Are Analysts Cash Flow Forecasts Important? Another Examination

Transcription

1 Are Analysts Cash Flow Forecasts Important? Another Examination Sandip Dhole Faculty of Business and Economics University of Melbourne Ananda Mohan Pal Department of Business Management University of Calcutta September 3, 2014 Sagarika Mishra Centre for Financial Econometrics Deakin University Abstract Analysts cash flow (CPS) forecasts have been the topic of much recent research. While some prior studies suggest that these forecasts have limited usefulness, others find evidence to the contrary. We find that the presence of CPS forecasts significantly reduces (increases) the bid-ask spread (abnormal trading volume) around the earnings announcement date, suggesting that these forecasts convey additional information to the market over and above analysts earnings forecasts. We also examine how investors use the information in CPS forecasts and find that the abnormal market reaction to an earnings surprise is much stronger when the associated cash flow surprise is high. Further, the incremental abnormal market reaction is only driven by those instances where the earnings forecast error is positive. Finally, we show that a higher level of cash flow in the current earnings surprise improves future stock liquidity. Our results thus show that analysts CPS forecasts are incrementally important in shaping market participants assessment of firms financial performance. Keywords: Analysts Cash Flow Forecasts, Analysts Earnings Forecasts, Information Asymmetry, Bid- Ask Spread, Abnormal Trading Volume, Liquidity JEL Classification: G12, G14, M41 We thank Belén Blanco, Gerry Lobo, Santosh Mishra, Matt Pinnuck, Bhimasankaram Pochiraju, Brad Potter, Amedeo Pugliese, Tharindra Ranasinghe, Arpita Shroff, Shiva Sivaramakrishnan, Doug Skinner, Naomi Soderstrom, Nasser Spear, Ram Thirumalai and workshop participants at the University of Melbourne for their helpful comments and suggestions. Please address correspondence to sandip.dhole@unimelb.edu.au.

2 Are Analysts Cash Flow Forecasts Important? Another Examination 1 Introduction Analysts earnings forecasts (EPS forecasts, henceforth) have long been regarded as reliable benchmarks for corporate performance and have been the topic of a vast volume of accounting and finance literature. Although most of the financial analyst literature focuses on earnings forecasts, it is to be noted that analysts also provide other information (such as revenue forecasts, forecasts of long-term growth, etc.) that can be useful in evaluating a firm s performance. Recently, analysts have been increasingly providing cash flow forecasts (CPS forecasts, henceforth). 1 In this study, we focus on CPS forecasts because like earnings, cash flows are important indicators of firm value. CPS forecasts presumably speak to the sustainability of earnings because cash flows are more persistent than earnings. Therefore, if analysts present a good (poor) outlook for operating or free cash flows, it has potentially positive (negative) implications for firm value. The importance of CPS forecasts can be seen from articles in the business press. Consider, for example, the following excerpt from an article in The New York Times: To individuals and businesses, reaching out to those ever-expanding networks is immensely appealing, and that helped drive LinkedIn s strong second-quarter financial performance, which blew past Wall Street s expectations on revenue, profit and cash flow...investors were certainly eager to connect with that kind of growth, sending shares of the company up more than 7 percent in after-hours trading, to around $228 a share. (And that was after a 4.5 percent gain during regular trading.) (The New York Times, August 1, 2013) 2 As this excerpt suggests, the market regarded LinkedIn s performance to be good on account of the company beating analysts expectations of earnings, revenues and cash flows. Consider another example: In New York, Vivendi s American depository receipts were down $3.67, closing at 1 See DeFond and Hung (2003) and Mohanram (2013)

3 $ The credit rating agency Standard & Poor s cut Vivendi s long-term corporate rating to junk status, citing the company s cash-flow forecasts that were lower than most analysts had expected. The agency said further downgrades were possible. Vivendi is laboring under $18.6 billion of debt, much of it amassed by Mr. Messier s acquisition spree. (The New York Times, August 15, 2002) 3 As the above example suggests, Vivendi s failure to match up to analysts expectations of cash flows caused the price of its ADRs to fall and led to Standard & Poor s downgrading its bonds to junk status. The examples above thus clearly show that the market attaches importance to analysts CPS forecasts. Consistent with the idea that CPS forecasts convey useful information to market participants, Call et al. (2013) show that there is a significant market reaction to analysts CPS forecast revisions that is incremental to EPS forecast revisions, suggesting that CPS forecasts convey additional information to the market place over and above that conveyed by EPS forecasts. Recent studies by Mohanram (2013) and Radhakrishnan and Wu (2013) further argue that CPS forecasts provide investors with estimates of accruals and thus help correct the market s mispricing of accruals. Yet, Givoly et al. (2009, 2013) and Bilinski (2014) argue that CPS forecasts are unsophisticated extensions of earnings forecasts and that when CPS forecasts are announced with other announcements, there is no incremental market reaction to such announcements. Thus, clearly there is disagreement in the literature on the usefulness of analysts CPS forecasts. One probable reason for the disagreement seems to be the assumption that the market views cash flow forecasts as independent performance benchmarks. We believe that this is unlikely to be true, since cash flow forecasts are generally less accurate than earnings forecasts (Givoly et al., 2009) and it is well known that the market fixates on earnings (see Darrough and Rangan, 2005; Shi and Zhang, 2012; etc.). Therefore, we move away from this contentious issue and view cash flow forecasts as providing information that supplements the information in earnings forecasts. Our approach is based on the recent empirical work of Mohanram (2013) and McInnis and Collins (2011), described above billion.html 2

