Accounting Conservatism and Corporate Investment *

Transcription

1 Accounting Conservatism and Corporate Investment * Paul Brockman College of Business and Economics Lehigh University 407A Rauch Business Center 621 Taylor Street Bethlehem, PA pab309@lehigh.edu Ming Liu Department of Accountancy and Law Hong Kong Baptist University 34 Renfrew Road Kowloon Tong, Kowloon Hong Kong, China Tel: (852) mliu@hkbu.edu.hk Tao Ma Moore School of Business University of South Carolina 1401 Greene Street Columbia, SC Tel: (803) tao.ma@moore.sc.edu * Tao Ma gratefully acknowledges financial support from the Center for Research in Economics and Strategy (CRES), in the Olin Business School, Washington University in St. Louis. We wish to thank Rich Frankel, Xiumin Martin, and Douglas Skinner for their insightful comments. We also thank Gauri Bhat, Jimin Ding, Tat Chan, Radha Gopalan, Walied Keshk, Ohad Kadan, Carl Viereger, and participants at 2010 AAA annual conference and Washington University workshop for their helpful comments and suggestions. Any remaining errors are our own.

2 Abstract: This paper documents a negative relationship between accounting conservatism and corporate investment for firms most likely to suffer from overinvestment as well as for firms most likely suffer from underinvestment. These negative correlations are consistent with the accounting literature s prediction that accounting conservatism can reduce overinvestment but can also lead to dysfunctional investment incentives whereby managers forego positive net present value (NPV) projects (Leuz, 2001; Watts, 2003a; Guay and Verrecchia, 2006). Consistent with expectations, our results also show that the negative relationship between conservatism and firm investments for underinvestment firms is concentrated in firms with high return volatility and firms with short CEO investment horizons. JEL Classification: G31; M41 Keywords: Accounting conservatism, Underinvestment, Investment efficiency.

3 Accounting Conservatism and Corporate Investment Abstract: This paper documents a negative relationship between accounting conservatism and corporate investment for firms most likely to suffer from overinvestment as well as for firms most likely suffer from underinvestment. These negative correlations are consistent with the accounting literature s prediction that accounting conservatism can reduce overinvestment but can also lead to dysfunctional investment incentives whereby managers forego positive net present value (NPV) projects (Leuz, 2001; Watts, 2003a; Guay and Verrecchia, 2006). Consistent with expectations, our results also show that the negative relationship between conservatism and firm investments for underinvestment firms is concentrated in firms with high return volatility and firms with short CEO investment horizons. JEL Classification: G31; M41 Keywords: Accounting conservatism, Underinvestment, Investment efficiency. 1

4 1. Introduction This paper examines the effects of accounting conservatism on managers investment incentives. Extant accounting literature is mixed on the predicted relationship between these two issues. One strand of accounting literature argues that timely loss recognition serves as good corporate governance by preventing managers from investing in projects with negative net present value (NPV) projects. Under this argument, accounting conservatism can reduce managers' incentives to overinvest (Ball, 2001; Ball and Shivakumar, 2005). The other strand of literature argues that accounting conservatism magnifies managers risk aversion and investment horizon problems and induces them to forego long-term, risky investments even if such investments are positive NPV projects. As such, accounting conservatism can distort managers investment incentives, resulting in underinvestment (Leuz, 2001; Watts, 2003a; Guay and Vierrecchia, 2006; Roychowdhury, 2010). However, a third line of literature suggests that accounting conservatism helps alleviate managers underinvestment incentives. This is so because accounting conservatism helps firms to obtain cheaper credit to finance investment projects and thereby alleviates underinvestment problems arising from limited access to capital markets (Ahmed et al., 2002; Watts and Zimmerman, 1986; Watts, 2003a; Zhang, 2008). To date, most empirical evidence suggests that conservatism improves investment efficiency and constrains managers incentives to overinvest (Ahmed and Duellman, 2011; Francis and Martin, 2009; Bushman, Smith and Piotroski, 2011), while little empirical research examines whether accounting conservatism leads to underinvestment. Our study extends this line of literature by examining whether accounting conservatism solves an overinvestment problem by creating an underinvestment problem. 2

5 We examine the association between a firm s capital investment and its accounting conservatism conditional on the firm s likelihood of overinvestment or underinvestment as specified in Biddle et al. (2009) and Cheng et al. (2013). For firms that are more likely to overinvest, a negative association between accounting conservatism and capital investments is consistent with prior studies findings that accounting conservatism constrains overinvestment. We focus on firms that are prone to underinvestment and examine how accounting conservatism affects managers investment incentives for these firms. In contrast to overinvesting firms, a negative association between accounting conservatism and capital investments for underinvesting firms would suggest that conservatism distorts managers investment incentives. Such negative incentives are likely to dominate any positive effects related to cheaper debt financing. On the other hand, a positive association between accounting conservatism and capital investments indicates that accounting conservatism can help mitigate underinvestment as it allows underinvesting firms to obtain debt financing. Our empirical analysis is based on three commonly used measures of accounting conservatism, as well as a composite measure of conservatism based on these three individual measures. Following prior studies (Richardson, 2006; Denis and Sibikov, 2010), firm investments are measured as firm capital expenditures net of depreciation expenses obtained from cash flow statements. We find that accounting conservatism is negatively related with investments for two distinct groups of firms: firms with a propensity to underinvest, and firms with a propensity to overinvest. The negative association between conservatism and investments for firms with a propensity to overinvest is consistent with prior studies showing that accounting conservatism can constrain overinvestment, while the negative association for firms with a propensity to underinvest indicates that accounting conservatism can lead to underinvestment. 3

