Higher ERC or Higher Future ERC from Income Smoothness? The Role of Information Environment

Transcription

1 Higher ERC or Higher Future ERC from Income Smoothness? The Role of Information Environment ABSTRACT We examine the differential effects of income smoothness on value-relevance of current future earnings (termed as earnings response coefficient, ERC, and future earnings response coefficient, FERC, respectively) as affected by different information environments. In theory, if reported earnings are permanent, earnings should explain concurrent returns fully. In this case, the FERC should approach zero. However, a rich information environment provides more information that helps the market better assess the firm s future profitability. To the extent that current permanent earnings cannot fully predict future realized earnings, we should find a positive FERC in rich information environment. Based on the premise that smoothing brings reported earnings closer to permanent earnings, we predict and find that for firms with poor information environment, income smoothness improves ERC but not FERC. For firms with superior information environment, income smoothness improves FERC. We also find higher FERC when earnings are less smoothed for firms in poorer information environment. Many recent studies have focused solely on FERC in assessing the informativeness of earnings. We remind researchers that theoretically more value-relevant earnings do not effectively improve FERC in an environment with little or no other information. Keywords: Income Smoothness; ERC; FERC; Information Environment Data Availability: Data used are from public sources identified in this paper. 1

2 Does Income Smoothness Increase both Current and Future Earnings Response Coefficients 1. INTRODUCTION This paper investigates the differential effects of income smoothness on the valuerelevance of current earnings (termed as earnings response coefficient, ERC) and the valuerelevance of future earnings (termed as future earnings response coefficient, FERC) under different information environments. Collins et al. (1994) suggest that inclusion of future earnings should increase the explanatory power of return-earnings relationship due to the lack of timeliness of earnings. If some of the current earnings, which the market has already responded to, have to be delayed to future, current returns should be associated with future earnings. This implies that if earnings are timelier, the ERC should be higher while the FERC should be lower. However, regardless of the informativeness of earnings, the firm s information environment should have an effect on firm valuation. If more information is available in the market and the market uses this information to predict future profitability not reflected in current earnings, we should observe a positive FERC. In evaluating the effect of income smoothness on earnings informativeness, Tucker and Zarowin (2006) propose to evaluate FERC and they find income smoothness increases FERC. 1 In theory, if permanent earnings are reported, then the market should respond mainly to current earnings since no current earnings will be delayed to future reporting; accordingly, we should observe positive ERC and no or small FERC. Alternatively speaking, higher FERC can mean a lower earnings quality if we define earnings quality based on the closeness between reported 1 Tucker and Zarowin (2006) include past and current earnings in their model; however, they only discuss the effect of income smoothness on FERC, but did not discuss the effect on ERC. 2

3 earnings and permanent earnings. Accordingly, one cannot conclude that higher FERC implies better earnings informativeness. However, evidence from prior studies on FERC indicates that better information environment, as proxied either by better disclosure quality (e.g., Lundholm and Myers, 2002; Gelb and Zarowin, 2002; Ettredge et al., 2005) or by better information intermediaries (e.g., Ayers and Freeman, 2003; Lee et al., 2007; Chou, 2013), improves FERC. This is because better information environment will generate more value-relevant information which will facilitate investors to predict firm s future performance. As a result, current market return should be associated with future earnings. If better earnings property improves the ability for the market to better predict future profitability, we shall also see a positive effect of earnings property on FERC. The purpose of our paper is to illustrate that information environment is a key determinant for researchers to choose FERC in evaluating informativeness affected by an earnings property. More permanent earnings should improves current ERC even if information environment is low in providing other information for predicting future profitability. In a good information environment, if the earnings property is associated with higher earnings predictability, then the FERC should be higher. As a result, a good earnings property can have positive impacts on both ERC and FERC; however, the sources of the impacts are different. The impact on ERC is due to the direct effect that timelier earnings contain key relevant information so that ERC should be higher while FERC should be lower. The impact on FERC is due to the indirect effect that this earnings property is associated with higher predictability of future earnings. To distinguish these two effects, we propose that an investigation of informativeness of earnings characteristics has to consider the role of the information environment. 3

4 As one of the key indicators of earnings quality, income smoothness has been widely examined for decades. Prior studies have documented two primary incentives that managers use their reporting discretions to smooth earnings: (1) efficiently communicating private information about the firm (e.g., Chaney and Lewis, 1995; Sankar and Subramanyam, 2001); and (2) opportunistically masking the real performance of the firm (e.g., Bhattacharya et al., 2003; Leuz et al., 2003; Myers et al., 2007; Jayaraman, 2008). These two underlying incentives lead to opposite predictions that income smoothness could either improve or distort financial reporting informativeness, depending on which incentive dominates. Extant studies provide large sample evidence from US firms that on average income smoothness is associated with more informative earnings, as reflected by higher ERC (Hunt et al., 2000) and higher FERC (Tucker and Zarowin, 2006). As discussed above, ceteris paribus, timeliness of earnings caused by income smoothness should increase ERC and at the same time decrease FERC. We propose the reason for the evidence of higher FERC for smoothed earnings is due to a better information environment that the investors can use more concurrent information to predict future earnings and income smoothness enhances predictability. In sum, we expect that timeliness of earnings will increase ERC through the earnings ability to capture the underlying profitability (e.g. the permanent earnings concept in Ohlson, 1995) but will reduce FERC. We also expect that good information environment will increase FERC through the ability of other information in predicting future earnings. If smoother earnings series are easier to predict and current earnings do not capture all this predictability, we will also observe a positive effect of income smoothness on FERC. Combining these two aspects, we expect that in a poor information environment, income smoothness will mainly increase ERC but not FERC; while in a good information environment, income smoothness will increase ERC and also increase 4

