Managing specific accruals vs. structuring transactions: Evidence from banking industry

Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

Contents lists available at SciVerse ScienceDirect

Advances in Accounting, incorporating Advances in International Accounting journal homepage: www.elsevier.com/locate/adiac

Managing specific accruals vs. structuring transactions: Evidence from banking industry Xiaoyan Cheng ⁎ College of Business Administration, University of Nebraska–Omaha, Mammel Hall, 67th & Pine Streets, Omaha, NE 68182, United States

a r t i c l e

i n f o

Keywords: Loan loss provisions Gains Securitization and loan sales SFAS No. 140 Earnings benchmarks

a b s t r a c t This study investigates earnings management through managing specific accruals vs. structuring transactions in the banking industry. This paper explores the circumstances under which banks manipulate loan loss provisions vs. circumstances that lead banks to structure loan sales and securitizations for the purpose of achieving earnings benchmarks. Empirical results show that banks manage earnings through loan loss provisions, before resorting to structuring transactions, to avoid small earnings decreases and or just meet or beat analysts' forecasts. The findings imply that structuring loan sales and securitizations is more likely to be used as a secondary instrument. In addition, I find that the earnings of banks with lower discretionary loan loss provisions and higher discretionary gains from loan sales and securitizations are priced more negatively, suggesting that investors impose incremental penalties on the joint use of loan loss provisions and gains from loan sales and securitization to meet or beat earnings benchmarks. © 2012 Elsevier Ltd. All rights reserved.

1. Introduction Existing literature has consistently documented evidence that companies manage earnings through manipulating specific accruals. Examples abound where companies adjust accounting estimates, such as bad debt expense (McNichols & Wilson, 1988), loan loss provisions in the banking industry (Beatty, Chamberlain, & Magliolo, 1995; Beatty, Ke, & Petroni, 2002; Kanagaretnam, Lobo, & Mathieu, 2003), and insurance loss estimates in the insurance industry (Adiel, 1996; Beaver, McNichols, & Nelson, 2003), to manage earnings. Alternatively, companies can manipulate earnings through real activities manipulation (Roychowdhury, 2006). For example, companies structure issues of contingent convertible bond to manage diluted earnings per share (Marquardt & Wiedman, 2005). In addition, gains from securitizations have been used to manage earnings (Dechow, Myers, & Shakespeare, 2008; Karaoglu, 2005; Niu & Richardson, 2004).1 A recent article in New York Times cites industry specialists who describe securitizations as “the thing about gain on sale

⁎ Tel.: + 1 402 554 3139; fax: + 1 402 554 3747. E-mail address: [email protected]. 1 Securitization is the process of transferring loans to qualifying special purpose entities through the issuance of debt. A unique feature of securitization is that the issuer receives cash instantly from outside investors and pays back this obligation when the securitized financial assets are collected. Securitization has several advantages over traditional bank financing. First, securitization enables banks to transfer illiquid assets to third parties and therefore firms obtain cash flows without waiting for customers to pay. Second, securitization is viewed as off-balance sheet financing, which favorably affects leverage ratio. 0882-6110/$ – see front matter © 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.adiac.2012.02.001

accounting is that you can create a machine that just manufactures earnings out of thin air”. Although previous studies suggest that companies may have opportunities to influence earnings via the two available instruments – managing specific accruals and structuring transactions, no study to date has explicitly investigated how managers trade off real and accrual manipulations in a regulated industry – the banking industry. I fill this gap in the literature. Graham, Harvey, and Rajgopal (2005), Zang (2007), and Cohen and Zarowin (2010) examine the relation between accrual-based and real earnings management in non-banking sector.2 The motivations to manipulate earnings may differ in the regulated industry. For instance, bank regulators establish the minimum capital requirements and intervene in the operations of banks with inadequate capital (Beatty et al., 1995). Furthermore, banking companies provide a fertile ground for research to isolate the specific accruals from structuring transactions. Focusing on a single regulated industry enables me to better measure earnings management activities. This research design helps me compare the differential costs and benefits associated with each instrument, thus providing insight into how earnings management is achieved. Following prior research (Beatty et al., 1995; Beatty et al., 2002), I use discretionary loan loss provisions as a proxy for managing specific accruals, given that loan loss provisions are a very important accrual