6 After documenting the negative relationship for likely-to-underinvest firms, we further examine cross sectional variations in the negative relation based on the previous literature s predictions. Consistent with these predictions, we show that for likely-to-underinvest firms, the negative relationship between accounting conservatism and investments is more pronounced for firms with high stock return volatility and for firms with myopic CEOs displaying short investment horizons. These cross sectional findings lend further support to the underinvestment hypothesis. It is possible that accounting policies and corporate investments are determined simultaneously. We take several steps in the empirical design to address such potential endogeneity problems. For example, we measure conservatism at time t whereas our measure for capital investment is at t+1. To further alleviate any remaining concerns, we estimate a system of two equations simultaneously using two step least squares. Our simultaneous equations estimation yields similar results and provide consistent evidence that accounting conservatism both constrains overinvestment and leads to underinvestment. This paper makes significant contributions to the literature examining the economic impact of accounting conservatism. Extant studies argue that accounting conservatism can mitigate agency conflicts between lenders and borrowers (Watts, 2001; LaFond and Watts, 2008; LaFond and Roychowdhury, 2008; Caskey and Huges, 2012), and can reduce the cost of capital (Ahmed et al., 2002; Wittenberg-Moerman, 2008; Zhang, 2008; Garcia Lara et al, 2011; Balakrishnan et al., 2014). 1 However, none of these prior studies investigates whether the reduced costs of capital leads to increased corporate investments, especially for financially constrained firms. Given that the accounting literature also argues that accounting conservatism 1 Please refer to Watts (2003a, 2003b) for a detailed survey of empirical papers examining the demands and supply of accounting conservatism prior to

7 can distort managers investment incentives, it is possible that managers may decide not to take advantage of the reduced cost of capital and forego positive NPV projects. Our study provides evidence that accounting conservatism can cause dysfunctional managerial incentives by encouraging just this sort of underinvestment (i.e., forgoing positive NPV projects). The remainder of the paper is organized as follows. We develop our testable hypotheses in Section 2, and discuss the sample selection criteria and research methodology in Section 3. Sections 4, 5, and 6 present the main empirical results. Robustness tests are presented and discussed in Sections 7 and 8. The paper concludes in Section Literature review and hypotheses development 2.1. Accounting conservatism and corporate investment The separation of ownership and control causes agency problems (Jensen and Meckling, 1976), and one such agency problem concerns firm investment decisions. Shareholders want managers to undertake every positive NPV project to maximize firm value. However, managers whose utility is closely tied to accounting earnings also compare their personal costs and benefits when deciding whether to undertake a project. Managers may decide to forego positive NPV projects or undertake negative NPV projects if their personal benefits of undertaking or foregoing such projects are greater than their personal costs, resulting in both overinvestment problems and underinvestment problems. Prior literature argues that accounting conservatism is an important governance mechanism in deterring managers from undertaking negative NPV projects by accelerating future investment losses into current earnings (see e.g., Ball, 2001; Ball and Shivakumar, 2005; Watts, 2003a; Francis and Martin, 2010; and Ahmed and Duellman, 2011). Managers are less 5

8 likely to undertake negative NPV projects when they have to book subsequent losses on a more timely basis (i.e., under accounting conservatism). Rapid recognition of losses from negative NPV projects will tarnish managerial reputations and reduce income linked compensation. In addition, timely loss recognition can also alert corporate boards on a timely basis to investigate the underlying causes of such losses and threaten managers job security. Hence, ex ante, accounting conservatism will deter managers from undertaking negative NPV projects, and alleviate overinvestment problems. Accordingly, our first hypothesis is stated as follows: H1: Accounting conservatism is negatively related to overinvestment. However, another strand of accounting literature argues that accounting conservatism can cause managers to abandon positive NPV projects. Leuz (2001) and Guay and Vierrecchia (2006) argue that untimely (i.e., deferred) gain recognition can cause dysfunctional incentives for managers to abandon positive NPV projects. In particular, executives responsible for designing long-run investment policy may no longer be in office by the time the gains are realized if gains are not recognized on a timely basis. Even if managers have sufficiently long tenure horizons to benefit from the gains, they may still find that the benefits of undertaking the projects do not offset the efforts of implementing the projects. This is so because managers are risk averse, and they typically require a higher discount rate on future gains in deriving the present value of expected payoffs (Reichelstein, 2000). Therefore, a positive NPV project to shareholders may become a negative NPV project to managers if the gains are not recognized on a timely basis. Watts (2003a) similarly argues that both timely gain and loss recognition can avoid dysfunctional outcomes related to limited tenure horizon. Watts viewpoint is clearly conveyed in the following excerpt (Watts, 2003a, p. 211): Ceteris paribus, managerial performance measures in compensation contracts, such as earnings, are more effective when they are timely and reflect the effects of the managers actions on firm value in the period in which the actions are taken. Timeliness avoids dysfunctional outcomes 6