5 FERC. While we can predict the effect of income smoothness on ERC is larger than that on FERC for poor information environment, we cannot predict if the effect on ERC is larger or smaller than that on FERC for good information environment. We leave this as an empirical question. Our full sample consists of 47,990 firm-year observations for 6,542 US firms over the period from 1991 to We collect all data from the 2012 versions of COMPUSTAT, CRSP, and I/B/E/S databases on WRDS. Our sample period ends at 2009 because our return-earnings relation model requires future three-year annual earnings and returns data. We proxy for income smoothness using two alternative measures. Specifically, IS_FLOS is measured as the volatility of cash flows with respect to the volatility of income. The more income smoothing activities a manager engages in, the higher the variability of cash flows with respect to the variability of income will be (Francis et al., 2004). Alternatively, we construct the second measure IS_TZ as the negative correlation between the change in discretionary accruals and the change in prediscretionary income, because this measure assumes that there is an underlying pre-managed income series and that managers use discretionary accruals to make the reported series smooth (Tucker and Zarowin, 2006). To delineate ERC and FERC in the same model, we follow Collins et al.'s (1994) FERC framework with some adjustments. In particular, as the inclusion of future earnings in Collins et al.'s (1994) framework confounds with the traditional interpretation of ERC (Lundholm and Myers, 2002), the ERC and FERC estimates based on earnings levels may not be accurately associated with current and future earnings surprises. Therefore, we focus on changes in current earnings and changes in future earnings as our key variables to identify the ERC and FERC in the same model. For robustness, we also provide results using the earnings-levels model. Our baseline results are consistent with those concluded in the prior studies that income smoothness on average improves 5

6 both ERC and FERC. In addition, we find, on average, the effect on ERC is stronger than that on FERC regardless of measures of income smoothness. To test whether the impact of income smoothness on ERC and FERC depends on the firm's richness of information environment, we estimate our baseline models using subsamples of firms with rich versus poor information environment. We measure information environment by firm size, number of analyst following, analyst forecast errors, or analyst forecast dispersions. We also use multiple regression analyses including three-way interaction variables among information environment measures, income smoothness measures and ERC/FERC. Our evidence indicates that for firms with poor information environment, income smoothness has a significantly positive impact on ERC but not much on FERC. On the other hand, for firm with rich information environment, income smoothness has a significantly positive impact on FERC. The information environment does not have a strong incremental effects on the association between income smoothness and ERC. We find positive FERC when income smoothness is low for firms in poor information environment but not for firms in rich information environment. These results are consistent with our predictions that higher FERC does not necessarily indicate better earnings informativeness. In additional analyses, we first control for potentially omitted correlated variables by including either loss firms, firm growth, or volatility of future earnings together with their corresponding interactions with earnings changes and return variables. Our inferences remain qualitatively similar. Second, we re-conduct our main analyses using the traditional returnearnings model based on earnings-level variables, which is widely applied in the existing FERC studies. While the baseline results are in line with those reported in Tucker and Zarowin (2006), 6

7 the results based on subsamples of rich and poor information environment are largely consistent with the results using earnings-changes models. We contribute to the literature in a significant way. Prior studies on the return-earnings relation mostly focus on either ERC or FERC separately with many recently studies focus only on FERC. In this study, we examine both ERC and FERC under the same setting. We provide theoretical discussion and empirical support for the differential effects on ERC and FERC from information quality (the income smoothness) and firm s information environment. One important implication of our study is that when we evaluate the effect of financial reporting quality, it is crucial to focus on evaluating both ERC and FERC; specifically, we find that it is more important to evaluate ERC for firms with poor information environment. Extant studies on financial reporting informativeness mainly emphasize the importance of FERC, because more informativeness of prices about future earnings leads to more efficient resource allocation in the economy (Durnev et al., 2003). However, our study suggests that if we want to evaluate informativeness of earnings characteristics, we need to interpret the existing findings with respect to FERC with caution. Especially, under circumstances where firms' information environment is poor, we may obtain biased results and draw wrong conclusions if we apply FERC as the only proxy for earnings informativeness. To our knowledge, this is the first paper combining ERC and FERC in evaluating informativeness of earnings characteristics; our methodology and findings should have implications to investigate informativeness of other earnings characteristics. The remainder of this paper is organized as follows. In section 2, we discuss related literature and our hypothesis. Section 3 presents a description of how to measure our key variables. In section 4, we describe our sample, present descriptive statistics, and report our main regression results. Section 5 presents our additional analyses and we conclude in Section 6. 7

8 2. PRIOR LITERATURE 2.1. Background on Informativeness of Earnings Prior accounting literature has been interested in assessing the informativeness of earnings since the publication of two seminal research papers by Ball and Brown (1968) and Beaver (1968). Although early empirical evidence shows a clear statistical association between returns and contemporaneous earnings, the explanatory power of current earnings with respect to stock returns is quite low. In a review paper, Lev (1989) documents both low values of earnings response coefficient (i.e., ERC) and low R 2 values obtained from regressing current stock returns on earnings levels or earnings changes. He attributes these weak results to the low accounting earnings quality. Collins et al. (1994) further argue that the primary reason for the weak return-earnings relation is the lack of timeliness of reported earnings, due to the conventional accruals model that causes current returns to incorporate information that will be reflected in earnings of future periods. 2 This notion is supported by a large body of prior literature (e.g., Beaver et al., 1980; Kothari and Sloan, 1992; Warfield and Wild, 1992). To proxy for the stock market's ability to anticipate future earnings, Collins et al. (1994) develop the future earnings response coefficient (i.e., FERC) framework by adding future earnings into the regression of current returns on current earnings. They find that the explanatory power of the FERC model is three to six times greater than that of the traditional ERC model. Moreover, they find the FERC is large relative to ERC. 2 As conventional accounting model trades off timeliness in favor of objectivity, verifiability, and/or conservatism, reported current earnings cannot capture all the value relevant events that cause revisions in the investors' expectations about future earnings. However, since stock market has access to various available information other than reported earnings, stock returns can capture value relevant events and thus reflect the investors' expectations about future earnings. Therefore, stock returns will be related not only to current earnings but also to future earnings, resulting in low contemporaneous return-earnings relation. 8

