Taking a Long View: Investor-Trading Horizon and Earnings Management Strategy

Transcription

1 Taking a Long View: Investor-Trading Horizon and Earnings Management Strategy Yeejin Jang Purdue University jang67@purdue.edu Kailey (Kyung Yun) Lee Purdue University lee1428@purdue.edu First draft: September 2016 This version: March 26, 2017 Abstract This paper studies how the investment horizon of institutional investors affects firms earnings management strategies. We find that firms largely held by long-term investors are more likely to manage earnings through adjusting operational decisions than through manipulating accruals. The impact of an investor s trading horizon on real activities manipulation is stronger when long-term investors face high performance pressures with low fund flows and high market uncertainty and when they have strong influence on managers with large holdings. We further document that adverse future consequences of operational adjustment are relatively less severe for the firms with long-term investors than for those with short-term investors. Overall, the evidence suggests that firms choose earnings management methods to meet earnings expectations of institutional investors who have different earnings target windows. Our identification strategy exploits the Russell 2000 Index inclusions as an instrumental variable for the investor horizon and confirms our results are robust to endogeneity concerns. Keywords: Investor horizon, Institutional investors, Earnings management, Real activities manipulation, Accruals JEL Codes: M41, G23, G30, G31 *We thank Wonik Choi, Mitchell Johnston, Michael Woeppel, and seminar participants at Purdue University and the 2017 Midwest Financial Association conference for helpful comments.

2 There is a constant tension between the short-term and long-term objectives of the firm. - From the survey of CFOs by Graham, Harvey, and Rajgopal (2005) 1. Introduction Firms make decisions on how to report earnings in financial statements, and sometimes they trade short-term profits for long-term value or vice versa. Managers often cite that the main goal of earnings management is to meet the earnings expectations of existing and potential investors in financial markets. Institutional investors differ on trading horizons for various reasons (e.g., types of funds or tax purposes), and their investments may have different windows for achieving their target returns. Consequently, managers might consider whether to meet long-term or short-term expectations of their existing investors, and this consideration affects earnings management choices. In this paper, we study the effect of institutional investors trading horizons on firms earnings management strategies; in particular, we study the trade-off decisions between real activities manipulation and accrual-based earnings management. Although the common goal of earnings management is to avoid accounting losses and increase reported earnings, firms manipulate earnings in various ways. A large volume of accounting studies focuses primarily on accrual-based earnings management (AEM) (Healy and Wahlen 1999; Dechow and Skinner 2000). However, in practice, as long as these actions do not violate regulations and actual sacrifices are not too substantial, many firms engage in real activities manipulation (RAM) by adjusting actual operating decisions, (see the survey of 401 CFOs by Graham, Harvey, and Rajgopal (2005)). For example, to lower reported expenses in current financial statements, firms postpone valuable investment projects, such as R&D, that will only return profits in the long term. Furthermore, firms may provide customers with aggressive price discounts or lenient credit terms to temporarily boost reported sales. 1 Recent studies find that AEM and RAM act as substitutes for the purposes of earnings management, and those also show that relevant costs and benefits of each alternative, such as the probability of being scrutinized by auditors and regulators, can explain firms choices of earnings management strategy (Cohen and Zarowin 2010, Kothari, Mizik, and Roychowdhury 2012, Zang 2012). The following distinct differences between the two earnings management strategies provide predictions on the shareholders trading horizon effect on how firms manage earnings. First, managing earnings via accounting procedures should be executed in the short period between the fiscal year-end and the financial report filing date. In contrast, managing earnings via adjusting actual operations can be planned in advance and executed over the full fiscal year. Second, while managers face uncertainty on whether 1 Prior studies have documented the evidence that firms engage in real activities manipulation to avoid accounting losses (Bartov 1993; Baber, Fairfield, and Haggard 1991; Roychowdhury 2006). 1

3 AEM will be allowed by auditors, changing operational decisions is largely under manager s control. Lastly, both types of earnings management methods are costly, leading to negative future consequences, but they bear different types of risks. AEM is subject to future scrutiny by regulators and auditors. 2 In addition, AEM has a reverting nature so that managing earnings through accruals is more difficult to maintain in consecutive years. On the other hand, RAM involves changing the firm s actual operations in the current period, potentially destroying future cash flows. However, RAM can be used more flexibly to smooth earnings. Since long-term investors expect to hold the company for long horizons, they have greater influence on managers decisions than do short-term investors. Furthermore, if long-term investors do not quickly exit their positions, they might want to resolve any uncertainty on earnings performance in advance and smooth earnings over long periods of time. Therefore, if RAM takes longer planning and execution over the fiscal year than AEM, we expect that the presence of long-term investors makes firms more likely to engage in RAM than in AEM. On the other hand, if making a suboptimal operational decision harms future cash flows more than manipulating accounting numbers, i.e. if RAM is more costly than AEM in terms of long-term firm value, firms with long-term investors would prefer AEM to RAM. To what extent shareholders trading horizon would affect firms tradeoff decision between AEM and RAM is thus an empirical question. We use a sample of firm-year panel data of 6,988 U.S. firms that were traded on the NYSE, NASDAQ, or AMEX from 1988 to 2014 and find evidence that the investor-trading horizon affects firms earnings management strategies between AEM and RAM. Using the quarterly holdings information from 13F filings from 1987 to 2014, we follow previous studies (e.g. Gasper, Massa, and Matos 2005) to estimate investor horizon. We calculate the value-weighted average of the turnover ratios of the institutional investors that held a specific firm over four quarters. Next, we follow conventional methods of estimating discretionary accruals to measure AEM (Jones 1991), and we employ three RAM measures following Roychowdhury (2006): abnormal cash flows, abnormal discretionary expenses, and abnormal production costs. The goal of this paper is to examine the effect of investor-trading horizon on managers relative use of AEM and RAM, holding the incentive of managing earnings constant. To do so, we focus on a sample of firms that are likely to manage reported earnings through high levels of earnings management regardless of the use of two methods. Specifically, we estimate a two-stage model, as in Cohen and Zarowin (2010), to control for the firms incentive to manage earnings in the first step. Previous studies argue that 2 Firms tend to prefer RAM over AEM, especially when being audited by high quality auditors and facing tougher scrutiny in accounting practices after the passage of the Sarbanes-Oxley Act (SOX) (Cohen, Dey, and Lys 2008; Zang 2012). 2