9 associated with managers limited tenure with the firm, often referred to as the manager s limited horizon. For example, a manager may forego positive net present value projects with near-term negative earnings because future earnings will reflect the benefits of the project after the manager has retired or left the firm. Roychowdhury (2010) contends that since risky projects are more likely to become negative NPV projects, timely loss recognition can also deter risk-averse managers from undertaking risky projects even though the projects have positive NPV to shareholders. Specifically, in pursuing risky projects, managers are exposed to a higher probability of large losses occurring in the future and consequently their reputation and income-linked compensation will be affected adversely. Therefore, they are less likely to undertake these investments ex ante even though those projects may have positive NPV to shareholders. In contrast, there is also a long line of literature which argues that accounting conservatism facilitates debt contracting and lowers firms cost of capital (Watts and Zimmerman, 1986; Watts, 2003a). Consistent with this argument, Brockman et al. (2014) and Kravert (2014) find that conservative financial reporting constrains managers from investing in risky projects and mitigates debtholder-shareholder conflicts over assets substitution. Ahmed et al. (2002) find that firms with more conservative accounting enjoy better credit ratings and lower costs of debt. Similarly, Zhang (2008) finds that conservative accounting firms incur lower interest rates when they borrow from banks. According to this argument, accounting conservatism should be positively associated with capital expenditures for firms that are more likely to suffer underinvestment problems due to financial constraints, resulting in a negative association between corporate underinvestment and accounting conservatism. In view of the two competing arguments (i.e., the underinvestment argument and the lower cost of capital argument) on how accounting conservatism affects the firm s tendency to underinvest, our second hypothesis is stated in its null form as follows: 7

10 H2: Accounting conservatism is unrelated to underinvestment Further discussion One valid concern of the underinvestment argument is that managers should be able to voluntarily disclose good news and potentially offset the negative impact of untimely gain recognition in earnings on firm investments. Kothari et al. (2009) document that voluntary disclosures appear to favor the release of good news in a more timely manner than the release of bad news. The authors suggest that managers use the voluntary disclosure of good news to offset the timely loss recognition in earnings. However, there are two reasons why the voluntary disclosure of good news would not completely solve the underinvestment problem induced by accounting conservatism. First, the voluntary disclosure of good news is not always credible and investors tend to discount the economic value of the supposed good news. More importantly, the underinvestment problem documented in this paper is related to managers earnings-based incentives rather than their stock-based incentives. Although stock prices can fully incorporate future good news, the impact of accounting conservatism on earnings remains unchanged by the voluntary disclosure of good news. The argument that accounting conservatism can cause underinvestment does not necessarily mean that firms are behaving irrationally by setting up conservative accounting policies. For example, it is very costly for financially constrained firms to waste limited resources in negative NPV projects, and the benefits of imposing accounting conservatism might outweigh the potential costs of forgoing certain positive NPV projects. Alternatively, because the demands for conservatism come from multiple sources such as creditors, auditors, managers legal liability concerns, etc., conservative reporting is not completely at the discretion of managers or corporate boards. Since the firm s conservative reporting can also be exogenously 8

11 imposed upon managers by outside forces (such as legal liabilities or regulations), it is possible that conservatism may not be optimally set from the shareholders perspective for all firms (e.g., financially constrained firms). 3. Sample selection and empirical design 3.1. Sample selection The initial sample is drawn from the intersection of COMPUSTAT and CRSP from 1989 to All observations included in the sample are required to have sufficient data to calculate accounting conservatism, firm investments, and other control variables. CEO ownership of stock options and restricted stocks is obtained from EXECUCOMP data file. We delete financial firms (SIC codes between 6000 and 6999) and utility firms (SIC code between 4910 and 4939). The final sample for the empirical analyses consists of a total of 79,523 firm-year observations from 9,805 individual firms. All the variables are winsorized at the 1st and 99th percentile values in order to reduce the influence of extreme observations Measure of accounting conservatism We use three different measures of accounting conservatism, as well as a composite measure based on these three individual accounting conservatism metrics to show the robustness of our results. Following Givoly and Hayn (2000), we use negative non-operating accruals and earnings skewness relative to cash flow as two measures of conservatism. Negative nonoperating accruals, NACCR, is calculated as the average of the prior five years non-operating accruals scaled by total assets (with a minimum of two years). We multiply this measure by negative one so that accounting conservatism increases in the value of NACCR. The skewness of earnings relative to cash flow, SKEW, is calculated as the difference between cash flow skewness 9