9 Following Collins et al. (1994), a number of studies have applied the FERC framework to show that better information environment can "bring the future forward" into current returns. There are mainly two strands of research under this line of literature. The first strand of research investigates whether variations in firms' disclosure practices affect the association between current returns and future earnings. For example, Gelb and Zarowin (2002) and Lundholm and Myers (2002) find that current returns incorporate more information in future earnings for firm with more informative disclosures, as proxied by disclosure scores assigned by the Association for Investment Management and Research (i.e., AIMR). By focusing on a specific type of disclosure, Ettredge et al. (2005) investigate the effect of the firms' adoption of SFAS No.131 segment disclosure rules on FERC. They find that firms starting to disclose multiple segments after adopting SFAS No.131experienced a significant increase in FERCs. Similarly, Orpurt and Zang (2009) explore the predictive value of direct method cash flow disclosures. They document that firms utilizing direct method disclosures have higher FERCs than firms utilizing indirect method disclosures. More recently, Choi et al. (2011) focus on management earnings per share forecast and find that firms with more frequent and more accurate management forecasts have higher FERCs. The second strand of studies investigates the role of information intermediaries (e.g., financial analysts, institutional investors, auditors and rating agencies) that helps investors to anticipate future earnings. As financial analysts and institutional investors have advantages of gathering and processing information, Ayers and Freeman (2003) show that stock prices incorporate future earnings earlier for firms with heavy analyst following or high institutional holdings than for other firms. Moreover, Lee et al. (2007) find that stock prices better anticipate future earnings when financial statements are audited by the big accounting firms. Most recently, 9

10 Chou (2013) finds that firms with credit ratings have higher FERCs than non-rated firms, and that FERCs are even higher for higher-rated firms. Overall, evidence from both disclosure practices and information intermediaries suggests that FERCs are higher for firms with better information environment Income Smoothness Income smoothness represents managers' attempts to use their reporting discretions to "intentionally dampen the fluctuations of their firms' earnings realizations" (Beidleman, 1973). It has been the subject of extensive practical debates and academic research for decades. In practice, as the outcome of income smoothness is to reduce the inherent variability of earnings, investors generally have strong preferences towards smoothed earnings. Graham et al. (2005), in their survey of CFOs, find that an overwhelming percentage of the survey respondents indicate that they prefer a smooth earnings path (96.9%), smoothed earnings are perceived less risky by investors (88.7%), and smoothed earnings should make it easier for investors to predict future earnings (79.7%). Managers use their discretions to smooth earnings across periods, and thus the firms' reported earnings in any period may not reflect the economic earnings of that period (Trueman and Titman, 1988). When managers smooth earnings over time, such behaviors can either improve or impair the informativeness of earnings depending on the managerial incentives to smooth earnings. Two streams of academic research exist regarding different managerial incentives underlying income smoothness activities. On the one hand, managers use their reporting discretions to smooth earnings to efficiently communicate private information about the firm. For example, Ronen and Sadan (1981) apply the signaling theory by Spence (1973) and argue that only firms with good future prospects smooth earnings since borrowing from the future could be disastrous to a poorly performing firm. In a similar vein, Chaney and Lewis (1995) develop a model in which the 10

11 managers of high quality firms smooth earnings to convey their private information, allowing investors to better predict future earnings. In addition, Sankar and Subramanyam, (2001) model a two-period pure exchange economy where the manager has private information regarding future earnings. They show that managers smooth earnings in the first period to communicate their private information through reported earnings. On the other hand, managers may smooth earnings opportunistically to mask the real performance of the firm, and thus income smoothness is harmful to investors. For example, Bhattacharya et al. (2003) and Jayaraman (2008) show that discretionary income smoothness distorts the contemporaneous information content of earnings and cash flows, leading to greater earnings opacity. Leuz (2003) suggests that insiders of firms smooth income to conceal true firm performance so as to protect the private benefits of control. Myers et al. (2007) further offer evidence that firms use income smoothness as an earnings management tool to artificially maintain long strings of earnings increase trend. Since the two alternative income smoothness incentives lead to opposite predictions on the informativeness of earnings, it is an empirical question of which incentive dominates in a crosssectional setting. Two prior studies directly investigate the implications of income smoothness on the informativeness of either current earnings or future earnings. Hunt et al. (2000) find that income smoothness enhances the contemporaneous return-earnings relation, suggesting that income smoothness improves ERC. More recently, Tucker and Zarowin (2006) apply the FERC framework and document that income smoothness on average improves FERC. Both studies provide evidence in supporting the incentive of efficient private information communication. However, this evidence is based on average results, and it is likely that income smoothness may not play a positive role at different information environments. Different from previous papers, our 11

12 paper examines ERC and FERC simultaneously and use the differential effects of income smoothness on ERC and FERC in different information environments to infer the source of informativeness of income smoothness. We suggest that if income smoothness increases ERC in a poor information environment, the reason is more due to its timeliness in representing underlying profitability (the permanent earnings concept as discussed in Ohlson, 1995). From this perspective, higher FERC actually means worse earnings quality. On the other hand, if income smoothness increases FERC in a rich information environment, the reason is more due to the ability of other information in predicting future earnings Income Smoothness, Information Environment, and Informativeness of Earnings Extant research on the informativeness of income smoothness generally overlooks the impact of the information environment, with one noticeable exception by Allayannis and Simko (2009). They argue that the under a poor information environment where there is limited information about the firm, the improvement of informativeness of earnings caused by income smoothness should be valued more by investors. Using analyst following as the proxy for a firm's information environment, they find that the market premium for firms that smooth earnings is concentrated among firms with low or no analyst following, while there is no such premium for firms with a high analyst following. Allayannis and Simko s (2009) finding suggests that smoothed current earnings are important in firm valuation only for firms with poor information environment, which is consistent with our prediction that income smoothness improves ERC for firms with poor information environment. However, their conclusion is incomplete in that they do not investigate how a firm s information environment will affect the role of earnings smoothness on FERC, which is even more important in firm valuation. Our study complements Allayannis and Simko (2009) 12

