Bank Corporate Governance and Future Earnings Predictability

This study examines the impact of corporate governance on earnings predictability in large banks from 35 countries over the period 2004-2010. We find that board structure and CEO power have a significant positive influence on earnings predictability of future cash flows although these findings vary for emerging markets and between common and civil law countries. Board structure and CEO power are more effective in predicting future cash flows in civil law countries whereas in common law countries risk governance is a more accurate predictor of earnings. Similarly, we find differences between developed and emerging countries. While in both domains there is no qualitative difference in CEO power to predict future cash flows, board structure and risk governance are less effective in emerging countries.


Introduction
An extensive literature examines how corporate governance and reporting quality impact on earnings, but these issues have little been addressed in the banking area. The extant non-bank literature examines earnings management and/or quality typically linking these to such factors as: institutional ownership (Velury and Jenkins, 2006); book-tax conformity (Atwood et al., 2010); gender diversity (Ye et al., 2010); board characteristics (Mashayekhi and Bazaz, 2010); employee expenses (Schiemann and Guenther, 2013); the features of chief financial officers (Dichev et al., 2013); and country-level institutional factors (Kanagaretnam et al., 2014). Our study extends this literature by using bank governance indicators such as, board structure, CEO power, and risk governance, to examine the determinants of future cash flow (CFOt+1) of large banks from 35 countries. We measure earnings' ability to predict future cash flows of banks as the coefficient from a regression (both GLS and GMM techniques) of one-period-ahead cash flows (CFOt+1) on current period earnings and governance indicators.
Earnings quality has received considerable policy attention since the global financial crisis (2008)(2009) when banking industries in several Western economies were substantially re-shaped (Dechow et al., 2010;Prior et al., 2014) and post-crisis regulators have introduced new regulations on bank financial reporting quality in order to re-build the financial strength of large banks (Kanagaretnam et al., 2014). The concept of quality earnings is fundamental in accounting and financial economics, yet there are broad disagreements about how it should be defined and measured (Dichev et al., 2013) 2 . Complications with measurement issue debate as to whether 'accounting-based' (accruals quality, earnings management, persistence, predictability, and smoothness) or 'market-based' (relevance and timeliness) measures are the most appropriate to gauge quality. Conceptually, however, earnings quality is deemed 'high' if current earnings can better predict future cash flows as well as the long-run profits of the firm (Valury and Jenkins, 2006;Dichev et al., 2013;Schiemann and Guenther, 2013). Earning quality is regarded as 'low' if managers have an incentive to manipulate the figures opportunistically (Healy and Wahlen, 1999;Dechow and Skinner, 2000;Rosenfield, 2000;Dechow et al., 2010). Both agency theory and signalling theory deal 2 High-quality earnings have been defined/measured in the literature as those that are: persistent and hence the best predictor of future long-run sustainable earnings; smooth; predict future earnings; backed by past, present, or future cash flows; have, small changes in total accruals that are not linked to fundamentals (Dechow and Dichev, 2002;Schipper and Vincent, 2003;Dechow and Schrand, 2004;Francis et al., 2004;Kothari et al., 2005;Melumad and Nissim, 2009). with such manipulation issues in the context of an asymmetric information environment and deterrent measures through enhanced corporate governance and financial reporting standards.
The overall quality of the accounting information environment derives from the current US Generally Accepted Accounting Principles (GAAP) and International Financial Reporting System (IFRS) standards bringing incremental information benefits to investors. Under the GAAP/IFRS framework, the current earnings of firms are reflected in their future operating cash flows (CFOt+1) 3 . Earnings predictability is an important measure as it deals with how well past earnings can explain current and future earnings and this can lead to more accurate valuation as it enables investors to more accurately anticipate expected future cash flows (Valury and Jenkins, 2006;Schiemann and Guenther, 2013). Earnings predictability can not only provide a better understanding of expected future cash flows (CFOt+1) but also determines future bank investment and lending activities (Hasan et al., 2012). Despite the importance of earning quality and predictability to banks this has been barely considered in the empirical literature.
Prior banking literature documents that managers use their discretion in financial reporting for various reasons to: signal private information; manage risk; meet or beat earnings benchmarks; and to avoid adverse compensation and career consequences (Kanagaretnam et al., 2004 and2010;Bushman and Williams, 2012;Dechow et al., 2012;Dichev et al., 2013). Cornett et al. (2009) show that US banks use their discretion to smooth earnings during periods of low profitability by delaying the reporting of loanlosses and increasing the realization of securities gains despite monitoring and oversight by regulators.
While the purpose of loan-loss provisions is to adjust banks' loan-loss reserves to reflect expected future losses, bank managers may also have incentives to use them to manage earnings and regulatory capital (Ahmed et al., 1999;Prior et al., 2014). As noted by Pérez et al. (2008) for banks the accrual of loan-loss provisions is left to managers' discretion. However, the literature is somewhat inconclusive as to the links between loan-loss provisioning and earnings. Collins et al. (1995) find evidence of a positive relationship whereas Beatty et al. (1995) find no evidence of earnings smoothing.
It is evident that managerial discretion to manipulate earnings has declined overtime due to the current US GAAP and IFRS standards and their convergent objectives aimed at improving earnings quality (see Dichev et al., 2013). The introduction of a 'conceptual framework' under US GAAP and IFRS standards is an important milestone in this regard as these aim to address the concern of regulators and investors on financial reporting manipulation and the quality of earnings. Both US GAAP and IFRS standards are more compatible than ever, as Jamal et al. (2009) suggest allowing US companies between using US GAAP and IFRS to promote the likelihood of generating high quality financial reporting and earnings, while US adoption of IFRS or converging US GAAP with IFRS improves earnings quality further (Dichev et al., 2013). Barth et al. (2008) contend that IFRS limits the opportunity of management to engage in opportunistic behaviour by restricting accounting options available to them. Leventis et al. (2011) also find that IFRS has been beneficial to users of financial reports, as it has reduced opportunistic behaviour in earnings management. Although, the occurrence of earnings manipulation is hard to unravel (Dechow et al., 2010) a lack of correspondence between earnings and cash flows and deviations from industry and other peer experience can provide helpful indication of earnings management practice (Dichev et al., 2013). Kanagaretnam et al. (2014) suggest a strong association between measures of earnings quality (earnings persistence or cash flow predictability) and strong country-wide legal, extra-legal and political institutions. They also find that strong institutional factors constrain opportunistic earnings management by managers of banks, which increases the information value of bank earnings and the credibility of their financial statements.