2 Zang (2007) and Cohen and Zarowin (2010) provide evidence suggesting that managers prefer real activities manipulation compared to accrual-based earnings management. This is because real earnings management is less likely to draw auditor or regulatory scrutiny, and therefore has a higher probability of being undetected.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

for banks. Discretionary gains or losses (gains, hereinafter) from loan sales and securitizations are used as a proxy for structuring transactions at least for two reasons.3 First, managers have high level of judgment and discretion in reporting gains on loan sales and securitizations. 4 Second, loan sales and securitizations are of particular interest, because they have expanded dramatically over the last decade. Securitizations give banks significant opportunities to affect accounting outcomes, as loans constitute over half of the balance sheet in commercial banks, and securities resulting from securitizations were the largest segment in debt market in 2005, totaling $7.4 trillion (Dechow et al., 2008). 5 I build on previous research (Cohen & Zarowin, 2010; Zang, 2007) by examining how and when bank managers use loan loss provisions and securitizations to manage reported earnings based on the respective costs and benefits. I argue managers' choice of each instrument is a function of the firms' ability to use the instrument and the costs of doing so. As compared to managing earnings through specific accruals, structuring transactions such as securitizations is more complex. Securitization is a timing consuming process and incurs higher direct costs. Therefore, managing earnings through accruals provides a cheaper way to influence financial statements. However, managing loss loan provisions has its own costs. A reduction on loan loss provisions increases reported earnings, but it may signal poor future prospects or even result in a violation of regulatory standards (Kanagaretnam et al., 2003). The conflicting directions arising from multiple objectives restrict the use of loan loss provisions. Managers are more likely to utilize loan loss provisions to meet earnings targets if the magnitude of earnings manipulation is close to the limit of loan loss provisions. This study I focus my attention on firms that slightly meet earnings targets, since these firms have a strong motivation to manage earnings. In particular, firms just meeting earnings targets have intrinsic earnings that are close to earnings targets (Lee, 2007) and therefore they have the ability to manage earnings by picking a cheaper instrument first. Consistent with the predictions, I find that banks use loan loss provisions, before resorting to structuring transactions, to avoid small earnings decreases and to just meet or beat analysts' forecasts. Specifically, gains from loan sales and securitizations are only used to manage earnings when pre-securitization earnings (i.e., earnings before the gain on loan sale and securitization) are lower than prior year's level and when pre-securitization earnings per share are less than analysts' forecasts. The findings of this study suggest that structuring transactions are more likely to be used in specific circumstances as a secondary tool. Additional analysis indicates that the results are robust after controlling for the simultaneity of discretion choices. The second part of this study is to examine how investors in the market react to earnings management when banks disclose loan loss provisions and loan sales and securitizations. Previous research (Niu & Richardson, 2004; Wahlen, 1994) has examined the financial reporting consequences of these two instruments independent of 3 Besides loan sales and securitization, I assert that banks structure other transactions for earnings management. For instance, Beatty et al. (1995) documents that gains on sale of securities, gains on sale of fixed assets, and gains on pension settlement also can be used for earnings manipulation. As compared to securities sales and fixed asset sales, securitization can be more complicated and/or more costly to structure. For example, securities sales typically do not involve setting up qualifying special purpose entities. 4 In a typical securitization, firms retain some interest in the receivables in order to obtain the desired credit rating. Thus, less than 100% of the receivables are sold. The accounting rule SFAS140 requires the retained interest to be recorded at the fair value. In contrast to securitization, loan sales are the whole “pure” assets sales without any future involvement by the transferor. In loan sales, managers exercise discretion over the timing and selection of loans to be sold. But in the case of securitizations, managers have additional discretion in reporting gains from retained interests due to the fair value treatment (Karaoglu, 2005). As a result, the impact on financial reporting is higher for securitizations than for loan sales. 5 Specifically, Dechow et al. (2008) find that 13% of their sample firms report securitization gains, and the magnitude of securitization gains is sufficient to convert an accounting loss into a profit.