13 by directly testing the different roles of information environment in affecting the effectiveness of income smoothness to improve ERC and FERC. In our study, we develop two competing hypotheses for the impact of income smoothness on ERC and FERC: earnings timeliness hypothesis and information environment hypothesis. Figure 1 below shows the theoretical framework underlying these two hypotheses. Specifically, we first model current returns as a function of unexpected current earnings: Ret t = α UE t + ε t. Then, the theoretical unexpected earnings can be proxied by change in current earnings with noise: UE t = β E t + μ t. According to Collins et al. (1994), if some of current earnings are not timely reflected, then current returns should capture information in future earnings. Therefore, the noise in current unexpected earnings can be modeled as a function of change in future earnings: μ t = γ E t+k + δ t. <<Insert Figure 1 here>> On the one hand, since one of the main purposes of the income smoothness is to make current earnings more permanent, investors tend to value current earnings more for smoothing firms. In other words, a timelier and more informative accounting system caused by income smoothness makes E t a better proxy for UE t. Thus, the market should respond more to current earnings (i.e., higher β or ERC), leaving less information being revealed by future earnings (i.e., lower γ or FERC). Therefore, under earnings timeliness hypothesis, we expect income smoothness to increase ERC but decrease FERC through the direct earnings timeliness effect. On the other hand, a rich information environment can provide other value relevant information (e.g., current unreported accrued earnings and future earnings arising from investment and growth). As more other information is available in the market and the market uses this information to predict future profitability (i.e., UE t+k ), a better information environment will improve FERC. If smoothed 13

14 earnings series enhances the ability of other information in predicting future earnings, it may also have a positive impact on FERC. Therefore, under information environment hypothesis, we expect that income smoothness increases FERC through the indirect future earnings predictability effect. In sum, earnings timeliness hypothesis and information environment hypothesis jointly determine how income smoothness affects both ERC and FERC. Which hypothesis dominates depends on the richness of the firm s information environment. Given the predictions under the two hypotheses, we expect that in poor information environments, income smoothness improves ERC but has little effect on FERC; in rich information environment, income smoothness improves both ERC and FERC. 3. MEASUREMENT OF VARIABLES 3.1. Measures of ERC and FERC Our approach to investigate the ability of returns to reflect information in current and future earnings follows models applied by Collins et al. (1994) and Gelb and Zarowin (2002) as follows: 3 R t = β 0 + β 1 UX t + k=1 β k+1 E t (X t+k ) + e t (1) where R t is the continuously compounded return for fiscal year t, X t is the continuously compounded growth rate of earnings, UX t = X t E t 1 (X t ) is the unexpected earnings growth rate, and E t is the change in market expectations from the beginning to the end of period t. Under the assumption that earnings follow a random walk, Collins et al. (1994) use the realized earnings for year t+k as the proxy for the earnings expectation formed at the end of year t, and use past earnings to form an expectation at the beginning of the year t. Therefore, the change of earnings proxies for the unexpected earnings, and equation (1) can be re-written as: 3 R t = β 0 + β 1 X t + k=1 β k+1 X t+k + e t (2) 14

15 As Collins et al. (1994) point out, using the realized future earnings to proxy for investors expectation introduces an error in variables problem. To reduce the measurement error, they include future returns as the instrument variable in the model. Moreover, Lundholm and Myers (2002) modify Collins et al. s (1994) model using earnings levels for a more general form, and they aggregate the three future years' earnings (returns) into one variable X t3 (R t3 ). We follow this method and implement our model as follows 3 : where R t = β 0 + β 1 X t + β 2 X t3 + β 3 R t3 + e t 4 (3) Rt = the cumulative return for fiscal year t, as measured over the fiscal year from the fourth month after the prior fiscal year end; ΔXt = X t X t 1 ; ΔXt3 = X t3 X 3 t; Xt-1 = Income available to common shareholders before extraordinary items for year t-1 deflated by the market value of equity at the beginning of fiscal year t; Xt = Income available to common shareholders before extraordinary items for year t deflated by the market value of equity at the beginning of fiscal year t; Xt3 = The sum of income available to common shareholders before extraordinary items for years t + 1 through t + 3 deflated by the market value of equity at the beginning of fiscal year t; Rt3 = the cumulative return for fiscal years t + 1 through t + 3; Consistent with Collins et al. (1994) and Gelb and Zarowin (2002), the coefficient β 1 is the ERC proxy and the coefficient β 2 is the FERC proxy. Both coefficients are predicted to be positive Measures of Income Smoothness We construct measure of income smoothness (IS_FLOS) following Francis et al. (2004). 3 We also apply Lundholm and Myers' (2002) return-earnings relation model based on earnings level variables in the additional analyses. 4 In appendix, we show the derivation of Lundholm and Myers' (2002) return-earnings relation model based on earnings level variables from the model based on earnings change variables. 15

16 In particular, IS_FLOS is measured as the volatility of income with respect to the volatility of cash flows over the most recent five-year rolling window (Equation 4). The more income smoothing a manager engages in, the higher the variability of cash flows with respect to the variability of income will be. Thus, a higher ratio signifies a smoother income stream. IS_FLOS it = StdDev ( CFO it ) TA it 1 StdDev ( NI it TA it 1 ) (4) where NIit = Net income before extraordinary items (IB) at year t; CFOit = Cash flows from operations less cash flows from extraordinary items, (OANCF XIDOC), at year t, following the approach in Hribar and Collins (2002); = Total assets (AT) at year t-1. TAit-1 Alternatively, we follow Tucker and Zarowin (2006) and measure income smoothness (IS_TZ) as the negative correlation between the change in discretionary accruals and the change in pre-discretionary income based on the Modified Jones (1991) model, adjusted for firm s performance (Kothari et al., 2005). TAcc t 1 Sales = a TA 1 + a t PPE t 1 TA 2 + a t t 1 TA 3 + a t 1 TA 4 ROA t + ε t (5) t 1 where TAcct = Total accruals, the dependent variable, measured as income before extraordinary items (IB) less cash flows operating less cash flows from extraordinary items (OANCF XIDOC), at year t, following the approach in Hribar and Collins (2002); ΔSalest = Change in sales, sales revenue at year t less sales revenue at year t-1; PPEt = Net property, plant and equipment, at year t; ROAt = Return on assets at year t, measured as net income before extraordinary (IB) at year t, scaled by total assets at year t-1; = Total assets at year t-1. TAt-1 Equation (5) is estimated cross-sectionally each year within the same industry group 16