The recent global financial crisis of 2008-2009 raises questions with respect to the link between corporate governance and bank earnings. For non-bank firms it has been shown that weak corporate governance structures result in deterioration in earnings quality (du Plessis et al., 2005;Hashim and Devi, 2007;Jo and Kim, 2007). However, the development in corporate governance mechanisms and the introduction of GAAP/IFRS standards has generally improved earnings quality and reduced opportunistic behaviour (via tougher monitoring systems) (Dechow et al., 1996;Klein, 2002;Xie et al., 2003;Cormier and Martinez, 2006;Jo and Kim, 2007;Shen and Chih, 2007;Kent et al., 2010 andPrencipe andBar-Yosef, 2011). Some of these monitoring mechanisms relate to strengthening inside corporate governance via legal and disclosure related legislation (macro-level), and also through improved systems and procedures at the firm-level (micro-level) (Agrawal and Chadha, 2005;Hope and Thomas, 2008;Holm and Schøler, 2010;Kanagaretnam et al., 2014). Therefore, in conjunction with compliance to GAAP/IFRS standards, various governance mechanisms such as, strong board structures, CEO power and risk governance can accelerate the quality of earnings by restraining managerial manipulations detrimental to firm value. This paper examines the impact of corporate governance on the predictability of cash flows (CFOt+1) using a sample of 179 large commercial banks and bank holding companies (BHCs) chartered in 35 countries over the period 2004-2010. Our approach is similar to Cornett et al. (2009) andKanagaretnam et al. (2014). Cornett et al. (2009) study the relationship between earnings management and corporate governance indicators for large US bank holding companies (BHCs). They use data from the 100 largest publicly traded BHCs from 1994 to 2002 and find that reported earnings, board independence, and capital are negatively related to earnings management. On the other hand, Kanagaretnam et al. (2014) study the impact of international institutional factors on earnings quality. Using a large sample of non-US banks from 35 countries over 1993-2006 they report that cash-flow predictability is higher in countries with stronger legal, extra-legal and political institutional structures. Our study extends Cornett et al. (2009) and Kanagaretnam et al. (2014), rather than focusing on earnings management and institutional factors directly, we investigate earnings predictability of CFOt+1 and corporate governance relationships. We find that board structure and CEO power have a significant influence on the earnings predictability of large banks. However, the results vary for emerging and developed countries, and common law and civil law countries.
Our study contributes to existing research in several ways. Firstly, to the best of our knowledge this is the first study using an international sample of commercial banks and BHCs that investigate the relationship between corporate governance and earnings predictability. Our primary result is that a small board with more independent directors helps predict future cash flow (CFOt+1). We also show that CEOs with role duality (CEO and Chairperson) and those appointed internally are more likely to provide positive signals for future earnings growth. We also report that having risk officers on the board as well as female directors is positively associated with CFOt+1. We, further, show a positive and significant relationship between current earnings and CFOt+1, indicating that current year's reported earnings is an important predictor of future earnings. These findings are consistent with both Velury and Jenkins (2006) and Atwood et al. (2010) who show that future cash flows are positively associated with current earnings.
Secondly, in-line with Pathan (2009), Pathan andFaff (2013) and Mollah and Zaman (2015) we argue that governance variables are effective in enhancing performance and reducing risks in banking. Since earnings' are positively associated with performance and negatively with higher risk, it can be argued that governance variables also have a bearing on current and future earnings. Moreover, Cornett et al. (2009), suggest that corporate governance plays some role in earnings and earnings management at large U.S. banks and they find a significant negative relationship between earnings management and board independence. Our focus on the link between corporate governance indicators and CFOt+1 interactions with current earnings is motivated by the GAAP/IFRS conceptual framework, which suggests that users and investors should get quality information from financial reports to assess CFOt+1. We find a significant influence of board structure and CEO power on the earnings predictability of large banks (but no relationship with our risk governance variable). Our study therefore, provides empirical evidence consistent with the signalling theory built on financial reporting under the GAAP/IFRS regime as well as agency theory predictions from the regulatory environment concerning board structure and CEO-power in explaining earnings predictability.
Thirdly, banks are highly regulated, as compared to non-banks, to comply with not only accounting standards or listing requirements on disclosure, but also to meet regulatory requirements (Basel requirements and so on). Under these standards, banks are monitored by a number of regulatory authorities. Kousenidis et al. (2013) examine whether and to what extent the recent crisis in the European Union (EU) had an impact on the quality of the reported earnings of listed firms in PIIGS (Portugal, Ireland, Italy, Greece and Spain) countries. They find that on average, earnings quality improved over the crisis period but in the presence of incentives for earnings management, earnings quality deteriorated despite increased monitoring and oversight by regulators. Among explanations as to why banks exposed themselves to excessive risks during the run-up to the financial crisis was that there was a failure of risk management (Ellul and Yerramilli, 2013). This motivates us to investigate whether predicted earnings of banks were reflected in the current earnings reported under GAAP/IFRS. If this was the case then the responsibility for the banking collapse may be directed more toward governance factors rather than accounting numbers or earnings quality. We find no evidence that corporate governance features had a different impact on earnings predictability during the global crisis period (2008)(2009).
Finally, since the sample of banks operate in different countries with varying institutional featurescommon and civil law; developed and emerging countries -it is of interest to observe whether these factors have any bearing on the relationship between earnings, governance and predicted earnings. Kanagaretnam et al. (2014) find that banks' cash-flow predictability is higher in countries with stronger legal, extra-legal and political institutional structures. We find that board structure and CEO power are more effective in predicting future cash flows in civil law countries whereas in common law countries risk governance is a more accurate predictor of future earnings. Similarly, we find differences between developed and emerging countries. While in both domains there is no qualitative difference in CEO power to predict future cash flow, board structure and risk governance are less effective in emerging countries.
The remainder of the paper is structured as follows: Section 2 provides an overview of literature and hypotheses on board structure, CEO power and risk governance while Section 3 discusses the methods, data and model specifications; Section 4 focuses on empirical results and robustness tests and finally Section 5 provides concluding remarks.