12 and earnings skewness. The measure of skewness is based on a maximum of 20 quarters and a minimum of 5 quarters of data. Our third measure is Basu s (1997) conditional conservatism specification as modified in Beatty et al. (2008). Specifically, we first measure credit quality for each COMPUSTAT firm with values ranging from 1 to 24 and then estimate Basu s (1997) model each year for every credit quality class using the prior five years of observations with the following equation: 2 Xi,t = β0 + β1 Di, t + β2 Ri, t + β3 Di, t * Ri, t + ei,t where X is earnings per share before extraordinary items scaled by the stock price at the beginning of the fiscal year; R is the twelve month compounded returns starting from the fourth month after the previous year s fiscal year end, and ending three months after the current year s fiscal year end; and D is a dummy variable equal to 1 if the stock return is negative and 0 otherwise. The coefficient β3 captures the incremental timeliness of loss recognition in earnings, and we use β3 as the degree of accounting conservatism denoted as BASU. All firms within the same class of credit quality have the same measure of conservatism for each year. Our last measure of conservatism, ACRANK, is the sum of annual decile ranks of the three individual conservatism metrics. We rescale this composite measure every year so that its value ranges between zero and one Empirical model Following Biddle et al. (2009) and Cheng et al. (2013), we use an index derived from the value of firms leverage and cash holding as a measure of the likelihood that a firm suffers from underinvestment or overinvestment. Specifically, we use firms cash holding and leverage as 2 Consistent with Barth et al. (2008), we estimate a pooled cross-sectional regression of debt ratings on total assets, return on assets, debt to assets, a dividend indicator, a subordinated debt indicator, and a loss indicator. The estimated regression parameters are used to estimate credit ratings for firms without debt ratings. 10

13 two proxies for firm liquidity as firms that are highly leveraged and carry lower balances of cash are more likely to face liquidity constraints and suffer from underinvestment. In contrast, firms with large cash balances and low leverage face high agency problems and are more likely to overinvest. Our measure UnderFirm is computed as the average of decile ranks of the two liquidity proxies, cash holding multiplied by negative one and firm leverage. We rescale the average rank so that UnderFirm ranges between zero and one, where the value of one (zero) indicates that the firm is most likely to suffer from underinvestment (overinvestment). To test our two hypotheses, we estimate the following equation: Investmenti,t+1 = β0 + β1 Consi,t + β2 Consi,t * UnderFirmi, t+1 + β3 GOVi,t + β4 GOV * UnderFirmi, t+1 + Σβj Controlsj,i,t + Yeart + indk +ei,t+1 (1) where Investment is the capital expenditure net of depreciation expenses scaled by lagged assets (Richardson 2006; Denis and Sibilkov 2010). The coefficient β1 captures the effect of accounting conservatism on firm investments for firms that are likely to suffer from overinvestment; β2 is the incremental relation between accounting conservatism and firm investments as underinvestment becomes more likely; and (β1 + β2) therefore caputures the effect of accounting conservatism on firm investments for firms that are most likely to suffer from underinvestment. Under our first hypothesis (H1), we expect β1 to be negative and significantly different from zero, suggesting that accounting conservatism can reduce firm investments for likely-tooverinvest firms. Under the prediction of the underinvestment argument, (β1 + β2) will be negative and significant, suggesting that accounting conservatism can reduce firm investments for firms that suffer from underinvestment; On the other hand, if (β1 + β2) is positive and significant, then the lowered cost of capital argument, which advocates that accounting conservatism alleviates underinvestment, would be supported. 11

14 Following Biddle et al. (2009) and Cheng et al. (2013), we use analysts following (Analysts), institutional investors (Institutions), and the inverse G-index (InvG-Score) as three proxies for corporate governance (GOV). Analysts is the number of analysts following the firm obtained from IBES, and zero if there is no analyst following. Institutions is the the ownership of the firm by institutional investors obtained from Thomson Financial. InvG-Score is the inverse of the anti-takeover protection index compiled by Gompers et al. (2003). The higher the value of the inverse index, the stronger is the firm s corporate governance. Because G-scores are missing for more than 50% of the sample, we also use a dummy variable, G-ScoreDum, which equals one when the G-score is missing. We interact our accounting conservatism measures with each of the three corporate governance proxies in order to exclude the possiblity that corporate governance mechanisms are simultaneously driving both accounting conservatism policies and (under)overinvestment problems. In addition, we control for other detreminants of firm investments, including UnderFirm, firm size (LogAsset), growth options (TobinQ), the standard deviation of cash flow (σ(cfo)), sales (σ(sales)), and investment (σ(i)), firm bankruptcy risk (Z-Score), tangibility (Tangibility), industry leverage (IndK-Structure), operating cash flow (CFOSale), dividends (Dividend), firm age (Age), and operating cycle (OperatingCycle). Furthermore, we include year and two-digit SIC industry dummy variables to control for time series and industry effects. 3 We estimate Equation (1) using ordinary least square (OLS) and adjust the standard errors for heterskedasticity and serial correlation at the firm level. 4. Empirical results 3 The Appendix provides more detailed descriptions of individual variable definitions. 12