17 (industrty is defined by two-digit SIC) to obtain the expected (non-discretionary) accruals, and the difference between the observed value and the fitted value (i.e., the residual ˆ t ) is the discretionary accruals predicted (DAP). Pre-discretionary income (PDI) is then defined as net income minus discretionary accruals. As the volatility of earnings consists of three components (namely, the volatility of operating cash flows, the volatility of accruals and the correlation between operating cash flows and accruals), the volatility of earnings will be lower when the correlation between operating cash flows and accruals is more negative. Therefore, the more negative the correlation the smoother the income stream should be. IS_TZ measure is the negative correlation between a firms change in discretionary accruals and its change in pre-discretionary income using a fiveyear rolling window, IS_TZit = Corr(ΔDAPit, ΔPDIit) Measures of Information Environment and Other Control Variables We follow prior studies and use firm size (e.g., Collins et al., 1987; Freeman, 1987), the number of analysts following (e.g., Easley et al., 1998; El-Gazzar, 1998; Callen et al., 2006), analyst forecast errors (e.g., Lang and Lundholm, 1996), analyst forecast dispersions (e.g., Abarbanell et al., 1995; Diether et al., 2002) as our proxies for firms' information environment. Specifically, firm size is measured as the market value of common equity at the beginning of fiscal year t. Number of analyst following is measured as the natural log of (1 plus the number of analysts following) at the firm in the month prior to the earnings announcement for fiscal year t. Analyst forecast error is measured as the magnitude of one-year ahead earnings per share minus the consensus estimate in the third month following the fiscal year end scaled by the price per share when the estimate was made. 5 Analyst forecast dispersion is measure as the standard deviation of 5 Using actual one-year ahead earnings necessarily creates a look ahead bias. However, note that forecast dispersion does not contain such bias. 17

18 the around the consensus estimate scaled by price. 6 To control for industry and time effects, we use the fractional ranking variables of each information environment proxy within its industryyear (two-digit SIC). In addition, we use the reversed measures of analyst forecast errors and analyst forecast dispersions (i.e., multiplying each of the two measures by negative one) to make higher values of AF_ERR and AF_DISP represent better information environment. In the additional analyses, to control for other potentially omitted correlated variables, we follow prior studies and include the following control variables with their corresponding interactions with earnings change variables: (1) Loss firms (LOSSt) is measured as an indicator variable equals 1 if ΔXt3 is negative and 0 otherwise; (2) Firm growth (GROWTHt) is measured as the percentage growth in total assets from year t-1 to year t+1; (3) Volatility of future earnings (EARNSTDt) is measured as the standard deviation of X for year t through t DATA AND MAIN EMPIRICAL RESULTS 4.1. Sample Selection We collect financial statement data from COMPUSTAT, stock returns and prices from CRSP, and analyst information from I/B/E/S unadjusted summary file. Table 1 summarizes the data availability for our sample. We start with all firms in COMPUSTAT and CRSP with sufficient data for FERC model from 1991 to 2009, excluding firms from the financial (SIC ) and utility (SIC ) industries. The sample period starts from 1991 because 1987 is the first year that firms are required to report statement of cash flows and we use the most recent five-year rolling window to calculate our two alternative income smoothness measures. The period ends at 2009 because the FERC model requires future three-year (i.e., from t+1 to t+3) annual earnings and returns data. We drop observations with insufficient data to compute income smoothness 6 Per share numbers in AF_ERR and AF_DISP are adjusted to the end of the returns accumulation period. 18

19 measures and other control variables. This results a final sample of 47,990 firm-year observations from 6,542 firms. We mitigate the effect of outliers by winsorizing observations that are in the top or bottom 1% of the distributions of the following variables: past, current and future 3 years' earnings, as well as some control variables including GROWTH and EARNSTD. <Insert Table 1 here> 4.2. Descriptive Statistics and Univariate Correlations Table 2 reports descriptive statistics of the 47,990 sample observations. Our first income smoothness measure IS_FLOS has a mean (median) value of (1.139), which is in line with that reported in Francis et al. (2004) and McInnis (2010). Our second income smoothness measure IS_TZ has a mean (median) value of 0.77 (0.947), similar to that reported in Tucker and Zarowin (2006). The descriptive statistics of the return and earnings variables are also similar to those reported in prior studies on informativeness of earnings. For example, the mean (median) values for current returns, future returns are (0.045), (0.155), and those for prior earnings, current earnings, future earnings are (0.038), (0.042), (0.124), respectively. These results are highly comparable with those in Lundholm and Myers (2002) <Insert Table 2 here> Table 3 presents Spearman correlations among our income smoothness proxies, returnearnings variables, information environment proxies, and other control variables. Our two income smoothness measures are positively correlated, with a correlation coefficient of Both IS_FLOSt and IS_TZt are negatively associated with current and future unexpected earnings. These results are consistent with Hunt et al. (2000) and Tucker and Zarowin (2006), suggesting that firms 19

20 with better performance smooth earnings to signal private information and thus make earnings more informative. In addition, both IS_FLOSt and IS_TZt are positively associated with SIZEt (with correlations of and 0.163, respectively), NANALt (with correlations of and 0.106, respectively), AF_ERRt (with correlations of and 0.183, respectively), and AF_DISPt (with correlations of and 0.198, respectively), suggesting that higher-smoothing firms tend to be those with better information environment. <Insert Table 3 here> 4.3. Regression Analyses The Effect of Income Smoothness on Informativeness of Earnings In this section, to first establish a baseline result that can be compared with prior work (e.g., Collins et al., 1994; Lundholm and Myers, 2002; Tucker and Zarowin, 2006), we re-examine the effect of income smoothness on ERC and FERC using our refined model with earnings changes variables. In particular, we extend Equation (3) by including income smoothness measure and its corresponding interactions with earnings changes and future return variables. The empirical model is presented as follows: R t = β 0 + β 1 X t + β 2 X t3 + β 3 R t3 + β 4 IS t + β 5 IS t X t + β 6 IS t X t3 + β 7 IS t R t3 + e t (6) In our tests, we measure ISt as a firm's fractional ranking of either IS_FLOSt or IS_TZt within its industry-year (two-digit SIC) to control for industry and time effects. In addition, we correct for heterokedasticity following White (1980) and perform firm-level clustering to control for correlation that may exist due to the multiple observations from the same firm in our dataset (Petersen 2009). 20