Related Literature and Hypotheses Development
A number of studies measure earnings quality by assessing the ability of earnings to predict future cash flows (Doyle et al., 2003;Cohen et al., 2004;Francis et al., 2004;Van der Meulen et al., 2007;Melumad and Nissim, 2009;Atwood et al., 2010 and. Researchers use different proxies for earnings quality such as, earnings management, accrual quality, earnings predictability, smoothness, persistence, earnings informativeness, benchmark beating (Francis et al., 2004;Dichev et al., 2013), but one of the proxies of earnings' quality, earnings predictability, refers to earnings with a high predictive value, namely, a strong association with future cash flows. A stronger relationship between earnings and future cash flows is a better indicator of earnings quality. Lipe (1990) considers earnings predictability as the ability of earnings to explain itself. If past earnings are good estimates of current earnings then predictability is said to be high. The quality of firms' earnings relates to the usefulness of accounting information to financial statement users. While quality of earnings is a concept having multidimensional constructs, the foundation of earnings quality derives from the 'conceptual framework' of accounting standards, namely the GAAP/IFRS regime. GAAP/IFRS promotes quality disclosure and reporting and hence, financial reports are designed to provide value relevant information. Therefore, earnings numbers under GAAP/IFRS are viewed as high quality and should better be able to estimate a firm's future prospects.
Again, as part of accounting information, earnings numbers should also have qualitative characteristics to be useful in the firm decision-making process. In the GAAP/IFRS conceptual framework, both relevance and reliability are viewed as the two principle qualitative characteristics of earnings numbers. To be relevant, among other things, earnings numbers must have predictive value. The predictive nature of accounting earnings is also manifest in valuing firm's equity which also requires investors to anticipate firm's expected future cash flows (Valury and Jenkins, 2006). Dechow (1994) contend that current earnings generally produce better predictions of future cash flows. Current earnings figures are believed to be of higher quality, the more they are able to predict future cash flows, and the quality of current accounting information is firmly rooted into strict compliance of GAAP/IFRS based accounting standards.
Furthermore, firms with higher earnings quality are more likely to have a good governance system that signals the reliability of their financial reporting process (Engle, 2005). The direct and indirect agency costs relating to managerial incentives can be mitigated through a number of governance mechanisms. It is argued that governance mechanisms assist to align managers' interests with those of shareholders and reduce agency costs. Similarly, the CEO's power and background also serve the integrity of a company's financial statement and hence firm's financial statement quality (Zhang and Wiersema, 2009). Agency theory predicts that powerful CEOs tend to be entrenched and operate detrimentally (to shareholders) by extracting private benefits of control and empire building (Jensen, 1986). So as to reduce this likelihood improved governance recommends the separation of the dual Chairman and CEO roles, as well as to increase the number of independent board members (Shleifer and Vishny, 1997). In the same vein, recent banking literature (Mongiardino and Plath, 2010;Sabato, 2010;Aebi et al., 2012) suggest the creation of risk management mechanisms in the governance structure so as to emphasize reporting line to the board, rather than the CEO, and to check the power of CEOs thus reducing possibilities for opportunistic behaviour. Our plan is to incorporate these issues in our methodology on bank earnings predictability outlined below.