15 4.1. Descriptive statistics Table 1 reports summary statistics for the dependent and independent variables used in the empirical analyses. The mean ratio of net capital expenditures to firm assets (Investment) is 1.9%, which is close to that reported in Denis and Sivilkov (2010). The mean value of NACCR, SKEW, and BASU, the three measures of conservatism, are 0.020, and 0.222, respectively. These values are consistent with prior studies (Givoly and Hayn, 2000; Beatty et al., 2008). Table 2 reports the Pearson correlation matrix between selected variables. As shown in the table, firm investments are negatively and significantly correlated with all three measures of accounting conservatism. As discussed earlier, this negative correlation is consistent with both accounting conservatism deterring overinvestment and causing underinvestment. Furthermore, the three measures of accounting conservatism are positive and significantly correlated with each other. The maximum (minimum) correlation is (0.008) between negative non-operating accruals and skewness (skewness and Basu coefficient). This suggests that each of the three measures of accounting conservatism captures different aspects of conservative accounting policies Accounting conservatism and investment Table 3 provides results from multivariate analyses testing the relationship between accounting conservatism and firm investments. Columns (i) through (iv) provide results when accounting conservatism is measured as negative non-operating accruals (NACCR), earningscash-flow skewness (SKEW), Basu s asymmetric timeliness (BASU), and the composite measure (ACRANK), respectively. The estimated coefficients on accounting conservatism are all negative and significant except for BASU. The negative and significant coefficients are consistent with the argument that accounting conservatism constrains overinvestment in firms that are more likely to 13

16 overinvest. In addition, the sum of the coefficients on accounting conservatism and the interaction term are all negative and significant at the 1% level except for column (i) where accounting conservatism is measured as negative non-operating accruals. Except for column (i), the joint test (the sum of accounting conservatism measures and the interaction between accounting conservatism and underinvestment firms) rejects the hypothesis that the sum of the two coefficients is indistinguishable from zero with p-values less than 1%, indicating that accounting conservatism is negatively related to investments for firms that are most susceptible to underinvestment. This result is consistent with the argument that accounting conservatism can lead to underinvestment. For example, in column (iv) when the composite measure of conservatism is used, the coefficient on ACRANK is with the t-value of -4.17, suggesting that for firms that are likely to overinvest, a one standard deviation increase in accounting conservatism can lead to 3.7% ( * 0.1 / 0.019) reduction of the mean value of capital investments. The coefficient on ACRANK_UnderFirm is , and the sum of the two coefficients (ACRANK and ACRANK_UnderFirm) is The related p-value testing joint significance is less than 0.1%. Hence, a one standard deviation increase in accounting conservatism leads to 6.3% ( * 0.1 / 0.019) reduction of the mean value of capital investments among firms most likely to suffer from underinvestment. Overall, the results in Table 3 provide evidence consistent with prior studies that accounting conservatism reduces overinvestment. More importantly, the results also show that accounting conservatism is positively rather than negatively related to underinvestment. This new result suggests that the underinvestment argument dominates the debt facilitation argument. For the corporate governance variables, we find that firms with high institutional investor ownership (Institutions) invest more and that more analysts following (Analysts) tend to reduce 14

17 firms investment. The coefficients on the interaction terms of the three proxies for governance and accounting conservatism are mostly significant for institutional investor holdings and analysts following, but not for inverse G-score. For the other control variables, consistent with prior studies, we find that firms with higher growth options (TobinQ), firms with high levels of tangible assets (Tangibility), firms generating high cash flows (CFOSale) from operations, and firms with lower bankruptcy risk (Z-Score) all invest more in capital expenditures. On the other hand, large and old firms, dividend paying firms, firms in highly leveraged industries (IndK- Structure), firms with high volatility in large cash flows and sales all invest less in capital expenditures Investment risk, accounting conservatism, and investment Roychowdary (2010) suggests that the negative consequences of accounting conservatism on firm investments should be more prominent for riskier firms because the projects of such firms are more likely to turn into negative NPV projects. The acceleration of loss recognition exposes managers to higher probabilities of large near future losses and subsequent negative compensation and reputation consequences. Hence, ex ante, risk-averse managers are less willing to invest in risky projects even though these projects have positive NPVs, resulting in firm underinvestment. For firms that are likely to suffer from underinvestment, we expect the negative relationship between accounting conservatism and firm investments will be stronger when a firm s investment set contains more risky investments. Likewise, among firms that are more likely to overinvest, if accounting conservatism is effective in preventing managers from overinvesting in risky negative NPV projects, we expect the 15

18 negative relation between accounting conservatism and firm investments to be stronger for firms with riskier investment sets. 4 To test these predictions, we use daily stock return volatility to proxy for the riskiness of a firm s investment set. We create a dummy variable HVol based on the annual medians of daily return volatility. The variable HVol takes the value of one if a firm s return volatility for a year (RetVol) is higher than the median value of return volatility, and zero otherwise. We focus on specifically on those firms that are likely to suffer from underinvestment (UNDERFIRM =1) and overinvestment (UNDERFIRM = 0) problems. We estimate the following equation separately for under and overinvestment firms: Investment i,t+1 = β 0 + β 1 ACRANK i,t + β 2 HVol i,,t + β 3 ACRANK i,,t * HVol i,t + β 4 GOV i,t + Σβ j CONTROL j,i,t + Year t + Ind k +e i,t+1 (2) For brevity, we focus on the composite measure of accounting conservatism and all the other variables are as defined in the previous section. The results from testing the above model are reported in Table 4. For firms that are likely to underinvest (UnderFirm = 1), we find that the coefficient on ACRANK is with a t-value of -1.05; and the coefficient on ACRANK _HVol is negative with a t-value of The sum of the two coefficients is with a p-value less than 0.001%. These results show that for underinvesting firms, the negative effect of accounting conservatism on firm investments is mostly concentrated in firms with riskier investment sets. For firms having less risky investments, accounting conservatism does not exert a significant effect on firm investments. Hence, the results here are consistent with Roychowdhury s (2010) argument that accounting 4 One caveat is that we cannot observe whether the risky projects are positive or negative NPV. We rely on a reasonable assumption that likely-to-underinvest firms tend to have more (unrealized) positive NPV projects; in contrast, likely-to-overinvest firms tend to have fewer positive NPV projects. If this assumption is not correct, then the negative association between accounting conservatism and firm investments for risky firms also can be interpreted as a general reduction in risky positive NPV projects. 16