4 long-term investors, who are usually categorized as sophisticated investors, have strong incentives to monitor firms, which leads to lower earnings management (Bushee 1998). Thus, it is important to consider the first-order effect of investor horizons on the aggregate level of earnings management. Conditional on engaging in high overall earnings management, in the second stage, we estimate whether investor-trading horizon matters in determining firms preference of adjusting operational decisions over manipulating accounting figures. Our main finding is that firms with longer investor horizons prefer adjusting real operations to manipulating short-term accruals when firms have strong incentives to manage earnings. Firms primarily held by long-term investors have lower abnormal accruals compared to firms primarily held by short-term investors. In contrast, firms with long-term investors are more likely to engage in RAM than those with short-term investors. Specifically, the investor horizon is positively associated with lower abnormal discretionary expenses and higher abnormal production costs. One of the empirical challenges in our analysis is that the investor-trading horizon and firms earnings management strategies can be endogenous. It is possible that institutional investors with long trading horizons prefer to buy and hold firms with specific unobservable characteristics (e.g. good governance), which can also affect the firms earnings management strategies simultaneously. To establish the causal effect of the investor-trading horizon on earnings management strategies, we exploit the Russell 2000 Index inclusions as an instrumental variable to measure the plausibly exogenous changes in investors trading horizons. Since the reconstitution of Russell indices is based on the order of stocks market capitalizations, which cannot be solely determined by an individual firm s policies, several papers use a similar identification strategy to measure exogenous changes in institutional ownership. 3 Since stocks included in the Russell 2000 Index attract funds that track indices, which usually trade in short runs, we find that inclusion in the index negatively predicts investor horizon, especially in the year when they are added to the index, which is a finding consistent with Cremers, Pareek, and Sautner (2017). Our main specification instruments investor horizon and estimates a two-stage least squares regressions (2SLS), and our main finding that firms held by long-term investors prefer RAM over AEM holds after addressing this endogeneity problem. We further develop our analysis to explore the two underlying mechanisms through which longterm investors pressure firms to manage earnings through RAM. First, we expect that long-term investors that experience performance pressures and face high market uncertainty would have strong incentives to affect firms earnings management strategies so that they can benefit from increased earnings. We find evidence that supports the incentive channel; firms held by long-term investors are more likely to boost 3 For example, Aghion, Van Reenen, and Zingales (2013) and Cremers, Pareek, and Sautner (2017) use the inclusion of a firm into the S&P 500 or Russell 2000 index as an instrumental variable for institutional ownership holdings. 3

5 earnings through RAM, especially when long-term investors experienced low fund flows in the previous year. Our analysis further documents that long-term investors are more likely to pressure managers into RAM when market uncertainty, as measured by the Volatility Index (VIX), is high. Second, we look at the ability of long-term investors to impact firms earnings management strategies. In this influence channel, we expect the effect of the investor-trading horizon on earnings management strategies to be more pronounced if the long-term investors have the ability to put more pressure on firms managers. We observe that long-term investors are more able to influence smaller firms RAM decisions. Often, larger firms are considered to have more numerous investor interests and greater analyst and media coverage (Bhushan 1989; O Brien and Bhushan 1990; Shores 1990; Piotroski and Roulstone 2004). Thus, it is more costly for long-term investors to influence the earnings management strategies of larger firms that receive intense market attention. In addition, we find that the likelihood of the use of RAM over AEM is higher when long-term investors hold higher ownership in the firms. Taken together, our results consistently indicate that the effect of the investor-trading horizon on earnings management strategies largely depends on long-term investors incentives to boost earnings and the ownership power they can exert. Given that making a suboptimal decision in the current period can deteriorate long-term future value, our finding that firms with long-term investors tend to manage earnings by adjusting operational decisions rather than by manipulating accruals naturally invites the question on the economic consequences of the earnings management choice in terms of long-term value. Next, we investigate the implications of investor horizons and the choice of earnings management on future performance by comparing the changes in future operating performance and long-run stock returns. If earnings management strategies are selected in order to meet investors earnings targets, then we expect that managing earnings through RAM should be associated with less severe losses in long-term value for the firms held by long-term investors than those by short-term investors. We find that when firms manage earnings through operational manipulation, their long-term investors do not benefit in the short run, as measured by five-day abnormal returns around earnings announcement. On the other hand, firms that engage in high RAM experience higher industry-adjusted ROA in the next fiscal year when they are mostly held by long-term investors than by short-term investors. However, earnings management through AEM is associated with more severe decrease in future ROA and long-run stock returns as investor horizon increases. Further, we find evidence that firms that engage in extensive RAM consistently use relatively high RAM in the following consecutive years, but AEM is used in a specific year only. We also document that firms that manage earnings through operational changes tend to have smoother earnings than those through accrual management. 4