Board structure and earnings predictability:
Agency theory suggests that board independence acts as a watchdog over a firm's operations and provides monitoring incentives for reducing agency costs. The existence of independent directors on the board will enhance earnings ability to predict future cash flows. Chen and Jaggi (2000) provide evidence that board of director independence is crucial in influencing the level of good quality reporting. Cheng and Courtney (2006) document a significant positive association between the proportion of independent directors and the extent of quality of reporting.
Again, board size is also one of the factors influencing the effectiveness of board oversight duties. There has been continued debate on the role of strong boards (smaller size), although agency theory suggests that larger boards support effective monitoring (Coles et al. 2008;and Pathan, 2009). Small boards are favoured for being easier to co-ordinate, quicker in making decisions, less likely to have free-rider problems, and less likely to oppose innovation (Dimitropoulos and Asteriou, 2010; Mollah and Zaman, 2015;Pathan, 2009). Xie et al. (2003, p300) point out that "A smaller board may be less encumbered with bureaucratic problems and may be more functional. Smaller boards may provide better financial reporting oversight". However, Rahman and Ali (2006) point out that large boards are less effective in reducing managerial manipulation of earnings numbers and enhancing earnings quality. Vafeas (2000) and Cho and Rui (2009) indicate that earnings numbers of firms with small boards are more informative. However, there are opposing views about large boards and their role in monitoring. Empirical studies conclude that a greater number of board members will likely lead to more independent directors and provide more expertise, experience and knowledge and diversity and increase the board's monitoring capacity and hence, they are able to delegate more responsibilities to board committees than smaller boards (Dalton et al., 1998;John and Senbet, 1998;Dalton et al., 1999;Linck et al., 2008). Thus, given the mix of findings on both board independence and size, we hypothesize that board structure 4 , especially strong board structure (smaller board size and higher independence) help improve earnings predictability of banks.

H1: Board structure has an effect on earnings predictability.
Or-Small and independent board has a positive effect on earnings predictability.

CEO power and earnings predictability
The governance literature defines CEO power in a variety of ways: CEO-chair role duality; internallyhired CEO and CEO tenure (May, 1995;Hermalin and Weisbach, 1988;Morck et al., 1989;Adams et al., 2005;Pathan, 2009;Fracassi and Tate, 2012;Mollah and Zaman, 2015). A CEO becomes more powerful when they chair the board and such power grows if the CEO is internally-hired, indicating CEO's longstanding involvement with the firm (Pathan, 2009;Mollah and Zaman, 2015), and has been employed for a longer period (Fracassi and Tate, 2012). CEO power can be entrenched in restricting information flows to other board directors, influencing board decisions and undermining the board's independence to oversee managers. Pathan (2009) document that CEO power is negatively related to bank risk-taking, namely, bank CEOs have incentives to take less risk to secure their jobs and human capital investment by accepting some safe, value-reducing projects, and rejecting some risky but value-increasing projects (May, 1995;Saunders and Cornett, 2006). Similarly, in the earnings informativeness literature, CEO duality seems to have a detrimental effect on the usefulness of earnings numbers (Gul and Lai, 2002;Anderson et al., 2003;Firth et al., 2007). Combining two roles (chairperson and CEO) along with longer tenure exacerbates the potential for managing earnings, thus impairing the quality of reported earnings (Saleh et al., 2005;Prencipe and Bar-Yosef, 2011) and leading to earnings with less predictive value. Similarly, other CEO characteristics, such as whether the CEO was an internal appointment and length of tenure also have a negative bearing on firm's financial statement quality and accordingly on earnings predictability (Pathan, 2009;Fracassi and Tate, 2012). Thus, following the concepts of agency theory, we propose a hypothesis as: H2: CEO power has a negative effect on earnings predictability.
Or-CEO-duality, CEO-internal and CEO tenure have negative effect on earnings predictability.