4 We start by examining whether the presence of CPS forecasts affect the information environment around the earnings announcement, by examining changes in the bid-ask spread surrounding the earnings announcement, based on prior research. 4 If CPS forecasts improve the firm s information environment, one should observe lower information asymmetry and hence, a lower bid-ask spread around the earnings announcement. We find evidence consistent with our expectation. We next compare the abnormal trading volumes around the earnings announcement date for firms with and without CPS forecasts in order to test whether CPS forecasts convey additional information to market participants. As Bamber et al. (2011) note, trading in response to a disclosure arguably provides the most direct evidence that the disclosure has affected individual investors expectations and investment decisions. Trading volume reflects differences in investors interpretation of the information content of a signal. High trading volume thus indicates that a signal has significant information content. Consistent with the notion that CPS forecasts convey additional information, we find that the abnormal trading volume is significantly greater for firms with CPS forecasts, after controlling for the potential endogeneity involving analysts decision to provide cash flow forecasts. Having established that CPS forecasts improve the firm s information environment, we next focus on understanding how these forecasts improve the information environment. Specifically, we study whether the magnitude of the CPS forecast error affects the market s perception of the associated earnings surprise. 5 As McInnis and Collins (2011), Mohanram (2013) and Call et al. (2013) argue, CPS forecasts serve an important purpose by providing information about operating accruals. Given the lower persistence of accruals (Sloan, 1996), it is likely that the market would price earnings surprises with greater cash flow content higher than earnings surprises with greater accrual content. Accordingly, we test whether the level of the cash flow surprise affects the mar- 4 We examine the bid-ask spread both before and after the earnings announcement. 5 Brown et al. (2013) show that the direction of the cash flow surprise affects the market s perception of the earnings surprise. However, they do not examine whether the magnitude of the cash flow surprise has any incremental effect on the market s pricing of the earnings surprise, which is an important part of our study. Therefore, our results are more generalizable in that they can address questions relating to whether the magnitude of the cash flow surprise affects market prices and trading behavior in response to an earnings surprise. For example, the market s reaction to the same earnings surprise could be very different for low and high levels of the associated cash flow surprise. We elaborate on this point in Section 2. Brown et al. s research design would not allow the researcher to study this. 3

5 ket s reaction to an earnings surprise. Our results show that the abnormal return surrounding the earnings announcement is significantly greater for higher levels of cash flow surprise, controlling for the level of earnings surprise. We next extend this analysis by conditioning the ERC on the sign of the earnings surprise. This is based on the idea that a positive earnings surprise is less credible than a negative earnings surprise, since a manager have incentives to report a positive earnings surprise. 6 Therefore, we argue that when a firm reports a positive earnings surprise, the market is likely to benchmark this surprise against the information conveyed by the cash flow surprise. In contrast, a negative earnings surprise would likely convey a more credible signal to the market, making the cash flow forecast error less relevant. We find support for our arguments. Specifically, we provide evidence that the incremental ERC associated with the firm reporting a higher cash flow surprise is significant only for the sample of positive earnings surprises. Our results thus show that the market uses the cash flow forecast error to assess earnings quality when the earnings surprise is positive. Finally, we examine whether the level of the current cash flow surprise impacts future changes in stock liquidity. Our hypothesis is motivated by the argument that the level of the cash flow surprise conveys important information about the sustainability of earnings. Thus, as the level of the cash flow surprise rises, it is more likely to indicate sound firm fundamentals to the market. This would likely improve stock liquidity. Consistent with our expectation, we find that higher levels of cash flow surprise improve liquidity upto four quarters ahead, after controlling for the current level of the earnings surprise. We make several important contributions to the literature. First, to the best of our knowledge, our study is the first to examine the impact of the presence of CPS forecasts and the level of cash flow surprise on abnormal trading volume and stock liquidity. As mentioned above, trading volume is a reflection of market participants interpretations of a signal and is thus an important measure of the information content of the signal. In fact, many scholars analyze both price and 6 Much prior research has established that there is a significant equity market premium (penalty) for meeting and exceeding (missing) an earnings expectation. 4