6 This result implies that firms held by long-term investors may engage in RAM only if managers are confident in high future growth of the firm or if changing operational decisions in an effort to meet long-term investors expectation is relatively less costly. Less volatile future earnings of firms undertaking high RAM suggest that long-term investors might be willing to trade long-term value for smoother earnings. Our study contributes to the literature on the effect of institutional investors, which has primarily focused on their governance role. In particular, among institutional investors, those with longer trading horizons have stronger incentives to monitor the firms they hold and improve their long-term value. Consistently, previous studies show that investors trading horizons affect corporate policies; the existence of long-term investors is associated with higher investment and lower payouts (Derrien, Kecskes, and Thesmar 2013), better governance (Harford, Kecskes, and Mansi 2017), better takeover decisions (Gasper, Massa, and Matos 2005; Chen, Harford, and Li 2007), and a lower frequency of corporate misconduct (Harford, Kecskes, and Mansi 2017). Rather than examining the monitoring effect of long-term investors on the level of earnings management, our study focuses on another incentive of long-term investors in terms of firms tradeoff decisions on earnings management strategies. Our finding suggests that holding the incentive of managing earnings constant, firms mostly held by long-horizon investors prefer to adjust operational decisions as long as changes in operations do not harm future performance much severely in the long run. Our paper is also related to the stream of literature that shows how different institutional characteristics are associated with corporate accounting policies. Using the classification of institutional investors from Bushee and Noe (2000) and Bushee (2001), previous studies show that firms owned by dedicated investors and quasi-indexers, who typically have long trading horizons, report lower levels of discretionary accruals compared to firms owned by transient investors (Koh 2007; Lin 2012). Our contribution is threefold. First, instead of using a simple classification of institutional investors, we employ the more nuanced measure of institutional investors trading horizons, which is also timevarying. In addition, by incorporating investors detailed portfolio information from 13F filings, we are able to exploit the characteristics of institutional investors, such as previous fund flow. Second, previous studies on the effect of the investor-trading horizon on earnings management focus on AEM. Our focus is on the choice of earnings management strategy after controlling for a firm s overall level of earnings management. We provide another insight on long-term investors role in earnings management: long-term investors, who not only have strong incentives to monitor but also need to bear the uncertainty on the possibility of earnings management through accruals, pressure managers to conduct real activities manipulation to meet long-term earnings expectations and to smooth earnings. Lastly, our identification strategy for investors trading horizons confirms the causal effect of the existence of long-term investors on different earnings management measures. 5

7 The remainder of this paper is organized as follows. Section 2 describes the sample and main variables, and Section 3 explains our empirical methodology. In Section 4, we test the effect of investor horizons on firms earnings management strategies. Section 5 explores the underlying mechanism of the long-term investor effect using cross-sectional and time-series variations in firms and institutional investors. Section 6 examines the implication on future performance, and Section 7 offers some concluding remarks. 2. Data and Variables 2.1. Data We begin with all publicly traded firms in Compustat and the Center for Research in Security Prices (CRSP) from 1988 to We limit our sample to firms incorporated and headquartered in the United States and to common stocks traded on the NYSE, AMEX, or NASDAQ stock markets. We exclude firms in the financial (SIC ) and utility industries (SIC ). We further require our sample firms to have positive assets and positive sales. The accounting data are from Compustat, monthly stock price data are from CRSP, and analyst forecast data are from I/B/E/S. The data on institutional investor holdings are from the Thomson Reuters Institutional Holdings 13F database. All institutional investors with more than $100 million of securities under management are required to report their quarterly holdings to the Securities and Exchange Commission (SEC) on form 13F. We merge institutional investor holdings data with the sample of firms described above. Finally, we require firms to have non-missing earnings management measures, an investor horizon measure, and firm-level control variables. Our final sample includes 59,033 firm-year observations of 6,988 firms Variables Investor Horizon To determine how frequently investors buy or sell their investments, we rely on the quarterly portfolio holding data from 13F filings. Following Gaspar, Massa, and Matos (2005), we calculate churn rates (CR) for each individual investor i in each quarter q as follows: CR i,, t q j J N P N P N P j, i, q j, q j, i, q 1 j, q 1 j, i, q 1 j, q j J N P N P j, i, q j, q j, i, q 1 j, q 1 2, (1) 4 Our sample period begins in 1988 because the variables used to calculate earnings management measures became available after 1987 when the statement of cash flow was introduced. 6