Risk governance and earnings predictability
In recent years, risk management has become an integral part of governance in the banking sector. Mongiardino and Plath (2010) outline best practice in banking risk governance and highlight the need to have at least a dedicated board-level risk committee, of which a majority should be independent, and that the Chief Risk Officer (CRO) should be part of the bank's executive board (Aebi et al., 2012). Both Aebi et al. (2012) and Ellul and Yerramilli (2013) note that risk governance (having a CRO on the board as well as a Risk Management Committee) matters for bank risk-taking as they perform risk monitoring functions. As banks are at the centre of taking risky business, the main purpose of the risk management function is to mitigate the risk of large losses, known as tail risk (Ellul and Yerramilli, 2013). Ellul and Yerramilli (2013) suggest that a strong and independent risk management function in banks can curtail such tail risk exposures. Kashyap et al. (2008) and Stulz (2008) contend that the presence of a strong and independent risk management team is necessary to control tail risk exposures of financial institutions.
Policy makers around the globe have emphasized the importance of appropriate risk management practice being applied within financial institutions, including the existence of a risk management committee and appointment of chief risk officers (Brancato et al., 2006;Sabato, 2010). Therefore, risk governance mechanisms, such as the presence of a risk committee, Chief Risk Officer (CRO) on a firm's executive board and clear CRO reporting lines are important components of bank corporate governance.
A strong risk management function is necessary to correctly identify risks and prevent excessive risktaking (Kashyap et al., 2008;Stulz, 2008) that cannot be controlled entirely by regulatory supervision or external market discipline (Ellul and Yerramilli, 2013). Similarly, the role of the CRO and other risk managers is also important to evaluate the viability of bank's loans and other investments. Risk managers or CRO's can restrain banks taking excessive tail risks if an appropriate level of accessibility to the executive board and clear lines of reporting to the board are established.
Again, female directors are "more risk-averse to fraud and opportunistic earnings management" (Man & Wong, 2013, p. 391) than their male counterparts. In suggesting that women are generally more conservative and less inclined to take excessive risks in banks (Palvia et al., 2014), Sunden and Surette (1998) also reveal that female directors are more risk-averse in making decisions, thus suggesting that female directors are unlikely to risk themselves by committing or colluding in fraud given that the risk of litigation and reputation loss post-fraud are costly (Srinidhi et al., 2011). Srinidhi et al. (2011) demonstrate that the participation of woman directors on boards increases the quality of reported earnings. But, Sun et al. (2010) find no relationship between female directors and earnings management and Berger et al. (2014) indicate that a higher proportion of female board members increases bank portfolio risks. Despite diverse results of risk governance and earnings predictability, we propose a hypothesis following the concepts of agency theory and combining a risk committee, CRO and female directors of banks as a risk governance indicator, as:

H3: Risk governance has a positive effect on earnings predictability
Or-Risk committee, chief risk officer and female directors have positive effect on earnings predictability.

Data
We primarily focus on a sample of large listed commercial banks and BHCs with accounting data available in Bankscope and market data available in Datastream for the period 2004-2010. The sampled banks have been selected from an initial list of 500 publicly quoted large commercial banks and bank holding companies in terms of total assets at the end of the fiscal year 2004 as available in Bankscope.
From this list we retained in the sample only institutions with an independent ownership structure defined by the database (non-independent banks such as subsidiaries can be influenced by the parent).
The remaining banks have to have at least four years of accounting, market and governance data. This criterion gives us the possibility to exploit the panel structure of the dataset, especially to estimate a dynamic panel model, which help minimise potential endogeneity problems (see Wintoki et al., 2012).
The sample selection criterion leads to a sample of 185 banks. Among the 185 banks, six (6) banks are excluded due to data unavailability for earnings predictability (CFO(t+1)) -the key dependent variable of the study. Therefore, the final sample stands at 179 commercial banks and BHCs chartered in 35 countries incorporating a total number of 1015 observations over the period 2004-2010. We present the sample distribution in terms of country and year in Table 1 [Insert Table 1 about here]