6 volume reactions to an event in order to study the information content of an event. 7 By showing that the presence of cash flow forecasts increases the abnormal trading volume around the earnings announcement, we provide powerful evidence that these forecasts convey useful information that presumably leads to investors updating their beliefs. Second, we contribute to the debate on the usefulness of CPS forecasts by showing that the existence of CPS forecasts significantly reduces the bid-ask spread around the earnings announcement. Our evidence is thus consistent with the notion that CPS forecasts convey valuable information to the market. In this regard, our study joins Call et al. (2013), McInnis and Collins (2011), Mohanram (2013) and Radhakrishnan and Wu (2013), who show that CPS forecasts convey useful information. Finally, we also provide strong evidence that investors use CPS forecasts in their assessment of the earnings surprise by documenting that the market reaction to an earnings surprise is conditional on the level of the cash flow surprise when analysts provide estimates of cash flows, thereby addressing the concern raised by Ecker and Schipper (2014) that there is a paucity of research examining how investors use the information in cash flow forecasts. Our results provide evidence that investors use the cash flow forecast error to judge the quality of a positive earnings surprise, thereby indicating the important role of cash flow forecast errors in helping investors assess earnings quality. Our results thus provide a more comprehensive analysis of the information role of CPS forecasts. The paper is organized as follows. We build our hypotheses in Section 2; Section 3 describes the data and the results of the empirical estimation. Section 4 concludes. 7 Beaver (1968) points out that while price changes reflect investors average response to an information event, trading volume reflects idiosyncratic beliefs. Karpoff (1987) and Kothari (2001) present excellent discussions of studies using price and trading volume to illustrate the importance of both approaches to test information content. 5

7 2 Hypotheses and Research Methodology 2.1 Hypotheses It is now well established that analysts EPS forecasts shape market expectations. However, while EPS forecasts are important, these are not the only indicators of firm performance that analysts forecast. Recently, an increasing number of analysts has been providing cash flow forecasts and a recent body of research examines a range of questions related to these forecasts, ranging from issues related to their accuracy and sophistication (Givoly et al., 2009, 2013) to why analysts issue cash flow forecasts (DeFond and Hung, 2003). In an early study, Givoly et al. (2009) argue that analysts CPS forecasts are naïve extensions of EPS forecasts and are limited in their usefulness, as they tend to be less accurate than EPS forecasts and provide no incremental information about changes in net working capital. Givoly et al. (2009) further demonstrate that CPS forecast errors are weakly associated with stock price movements. However, DeFond and Hung (2003) and Call et al. (2013) suggest that analysts make sophisticated adjustments for changes in working capital and that their CPS forecasts are not simple extensions of their EPS forecasts. Call et al. (2013) also show that there is a significant stock market reaction to analysts CPS forecast revisions, even after controlling for analysts EPS forecast revisions, suggesting that CPS forecasts convey incremental meaningful information to the market. Further, two recent studies by Mohanram (2013) and Radhakrishnan and Wu (2013) show that CPS forecasts help reduce the mispricing of accruals. Subsequently, however, the evidence in Call et al. (2013) and DeFond and Hung (2003) has been questioned by Givoly et al. (2013) and Bilinski (2014), who argue that CPS forecasts are not sophisticated and do not convey additional information to the market. Based on the above discussion, it is clear that there is considerable disagreement among scholars on the usefulness of CPS forecasts as a source of information, suggesting that there is a necessity to further examine this issue. As mentioned above, one reason for the disagreement could be that many extant studies view cash flow forecasts as independent performance benchmarks. We 6

8 disagree with this view for the following reasons. First, cash flow forecasts are strictly less accurate than earnings forecasts (Givoly et al., 2009). This would make it unlikely that investors would base their expectations solely on these forecasts. Second, cash flow forecast errors are significantly correlated with earnings forecast errors, thus introducing potential research design issues. 8 We thus follow a different approach in this study and view cash flow forecasts as providing information that supplements, rather than substitutes the information in earnings forecasts. Our approach is motivated by the evidence in Mohanram (2013), Radhakrishnan and Wu (2013) and McInnis and Collins (2011), who show that cash flow forecasts provide investors with an estimate of accruals. We address two main issues in this study. First, we examine whether the presence of CPS forecasts affects the nature of the information environment, by studying the impact of the presence of CPS forecasts on the bid-ask spread and abnormal trading volume surrounding the earnings announcement. Second, we examine how investors use the information contained in the CPS forecasts to evaluate firm performance. We elaborate on these points in the next few paragraphs. A large volume of prior research (for example, Chiang and Venkatesh, 1988; Chung et al., 1995; Kalimipalli and Warga, 2002; Attig et al., 2006; etc.) has established a link between information asymmetry and the bid-ask spread. The general notion is that higher bid-ask spreads denote higher information asymmetry and thus poorer information environments. Empirical evidence (for example, Roulstone, 2003; Kanagaretnam et al., 2005; etc.) indicates that analysts can help reduce the bid-ask spread by providing useful information about firm performance. Providing CPS forecasts, in addition to EPS forecasts, could be one way to further reduce the information asymmetry if these forecasts convey additional value relevant information. As mentioned above, while Givoly et al. (2009, 2013) suggest that analysts CPS forecasts have limited usefulness, there is also emerging evidence to the contrary. Specifically the evidence in McInnis and Collins (2011), Mohanram (2013) and Radhakrishnan and Wu (2013) suggests that CPS forecasts indirectly provide information about the accrual content of earnings and help reduce the mispricing of accruals. 8 Untabulated results suggest that the correlation between earnings forecast errors and cash flow forecast errors is significant at 1 percent. 7