8 in which N j,i,q represents the number of shares of company j held by investor i at quarter q, and P j,q represents the price of shares of company j at quarter q. 5 Long-term investors buy and sell their stocks less frequently, so their churn rates are expected to be lower than those of short-term investors. 6 Next, we aggregate institutional investors churn rates to construct our annual investor horizon variable at the firm level. In Figure 1, we illustrate how we estimate our firm-level annual investor horizon measure using the quarterly institutional holding data. Investor turnover for company j in quarter q is the weighted moving average of investors churn rates over four quarters as of one quarter before fiscal yearend. 4 INVHOR w 1 j, q i I j, i, q i, q r 1 4 CR r 1 in which the weight assigned to investor i by firm j at quarter, w j,i,q, is measured by the percentage of shares held by investor i among all institutional holdings in firm j at quarter q. Then, we construct our main investor horizon measure, INVHOR, by multiplying the investor turnover measure by -1 to facilitate interpretation. 7 Thus, a higher value of INVHOR indicates that the average holding period of the firm s institutional investors is longer Accrual-based Earnings Management and Real Activities Manipulation To measure firms AEM, following prior literature, we use discretionary accruals, the difference between firms actual accruals and the predicted levels of accruals. Jones (1991) model is used to estimate the normal levels of accruals for each two-digit SIC-year: (2) Accruals 1 t Salest Sales t 1 PPE t , Assetst 1 Assetst 1 Assetst 1 Assetst 1 (3) in which Accruals t is the income before extraordinary items less operating activities net cash flow, Sales t is net sales, and PPE t is property, plant, and equipment. The residuals from the regression model above are our proxy for accrual-based earnings management (AEM). 5 Following Gaspar, Massa, and Matos (2005), we exclude investors or firms who enter the Thomson Reuters Institutional Holdings 13F database for the first time, because, by definition, churn rates are 2 for both cases. 6 In this study, we assume that institutional investors hold stocks for at least three months if they appear in the 13F at least once, as 13F filings only provide snapshots of institutional holdings at the end of each quarter. On average, investors churn rates are about 0.23, with a median value of INVHOR is set as missing for firms that do not appear in the Thomson Reuters Institutional Holdings 13F database for the quarter (assuming those stocks are not held by institutional investors). 7

9 Using the models in Roychowdhury (2006), we employ three RAM measures: abnormal levels of cash flows from operations, discretionary expenses, and production costs. First, firms may provide sales discounts or generous credit terms to clients to temporarily boost their earnings, which leads to lower cash flows from operations compared to overall sales generated. Second, firms often reduce discretionary expenses, such as advertising, R&D, and selling, general, and administrative (SG&A) expenses to temporarily increase their earnings. Third, to manage earnings upward, managers can lower firms fixed costs by overproducing. This will increase production costs, as defined by the sum of the cost of goods sold (COGS) and changes in inventories. For each RAM measure, for each two-digit SIC-year, we run the following cross-sectional regressions to estimate the normal levels of cash flows from operations, discretionary expenses, and production costs: CFO 1 t Sales t Salest Sales t , Assetst 1 Assetst 1 Assetst 1 Assetst 1 DISEXP 1 t Sales t , Assetst 1 Assetst 1 Assetst 1 PROD 1 t Sales t Salest Sales t 1 Salest 1 Sales t , Assetst 1 Assetst 1 Assetst 1 Assetst 1 Assetst 1 (4) (5) (6) in which CFO t is the net cash flow from operating activities, DISEXP t is the discretionary expenses (defined as the sum of advertising, R&D, and SG&A expenses), and PROD t is the production cost calculated by adding COGS and changes in inventories. The residuals from the regression models above are used to construct our proxies for abnormal cash flows from operations, discretionary expenses, and production costs, respectively. We multiply the residuals in model (4) and (5) by -1 to construct variables AbCFO and AbEXP such that higher values of AbCFO and AbEXP indicate lower abnormal cash flows from operations and lower abnormal discretionary expenses, respectively, which are indicators of high RAM. We use the residuals from the regression model (6) to proxy real activities manipulation through overproduction (AbPROD). To measure the overall real activities manipulation activities, following Bartov and Cohen (2007), we aggregate three manipulation activities into one real activities manipulation proxy, RAM. Thus, our combined proxy, RAM, increases as firms engage in more aggressive and greater RAM. 8

