the type of earnings forecasts made by managers, overall and by forecast period. As is conventional in the disclosure literature, we divide the forecasts into point forecasts, range forecasts, minimum forecasts, maximum forecasts and qualitative forecasts (e.g., Pownall et al., 1993). 12 Most of the sample forecasts are point (27%) or range forecasts (51%), with about 10% of the sample being minimum or maximum forecasts and 11% being other, qualitative forecasts. This suggests that First Call s collection efforts focus on more quantitative guidance and excludes some qualitative guidance. Managers sometimes release earnings forecasts at the same time as earnings announcements, consistent with previous research (Waymire, 1984; Hoskin, Hughes and Ricks, 1986; Hutton et al., 2003). Panel D of Table 2 reports that nearly one-quarter (23.8%) of quarterly forecasts are made at earnings announcement dates and that about 11% of annual forecasts are made at earnings announcement dates. There is a strong increase in this tendency during the sample period. The fraction of quarterly forecasts released at the previous quarter s earnings announcement date increases from around 5% in 1996 and 1997 to 16% in 2000, 34% in 2001, and 42% in Because forecasts released at earnings announcement dates may be 12 The appendix describes how this is done for the First Call data in our sample. 13

15 systematically different from other forecasts, we exclude them from many of our tests, although our results are generally robust to whether or not we retain these observations. 4. The Informativeness of Firm-Level Earnings Guidance Table 3 reports on the nature of the earnings news conveyed by earnings guidance in each year of the sample period. We divide forecasts into those that convey downward, upward and neutral guidance by either (in the case of quantitative forecasts) comparing the management forecast to the most recent median analysts forecast available from First Call or (in the case of qualitative forecasts) using information provided about the forecast by First Call. 13 Panel A of Table 3 reports on forecasts of annual EPS. Overall, forecasts of annual earnings are slightly more likely to convey downward guidance than upward guidance: 40% of annual forecasts convey downward guidance, 36% convey upward guidance, and 24% convey neutral guidance. There is some variation across sample years in the relative frequency of the three types of guidance, although the only clearly discernible trend is that neutral guidance is less frequent since 2000 while upward guidance increases. Panel B reports on forecasts of quarterly EPS. There is a strong tendency for quarterly forecasts to convey downward guidance: 56% of quarterly forecasts convey downward guidance while 22% convey neutral guidance and another 22% convey upward guidance. This confirms the idea that an important motivation for quarterly earnings guidance is to manage expectations downwards in the presence of adverse earnings news (e.g., Skinner, 1994). To get a clearer picture of variation in the relative extent of upward and downward guidance, Figure 1 plots the ratio of downward to upward earnings guidance for each quarter of 13 First Call provides a very detailed coding scheme (their CIG code variable) which facilitates this process. See the appendix for details of how we classify the forecasts into upward, downward and neutral guidance. 14

16 the sample period. The ratio is computed separately for annual and quarterly guidance. To compute this ratio, we divide the number of forecasts that convey downward guidance by the number that convey upward guidance in each calendar quarter of the sample period (Q to Q4 2002). 14 Figure 1 shows that there is a good deal of quarter-to-quarter variation in the relative extent of downward and upward guidance, especially for quarterly forecasts, making it more likely that variation in this measure captures aggregate earnings news. The ratio of downward to upward guidance formed using annual forecasts (the annual guidance ratio ) typically varies between 1 and 2, with a mean of 1.4 and a standard deviation of.6. In contrast, the quarterly guidance ratio has a mean of 3.2, a standard deviation of 1.3 and never falls below 1.2, and so is both more variable and substantially larger than the annual guidance ratio, consistent with previous evidence that quarterly forecasts are more likely to convey bad news and conventional wisdom that the average ratio exceeds one. 15 Table 4 reports the stock price reaction to management earnings forecasts. To conduct these tests we compute market-adjusted returns for the three trading days centered on the date of the management earnings forecast (Brown and Warner, 1985). The sample size for these tests declines to 29,611 observations because we lose observations with no ready CRSP match or with missing returns One complication with this analysis arises in matching forecasts to calendar quarters to do this we include all forecasts that pertain to the calendar quarter; that is, that are made in the four subperiods shown in Figure 3. Thus, we remove forecasts for firms that do not have (end of) March, June, September or December fiscal year ends. We also exclude forecasts made in subperiod 1, which mainly contains forecasts issued in conjunction with the previous quarter s earnings announcement. This reduces the sample size to 13,843 quarterly forecasts and 10,410 annual forecasts. 15 These ratios also have different time series properties. While the quarterly ratio appears to be non-stationary the annual ratio displays little autocorrelation in the levels but is negatively autocorrelated in the changes. 16 The table reports results that use CRSP equal-weighted index returns as the market return. Results are similar if we use the CRSP value-weighted index instead. We do not report statistical significance tests. Given the sample 15