19 conservatism can induce managers to avoid undertaking risky projects, resulting in underinvestment. For firms that are more likely to overinvest (UnderFirm = 0), the coefficients on ACRANK and ACRANK_HVol are negative but insignificant. However, the sum of the two coefficient (ACRANK and ACRANK_HVol) is , which is marginally significant with a p- value of Hence, for firms with a high level of risky investments, accounting conservatism is negatively correlated with firm investments but this negative relationship is statistically significant only when firms investment sets contain a high level of low-risk projects Managers investment horizon, accounting conservatism and investment Leuz (2001), Watts (2003a), and Guay and Vierrecchia (2006) suggest that myopic managers with short-term investment horizons will be particularly susceptible to underinvestment arising from accounting conservatism. Specifically, if managers have to accelerate future losses into the current period and defer gains into future periods, then managers with limited tenure horizons may not be able to reap the benefits of firm investments if the payoffs are far in the future. Alternatively, managers with short investment horizons value shortterm gains more than future gains as they require a higher discount rate on future gains in deriving the expected payoffs (Reichelstein, 2000), thus making long-term projects less attractive. The effect of accounting conservatism on underinvestment is therefore expected to be more pronounced when managers investment horizons are short. On the other hand, because accounting conservatism is equally likely to deter managers from overinvesting in both longterm and short-term negative NPV projects, it is less clear on how investment horizons affect the relationship between overinvestment and accounting conservatism. 17

20 We use CEOs ownership of unexercised stock options and restricted stock as an empirical proxy for managers investment horizons. In particular, stock options and restricted stock can only be exercised in the future when managers meet certain predetermined benchmarks. As such, managers investment horizons are extended because long-term investments are important for their future compensation. We obtain CEO ownership of unexercised stock options and restricted stock from EXECUCOMP. After merging with the EXECUCOMP, the sample is reduced to 23,149 firm-year observations with available data to measure CEO stock and option ownership. Untabulated results show that the average CEO ownership of unexercised stock options and restricted stock is 3.2%. We create a dummy variable HOwn, which equals one if CEO ownership of unexercised stock options and restricted stock (CEOOwn) is higher than the median ownership for a given year and zero otherwise. We focus on firms that are likely to suffer from underinvestment (UnderFirm =1) and overinvestment (UnderFirm = 0) problems. We estimate the following equation separately for underinvestment and overinvestment firms: Investment i,t+1 = β 0 + β 1 ACRANK i,t + β 2 HOwn i,,t + β 3 ACRANK i,,t * HOwn i,t + β 4 GOV i,t + Σβ j CONTROL j,i,t + Year t + Ind k +e i,t+1 (3) Consistent with the previous section, we focus on the composite measure of conservatism and all the other variables are as defined in the previous section. Table 5 reports the results. For underinvestment firms, the coefficient on ACRANK is with a t-value of -1.69, which is marginally sigificant. The coefficient on ACRANK_HOwn is with a t-value of More importantly, the sum of the coefficient on ACRANK and ACRANK_HOwn is close to zero and the joint test of these two variables cannot reject the hypothesis that the sum is equal to zero. This finding suggests that when CEOs ownership of unexercised stock options and restricted stock is low (investment horizon is short), accounting 18

21 conservatism is negatively associated with firm investments; however, when the ownership is high (investment horizon is long), accounting conservatism is no longer associated with firm investment. This finding is consistent with the prediction that ownership extends CEOs investment horizons and alleviates the negative effect of accounting conservatism on firm investments for likely-to-underinvest firms. For overinvesting firms, we fail to find any statistically significant effects for either accounting conservatism or the interaction between accounting conservatism and CEO ownership. This insignificant result may be due to the reduced power of the tests because of small sample size. 5. Endogeneity of accounting conservatism and firm investment It is possible that firms make decisions regarding corporate financial reporting policies and investment policies jointly. If accounting conservatism and firm investments are jointly determined, then the OLS estimation will be biased. To alleviate this endogeneity concern, we estimate a system of two equations allowing for the joint determination of conservatism and investment. Specifically, we estimate the following two equations: ACRANKi,t = β0 + β1 Investmenti,t+1 + β2 LogAsseti,,t + β3 TobinQi,,t + β4 Leveragei,t + β5 Litigationi,t + β6 Institutionsi,t + β7analystsi,t + β8 InvG-Scorei,t (4) + β9 G-ScoreDumi,t + β10 Ratingi,t + β11 TaxRatei,t + Yeart + Indk +ei,t Investmenti,t+1 = β0 + β1 ACRANKi,t + β2 ACRANKi,t * UnderFirmi, t+1 + β3 GOVi,t + β4 GOV * UnderFirmi, t+1 + Σβj Controlsj,i,t + Yeart + INDk +ei,t+1 (5) We rely on prior studies in selecting our control variables for the accounting conservatism model (equation 4) to reflect the fact that accounting conservatism can be shaped by various economic forces such as litigation risk, tax considerations, shareholders demands, and debt holders preferences. We include four commonly-used controlled variables based on 19