9 Thus if CPS forecasts convey additional value relevant information, they will further reduce the information asymmetry (and hence, the bid-ask spread) surrounding the earnings announcement date. Accordingly, we hypothesize that: 9 Hypothesis 1A (H1A): The bid-ask spread is lower when analysts provide cash flow forecasts, ceteris paribus. While a price change (and hence, change in the bid-ask spread) can help identify the information content of a signal, the absence of a price change does not necessarily mean that the signal has no information content. Indeed, it is possible that investors interpret a signal differently and form offsetting trading positions that cancel any price effect of the signal. 10 Therefore, it is important to examine both price and volume reactions to understand how investors react to information (Morse, 1981). Accordingly, we next focus on the impact of the presence of CPS forecasts on the abnormal trading volume surrounding the earnings announcement date. Trading volume is an important test of the information content of a signal because it can provide insights into how investors use the information in the signal to form different beliefs (Morse, 1980; Jain, 1988; etc.) based on which, they trade. Recent empirical evidence (Cready and Hurtt, 2002) suggests that volume reactions are more powerful indicators of market response than price changes and that when supplementing return-based analyses with volume-based tests allows one to identify market response more accurately. A large volume of prior research has established that market participants trade on earnings announcements (Beaver, 1968; Landsman and Maydew, 2002; etc.), analyst forecast revisions and forecast errors (Collins et al.,2009, Ayers et al., 2011; etc.), public firm-specific news announcements (Nofsinger, 2001), etc., suggesting a positive association between the availability of information and trading volume. As mentioned above, CPS forecast errors provide information about the cash and accrual components of the earnings surprise. We argue that if this information 9 All hypotheses have been stated in the alternate form. 10 Bamber and Cheon (1995) present empirical evidence of this. 8

10 is useful to market participants, they would likely trade on it, leading to greater abnormal trading volume around the earnings announcement date. We state the hypothesis below: Hypothesis 1B (H1B): The abnormal trading volume around the earnings announcement is higher when analysts provide cash flow forecasts, ceteris paribus. While H1 above adds to our understanding of whether CPS forecasts reduce information asymmetry, it does not address the important question of how CPS forecasts affect the information environment. In other words, what is the role played by CPS forecasts? This is an important issue to investigate as it has implications for investors who use this information to invest in the stock market. Some research (for example, Barth et al. 2001) argues that cash flows convey incremental information over earnings, that aids in valuation. This is because cash flows are more persistent than accruals (Sloan, 1996). Therefore, if high earnings are supported by high cash flows, it indicates that the earnings are likely to be more persistent and, therefore, be of higher quality. Consistent with this reasoning, Brown et al. (2013) find that firms that meet or beat both analysts CPS forecasts and EPS forecasts are likely to have superior future performance than firms that just beat the EPS forecasts. We argue that if investors recognize the greater persistence of cash flows, they will react more strongly when the earnings surprise is driven more by cash flows, than when it is driven more by accruals. In order to test this conjecture, we examine whether the market s (price) reaction to an earnings surprise is influenced by the level of cash flow surprise. More specifically, we examine whether or not an earnings surprise achieved through a greater amount of cash flow elicits a stronger market reaction than the same level of surprise achieved through a smaller amount of cash flows. Evidence supporting with this would be consistent with the notion than investors use cash flow forecasts to judge the quality or credibility of the earnings surprise. We state the hypothesis below. Hypothesis 2 (H2): The earnings response coefficient (ERC) is higher when the same level of earnings surprise is accompanied by a higher level of cash flow sur- 9

11 prise. We next explore whether the influence of the magnitude of the cash flow surprise on the market s reaction to an earnings surprise is different for positive and negative earnings surprises, i.e. we examine the importance of cash flow forecasts separately in instances where the associated earnings surprise is positive and negative. Prior research (for example, Bartov et al., 2002; Skinner and Sloan, 2002; etc.) suggests that the stock market penalizes firms for failing to meet analysts expectations and that managers often manage earnings to avoid missing these expectations (Burgstahler and Eames, 2006). Given the importance of meeting analysts earnings forecasts and the fact that managers sometimes manipulate earnings in order to meet these forecasts, we argue that the magnitude of the cash flow surprise will play an important role in helping the market assess the quality of a reported positive earnings surprise. In a related study, Brown et al. (2013) show that the abnormal market return to a positive earnings surprise is higher when the accompanying cash flow surprise is also positive, than when it is negative, suggesting that the market s confidence in a positive earnings surprise increases when it is backed up by a confirmatory signal. In contrast, when the firm misses an earnings forecast, the market is likely less concerned about the credibility of the signal. This is because prior research has established that the credibility of bad news is higher than that for good news (Hutton et al., 2003; Mercer, 2004; Anilowski et al., 2007; etc.). In other words, when a firm reports a negative earnings surprise, it is probable that the market believes the news to be sufficiently credible. In such cases, the magnitude of the cash flow forecast error is likely to be less important. We formally state the hypothesis below: Hypothesis 2a (H2a): The earnings response coefficient for a positive (negative) earnings surprise is incrementally higher (not higher) when the accompanying cash flow surprise is high. Finally, we test the impact of the extent of cash flow surprise on the future liquidity of the firm. Prior research, for example, Ng (2011), Lang et al. (2012), etc. has shown a positive association between information quality and stock liquidity. In other words, an improved information envi- 10