Methodology
We adopt a random-effect GLS estimation technique (see Baltagi and Wu, 1999) based on the nature of the panel data sample. There are several reasons for choosing random-effects estimation. First, OLS ignores the panel structure of the data (Gambin, 2004). Second, the time-invariant parameters like CEO characteristics, risk governance proxies, GAAP/IFRS, Big4 cannot be estimated with fixed-effects. Third, the board structure variables (Board Size and Independence) and CEO power (CEO_chair and CEO_internal) do not vary much over time and hence applying fixed-effect estimations would lead to a substantial loss of degrees of freedom (Baltagi, 2005: 14;Wooldridge, 2002: 286) Therefore, GLS is more suitable than OLS and/or fixed effects to overcome potential autocorrelation and heteroskedastic issues from our panel dataset. We test different models with GLS based on the base regression model presented below.
However, since GLS cannot mitigate endogeneity/causality problem between governance and future cash flow variables, two-step system GMM technique are also applied on a few GLS models. System GMM can effectively address endogeneity/causality concerns between the variables used (Arellano and Bond, 1991;Blundell and Bond, 1998;Arellano and Honoré, 2001).

Model Specification
We use the following model to test our hypotheses in line with Cornett et al. (2009), Kanagaretnam et al. (2014) and others in the literature: CFOi,(t+1)=α0+α1* ROIAAi,t +β1* CGi,t + β2* ROIAAi,t * CGi,t +γ*Xi,t+*MEt+εi,t …… (1) Where, CFOi,(t+1) is the cash flow from operations at time t+1 (CFOi,(t+1) is the proxy used to test for earnings predictability). ROIAAi,t is the current profitability (return on operating income by average assets), CG,it is a matrix of firm level corporate governance variables of bank i at time t, ROIAAi,t * CGi,t is the interaction term, Xi,t is a matrix of firm-level control variables of bank i at time t, MEt is a matrix of country level macroeconomic variables at time t, i,t is the error term, α0 is the constant, and , ,  and  are the vectors of coefficient estimates. We use the model to analyse the effect of corporate governance on earnings predictability of banks. All hypotheses are linked to the CG matrix, which includes board structure (board size, board independence), CEO power (CEO-duality, CEO-internal and CEO tenure) and risk governance (risk committee, chief risk officer and female directors) related variables. In addition, IFRS, Big-4 audit firms, 1/Z-score (risk-taking), Tier-1 capital, deposit insurance, log total assets and GDP are used as control variables. The descriptions of these variables are shown in Table 2.

Descriptive Statistics and Correlation Matrix
[Insert Table 2 about here] We present summary statistics for earnings predictability, the corporate and risk governance, and firm and country-level control variables in Table 2 [Insert Table 3

Multivariate Regression Analysis
We begin our multivariate analysis by examining the effect of current earnings, corporate governance and their interactions on CFOt+1 as well as a number of specially selected controls, such as first time that would indicate a 'complementary' link between earnings and governance. In case there is no link between the governance and other variables and CFOt+1, it is of interest to see whether earnings can mediate the relation with CFOt+1 through the interaction terms. In the same vein, our tests extend to select controls to observe whether findings of base estimation models persist.

Corporate governance and earnings predictability
[Insert Table 4 about here] Table 4  In Model 1 board structure variables (board size and independence) are included and we find a significant negative relationship between board size and CFOt+1 illustrating that a small bank board is capable of predicting future cash flows. Conversely, independent boards have a positive significant impact on CFOt+1. Thus, we argue that a small board with more independent directors help predict future bank CFO. These findings remain unchanged in combined base estimation Model 4 as well as in all interaction and control models from Models 5-10, hence, our H1 is supported. These results are consistent with the findings of Al-Dhamari and Ismail (2013), but different from Mashayekhi and Bazaz (2010). Again, in Model 2, we include CEO power variables (CEO-duality, CEO-internal and CEO-tenure) and contrary to expectation, we find a positive relation between CEO-duality and CFOt+1 while CEO-internal and CEOtenure show positive but insignificant relations. The positive sign also extends to Model 4 and other control models (7-10) showing CEO-internal having a significant impact on CFOt+1 but not the CEO-duality and CEO-tenure. These findings do not support our H2, although it is evident that bank CEOs are more aligned to firms' interest rather than pursuing their own interests. It appears that banks where CEOs have role duality and are internally recruited are more likely to provide positive signals to the market to serve as the chief steward for future earnings growth. Again, in Model 3, we incorporate the risk governance variables (risk committee, chief risk officer and female director) and consistent with expectations, we report that both chief risk officer and female director variables are positively associated with CFOt+1, illustrating that they are capable of lowering bank risk, hence predicting future earnings.
This leads us to support marginally our H3, although the signs extend to Model 4 and other control models (7-10) but not the level of significance for any risk management variable. Finally, the combined base estimation Model 4 completely reflects the findings of Model 1 for board structure variables, but shows some variation from Model 2 and Model 3, respectively, for CEO power related and risk governance variables. Another plausible reason is the dominance of Japanese, US and Canadian banks in the sample who follow US GAAP equivalent to IFRS -they represent more than 50% of our sample (320+162+48) observations. However, if we run the regression without Japanese, US and Canadian banks, our IFRS dummy becomes insignificant 6 .