10 Control Variables We control for several firm characteristics that are known as the determinants of earnings management. Our control variables are firm size (SIZE), market-to-book ratio (MB), financial leverage (LEV), sales growth (SALEG), return on assets (ROA), and institutional ownership (INST%). Since our goal is to examine whether the investor-trading horizon affects firms earnings management strategies, controlling for total institutional ownership is important in order to isolate the investor horizon effect. Auditor characteristics are also included as control variables, as auditors play a significant role in monitoring firms earnings management activities. We include the number of years the auditor has audited the firm (TENURE) and an indicator variable that measures whether the firm is audited by a Big 8 auditor (BIG8). In our Heckman two-stage regression models, following prior studies (Cohen and Zarowin 2010; Zang 2012), we include additional control variables: the log number of shares outstanding (LOGSHR), a log of one plus the number of analysts following the firm (ANCOV), net operating assets (NOA), and a dummy variable for the firm being in a high-litigation industry (LIT). A detailed definition of all these variables is provided in Appendix Descriptive Statistics Figure 2 depicts the time trend of institutional holdings and their investor horizons over our sample period. Consistent with prior studies (e.g. Chen, Harford, and Li 2007), we find that there is a strong upward time trend in institutional holdings. The median institutional holding is around 10% in 1988 and increased to around 63% in On the other hand, in general, investor horizons do not show a strong time trend (the average of INVHOR is about -0.30). The constant level of the institutional horizon over the sample period demonstrates that the trading horizon of institutional investors is one of the specific institutional investor characteristics. Table 1 Panel A presents descriptive statistics for all variables included in the regression models. On average, our sample firms institutional ownership percentage is around 35%. Our main variable of interest, INVHOR, has a mean of and median of There is a large cross-sectional variation in investor horizons across firms, for the standard deviation is For Panel B and C, we focus on firms who engage in high earnings management, in which either RAM or AEM is above the industry-year 3 rd quartile. In Panel B, to see how firms with high investor turnover differ from firms with low investor turnover, we further divide the sample into two groups: firms primarily owned by long-term investors and those by short-term investors. The mean and median values of AEM of firms with long-term investors are significantly lower than those of firms with short-term investors, supporting their governance role. However, 8 Statistics of INST% are consistent with prior studies (e.g. Cremers, Pareek, and Sautner 2017). 9

11 with regards to RAM, the mean and median values of RAM (except for AbCFO) of firms with long-term investors are significantly higher than those of firms with short-term investors. 9 On the other hand, firms held by short-term investors have higher institutional ownership than firms held by long-term investors. In Table 1 Panel C, the correlations between INVHOR and RAM proxies are significantly positive (except for AbCFO), while that between INVHOR and AEM is significantly negative. This suggests that firms with more institutional investors with longer horizons restrict accrual-based earnings management, but actively engage in real activities manipulation. It is also important to note that the correlation between institutional ownership and the investor horizon measure is relatively low, implying that the average trading horizon of institutional investors is a distinct characteristic beyond the ownership percentage of institutional investors. 3. Investor Horizons and Earnings Management Strategies 3.1. Heckman Selection Model We begin our analysis by examining the relation between investor horizons and earnings management strategies. To examine the effect of investor horizons on the choice between RAM and AEM, we focus on the firm-year observations when firms engage in extensive earnings management. Because the decision to manage reported earnings is endogenous, however, using a sample of observations with high earnings management measures creates a potential problem of sample selection. To address this selection bias issue, we estimate a Heckman two-stage selection model (1979) as our main specification. Our approach of using a two-stage regression is similar to that of Cohen and Zarowin (2010). Although used for a sample of firms with seasoned equity offerings, they argue that their model can be used more generally to study a firm s choice of earnings management strategy. In the first stage, we estimate a probit model for whether a firm engages in high earnings management in a specific year either through RAM or AEM as follows: TOTEM 0 1INVHOR 2 INST % Controls IndFE YearFE, in which the dependent variable, TOTEM, takes a value of one if a firm s AEM or RAM is in the top 25% of the industry-year earnings management measures. We include investor-trading horizon and institutional ownership in addition to basic firm characteristics that are known as the determinants of earnings (7) 9 The prediction on abnormal operating cash flows can be ambiguous as the discretionary expenses and production costs, the other measures of RAM, can affect CFO. If discretionary expenses are paid in cash, then decrease in discretionary expenses would increase CFO. On the other hand, when a firm overproduces to lower COGS but has the similar level of sales, then the firm will have lower CFO. In this sense, the prediction on CFO is not clear if a firm engages in multiple RAM methods simultaneously. 10

12 management activities, such as firm size (SIZE), market-to-book ratio (MB), leverage (LEV), sales growth (SALEG), auditor size (BIG8), auditor tenure (TENURE), and return-on-assets (ROA). In addition, we add analyst coverage (ANCOV) and log the number of shares (LOGSHR) as proxies for capital market attention. The regression includes industry and year fixed effects. Investor horizon and institutional ownership are lagged by one quarter before the fiscal month, and the rest of firm-level control variables are lagged by one year. This selection model helps us address the concern that the investor-trading horizon might capture the monitoring incentive of institutional investors, reducing overall earnings management regardless of the earnings management strategy. In the second stage, conditional on a firm s decision to manage earnings, we estimate a firm s choice of real activities manipulation over accrual-based earnings management as a function of investor horizon as follows: EarningsManagement INVHOR INST % Controls ˆ IndFE YearFE, (8) in which the independent variable is INVHOR and ˆ is the inverse mills ratio estimated from the firststage probit model. We estimate the second stage regression as an OLS model for RAM as a dependent variable and as a probit model for HIGHRAM, an indicator variable of the firm-year when real activities manipulation is in the top 25% of the industry-year distribution. In the second stage, which estimates the choice between RAM over AEM, we additionally include net operating assets (NOA) and a dummy variable for the firm being in a high-litigation industry (LIT), which capture the cost of AEM. Industry and year fixed effects are included in all regressions, and standard errors are clustered at the firm level. The results of the selection model estimation are reported in Table 2. In the first stage regression for the determinants of total earnings management in panel A, we find that the coefficient of INVHOR is significantly negative. The coefficient of institutional investor holdings, one of the main control variables included, is negative but insignificant. Previous studies show that high institutional investor holdings can proxy for high monitoring effects, leading to lower earnings management (Bushee 1998; Chung, Firth, and Kim 2002; Roychowdhury 2006). Thus, INVHOR can be interpreted as the incremental effect of investor horizons, after controlling for the level of overall institutional ownership. The negative coefficient of INVHOR implies that firms primarily held by long-term investors engage in lower earnings management activities, which is consistent with previous studies on the monitoring effect of long-term investors (Bushee 1998; Chen, Harford, and Li 2007). We include auditor size and tenure information, for prior literature argues that big auditors constrain earnings management because they play a stronger monitoring role 11