23

each other. There is no evidence on how the joint use of these two instruments affects stock prices. A natural follow-up question is to investigate the incremental economic consequences of earnings management when both of these two instruments are used to meet earnings targets. Consistent with the hypotheses, I find that earnings with lower discretionary loan loss provisions and higher discretionary gains from loan sales and securitizations are priced more negatively. The results suggest that investors appear to recognize the tools for achieving earnings benchmarks and appropriately impose an incremental penalty when loan loss provisions and securitization gains are used simultaneously. In sum, this paper provides empirical evidence that investors may see through earnings management in the banking industry. A potential explanation is that, as a result of bank regulation, investors have access to extensive disclosures that are related to specific accruals and securitizations (Healy & Wahlen, 1999). 6 This paper makes several contributions to the current literature. First, this study extends the literature by examining how managers use multiple tools to manage earnings. It is one of the first to investigate how managers trade off real vs. accrual management in banking industry. Investigation of management manipulation using specific accruals vs. structuring transactions is an extension of earnings management research. This extension is important because banks provide a natural setting to isolate loan loss provisions from loan sales and securitization gains, thus enabling me to develop more reliable measures of managerial discretion over earnings. The findings of this study help us understand the rationale for managers' choices of these instruments to avoid small earnings declines and to just meet or beat analysts' forecasts. Second, this paper adds to a growing body of research that documents the capital market consequences of earnings management. My results suggest that market participants punish firms to a greater extent when managing specific accruals and structuring transactions are used simultaneously. In this sense, the findings of this study have implications in making investment decisions. In addition, evidence on market efficiency with the joint use of these two instruments is likely to be of interest to standard setters. 7 Finally, this paper presents evidence on how managers use the fair value accounting under SFAS140 to structure transactions in meeting earnings benchmarks, thus providing insights into the attributes that relate to the recent subprime crisis. 8 The remainder of this paper is organized as follows. Section 2 reviews the literature and develops the hypotheses. Section 3 introduces the empirical models. Section 4 describes the samples. Section 5 presents the results, and Section 6 summarizes the paper.

6 Such explanation is consistent to the theory that the key element for market discipline to play depends on information that is observable (Flannery, 2011). The ability of investors to interpret information on bank behavior is enhanced when extensive disclosures of loan loss provisions and securitizations are provided. 7 Healy and Wahlen (1999) review the academic literature on earnings management and indicate that standard setters are interested in a set of questions, including which specific accruals are used to manipulate earnings, the magnitude and frequency of earnings management, and the economic consequences of earnings management. Evidence in addressing these questions would help accounting regulators determine the need for financial standards to be revised. 8 The recent financial crisis has led to the debate about the use of fair value accounting rule SFAS140, which applies to financial instruments retained (retained interest) after a securitization. Critics argue that fair value accounting has significantly contributed to the financial crisis, given that it is difficult to evaluate the reasonableness and accuracy of the reported securitization gains and thus managers may take the advantage of using fair value to achieve desired accounting outcomes. Proponents argue that fair value increases the transparency of financial reporting and the undesirable consequences of fair value are driven by market inefficiencies, investor irrationality, or unreliable fair value model (Dechow et al., 2008). A recent study by Barth and Landsman (2010) indicates that although fair value accounting plays little role in the financial crisis, financial institutions need to improve the disclosures relating to asset securitizations. SFAS166, Accounting for Transfers of Financial Assets, and SFAS 167, Amendments to FASB Interpretation No. 46R were passed in 2010 to address the limitations of fair value accounting.