22 prior studies (see, e.g., Roychowdhury and Watts, 2007): firm size (LogAsset), growth option (TobinQ), leverage (Leverage), and litigation risk (Litigation). Litigation risk is a dummy variable equal to one if a firm is in a high litigation risk industry as identified by SIC industry codes (SIC codes , , , , and ) and zero otherwise. In addition, we also include the three proxies for corporate governance: institutional investor ownership (Institutions), analyst following (Analysts), and inverse G-score for antitakeover protection (InvG-Score), as well as the dummy variable indicating whether the G-score is missing (G-ScoreDum). Following Beatty et al. (2008), we include simulated tax income rates (TaxRate) and credit ratings (Rating) to control for factors that are related to tax considerations and access to credit markets, and how those considerations affect firms accounting conservatism. TaxRate is the simulated corporate marginal tax rate obtained from John Graham s website. Rating is the long term debt rating obtained from COMPUSTAT. If the rating is not available from the database, we estimate the credit rating for the firm following the method described in Barth et al. (2008) (see also footnote 3). The ratings for sample firms are between 1 (AAA) and 24 (C). Hence, the credit quality decreases as the value of rating increases. The investment model is the same as discussed in the previous section. We rely on earlier empirical studies to guide our selection of right hand side variables for the accounting conservatism and investments equations, assuming that the excluded variables are exogenous and satisfy the exclusion restrictions. Our simultaneous equation estimates can be biased if the related assumption fails to hold. To address these potential concerns, we run F-tests and partial R-square tests of excluded instruments and the results indicate that the instruments in the accounting conservatism and investment models are jointly significant in explaining the 20

23 endogenous variables and that the instruments are valid. Second, the results of tests for under identification or weak instrument problems reject the hypothesis that the instruments suffer from such problems. For example, the partial R-square for the accounting conservatism instruments in explaining the endogenous variable itself is 0.10 and the F-value is 315, which rejects the null hypothesis that the instruments are weak instruments. Finally, we run a Hausman test and reject the null hypothesis that accounting conservatism and investments are exogenous with the p-value of the test less than 1% (F-value is ). Panel A of Table 6 reports the results of estimating the accounting conservatism model (equation 4). The coefficient on investment is negative and significant, which is consistent with the results found in Table 3. The coefficients on firm leverage and credit rating are all positive and statistically significantly different from zero, consistent with the argument that long-term debt holders demand accounting conservatism. The coefficient on tax rate is positive and significant, as expected, suggesting that firms become more conservative when corporate tax rate is high. The coefficient on litigation risk is also positive and significant, suggesting that firms become more conservative when they are in a risky industry. Panel B of Table 6 reports the results of estimating the investment model (equation 5). Consistent with the results found in Table 3, the coefficient on accounting conservatism remains negative and statistically significant (the coefficient = and t-value = -2.65). The coefficient on the interaction between accounting conservatism and firms that are likely to underinvest is also negative (coefficient = ) and statistically significant (t = ). The joint test of the sum of the two coefficients (ACRANK and ACRANK_UnderFirm) easily rejects the hypothesis that the two coefficients are not different from zero (p-value < 0.001). Hence, after considering the joint determination of accounting conservatism and firm investments, we 21

24 continue to find that accounting conservatism is negatively related to corporate investments for firms that are likely to overinvest, as well as for firms that are likely to underinvest. These results confirm that while accounting conservatism can reduce overinvestment problems, it can also exacerbate underinvestment problems. 6. Alternative measure of accounting conservatism In this section, we use Basu s cross-sectional measure of conservatism, an alternative measure of accounting conservatism, to further check the robustness of our results. We estimate the following model: Xi,t = β0 + β1 Di,t + β2 Ri,,t + β3 Di,t * Ri,t + β4 Investment i,t+1 + β5 Investmenti,t+1 * Di,t + β6 Investmenti,t+1 * Ri,,t+ β7 Investmenti,t+1 * Di,t * Ri,t + β8 Di, t * UnderFirmi,,t+1 + β9 Ri,t * UnderFirmi,t+1 + β9 Di,t * Ri,t * UnderFirmi,t+1 + β10 Investmenti,t+1 * UnderFirmi,,t+1 + β11 Investmenti,t+1 * Di,t * UnderFirmi,t+1 + β12 Investmenti,t+1 * Ri,t * UnderFirmi,,t+1 + β13 Investmenti,t+1 * Di,t * Ri,t * UnderFirmi,t+1 + βi Controlsi, t + βj Controlsi,t * Di,t + β Controlsi,t * Ri,,t + β Controlsi,t * Di,t* Ri,t +Indk+ Yeart +ei,t (6) All variables are the same as defined in section 3.2 and section 3.3. Control variables (Controls) include firm size (LogAsset), firm leverage (Leverage), growth option (TobinQ), and litigation risk (Litigation). The coefficient β7 measures the extent to which accounting conservatism is associated with corporate investment for firms that are likely to overinvest, and β7 + β9 measures the extent to which accounting conservatism is associated with corporate investments for firms that are likely to underinvest. The results are tabulated in Table 7. Column (i) reports results without including UnderFirm and the interactions between UnderFirm and all the other variables. We find that the coefficient β7 is negative and statistically significant, suggesting that corporate investments and 22