17 The results in Table 4 largely accord with previous research (e.g., Baginski et al., 1993; Skinner, 1994; Hutton et al., 2003). For the overall sample, the mean (median) return is -3.5% (- 1.5%) and 58% of the returns are negative. The central tendency is negative because most forecasts (58%) are of quarterly earnings, which are more likely to convey bad news, and because bad news forecasts are more informative. 63% of the returns associated with quarterly forecasts are negative, and the mean (median) overall return for quarterly earnings guidance is - 4.9% (-2.6%). Forecasts of annual earnings are less informative, with a mean (median) return of -1.60% (-0.42%). The other results in Table 4 confirm that returns associated with downward and upward guidance are reliably negative and positive, respectively, a tendency that is again stronger for quarterly guidance than for annual guidance. The information content of both good and bad news forecasts is larger for quarterly forecasts than it is for the corresponding annual forecasts, with means (medians) of 4.2% (2.8%) and -10.1% (-7.1%) for quarterly forecasts of good and bad news, respectively, compared to corresponding means (medians) of 2.5% (1.7%) and -6.2% (-3.3%) for annual forecasts. Returns associated with the neutral forecasts are close to zero. 5. The Pervasiveness and Representativeness of Earnings Guidance The evidence in Table 2 shows that the frequency of earnings guidance has increased over time. In this section, we investigate: (1) whether this translates into an increase in the pervasiveness of guidance, and (2) the representativeness of earnings guidance. Table 5 and Figure 2a provide evidence on the number and proportion of firms in each year that provide either annual guidance or quarterly guidance. In each year, we define size most of the means and medians we report are likely to be reliably different from zero, in spite of some likely cross-sectional dependence. 16

18 forecasting firms as those that issue some type of guidance and that have certain basic data on annual Compustat, and the total number of firms as all firms on Compustat with the same available data. 17 The proportion of firms issuing guidance increases noticeably during the sample period, although the numbers in the earlier years may be understated if First Call s data collection is less complete in these years. The table shows that the proportion of firms issuing earnings guidance increases from around 5-10% in the mid 1990s to around 30% in 2001 and 2002, the last year with available data. Earnings guidance is more pervasive when we weight the observations by firm size. We use both total assets and market value to weight the observations because weighting by market value alone may distort the trends given the large increase in stock prices that occurred during the 1990s and the large subsequent reversal. 18 When we weight by total assets the proportion of forecasters is larger than the simple proportions, as expected given the well-known tendency that larger firms are more likely to provide guidance. The total asset-weighted proportion of forecasters increases from 13% in 1995 to 17% in 1996, 20% in 1997, and then to 43% in 1998, before falling to 33% in 1999 and 31% in 2000, and increasing again to 46% in 2001 and 47% in Thus, in recent years firms that represent nearly one-half of total assets on Compustat provide earnings guidance. The numbers are even more impressive when we value-weight using market capitalization, suggesting that forecasting is associated with relatively larger valuations. The value-weighted fraction increases from 17% in 1995 to 22% in 1996, 26% in 1997 and then to 44% in 1999, 42% in 2000 and 47% in 2001, before increasing sharply to 57% in 2001 and 56% in Thus, in 2001 and 2002 firms representing over one-half of the market 17 We require book value and prior year sales to be greater than or equal to zero, and stock price, shares outstanding, total assets and prior year total assets to be positive. For the logit regressions we also remove those observations with book-to-market, ROA or sales growth in the top and bottom 1% of the distribution. 18 We have also used earnings and positive earnings to weight the observations, with very similar results to those shown for total assets. 17

19 capitalization of Compustat provide earnings guidance. This is evidence that earnings guidance has become more pervasive, especially in value-weighted terms. To assess the representativeness of earnings guidance, Table 5 also reports the results of annual logit regressions of whether a firm provides guidance during the year on firm characteristics often associated with forecasting: firm size (log of total assets), book-to-market, ROA, sales growth, and a loss dummy. Consistent with previous research, we find that size is important the coefficient on size is strongly positive in all years. In addition, there is evidence of an increasing trend in the magnitude of the coefficient on size, from.22 in 1996 to.36 in We also find consistent evidence that book-to-market is significantly negatively associated with forecasting, although there is less of a trend in its coefficient. The ROA variable is reliably and consistently positive in the period, consistent with the idea that firms with good earnings performance are more likely to forecast earnings (Patell, 1976; Penman, 1980). However, the ROA variable is not significant in 2000 or 2001 and has a relatively small coefficient in The loss dummy is positive and significant from but not in later years. Thus, there is some evidence that accounting performance, as measured by ROA and reported losses, was important in explaining forecasting during the 1990s but that this is less true over when size becomes relatively more important. Sales growth is negatively associated with forecasting, but mainly in the latter part of the sample period. Overall, it seems that firm characteristics such as size, book-to-market, and growth have become more important than period-specific earnings performance in explaining forecasting. This implies that forecasts have become more representative over the sample period, in the sense that they are less likely to be made by firms when their earnings performance is unusually good or bad. These trends are also reflected in a fairly substantial increase in the explanatory power of the regression over the 18