24

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

2. Hypotheses development Avoiding earnings decreases and meeting or beating analysts' forecasts strongly motivate companies to manage earnings (Brown & Caylor, 2005; Burgstahler & Dichev, 1997; Degeorge, Patel, & Zeckhauser, 1999; Robb, 1998). Prior research suggests that firms have incentives to beat earnings benchmarks, because investors and creditors use earnings benchmarks to evaluate firm performance. Barth, Elliott, and Finn (1999) find that firms reporting continuous growth in earnings are rewarded at a premium price. Bartov, Givoly, and Hayn (2002) provides evidence that firms meeting or beating analysts' forecasts have higher earnings response coefficients than those missing analysts' forecasts. Skinner and Sloan (2002) show that firms failing to meet analyst expectations suffer “disproportionately large stock price declines” on the earnings announcement date. Boards of directors also rely on earnings benchmark to evaluate CEO performance. Matsunaga and Park (2001) find that failure to meet analysts' forecasts results in significant bonus cuts for the CEO. Overall, evidence from past research indicates that capital market pressures and institutional structures motivate mangers to manipulate earnings. Managing loan loss provisions and structuring securitizations provide two possible ways that managers can choose from. In addition to managing loan loss provisions and structuring securitizations, managers may use forecast guidance and classification shifting to achieve earnings benchmarks. 9 This study, I examine managers' discretion over reported earnings through manipulating specific accruals and structuring transactions. The reasons are as follows. First and most importantly, I investigate the research questions using bank holding companies (banks, hereinafter), because banks provide a fertile ground for research to isolate specific accruals from loan transfers and sales. 10 The unique accruals and structuring transactions in the banking industry enable me to develop more reliable measures of managerial discretion over earnings, thus mitigating measurement bias in testing managerial discretion over earnings (Cheng et al., 2011). 11 Second, past literature documents the potential limitations associated with classification shifting and forecast guidance. Specifically, Barua and Cready (2008) provide a competing argument that McVay (2006)'s classification shifting findings are based on a misspecified model, thus the evidence of classification shifting may be nothing. Consistent with Barua and Cready (2008)'s argument, Fan and Thomas (2011) do not find evidence that firms use classification shifting to just meet or beat analysts' forecasts or prior year earnings. In the case of forecast guidance, the SEC passed Regulation Fair Disclosure in October 2000, which has increased the cost of downward guidance. Recent studies on public policies (Koh, Matsumoto, & Rajgopal, 2008; Li, Rider, & Moore, 2009) find that the use of forecast guidance is decreasing in the post-Regulation period. 2.1. Managing earnings through loan loss provisions According to FAS No. 5 (FASB, 1975), Accounting for Contingencies, when the credit losses are probable and the amount can be reasonably estimated, banks should record an expense called the provision for loan losses and a contra-asset called the allowance for loan losses. In

9 Matsumoto (2002) documents that firms manage guided analysts' forecasts downward to just meet analyst expectations, a phenomenon often referred to as expectation management. Moreover, firms may engage in classification shifting to improve core earnings and hit the analysts' forecasts (McVay, 2006). 10 While it is appealing to compare the two types of earnings management activities in other industries, it is difficult to isolate the management activities in empirical analysis. 11 The limitation of using loan loss provisions is that it has been restricted to the banking industry. As a result, the findings from studying a specific accrual approach may not be generalizable to other industries where managers exercise discretion on several accruals and therefore the total accrual approach is more likely to detect earnings management.

the banking industry, loan loss provisions are one of the most discretionary accounting choices. Studies have shown that loan loss provisions are used for managing earnings (Adiel, 1996; Beatty et al., 2002; Beaver et al., 2003; McNichols & Wilson, 1988). Collins, Shackelford, and Wahlen (1995) documents similar evidence that banks report lower loan loss provisions in years with low nondiscretionary earnings. Liu and Ryan (2006) find further evidence that banks managed earnings downward by overstating loan loss provisions for homogenous loans during the boom period of the 1990s. In addition, Beatty et al. (2002) report that public banks are more likely to use loan loss provisions to avoid earnings decreases than are private banks. Overall, prior research has shown that loan loss provisions are a vehicle of earnings management. Bank managers exercise discretion over loan loss provisions for the purpose of earnings smoothing and for meeting or beating benchmarks. 12 2.2. Managing earnings through structuring transactions Securitization of assets has been used widely in recent years as a means of off-balance sheet financing. In a recent study, Dechow et al. (2008) document evidence that the gains from securitizations recognized under FASB, Financial Accounting Standard Board (2000), Accounting for Transfers and Servicing of Financial Assets and Extinguishments of Liabilities - a replacement of FASB Statement No. 125, are used to influence earnings. Appendix A illustrates the accounting regulation of securitization and treatment options. Dechow et al. (2008) find that derecognition, a removal of financial assets from the balance sheet, allows managers to exercise discretions to report gains, because gains are determined as the difference between fair value and book value of the receivables sold. The results from Dechow et al. (2008)'s paper indicate that the accounting rules under SFAS140 for securitizations provide ample opportunity for managers to manipulate earnings. Karaoglu (2005) documents similar evidence from the banking industry that gains from loan sales and securitizations recognized under FASB, Financial Accounting Standard Board (1996), Accounting for Transfers and Servicing of Financial Assets and Extinguishments of Liabilities, are used to manage regulatory capital and to meet or beat earnings benchmarks. Together, these studies document consistent evidence that gains from assets securitizations are subjective — bank managers may use loans for sales or for securitizations for accounting benefits. 2.3. Managing earnings through loan loss provisions vs. loan sales and securitizations Managing loan loss provisions and structuring transactions, such as loan sales and securitizations, provide two viable instruments that bank managers can exercise. However, these two instruments differ from each other in terms of implementation costs and potential ability to influence earnings.13 As compared to managing specific accruals, 12 While the evidence that banks abuse loan loss provisions to manage earnings is compelling, some studies have failed to document similar evidence (e.g., Ahmed et al., 1999; Beatty et al., 1995). 13 Marquardt and Wiedman (2004) classify the costs of managing specific accruals into two parts: detected and undetected costs. Costs incur when earnings management is detected through SEC enforcement actions, earnings restatements, shareholder litigation, qualified audit reports, or negative coverage in the press. Costs of undetected earnings management include reversal of accruals, constraints on future reporting flexibility, audit costs, and perceived earnings quality. Loan sales and securitizations may also have such similar detected and undetected costs. Dickson (2011) documents that a wave of securitization litigation against the financial institutions is becoming increasingly, given the fact that the financial parties failed to comply with regulations and fundamental requirements of property law. The negative coverage in the business press on securitization process has gained increased prominence. Cox, Faucette, and Lickstein (2010) indicate that securitization was criticized during financial crisis, because financial statement users were unable to assess the reasonableness of the reported gain. Moreover, Dechow et al. (2008) suggest that boosting securitization gains has the potential undetected cost of a future writedown, thus limiting future reporting flexibility. In this study I focus on the differential costs between managing specific accruals and structuring transactions.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