Model 5 presents the effects of the interaction terms of current earnings and governance variables on
Considering the Models 7-10 for the specific controls, respectively, for first time mandatory IFRS adoption in 2005, BHCs, financial crisis (2008)(2009) and Japanese banks, we find no significant relation between CFOt+1 and any of these indicator variables. They rather document identical findings as reported for our combined base estimation Model 4 for current earnings, corporate governance variables and other control variables. The only exception for the special control variables is Model 10 for the Japanese bank dummy which shows a negative significant relation with audit quality (Big-4 audit firms) and firm size (log total assets) variables to CFOt+1, and a positive significant relation between country GDP and CFOt+1.
These findings validate our findings from the base estimation models.
Additionally, in Model 8, covering BHC control, we add an additional variable the volatility in current earnings, which shows a positive influence on CFOt+1. Further tests of this variable in the other 9 models also consistently indicate positive effects on future earnings 7 , so volatility in current earnings is not an issue in determining future earnings. 6 The results are not reported due to space. 7 The results are not reported due to space.

Corporate Governance indices and earnings predictability
[Insert Table 5 about here] To check the validity of our earlier results (Table 4), we construct three corporate governance indices and conduct the regression analysis between the indices and earnings predictability (CFOt+1). We construct strong board, CEO power and strong risk governance based on the corporate governance variables 8 . Table 5 Model 1-3 shows the relationship for governance indices, their interaction variables and combined results, respectively.
In Model 1, as expected, the strong board index reports a positive effect on CFOt+1. This reinforces our findings in Table 4 and supports H1, implying that current earnings and the strong board index variable have sufficient explanatory power to determine future earnings. With regard to the CEO power index we also find a positive significant relation with CFOt+1, as in Table 4, contrary to our anticipation. This supports the argument that powerful bank CEOs tend to signal to the market about their aligned interest and to predict and sustain future earnings. Again, as for the strong risk governance index variable, we report a negative significant influence on CFOt+1, indicating risk governance mechanism are not effective in mitigating bank risk taking and thus hinder future earnings predictability. Such finding is different from that reported in Table 4. Therefore, as mentioned in the previous section, H3 cannot be supported. In Model 2, interaction variables for strong board, strong CEO power and strong risk governance with current earnings are used. Our results appear consistent in that both strong boards and strong CEO power have positive significant effects while strong risk governance has contained negative sign as before but the significant effect on CFOt+1 disappears. In Model 3, we observe qualitatively similar findings.
These are also consistent with the results reported in Table 4. As for control variables we find similar findings in Table 5 as reported in Table 4.

Corporate governance and earnings predictability in different legal jurisdictions (common law vs. civil law)
[Insert Table 6 about here] By following Table 4 Models 4-6, we re-estimate the same regression after splitting the sample based on the legal status of the sample countries. The results are reported in Tables 6 Models 1-3 of Panels A and Models 4-6 of Panel B. In Panel A, Model 1 for civil law countries reveals consistent findings as reported in Table 4 for current earnings, board characteristics, CEO power and risk governance variables. Again, in Model 2, the interactions variables for board characteristics, CEO power and risk governance with earnings show similar results as found in Table 4, except for the female director interaction variable which shows a negative significant relation to CFOt+1. Model 3 also shows qualitatively similar findings with variation in a couple of governance variables. In sum, we document that the results for our civil law countries generally confirm our earlier findings. Considering Table 6 Panel B for the results in common law countries, our findings in Model 1-3 are qualitatively different from Table 4 base estimation models as well as Table 6 Panel A for civil law countries with the exception of a few variables. This leads us to conclude that there are some regulatory differences between common law and civil law countries regarding the effectiveness of governance mechanisms aimed at enhancing bank earnings quality.