13 compared to small auditors (Becker, DeFond, Jiambalvo, and Subramanyam (1998). 10 The log number of shares and analyst coverage are negatively related to total earnings management, suggesting that capital market incentives play an important role in lowering earnings management. Our main interest is in the second stage regression, in which we estimate on the trade-off decision between two earnings management strategies. Out of a full sample of 59,033 observations from 1988 to 2014, 23,623 firm-year observations are included in the high earnings management sample in the second stage. In Panel B, the estimates in columns (1) and (2) show that INVHOR is negatively related to AEM, but positively to RAM; the coefficient on INVHOR is in the regression for the discretionary accruals in column (1) and for the aggregate real activity manipulation in column (2). Both coefficients are statistically significant at the 1% level. When looking at the effect of investor horizon on three separate components of RAM measures in columns (3) to (5), we find that INVHOR significantly increases the real activities manipulation measures except abnormal operating cash flows. These results imply that firms owned by long-term investors engage in lower discretionary accruals but higher real activity manipulation than those by short-term investors. In economic terms, a one standard deviation increase in investor horizons, which is equivalent to an increase in average investor holding period from 20 months to 33 months, is associated with a decrease in abnormal accruals of (0.235*0.119). Given that the median of AEM is 0.075, the impact of investor horizon on abnormal accruals is economically sizeable. On the other hand, a one standard deviation increase in INVHOR leads to a reduction in discretionary expenses of (0.519*0.119) and an increase in abnormal production costs of (0.277 * 0.119). After we control for the ownership effect in the selection regression, we find that the percentage of institutional ownership does not have a significant impact on the choice between RAM and AEM. 11 The coefficients on firm size (SIZE), growth opportunities (MB), leverage (LEV), performance (ROA), and auditor tenure (TENURE) are statistically significant. The coefficients of the factors that capture the relative costs and benefits of earnings management method have the opposite signs for AEM and RAM, which is consistent with the substitution effects between two earnings management methods documented in Roychowdhury (2006) and Zang (2012). 10 The relationship between auditor tenure and audit quality is controversial though. While the General Accounting Office (GAO) argues that longer auditor client relationships impair auditors independence and thus lower audit quality, some researchers conclude that auditor tenure can also be a measure of auditor scrutiny, for an auditor with a longer relationship with a client firm is often considered to be able to provide a higher quality audit (Stice 1991; Myers, Myers, and Omer 2003). Davis, Soo, and Trompeter (2009) argue that there is a non-linear relationship between auditor tenure and audit quality by showing that there is an increase in discretionary accruals to meet or beat analysts forecasts in both the early and later years of the auditor client relationship. 11 When we estimate the effect of institutional ownership on the magnitude of AEM or RAM using the full sample of firm-year observations, we find that INST% negatively predicts both AEM and RAM, which is consistent with the view that high institutional ownership plays a strong governance role. The results are not reported to save space. 12

14 Next, we also focus on the cases in which firms would not have met earnings benchmarks without managing earnings through AEM or RAM, i.e. firms that are highly suspected to intentionally manage earnings to avoid missing earnings benchmarks. We perform a similar analysis using a subsample of suspect firm-years with a stricter restriction. A suspect firm-year is defined as a firm-year in which a firm just beats/meets zero earnings, last year s earnings, or the analyst consensus forecast. 12 This approach to identify firms that are suspected to manipulate earnings to meet their benchmark is based on Roychowdhury (2006) and Cohen, Dey, and Lys (2008). 13 The main assumption of this definition for suspect firms is the following; if a firm manages earnings to avoid a loss, its reported income we observe ex post would only slightly beat the benchmarks. Using a sample of 8,932 suspect firm-year observations, we estimate Heckman two-step regressions, in which the first stage estimates the probit model predicting a selection of meeting or beating previous year s earnings or analyst earnings forecasts. The estimates from the regressions are presented in Appendix 2. The coefficient of INVHOR is insignificant for AEM, but firms with longer investor horizon engage in more RAM Endogeneity in Investor-Trading Horizon and Identification Strategy The Heckman two-step regression results above suggest that institutional investors long trading horizons are associated with low AEM and high RAM. However, the correlation is not sufficient enough to prove the causal impact of the existence of long-term investors on firm s earnings management strategies, for investors trading horizons could be endogenous to earnings management measures. In other words, there might be an unobservable factor that is simultaneously correlated with earnings management measures and investor horizons. For example, long-term investors may tend to hold firms with good governance, and at the same time, it is possible that better-governed firms prefer to manage earnings through real activities rather than through accruals. In this case, it is difficult to say whether the existence of long-term investors is the main driver behind firms decisions to manage earnings through adjusting real operations or through accruals. To confirm the causal effect of the existence of long-term investors, we employ an instrumental variables regression model. More specifically, we exploit the event of being included in the Russell 2000 Index as an instrumental variable for the investor-trading horizon measures. Inclusion in the Russell A suspect firm-year is defined as a year when a firm just beats or meets (a) zero earnings (reported income scaled by total assets in the current year is between 0 and 0.01), (b) previous earnings (change in reported income scaled by total assets from the last year to the current year is between 0 and 0.01), and (c) the analysis consensus forecast error is between 0 and Our results in Table 2 Panel B are quantitatively similar if we use alternative definitions of earnings management suspect firms. 13