21 The results of the refined baseline model (Equation 3), as reported in the first column of Table 4, are consistent with the prior studies on FERCs. The coefficient on X t (ERC) and the coefficient on X t3 (FERC) are significantly positive ( and 0.508, respectively), indicating that a significant amount of information about current and future earnings has been incorporated into current stock returns. The coefficient on R t3 is significantly negative (-0.096), which confirms the role of future returns as the instrument variable (Collins et al., 1994). These results justify the validity of our refined model. In column two of Table 4, we report the impact of income smoothness on both ERC and FERC using IS_FLOS as the measure of income smoothness. Consistent with the results reported in Tucker and Zarowin (2006), income smoothness improves both ERC and FERC, as evidenced by the significantly positive loadings on the interaction terms IS_FLOS*ΔXt (coefficient=1.288, t- stats=11.29) and IS_FLOS*ΔXt3 (coefficient=0.799, t-stats=7.73), respectively. To test the differential effects of income smoothness on ERC and on FERC, we compare the loadings on IS_FLOS*ΔXt and on IS_FLOS*ΔXt3. The results show that the improvement on ERC is significantly higher than the improvement on FERC (difference=0.489, F test=7.29). From column three of Table 4, we show that the results are robust when using IS_TZ as the measure of income smoothness, with significantly positive loadings on both IS_TZ*ΔXt (coefficient=0.785, t- stats=7.22) and IS_TZ*ΔXt3 (coefficient=0.523, t-stats=5.23), respectively. The differential test shows that the loading on IS_TZ*ΔXt is significantly higher than that on IS_TZ*ΔXt3. (difference=0.235, F test=2.45). Overall, in this section, we not only provide baseline results that are consistent with the results from the prior studies, but also justify the validity of our earningschanges models. <Insert Table 4 here> 21

22 4.3.2 The Effect of Income Smoothness on Informativeness of Earnings Conditional on Information Environment Our results so far indicate that on average income smoothness has a positive effect on both ERC and FERC. We expect this effect to be influenced by the firm's information environment. We first examine how the information environment affects ERC and FERC. In particular, we extend Equation (3) by including information environment proxy and its corresponding interactions with earnings changes and future return variables. The empirical model is presented as follows: R t = β 0 + β 1 X t + β 2 X t3 + β 3 R t3 + β 4 IE t + β 5 IE t X t + β 6 IE t X t3 + β 7 IE t R t3 + e t (7) In column one of Table 5, we report the impact of information environment on both ERC and FERC using the fractional ranking of firm size (SIZE) as the measure of information environment. The results show that better information environment increases both ERC and FERC (i.e., coefficient on IE t X t =0.469, t=4.27; coefficient on IE t X t3 =0.721, t=5.50), with the significantly higher loading on IE t X t3 (difference=-0.252, F test=4.17). From column two, three, and four of Table 5, we show results when using the fractional ranking of the number of analyst following (NANAL), analyst forecast errors (AF_ERR), and analyst forecast dispersions (AF_DISP) as the measure of information environment. The results are similar except there is no significant difference between the loading on IE t X t and that on IE t X t3. <Insert Table 5 here> We then divide our full sample into two sets of subsamples: (1) firms with rich information environment (Rich IE) and (2) firms with poor information environment (Poor IE), and re-estimate Equation (6) using these two subsamples. We define firms with rich (poor) information 22

23 environment as those within the top (bottom) quintile of both SIZE and NANAL, AF_ERR, or AF_DISP. In Table 6, we report the regression results for Equation (6) using subsamples of firms with rich and poor information environment based on the refined model with earnings change variables. Using IS_FLOS as the measure of income smoothness, we find results suggesting that income smoothness improves FERC when information environment is rich, but has little effect on FERC when information environment is poor. In particular, for subsample firms with rich information environment, the coefficients on IS_FLOS*ΔXt3 are all significantly positive across different information environment proxies: (t-stats=4.05), (t-stats=2.39), and (tstats=3.38), respectively, when the proxy for information environment is SIZE+NANAL, AF_ERR, or AF_DISP. In contrast, for subsample firms with poor information environment, the coefficients on IS_TZ*ΔXt3 are all insignificant: (t-stats=1.52), (t-stats=1.57), and (tstats=1.06), respectively, when the proxy for information environment is SIZE+NANAL, AF_ERR, or AF_DISP. These results provide initial evidence suggesting that the conclusions drawn by Tucker and Zarowin (2006) are conditional on the richness of information environment. In addition, our results also indicate that information environment affects the impact of income smoothness on ERC. In particular, for subsample firms with rich information environment, the coefficients on IS_FLOS*ΔXt are all significant for different information environment proxies: 1.504(t-stats=2.83), (t-stats=5.99), and (t-stats=2.74), respectively. For subsample firms with poor information environment, the coefficients on IS_FLOS*ΔXt are also all significantly positive for different information environment proxies: (t-stats=3.45), (tstats=4.33), and (t-stats=2.64), respectively. These results suggest that income smoothness improves ERC regardless of information environment of the firm. 23

24 The results using IS_TZ as the measure of income smoothness are robust. The effect of income smoothness on future (current) ERC is more pronounced for firm with rich (poor) information environment than the impact for firm with poor (rich) information environment. For firms with both large size and high analysts following, the coefficient on IS_TZ*ΔXt3 is significantly positive (1.732, t-stats=3.54) and on IS_TZ*ΔXt is also significant (0.969, t- stats=2.10). For firms with both small size and low analysts following, the coefficient on IS_TZ*ΔXt3 is insignificant (0.056, t-stats=0.29) and on IS_TZ*ΔXt is significantly positive (0.728, t-stats=3.68). These results are in line with the results estimated using IS_FLOS as the income smoothness proxy. When measuring information environment using AF_ERR, or AF_DISP, the results are robust. <Insert Table 6 here> In addition to the subsample analyses, we further extend the models by including the information environment variable to directly test its incremental impact. Specifically, we first generate an indicator variable (IE), which equals 1 if the firm is both in top SIZE quintile and in top NANAL quintile, and 0 if the firm is both in bottom SIZE quintile and in bottom NANAL quintile. Then we interact IE with all the variables in Equation (8) as follows: R t = β 0 + β 1 X t + β 2 X t3 + β 3 R t3 + β 4 IS t + β 5 IS t X t + β 6 IS t X t3 + β 7 IS t R t3 + β 8 IE t + β 9 IE t X t + β 10 IS t X t3 + β 11 IE t R t3 + β 12 IS t IE t + β 13 IS t IE t X t + β 14 IS t IE t X t3 + β 15 IS t IE t R t3 + e t (8) Our variables of interest are ISt*ΔXt, ISt*IEt *ΔXt, ISt*ΔXt3, and ISt*IEt *ΔXt3, which clearly show the incremental impacts of information environment on the effectiveness of income smoothness to improve ERC and FERC, respectively. Table 7 column one presents the regression results estimated using the earnings change model with IS_FLOS as the income smoothness 24