Emerging Countries)
[Insert Table 7 about here] In Table 7 Models 1-3 of Panels A and B, we re-estimate the same regression after splitting the sample based on countries' state of economic development (developed vs. emerging) based on the IMF definition, following a similar structure to Table 4 Model 4-6. In Panel A for developed countries, Model 1 reports findings consistent with Table 4 for current earnings, board characteristics, CEO power and risk governance variables. Again, in Model 2, the interactions variables for board characteristics, CEO power and risk governance with earnings show similar results as found in Table 4. Model 3 also shows qualitatively similar findings with variation in a few governance variables. In sum, we conclude that the results for our developed countries generally confirm our earlier findings. Considering Table 7 Panel B for the results for banks based in emerging countries, our findings in Model 1-3 are qualitatively different from the Table 4 base estimation models as well as Table 7 Panel A for developed countries with the exception of a few variables. Thus, we conclude that there are significant regulatory differences between developed and emerging countries that lead to poorer quality earnings that need to be addressed by more effective regulatory and governance mechanisms.

Robustness Checks:
4.2.5.1 Two-step system generalized method of moments (GMM) We use the two-step system GMM approach adopted by Arellano and Bover (1995) and Blundell and Bond (1998) for endogeneity tests in testing Eq. 1. This approach allows us to treat all the explanatory variables as endogenous and orthogonally uses their past values as their respective instruments. It also creates a matching equation of the first differences of all variables and estimates the model via GMM using the lagged values of the right-hand side variables. First differencing eliminates unobserved heterogeneity and omitted variable bias. This approach means that we treat all bank characteristics as endogenous covariates, while treating the country and macro controls as strictly exogenous. System GMM estimates were obtained using Roodman (2009)'s 'xtabond2' module in Stata. Table 8 Models 1-4 are presented in a similar fashion to those in Table 4 base estimations Models 1-4. In Model 1, the board structure variables document significant effect on CFOt+1 as expected, consistent with Table 4. This finding again reinforces our findings in Table 4 and supports H1. With regard to the CEO power related variables in Model 2 we find a positive significant relation to CFOt+1 for CEO_internal.
Again, for our risk governance variables in Model 3, both risk officer and female director appear to have significant negative and positive influences on CFOt+1,respectively. Model 4 also consistently reflects these findings, in addition to the positive effect of CEO_duality on CFOt+1. Consistent with Table 4, these results reconfirm our non-support for H2 while marginally supporting H3. Again, for our control variables we also find similar results as reported in Table 4 with the exception of our risk variable (Z) which has a negative effect on future earnings as per expectation. Table 8 show that the model is well fitted with statistically insignificant test statistics for both second-order autocorrelation in second differences (AR(2)) and the Hansen J-statistics of over identifying restrictions. The residuals in the first difference should be serially correlated (AR(1)) by way of construction but the residuals in the second difference should not be serially correlated (AR(2)). Accordingly, the model fit and diagnostics section in Table 8 shows the desirable statistically significant AR(1) and statistically insignificant AR(2). Likewise, the Hansen J-statistics of over-identifying restrictions tests the null of instrument validity, and the statistically insignificant Hansen J-statistics indicates that the instruments are valid in the two-step system GMM estimation. Overall, the 'system GMM' estimates in Table 8 support that even after controlling for unobserved heterogeneity, simultaneity and dynamic endogeneity, board structure, CEO power and risk governance have a significant influence on the earnings predictability of banks.

The diagnostic tests in
Insert Table 8 about here

Conclusion
We examine the impact of corporate governance on earnings predictability in large banks from 35 countries over the period of 2004-2010 in the light of GAAP/IFRS. High quality accounting numbers are important to banks (as well as other firms) because without accurate accounting information they are unable to precisely estimate their future earnings prospects and valuations. Under GAAP/IFRS frameworks we posit that higher quality reported bank earnings should be a better predictor of future earnings (CFOt+1). Corporate governance mechanisms can also mitigate agency problems and enhance bank performance. Since operating earnings are a key contributor to bank performance, governance mechanisms should be in place to promote disclosure quality and ensure sustained earnings. Effective governance provides positive signals to the market regarding the capacity of the firm to generate sufficient earnings over time, as such good governance should lead to quality disclosure and earnings and improved financial reporting that ultimately protect investors' interests and restrain managerial opportunism.
In particular, we evaluate whether current earnings, board features and CEO power as well as the risk governance features of large banks can effectively predict future cash flows (CFOt+1). Our results consistently suggest that CFOt+1 are positively associated with current earnings. Governance variables relating to board characteristics and interactive terms also generally show high explanatory power in influencing future CFO. However, such power tends to lessen in common law countries and emerging markets. Again, contrary to our expectation and theory, we find CEO power variables enhance the predictability of future bank cash flows (CFOt+1). Overall we find that risk governance variables have little impact on earnings predictability. Overall, these results suggest that a strong GAAP/IFRS based disclosure regime and powerful board as well as CEO power can contribute to determining future bank cash flows. These findings are robust to different sub-samples of country legal status (common law/civil law) and state of economic development (developed/emerging).