25 accounting conservatism are negatively related. In Column (ii), after controlling for UnderFirm, β7 remains negative and stastically significant, suggesting that the accounting conservatism is negatively related to investments for firms that are likely to overinvest. Hence, accounting conservatism can decrease overinvestment problems. The coefficient β9 is negative but not significant. However, the sum of the two coefficients (β7 + β9) remains negative and statistically significant with a p-value of as indicated in Joint Significance. Hence, using this alternative measure of accounting conservatism, we continue to find that accounting conservatism can cause underinvestment problems for firms that are likely to underinvest. 7. Alternative indicators for firms suffering from underinvestment (overinvestment) In the previous analysis, the variable UnderFirm, derived from the firms leverage and cash holdings, was used to estimate the probability of which a firm suffers from underinvestment (overinvestment). In order to enhance the robustness of this measurement, we constructed two additional indicators (Labis and Lqabi) with reference to Biddle et al. (2009) and re-estimated our equation by substituting UnderFirm with either of the additional indicators: Investmenti,t+1 = β0 + β1 Consi,t + β2 Consi,t * Labii, t+1 + β3 GOVi,t + β4 GOV * Labii, t+1 + Σβj Controlsj,i,t + Yeart + indk +ei,t+1 (7) Where Labi could be defined as either Labis or Lqabi. Both Labis and Lqabi derived from the residuals of the firm-specific investment model which can be stated as: Investmenti,t+1 = β0 + β1 SGROi,t + ei,t+1 (8) Where SGRO is the sales growth rate at year t. After running equation 8 for each industry-year based on the Fama and French 48-industry classification for all industries with at least 20 observations in a given year, the residuals (also known as abnormal investment) is obtained for 23

26 each firm in a given industry-year. Labis is a decile rank measurement that indicates the rank of the level of abnormal investment within a given industry-year. The rank was rescaled to range between zero (highest level of abnormal investment) and one (lowest level of abnormal investment). Lqabi is another variable that indicates the rank of the level of abnormal investment within a given industry-year. It takes the value of one if the level of abnormal investment is within the bottom 25% quartile, while it takes the value of zero if the level of abnormal investment is within the top 25% quartile. Table 8 presents the results of estimating equation 7 where UnderFirm was substituted by Labis (Panel A) and Lqabi (Panel B). Although the coefficients of accounting conservatism measures becomes positive in three cases, the coefficients of the interactions between acconting conservatism proxies and the probability indicators of a firm suffering from underinvestment (overinvestment) are moderately comparable to that of the main results in Table Multi-nominal logit estimation of conservatism-investment model With reference to the study of Biddle et al. (2009), we conducted a series of multinominal logit estimation of our conservatism-investment model in order to further support our hypothesis. Whilst previous analysis employs ordinary least square linear modeling technique to examine how the level of accounting conservatism affects the level of firm investment, this section is concerned with how the level of accounting conservatism would affect the probability in which a firm would underinvest or overinvest. In this section, Lqabi will be used as the indicator for firms suffering from underinvestment (overinvestment). The results shown in table 9 seem to enhance the robustness of our main analysis and confirm our hypothesis that accounting conservatism will contribute to underinvestment as well 24

27 as reduce overinvestment. In panel A, the significant positive coefficients in all the accounting conservatism proxies suggest that a higher level of accounting conservatism is associated with higher probability that the firm will suffer from underinvestment (with the level of abnormal investment lying within the bottom 25% quartile of the industry-year). Similarly in panel B, the coefficients for accounting conservatism proxies SKEW, BASU and ACRANK are negatively significant. This suggests that a higher level of accounting conservatism is associated with lower probability that the firm will suffer from overinvestment (with the level of abnormal investment lying within the top 25% quartile of the industry-year) 9. Conclusions The accounting literature has long debated the role of accounting conservatism on managers investment incentives. In spite of this long-running debate, extant empirical studies focus only on how accounting conservatism can reduce managers proclivity for overinvestment policies but, to the best of our knowledge, no previous study has yet investigated its potential underinvestment implications. This study provides direct empirical evidence that accounting conservatism can cause dysfunctional (under)investment incentives for managers. We rely on prior studies to identify ex ante firms that are likely to suffer from underinvestment and overinvestment and document evidence that accounting conservatism is negatively related to firm investments for both nonoptimal investment types (i.e., firms that are likely to underinvest and firms that are likely to overinvest). This negative relation between accounting conservatism and firm investments for both underinvesting and overinvesting firms is consistent with our predictions that accounting conservatism can reduce overinvestment problems, but at the same time can also magnify underinvestment problems. 25