25 measure and SIZE+NANAL as the information environment measure The coefficient on ISt *ΔXt is significantly positive (0.734, t-stats=3.45) while the coefficient on ISt*ΔXt3 is insignificant (0.316, t-stats=1.52). On the contrary, the coefficient on ISt*IEt*ΔXt is insignificant (0.771, t- stats=1.35) while the coefficient on ISt*IEt *ΔXt3 is significantly positive (2.006, t-stats=3.29). Taken together, these results are consistent with subsample analyses in that (1) for firms with poor information environment, income smoothness is effective in improving ERC but not FERC, while (2) better information environment can make the role of income smoothness on FERC more effective. These results are robust using alternative information environment proxies or income smoothness proxies, as reported in the rest of Table 7. <Insert Table 7 here> 5. ADDITIONAL ANALYSES In this section, we perform additional analyses to ascertain whether our results are robust to the inclusions of other control variables and alternative return-earnings models. Prior studies have documented various determinants of ERC and FERC, such as earnings persistence, growth, and volatility of earnings. Therefore, our results could be driven by alternative explanations that our income smoothness measures are merely proxies for these more fundamental determinants. To control for these potentially omitted correlated variables, we interact each of these control variables with the earnings change and returns variables in our main models, and add them individually to the regressions. We follow prior studies to proxy and control for loss firms (LOSS), firm growth (GROWTH), and volatility of earnings (EARNSTD). In Panel A and Panel B of Table 8, we report the regression results for Equation (7) with the control variables and Equation 25

26 (8) with the control variable, respectively. 7 The results are robust after controlling for these previously identified determinants of the ERC and FERC. <Insert Table 8 here> We also check if our results hold using the traditional models based on earnings level variables (e.g., Equation 6 in Lundholm and Myers, 2002: R t = b 0 + b 1 X t 1 + b 2 X t + b 3 X t3 + b 4 R t3 + e t ). Panel A of Table 8 first presents the baseline results, which are consistent with the results reported in Tucker and Zarowin (2006). Next, we separate the full sample into firms with rich versus poor information environment, and report the results in Panel B of Table 8. Even though the evidence is still consistent with our predictions, the effect of information environment is not as pronounced as that estimated using our refined model. Furthermore, we re-estimate Equation (8) using the earnings level variables and report the results in Panel C of Table 9. Taken together, our results still hold but become weaker if we apply the traditional model based on earnings level variables. The weaker results further support that our earnings-changes models better distinguish ERC from FERC. <Insert Table 9 here> 6. CONCLUSIONS In this study, we investigate how income smoothness affects both ERC and FERC, and whether the effects are conditional on the information environment. Using a sample of 47,990 firm-year observations for 6,542 US firms over the period from 1991 to 2009, we first establish a 7 For simplicity, we only report subsample results of rich and poor information environment using both SIZE and NANAL as the proxy for IE. The untabulated results are robust by using either AF_ERR or AF_DISP as the IE proxy. 26

27 baseline association between income smoothness and ERC/FERC in an unconditional context. Our baseline results are comparable to those from prior studies that income smoothness on average improves both ERC and FERC. More importantly, we show that the average associations between income smoothness and ERC/FERC are dependent on firm's information environment. In particular, we find that income smoothness increases ERC, not FERC for firm with poor information environment, while it significantly improves FERC for firms with rich information environment. We also find higher FERC when earnings are less smoothed for firms in poorer information environment. Our findings have one important implication that when evaluating the effect of financial reporting quality, we need to focus on evaluating both ERC and FERC, not just FERC as many concurrent studies focus on. Specially, we remind researchers that a good earnings quality can increase ERC but at the same time decrease FERC unless good information environment provides more information that is relevant to future profitability. 27

28 REFERENCES Abarbanell, Jeffery S., William N. Lanen, and Robert E. Verrecchia, 1995, Analysts' forecasts as proxies for investor beliefs in empirical research, Journal of Accounting and Economics 20, Allayannis, George, and Paul Simko, 2009, Earnings smoothing, analyst following, and firm value, Working Paper. Ayers, Benjamin C., and Robert N. Freeman, 2003, Evidence that analyst following and institutional ownership accelerate the pricing of future earnings, Review of Accounting Studies 8, Ball, Ray, and Philip Brown, 1968, An empirical evaluation of accounting income numbers, Journal of Accounting Research 6, Beaver, William H., 1968, The information content of annual earnings announcements, Journal of Accounting Research 6, Beaver, William, Richard Lambert, and Dale Morse, 1980, The information content of security prices, Journal of Accounting & Economics 2, Beidleman, C., 1973, Income smoothing: The role of management, The Accounting Review 48, Bhattacharya, Utpal, Hazem Daouk, and Michael Welker, 2003, The world price of earnings opacity, The Accounting Review 78, Callen, Jeffrey L., Joshua Livnat, and Dan Segal, 2006, The information content of sec filings and information environment: A variance decomposition analysis, The Accounting Review 81, Chaney, Paul K., and Craig M. Lewis, 1995, Earnings management and firm valuation under asymmetric information, Journal of Corporate Finance 1, Choi, Jong-Hag, Linda Myers, Yoonseok Zang, and David Ziebart, 2011, Do management eps forecasts allow returns to reflect future earnings? Implications for the continuation of management's quarterly earnings guidance, Review of Accounting Studies 16, Chou, Ting-Kai, 2013, Information content of credit ratings in pricing of future earnings, Review of Quantitative Finance and Accounting 40, Collins, Daniel W., S. P. Kothari, and Judy Dawson Rayburn, 1987, Firm size and the information content of prices with respect to earnings, Journal of Accounting & Economics 9,