20 sample period. The pseudo R-squared increases steadily from around 5% in the mid 1990s to 17% in Table 6 and Figure 2b provide further evidence on representativeness, this time on a quarterly basis for quarterly guidance. Because previous research suggests that quarterly forecasting is likely to reflect earnings surprises, we also include the Asurp variable in these regressions, which is the difference between realized quarterly earnings and the analyst consensus at the beginning of the quarter. There is a strong tendency for firms with adverse earnings news to provide quarterly guidance (e.g., Skinner, 1994). In fact, Asurp is the only consistently significant variable over the quarters from 1995 until early 2001, after which it largely fails to achieve significance. Firm size and book-to-market are generally not significant during the 1990s, but are significant beginning in early Consistent with the evidence in Table 5, this evidence suggests that firm characteristics, rather than period specific earnings news, are now relatively more important in explaining guidance. The increasing importance of size and book-to-market also helps explain the associated increase in the pervasiveness of guidance. 6. The Timeliness of Earnings Guidance The results to this point suggest that guidance has become increasingly pervasive and representative over the sample period. We next investigate the timeliness of quarterly earnings guidance. Table 7 reports the within-quarter timing of quarterly earnings forecasts. We divide forecasts into those that provide upward, downward, and neutral guidance to see whether managers forecast strategies vary as a function of the nature of the earnings news, as previously 19

21 suggested in the accounting literature (e.g., Francis et al., 1994; Skinner, 1994, 1997; Soffer et al. 2000). To assess timing, we classify forecasts of quarterly earnings into four subperiods, as shown in Figure 3: (1) the period from the end of the previous fiscal quarter to 50 calendar days before the end of the fiscal quarter; (2) the period from 50 days before the end of the fiscal quarter to 25 days before the end of that quarter; (3) the period from 25 days before the end of the fiscal quarter to the end of that quarter; and (4) the period from the end of the fiscal quarter to the earnings announcement date. We plot the results in Figure 4. Table 7, Panel A shows that well over half (65%) of the forecasts occur either late in the quarter (subperiod 3) or after the end of the quarter (subperiod 4), thus qualifying as earnings pre-announcements or warnings. We also find that many forecasts (25%) fall in the first subperiod, but that most of these are made in conjunction with the previous quarter s earnings release. After we remove these observations (Panel B), only about 5% of forecasts are made in the first subperiod, while 11% are made in the middle of the quarter (subperiod 2), and 38% are made in the last 25 days of the quarter (subperiod 3). Most forecasts (46%) are made after the end of the quarter, as we might expect if the rate of disclosure increases with the precision of managers earnings information. The intra-quarter timing of forecasts varies with the sign of the earnings news. Relatively more neutral news tends to be released at the beginning of the quarter, perhaps because managers have less information about earnings at that time: about one third of all guidance is neutral in this period. In the last part of the quarter (subperiod 3), however, more of the news is negative (63%) and relatively less is positive (16%) or neutral (20%). This tendency reverses after the end of the quarter when the amount of good news increases to about one-quarter (26%) while the amount of bad news decreases to a little over one-half (54%) and confirming news stays at 21%. Over half 20