12 ronment leads to the stock becoming more visible, thereby improving stock liquidity. Analysts likely contribute to the improvement in the quality of information available by generating private information (Lobo et al. 2012). DeFond and Hung (2003) argue that analysts provide CPS forecasts in response to investor demand for such forecasts. Presumably, the demand for CPS forecasts arises from greater information asymmetry. 11 The expectation is that CPS forecasts alleviate this problem. We argue that if the magnitude of the CPS forecast error conveys additional valuable information about the quality of an earnings surprise, it would likely improve the information environment, thereby improving stock liquidity in the future. Formally, we hypothesize that: Hypothesis 3 (H3): Future stock liquidity is higher when the same level of earnings surprise is accompanied by a higher level of cash flow surprise. 2.2 Research Methodology It is important to keep in mind that analysts do not randomly choose firms for which to provide CPS forecasts. If analysts only provide CPS forecasts for certain types of firms (DeFond and Hung, 2003; Bilinski, 2014), then it is possible that the results for H1 (impact of the presence of CPS forecasts on the bid-ask spreads) would be driven by inherent differences in firm characteristics, rather than the presence of the CPS forecasts. In order to make definite inferences about the impact of the presence of CPS forecasts on the bid-ask spreads, therefore, we need to explicitly model this potential endogeneity. We do so in this paper by using the Heckman (1979) two-stage methodology. 12 In the first stage, we model the analysts decision to issue a CPS forecast by estimating the following probit model: 11 Prior research (Moyer et al. 1989; Ahn et al., 2005; etc.) has shown that high information asymmetry increases analyst coverage and makes private information acquisition more valuable. 12 Following a recent trend in accounting research (see Lawrence et al., 2011; Armstrong et al., 2010; etc.), we also match on propensity scores to control for the selection bias. These results are discussed in detail under Robustness Analsyis in Section 3. 11

13 CP S it = α 0 + α 1 AccHet it + α 2 Size it + α 3 ZScore it + α 4 EarnV ol it + α 5 CapInt it + η it (1) In equation (1) above, the dependent variable, CP S it indicates whether there is a CPS forecast in a particular quarter or not. The independent variables in the model have been drawn from DeFond and Hung (2003) AcctHet it is accounting choice heterogeneity, constructed as an index ranging from 0 to 1 computed by assigning a value of one to each firm whose accounting choice differs from the most frequently chosen method in that firm s industry group, for each of the following five accounting choices: (1) inventory valuation; (2) investment tax credit; (3) depreciation; (4) successful-efforts vs. full-cost for companies with extraction activities; and (5) purchase vs. pooling. We code a choice as zero if information is missing. The score for each firm is summed, and then scaled by the number of accounting choices in the industry: 5 for firms in the petroleum and natural gas industry (because they are eligible for all 5 choices); 3 for firms in banking, insurance, real estate, and trading industries (because they have no inventory choice and are not extractive industries); and 4 for firms in all other industries (because they are not extractive industries). Size it is the natural logarithm of the market value of equity. ZScore it is the Altman (1968) Z-score, calculated as follows: ZScore it = 1.20 NW C it T A it RE it T A it EBIT it T A it MV E it BV L it +1.0 Sales it T A it 1 (2) In equation (2) above, NW C it is net working capital, defined as the difference between current assets and current liabilities, RE it is retained earnings, EBIT it is earnings before interest and taxes, MV E it is the market value of equity, BV L it is the book value of liabilities, Sales it is the net sales for the quarter and T A it 1 is total assets at the end of the previous quarter. All variables required to calculate the Z-score have been obtained from the Compustat North America Quarterly data. EarnV ol it is earnings volatility, defined as the absolute value of the ratio of the cross-sectional 12

14 standard deviation of earnings per share to the mean of earnings per share (obtained from I/B/E/S); CapInt it is capital intensity, defined as the ratio of gross property, plant and equipment and last quarter s sales (both obtained from Compustat). We obtain the Inverse Mills Ratio (IMR it ) from equation (1) above and use it as a regressor in the second stage model (presented below) to test H1A: Spread it = β 0 +β 1 Both it +β 2 P rice it +β 3 Ln(V ol) it +β 4 StdRet it +β 5 IMR it +Σ j Ind j +Σ t Y ear t +ɛ it (3) In equation (3) above, Spread it is the daily bid-ask spread measured over nine day windows before and after the earnings announcement; P rice it is the quarterly closing stock price and V ol it is the total number of shares traded during the quarter and StdRet it measures the standard deviation of daily returns, measured over the quarter. 13 The control variables have been chosen from prior research (for example, Benston and Hagerman, 1974; Coller and Yohn, 1997; etc.). Both it is a dummy variable set equal to one if both the EPS and the CPS forecasts are present in a given quarter; it is zero if only the EPS forecast is present. The coefficient of interest in the above model is ˆβ 1. This coefficient measures the impact of the presence of the CPS forecast on the bid-ask spread. H1A predicts that this coefficient will be significantly negative. We use the following model to test H1B: AbnV ol it = β 0 + β 1 Both it + β 2 UE it + β 3 Ln(T A) it + β 4 Growth it + β 5 Lev it + β 6 CAR it + β 7 IMR it + Σ j Ind j + Σ t Y ear t + ɛ it (4) In equation (4) above, AbnV ol it is the abnormal trading volume surrounding the earnings announcement date. We measure the abnormal trading volume over two windows three days and five days surrounding the earnings announcement date respectively. The abnormal trading volume is calculated by subtracting the normal volume during a forty two day period starting 7 days 13 The bid-ask spread is measured over two separate windows (-10,-1) and (1,10) days with respect to the earnings announcement date. 13