29 Collins, Daniel W., S. P. Kothari, Jay Shanken, and Richard G. Sloan, 1994, Lack of timeliness and noise as explanations for the low contemporaneuos return-earnings association, Journal of Accounting and Economics 18, Diether, Karl B., Christopher J. Malloy, and Anna Scherbina, 2002, Differences of opinion and the cross section of stock returns, Journal of Finance 57, Durnev, Artyom, Randall Morck, Bernard Yeung, and Paul Zarowin, 2003, Does greater firmspecific return variation mean more or less informed stock pricing?, Journal of Accounting Research 41, Easley, David, Maureen O'Hara, and P. S. Srinivas, 1998, Option volume and stock prices: Evidence on where informed traders trade, Journal of Finance 53, El-Gazzar, Samir M., 1998, Predisclosure information and institutional ownership: A crosssectional examination of market revaluations during earnings announcement periods, The Accounting Review 73, 119. Ettredge, Michael L., Kwon Soo Young, David B. Smith, and Paul A. Zarowin, 2005, The impact of sfas no. 131 business segment data on the market's ability to anticipate future earnings, The Accounting Review 80, Francis, Jennifer, Ryan LaFond, Per M. Olsson, and Katherine Schipper, 2004, Costs of equity and earnings attributes, The Accounting Review 79, Francis, Jennifer, and Katherine Schipper, 1999, Have financial statements lost their relevance?, Journal of Accounting Research 37, Freeman, Robert N., 1987, The association between accounting earnings and security returns for large and small firms, Journal of Accounting & Economics 9, Gelb, David S., and Paul Zarowin, 2002, Corporate disclosure policy and the informativeness of stock prices, Review of Accounting Studies 7, Graham, John R., Campbell R. Harvey, and Shiva Rajgopal, 2005, The economic implications of corporate financial reporting, Journal of Accounting and Economics 40, Hribar, P. and Daniel W. Collins, 2002, Errors in Estimating Accruals: Implications for Empirical Research. Journal of Accounting Research, 40, Hunt, Alister, Susan E. Moyer, and Terry Shevlin, 2000, Earnings volatility, earnings management, and equity value, Working Paper Jayaraman, Sudarshan, 2008, Earnings volatility, cash flow volatility, and informed trading, Journal of Accounting Research 46,

30 Jones, Jennifer J., 1991, Earnings management during import relief investigations, Journal of Accounting Research 29, Kothari, S. P., Andrew J. Leone, and Charles E. Wasley, 2005, Performance matched discretionary accrual measures, Journal of Accounting & Economics 39, Kothari, S. P., and Richard G. Sloan, 1992, Information in prices about future earnings: Implications for earnings response coefficients, Journal of Accounting and Economics 15, Lang, Mark H, and Russell J Lundholm, 1996, Corporate disclosure policy and analyst behavior, The Accounting Review 71, Lee, B.B., S. Cox, and D. Roden, 2007, Have the big accounting firms lost their audit quality advantage: evidence from the return-earnings relation, Journal of Forensic Accounting 8, Leuz, Christian, Dhananjay Nanda, and Peter D. Wysocki, 2003, Earnings management and investor protection: An international comparison, Journal of Financial Economics 69, Lev, Baruch, 1989, On the usefulness of earnings and earnings research: Lessons and directions from two decades of empirical research, Journal of Accounting Research 27, Lev, Baruch, and Paul Zarowin, 1999, The boundaries of financial reporting and how to extend them, Journal of Accounting Research 37, Lundholm, Russell, and Linda A. Myers, 2002, Bringing the future forward: The effect of disclosure on the returns-earnings relation, Journal of Accounting Research 40, McInnis, John, 2010, Earnings smoothness, average returns, and implied cost of equity capital, Accounting Review 85, Myers, James N., Linda A. Myers, and Douglas J. Skinner, 2007, Earnings momentum and earnings management, Journal of Accounting, Auditing & Finance 22, Ohlson, James A., 1995, Earnings, book values, and dividends in equity valuation, Contemporary Accounting Research 11, Orpurt, Steven F., and Zang Yoonseok, 2009, Do direct cash flow disclosures help predict future operating cash flows and earnings?, The Accounting Review 84, Petersen, Mitchell A., 2009, Estimating standard errors in finance panel data sets: Comparing approaches, The Review of Financial Studies 22, Ronen, Joshua, and S. Sadan, 1981, Smoothing income numbers: objectives, means, and implications, Boston, MA: Addison-Wesley Publishing Company. 30

31 Sankar, Mandira Roy, and K. R. Subramanyam, 2001, Reporting discretion and private information communication through earnings, Journal of Accounting Research 39, Spence, Michael, 1973, Job market signaling, Quarterly Journal of Economics 87, Trueman, Brett, and Sheridan Titman, 1988, An explanation for accounting income smoothing, Journal of Accounting Research 26, Tucker, Jennifer W., and Paul Zarowin, 2006, Does income smoothing improve earnings informativeness?, The Accounting Review 81, Warfield, Terry D., and John J. Wild, 1992, Accounting recognition and the relevance of earnings as an explanatory variable for returns, The Accounting Review 67, White, H, 1980, A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity, Econometrica, 48 (4),

32 Appendix We derive Lundholm and Myers' (2002) return-earnings relation model based on earnings level variables from the model based on earnings change variables as follows: R t = β 0 + β 1 X t + β 2 X t3 + β 3 R t3 + e t (Equation 3) = β 0 + β 1 (X t X t 1 ) + β 2 ( X t3 3 X t) + β 3 R t3 + e t Rearranging: R t = β 0 + β 1 X t β 1 X t 1 + β 2 3 X t3 β 2 X t + β 3 R t3 + e t = β 0 + ( β 1 )X t 1 + (β 1 β 2 )X t + β 2 3 X t3 + β 3 R t3 + e t Let b 0 = β 0, b 1 = ( β 1 ), b 2 = (β 1 β 2 ), b 3 = β 2 3, and b 4 = β 3, then R t = b 0 + b 1 X t 1 + b 2 X t + b 3 X t3 + b 4 R t3 + e t, which is Equation (6) in Lundholm and Myers (2002) 32

33 Figure 1 Theoretical Framework 33