15 Index is known for being relatively exogenous to the unobservable factors that impact earnings management measures, for its inclusion criteria is based on the size of market capitalization in the stock market as of an annual reconstitution date. Any specific firm cannot easily control its market capitalization, but on the other hand, the inclusion event significantly affects the types of investors who hold shares in the firm. Previous studies show that stocks attract funds that track indices after being included in the index, and these index funds usually trade on a short horizon (Lynch and Mendenhall 1996; Petajisto 2011; Cremers, Petajisto, and Zitzewitz 2013; Cremers, Pareek, and Sautner 2017). Our identification strategy is similar to the one used by Appel, Gormley, and Keim (2016), which uses membership in the Russell Index for institutional ownership. It is also similar to that of Cremers, Pareek, and Sautner (2017), which uses Russell index inclusion events for institutional investor s holding periods. In Figure 3 Panel A, using a sample of 1,756 firms that were newly added to the Russell 2000 Index from below, we document that inclusion events are accompanied by a sharp decrease in investor-trading horizons around the event year. 14 However, the exogenous shock in INVHOR is relatively temporary in that the decrease in INVHOR starts two years prior to the event year and then reverses to the original point two years after the inclusion event. On the other hand, the figures in Panels B and C illustrate that institutional ownership and analyst coverage also increase after inclusion events, but those changes are more likely to be permanent. Consistent with the exogenous increase in the holdings of long-term investors around inclusion events, in Panel D, we find preliminary evidence that the mean of AEM increases, but the mean of RAM sharply decreases around the event year. To incorporate our identification strategy in a regression setting, we estimate the two-stage least squares (2SLS) to instrument our main variable, INVHOR. First, we regress INVHOR on AddR2000, the indicator variable of the year when the firm is newly added to the Russell 2000 Index from below. INVHOR AddR2000 Controls ˆ IndFE YearFE, (9) We then estimate the second stage regressions of Heckman selection model, in which we regress each earnings management measure on the fitted value of investor horizon We focus on firms that are newly added to the Russell 2000 index due to an increase in their market capitalization rank. Firms that used to be in the Russell 1000 and are newly added to Russell 2000 index are not included, for those events are likely related to information-related events (e.g. default, acquisitions) 15 We estimate Heckman selection model including the instrumental variable, AddR2000, as a control variable in the first stage selection equation predicting a firm-year with high earnings management. In addition, the inverse mills ratio estimated from the Heckman selection model is also included the first stage of 2SLS. In this case we can assume that there are two sources of exogenous variation for INVHOR in equation (9): AddR2000 and the inverse mills ratio (Wooldridge 2010). 14

16 EarningsManagement INVHOR INST % Controls ˆ IndFE YearFE, (10) The regression results are reported in Table 3. In the first stage regression in column (1), the Russell 2000 Index addition event indicator significantly predicts the investor horizon. There are 20,562 Russell 2000 Index addition firm-year observations during our sample period of 1988 to When firms are newly included in the Russell 2000 Index from below, the average INVHOR decreases by The economic magnitude of these changes is equivalent to 32% of the standard deviation of INVHOR. The F- statistic for the weak identification tests of the first stage regression confirms that the instrument is strongly correlated with our main variable, INVHOR. 16 The second stage regression estimates in columns (2) to (5) suggest that the changes in INVHOR significantly affect the magnitude of earnings management after controlling for endogeneity. Consistent with the results in the regressions in Table 2, firms with longer investor horizons engage in less accrualbased earnings management but in more real activities manipulation. The economic significance of the long-term investor effect from the instrumental variable regressions is sizeable. A one standard deviation increase in INVHOR is associated with a increase in AEM and a decrease in RAM, which corresponds to the 46% and 32% changes in standard deviations of accrual-based and real activities manipulation measures, respectively. One thing to note is that the estimated magnitude in the instrumental variable regressions is almost five to ten times larger than those in the baseline regression. The reason is that by using the instrumented INVHOR, the regression estimates reflect the effect of INVHOR on the marginal firms that experience a sharp drop in investor horizons. 4. Economic Mechanisms Our results thus far show that in the presence of long-term investors, conditional on managing earnings, firms are more likely to adjust operating decisions rather than to manipulate accruals. In this section, we examine the potential underlying mechanisms through which long-term investors affect firms earnings management strategies. To do so, we further develop our analysis by exploiting the cross-sectional and time-series variations in firm and investor characteristics and market conditions. 16 One of the concerns of using the inclusion in Russell 2000 index as an instrument variable is that the exogenous change in investor horizon is driven by the increase in index funds, which usually do not have strong incentives to influence the earnings of the firms they hold. To address this issue, in Appendix 4, we test whether the main results in Table 3 would change if we exclude indexers. We find that there is a significant decrease in the trading horizons of non-indexers as well when a stock is included in the Russell 2000 index. The main effect of investor horizon on RAM also holds when we exclude indexers from our consideration. 15