15 prior to the earnings announcement, from the three day and five day trading volume surrounding the earnings announcement date. UE it refers to the earnings forecast error, calculated as the difference between the reported quarterly EPS and the consensus EPS forecast, calculated as the median of the last forecast issued by each analyst for the quarter, scaled by the closing price of the previous quarter. CAR it is the three-day size-adjusted cumulative abnormal returns. The other control variables included in the model are size (measured by the natural logarithm of total assets Ln(T A)), growth (measured by the closing market-to-book ratio), and leverage (measured as the ratio of long-term and short-term debt to the current market value of equity). H1B predicts that ˆβ 1, which measures the impact of the presence of the CPS forecast on the abnormal trading volume, will be significantly positive. In order to test H2 and H3, we first divide our sample into two parts, based on the relative rank of the CPS forecast error, i.e., we construct decile ranks of the CPS forecast error and assign observations lying above decile 5 (below) to the high (low) CPS surprise sample. 14 Following prior research, we use the following equation to test H2: CAR it = β 0 + β 1 High it + β 2 UE it + β 3 High it UE it + β 4 Ln(T A) it + β 5 Growth it + β 6 Lev it + Σ j Ind j + Σ t Y ear t + ɛ it (5) In equation (5) High it is a dummy variable equal to one, if the decile rank of the CPS forecast error is above 5; it is zero otherwise. The other variables are as defined above. High it thus measures the relative rank of the CPS forecast error. The coefficient on interest in H2 is the interaction coefficient ˆβ3, which captures the incremental price reaction to an earnings surprise, that is accompanied by a high CPS surprise. H2 predicts that ˆβ 3 will be positive. In order to test H2a, we modify (5) as follows: 14 In untabulated tests, we test the robustness of the results to the research design by replacing the dummy variable with the ranked CPS forecast error. We find that our results generally hold. 14

16 CAR it = β 0 + β 1 High it + β 2 UE it + β 3 High it UE it + β 4 NegUE it + β 5 High it NegUE it + β 6 NegUE it UE it + β 7 High it NegUE it UE it + β 8 Ln(T A) it + β 9 Growth it + β 10 Lev it + Σ j Ind j + Σ t Y ear t + ɛ it (6) In equation (6) above, NegUE it is a dummy variable equal to 1 if the earnings surprise is negative; it is zero otherwise. H2a predicts that the coefficient ˆβ 3, which measures the incremental ERC for a positive earnings surprise accompanied by a high cash flow surprise, will be positive and that the coefficient ˆβ 7, which measures the incremental ERC for a negative earnings surprise accompanied by a high cash flow surprise, will not be significant. We use the Amihud (2002) illiquidity measure as our proxy for liquidity and estimate the following model to test H3: 15 Liq it+n = β 0 +β 1 High it +β 2 UE it +β 3 High it UE it +β 4 Ln(V ol it )+β 5 Size it +Σ j Ind j +Σ t Y ear t +ɛ it In equation (7) above, Liq it+n is the measure of future liquidity, defined above. We estimate equation (7) above for upto four quarters into the future. The coefficient of interest in equation (7) above is ˆβ 1, which measures the impact of the magnitude of the CPS forecast error on future stock liquidity. H3 predicts that this coefficient will be negative, since the dependent variable is increasing in illiquidity. V ol measures the quarterly trading volume and Size is the natural logarithm of the market value of equity. Note that endogeneity is not a concern for H2 and H3 because we are interested in studying the effect of the magnitude of the cash flow surprise on the market s pricing of the EPS forecast error and future stock liquidity respectively when CPS forecasts are present. Therefore, we do not use the Heckman (1979) two-stage procedure for these tests. 15 The Amihud (2002) illiquidity measure is the ratio of the daily absolute return and the daily trading volume, averaged over the period (a quarter, in our sample). To aid interpretation, this ratio is multiplied by 10 6, following Amihud (2002). (7) 15

17 3 Data and Empirical Tests 3.1 Description of the Data We draw the data on analysts forecasts of EPS and CPS and actual EPS and CPS from the Quarterly Actuals and Quarterly Detail files maintained by Thomson Reuters s I/B/E/S data. We obtain data for the control variables used in this paper from Standard & Poor s Compustat Fundamentals Quarterly data. Finally, we obtain daily stock returns and trading volume from the Center for Research in Security Prices (CRSP) Daily Stock File. Our sample covers the sixteen year period from (both years inclusive). The initial sample in the intersection of I/B/E/S and Compustat consisted of 236,212 firm-quarters. This initial sample was constructed by removing observations with missing data on both analysts EPS forecasts and CPS forecasts and missing data on actual EPS and CPS. We have further restricted the initial sample to only those forecasts that are issued on or before the close of the fiscal quarter. 16 We have constructed the consensus analyst EPS forecast and CPS forecast by taking the median of the last EPS (CPS) forecast issued by a given analyst for a given firm in a given fiscal quarter. From the initial sample, 14,989 firm-quarters were lost because they could not be matched to CRSP or had missing company identifiers. Finally, we have removed outliers with respect to the key variables used in the analysis. 17 This leads to a final sample of 208,549 firm-quarter observations. Table 1 shows the sample selection process. 18 (Insert Table 1 here) Table 2 presents some descriptive statistics for our sample. Panel A (B) presents descriptive statistics for the full sample (sample with CPS forecasts). Panel A shows that the mean (median) EPS forecast error is smaller than the mean (median) CPS forecast error. The difference is particularly noticeable in the case of the mean forecast error (the mean EPS forecast error is , 16 Most one-quarter ahead forecasts are issued prior to the close of the fiscal period. Therefore, this selection criterion resulted in a loss of less than 5 percent of the sample. 17 Following convention, we examine observations lying above (below) the 99 th (1 st ) percentiles of the respective distributions to identify potential outliers. 18 Our regression samples in Tables 3 and 4 are smaller because some of the variables required to estimate accounting choice heterogeneity for the Heckman (1979) selection model have missing values. 16