22 (56%) of the good news that is released is released after the end of the quarter, perhaps because managers want to be very sure of themselves before releasing good news. The tendency of downward guidance to cluster in the weeks before the end of the quarter suggests that investors should know by the end of the quarter whether the quarter generally is going to be characterized by bad quarterly earnings news. Panel C of Table 7 reports that the information content of quarterly forecasts tends to increase as the quarter progresses, although bad news forecasts are again more informative in general. For forecasts made early in the quarter (first subperiod), the mean (median) return is - 9.4% (-5.1%) for bad news forecasts and 2.3% (.9%) for good news forecasts. 19 Returns associated with bad news forecasts then increase in magnitude to -10.3% (-7.6%) in the second subperiod, -11.9% (-9.2%) in the third subperiod, and to -12.7% (-9.8%) in the last subperiod. The magnitude of returns associated with good news forecasts also increases across the subperiods, with means (medians) of 3.1% (1.9%) in the second subperiod, 4.3% (3.0%) in the third subperiod, and 5.0% (3.4%) in the last subperiod. The results in Table 7 indicate that most of the information conveyed by bad news forecasts, in aggregate, has been conveyed by the end of the fiscal quarter, and that the information content of these forecasts is weighted more heavily toward the end of the quarter. This supports conventional wisdom that the relative number of earnings warnings observed by the end of each quarter is informative with respect to the earnings news that will be announced in the weeks to follow. 19 As in Panel B, these results exclude forecasts made in conjunction with earnings announcements. When these observations are included, the mean and median returns become more positive in all categories, indicating that returns at earnings announcement dates tend to be positive. 21

23 7. The Relation Between Aggregate Earnings Guidance, Aggregate Earnings News, and Market Returns. In the foregoing sections we report a number of regularities related to the pervasiveness, representativeness, and timeliness of earnings guidance. This section reports evidence on whether aggregate measures of guidance are related to measures of aggregate earnings news and to market returns. We conduct these analyses using both quarterly data (section 7.1) and monthly data (section 7.2). We also provide a comparison of intra-quarter return and guidance patterns for good and bad news quarters (Section 7.3) and an analysis of short-window market returns around the time that large firms (market bellwethers ) release earnings guidance (Section 7.4). 7.1 Quarterly data Table 8, panel A provides descriptive statistics for several measures of aggregate quarterly earnings news. First, we aggregate analysts earnings surprises for the quarter. To do this, we take the difference between realized EPS and the median analyst forecast from First Call at the beginning of the quarter for each firm, and then form an equally-weighted average across firms. We do this for all firms with available analyst forecasts data on First Call (ASurpA) and for those firms that issue earnings guidance in the quarter (ASurpF). Second, following Kothari et al. (2003), we construct earnings-based measures by taking realized earnings for the quarter, deflating by price, and either equal or value-weighting the results. 20 This yields the variables EP_agg (value-weighted by total assets) and EP_ew (equal-weighted). We then get measures of earnings changes by taking seasonal differences in these variables, yielding EP_agg and EP_ew, respectively. Because the changes are highly autocorrelated, we also calculate time series innovations in these variables (Innov_ EP_agg and Innov_ EP_ew, respectively) which 20 Following Kothari et al. we have also deflated by book value, with similar results. 22

24 are the residuals from time-series regressions that assume the changes follow an AR1 process. For all measures, we restrict the set of firms to those with fiscal quarters that coincide with calendar quarters. Finally, to gauge the extent to which the guidance and earnings news variables are related to GDP growth, we also include the logarithmic change in GDP for the quarter. 21 Panel B of Table 8 reports descriptive statistics for two sets of aggregate guidance variables, one for all firms and the other for S&P 500 firms, as well as for three market return measures, the CRSP equal-weighted market index, the S&P 500 index, and the NASDAQ Composite Index. 22 To be consistent with our timing convention (see Figure 3) we cumulate these returns over the three months beginning with the second month of each calendar quarter and ending with the month after the end of the quarter, which usually contains the earnings announcement (i.e., first quarter returns are cumulated over February, March, and April, etc.). This aligns our quarterly return window with the way that earnings news for a given quarter is typically disclosed and announced. The guidance variables, D, N, and U, are seasonal logarithmic changes in the number of forecasts of quarterly earnings that represent downward, neutral and upward guidance, respectively. Ratio is the ratio of downward to upward guidance for the quarter (as in Figure 1) and Ratio is the logarithmic seasonal change in this variable. As we might expect given the overall increase in guidance over the sample period, the seasonal changes in guidance are all 21 This is the real, seasonally-adjusted change in gross domestic product, obtained from the St. Louis Fed, 22 We have also restricted attention to guidance for tech firms, defined in two ways based on SIC codes. First, following Fama and French (1997), we define tech firms as those in the chip and computer industries (for computers, SIC codes , , 3695, 7373 and for chips, SIC codes 3622, , 3810, and 3812). Second, following Francis and Schipper (1999) we also include pharmaceutical and telecom firms (SIC codes 283, 357, , 481, 737, 873). The results are generally similar to those reported, although not surprisingly the tech guidance aggregates tend to be related to the NASDAQ Composite returns rather than to returns on the other two indices. Returns on the CRSP-value weighted index were also used, but the results are not reported because they are so similar to those obtained using the S&P 500 index return. 23