17 4.1. The Incentive Channel We expect that the effect of the trading horizon on earnings management strategies is more prominent when long-term institutional investors benefit from the increased earnings of firms that they hold. Given the potential adverse consequences on future performance of RAM, the possible benefits of meeting earnings benchmarks through managing current earnings may cancel out any negative future consequences of earnings management on cash flows. We call this mechanism the incentive channel. Since long-term investors need to internalize the economic consequences of RAM on future cash flows, they would not pressure firms to adjust operations to boost earnings over the fiscal year unless the benefit of managing earnings outweighs the cost. We focus on the fact that investment companies such as mutual funds and investment advisors manage funds that are sensitive to portfolio performance. A number of studies on mutual funds and hedge funds show that fund flows are significantly sensitive to previous performance (Berk and Green 2004). In addition, the compensation structure of fund managers is largely based on the size and performance of a fund (Edwards and Caglayan 2001). If missing earnings benchmarks would hurt investors performance, thus adversely affecting their fund flows, then long-term investors have strong incentives to encourage firms to manage reported earnings. We expect that long-term investors incentives to impact firms RAM are intensified when they face high performance pressures. To estimate how sensitive long-term investors are to the performance of their portfolios, we consider cross-sectional variation the prior fund flows of their portfolios and the time-series variation of market uncertainty. We estimate performance pressure from institutional investors using two measures: the fund flows of institutional investors and the market volatility index (VIX). To construct institutional investors fund flows, we first measure quarterly net flows at the fund level using 13F filings: 17 QuarterlyFundFlow ( Pj, qn j, i, q Pj, qn j, i, q 1) j J, j J P N j, q 1 j, i, q 1 (11) in which P j,q is the stock price for stock j at quarter q, and N j,i,q is the number of shares of stock j held by investor i at quarter q. 18 The quarterly fund flow measure reflects the price-change-adjusted growth in the 17 Fund flows might be largely affected by economic conditions and time. Including year dummies in our regressions partly captures the time effect. In addition, we also run a robustness check by measuring the fund flows adjusted for the overall market s fund flows and performance. The results are quantitatively similar. 18 It would be ideal to obtain the data on actual fund flows under management from actual transactions at the fund level. However, given the limited availability of data, we rely on the quarterly portfolio holding data from 13F filings to indirectly estimate the fund flows of each individual investor. 16

18 number of shares held by a fund and is scaled by total fund size in the prior quarter. We take the valueweighted moving average of fund flows over the previous year (four quarters) at the firm-year level, and we calculate this value-weighted moving average separately for long-term (FLOW_LT) and for short-term investors (FLOW_ST). Long-term (short-term) investors are defined as institutional investors whose churn rate (CR) is below (above) the quarterly median. We re-estimate the Heckman selection model, equation (8), by including the previous fund flows of long-term and short-term investors instead of investor horizon measure. The estimation results are presented in Table 4. In columns (1) and (2), we look at the effect of the previous fund flows of long-term and short-term institutional investors on AEM and RAM, respectively. Long-term investors fund flows do not affect the use of AEM, but they do negatively affect RAM. In contrast, short-term investors fund flows have the opposite impact; low fund flows of short-term investors pressure firms to engage in high abnormal accruals but low real activities manipulations. The result implies that the long-term investor effect on the preference for RAM becomes stronger when long-term investors face intense performance pressures. In columns (3) and (4), we test whether the effect of investor horizons on the trade-off decision between AEM and RAM varies by the market uncertainty condition. The coefficient on the market uncertainty index (VIX) is statistically significant and positive, which implies that firms tend to adjust their operational decisions when they expect high uncertainty in the market. The interaction term between investor horizon and the market uncertainty index is positive and significant. This result implies that when markets are more uncertain (i.e., when the VIX is higher), the effect of long-term investors on the firms RAM is more pronounced. Given that RAM can be done before the fiscal year end, this result suggests that firms adjust their operational decisions over the fiscal year before making earnings announcements to resolve high uncertainty on earnings performance. Consistent with the incentive hypothesis, we find evidence that firms with long-term investors that are performance-sensitive have strong incentives to engage in RAM. The effect of INVHOR on RAM is stronger when institutional investors suffer from low fund flows and when they are concerned about market uncertainty The Influence Channel As a potential mechanism through which INVHOR affects firms RAM as a main earnings management strategy, we examine the ability of institutional investors to influence firms earnings management strategies. We propose that the effect of long-term investors is stronger when it is easier for them to pressure firms to alter firms operations, which we call the influence channel. We expect that the ability of long-term investors to exercise influence over firms earnings management strategies is stronger 17