18 whereas the mean CPS forecast error is Note also that the standard deviation of the CPS forecast is 0.028, higher than that of the EPS forecast, which is 0.014). This suggests that, on average EPS forecasts tend to be less volatile and optimistically biased than CPS forecasts. This is not surprising because cash flows are inherently more difficult to predict than earnings. The mean (median) three-day size-adjusted cumulative abnormal return (CAR) around the earnings announcement date is 0.1% (0.1%), similar in magnitude to those reported in prior research. The descriptive statistics for the other variables are very similar to those reported in prior studies. From Panel B, we see that analysts tend to provide CPS forecasts for the larger firms (mean and median total assets and sales are significantly larger for Panel B, than for Panel A), suggesting that analysts might strategically choose firms for which to provide CPS forecasts. We, however, do not find any significant difference in terms of mean and median EPS forecast errors between Panels A and B, indicating that CPS forecasts do not influence the nature of bias in the EPS forecasts. (Insert Table 2 here) We next focus on the distribution of the three-day CAR surrounding the earnings announcement, separately for positive and negative earnings surprises, in order to get some preliminary insights into whether CPS forecasts affect the market s perceptions of these surprises. Figure 1 below focuses on those instances where the earnings surprise is positive. The figure charts out the CAR separately for cases where the accompanying cash flow surprise is also positive (dashed blue line) and negative (red line) respectively. The graph clearly shows that the CAR is more positive when a positive EPS error is supported by a positive CPS error, consistent with the empirical results in Brown et al. (2013). This presents strong evidence that when a firm reports a positive earnings surprise, the market judges the quality of the surprise by also considering the direction of the cash flow surprise, suggesting that analysts CPS errors have an important valuation role for positive earnings surprises. (Insert Figure 1 here) 17

19 Figure 2 presents the distribution of the three-day CAR surrounding the earnings announcement for negative earnings surprises and charts out the CAR separately for instances where the cash flow surprise is positive (dashed blue line) and negative (red line). This graph presents an interesting pattern. It suggests that the distributions of the abnormal returns are similar for both cases, suggesting that when the EPS forecast error is negative, the information role for the CPS forecast error is minimal. This result is not surprising, as it is established in the literature that bad news is more credible than good news (Langberg and Sivaramakrishnan, 2008). Thus when the earnings surprise is negative, the market does not need to consider the news conveyed by the associated cash flow surprise. Taken together, Figures 1 and 2 suggest that CPS forecasts have an incremental valuation role under particular circumstances. We examine the information role of CPS forecasts in more detail in the next section. (Insert Figure 2 here) 3.2 Empirical Tests Analysts CPS forecasts and the Firm s Information Environment Table 3 presents the results for H1A, which tests the impact of the presence of cash flow forecasts on the bid-ask spread around the earnings announcement date. The Table has two columns: Column 1 presents the results for the nine day window after the earnings announcement date (days 1 to 10) and Column 2 presents the results for the nine day window leading up to the earnings announcement date (days -10 to -1). 19 As mentioned above, we perform a two-stage Heckman (1979) procedure to test H1A. Panel A of Table 3 presents the results from the first stage model, which analyzes analysts decision to issue CPS forecasts. The signs on the explanatory variables are consistent with DeFond and Hung (2003). Specifically, we find that the coefficients on accounting choice heterogeneity, earnings volatility and capital intensity are positive and that on Altman (1968) Z-score is negative in both columns, suggesting that these factors influence analysts de- 19 We have also replicated our results using the (-5,-1) and (1,5) return windows as robustness checks and find that our results hold for each of these alternate windows. 18

20 cision to issue CPS forecasts. We also note that the Inverse Mills Ratio presented in Panel B is significant for the bid-ask spread measured after the earnings announcement (coefficient=0.097, t-statistic=2.72), indicating that there is potential endogeneity and thus justifying our decision to use the Heckman (1979) two-stage procedure. Panel B, which estimates the second stage model, shows that the coefficients on ˆβ 1 are negative and significant in both Columns 1 and 2 (coefficient=-0.078, t-statistic = in Column 1 and coefficient=-0.080, t-statistic= in Column 2), suggesting that the presence of CPS forecasts significantly reduces the bid-ask spreads surrounding the earnings announcement. This result is consistent with the prediction of H1A. The control variables generally have coefficients consistent with theory the coefficient on price is positive (coefficient=0.028 and 0.027, t-statistic= and respectively)and the coefficient on the standard deviation of returns is positive (coefficient=8.603 and respectively, t-statistic= and respectively). We find that the coefficient on the natural logarithm of volume is positive (coefficient=0.041 and respectively, t-statistic=51.52 and respectively). Although this result seems contrary to expectations, it is possible for bid-ask spread to be positively associated with trading volume if more information (owing to the presence of CPS forecasts) is associated with transaction size (Copeland and Galai, 1983). (Insert Table 3 here) Table 4 shows the results for the test of H1B, which predicts that the presence of CPS forecasts increases the abnormal trading volume around the earnings announcement date. Note that we again control for endogeneity in testing H1B. Accordingly, Panel A of Table 4 presents the results from the first stage model and Panel B presents the second-stage results. The two columns in Panels A and B respectively represent estimation results for the three-day and five-day window around the earnings announcement. Panel A of Table 4 shows that only the Altman (1968) Z-Score and size are significant in the first stage regression. As in Table 3 above, we find that the Inverse Mills Ratio is strongly significant (coefficient= and respectively, t-statistic= in both columns). 19

21 Moving to the second stage results in Panel B, we find that the coefficient on Both ( ˆβ 1 ) is positive for both the abnormal trading volume windows (coefficient=0.087 and respectively, t-statistic=3.21 and 1.70 respectively), consistent with the prediction of H1B. This result suggests that CPS forecasts increase the volume of trading activity around the earnings announcement date, indicating thereby that these forecasts convey additional information to the market. Consistent with expectations, we find that the coefficient on the earnings surprise is positive (coefficient=5.629 and respectively, t-statistic=9.97 and respectively). We also find that the coefficient on growth is positive (coefficient=0.066 and respectively, t-statistic=23.16 and respectively). (Insert Table 4 here) The Implications of the Magnitude of the CPS Forecast Error for the Market s Perception of the Earnings Surprise We now move to H2, which predicts that the earnings response coefficient (ERC) is higher if the cash flow content of the earnings surprise is higher. Table 5 shows that the coefficient on High ( ˆβ 1 ) is positive (coefficient=0.013, t-statistic=13.63). This suggests that the level of the CPS surprise is priced by the market. In other words, higher levels of CPS surprise are associated with a higher announcement period abnormal return. Consistent with expectations, we find that the coefficient on the EPS forecast error ( ˆβ 2 ) is positive (coefficient=1.01, t-statistic=9.33). The coefficient of interest in the model, ˆβ 3, is also positive (coefficient=0.321, t-statistic=1.97). This result indicates that the higher the level of the cash flow surprise in the current earnings surprise, the stronger is the abnormal market reaction to the earnings surprise, consistent with the prediction of H2. Table 5 shows that the coefficient on growth is negative (coefficient=-0.001, t-statistic=-3.33), consistent with prior studies (for instance, Skinner and Sloan, 2002; Cooper et al., 2008; etc.), who show a negative association between firm growth and abnormal returns. We also find a positive coefficient on leverage (coefficient=0.003, t-statistic=2.70). (Insert Table 5 here) 20

22 We present results for H2a in Table 6. H2a tests whether the magnitude of the cash flow surprise is more important when the earnings surprise is positive (negative) and predicts that the incremental ERC will be positive (not significant) when the earnings surprise is positive (negative). Table 6 shows that the coefficient ˆβ 3, which measures the incremental ERC for a positive earnings surprise accompanied by a high cash flow surprise, is positive (coefficient=0.363, t-statistic=1.78). In contrast, we see that the coefficient ˆβ7, which measures the incremental ERC for a negative earnings surprise accompanied by a high cash flow surprise is not significant (t-statistic=-0.71). These results are consistent with the prediction of H2a and are also consistent with the evidence presented in Figures 1 and 2 respectively and suggest that cash flow forecasts play an especially important validation role when the associated earnings surprise is positive. This suggests that when the earnings surprise is positive, the market compares the surprise with the associated cash flow surprise in order to judge the quality of the earnings surprise. In contrast, when the earnings surprise is negative, the market considers it to be sufficiently credible and does not appear to consider the associated cash flow surprise. These results thus present interesting insights into when market participants use the cash flow surprise to assess firm performance. (Insert Table 6 here) The Implications of the Magnitude CPS Forecast Error for Future Stock Liquidity Finally, in Table 7 we present results for H3, which tests whether the magnitude of the CPS forecast error improves stock liquidity in future quarters. H3 predicts that high levels of cash flow in the current earnings surprise will improve future liquidity by indicating high earnings quality. We present results using upto four quarters ahead values of liquidity (Columns 1-4 respectively) as the dependent variable in equation (7). The Table shows that the estimated coefficient ˆβ 1 is negative for all four columns (coefficient= in all columns, t-statistic=-3.68, -3.18, and respectively), consistent with the prediction of H3. 20 These results thus show that higher cash flows in current earnings improve 20 The ˆβ 1 estimates reported in the Table appear the same owing to approximation. They are not identical in the 21