structuring transactions is more complex. Securitizations enable the issuer to transfer non-collection risk to a large number of investors, thus diversifying risk and receiving cash instantly. Furthermore, securitizations are viewed as off-balance sheet financing, which favorably affects leverage ratio given that securities are not treated as debt. The process of securitizations, however, is time consuming and incurs higher direct costs. For example, securitizations involve legal and filing fees for the documentation, investment banking fees, accounting and consulting fees, and other fees associated with the credit rating agency and special purpose entity. As a result, securitizations less than $100 million are uncommon (Minton, Opler, & Stanton, 1997). Loan sales may not be as costly as securitizations, but maintaining qualifying special purpose entities may not be cheap. Because of the fixed costs associated with securitizations, many Wall Street firms have incentives to increase the volume of asset-backed security sales.14 Due to such significant direct and indirect costs, managers time securitization transactions to obtain accounting benefits. Dechow and Shakespeare (2009) find that managers wait until the last few days of the quarter to engage in securitization, since at the end of the quarter, manages know how much earnings are needed to meet analysts' forecasts.15 Given the higher implementation costs associated with structuring loan sales and securitizations, managing earnings through loan loss provisions provides a cheaper way to influence income statement and balance sheet. However, the impact of loan loss provision on earnings is limited to some extent, because further reduction on loan loss provisions may violate regulatory standards or signal poor future prospects. Previous research (Ahmed, Takeda, & Thomas, 1999; Kanagaretnam et al., 2003; Karaoglu, 2005) finds that, in addition to earnings manipulation, bank managers also have incentive to use loan loss provisions to manage regulatory capital as well as communicate or signal private information about future earnings. The potential interaction between alternative motivations limits the use of loan loss provisions. The logic is as follows: first, a reduction in loan loss provisions increases current reported income; however, it restricts bank managers to utilize loan loss provisions to meet regulatory capital. 16 The costly regulatory intervention motivates bank managers to manage capital. Under the new capital requirement in 1991, total regulatory capital must exceed 8% of risk-weighted assets. 17 Total regulatory capital is divided into tier I and tier II capital. Loan loss reserves are included as Tier II only up to 1.25% of risk-adjusted assets. 18 Although loan loss reserves make a limited contribution to Tier II capital, prior literature (Ahmed et al., 1999; Kim & Kross, 1998) documents that bank managers increase loan loss provisions to meet regulatory capital, especially when regulatory capital is low. Second, loan loss provisions can be used to convey managers' private information about future prospects of banks. Beaver, Eder, Ryan, and Wolfson (1989) suggests that an increase in loan loss provisions reflects the attitude that “management perceives the earnings power of the bank to be sufficiently strong that it can withstand a ‘hit to earnings’ 14 See “FBI probes accounting in subprime securitization” in Financial Week, January 30, 2008. 15 In addition, Moyer (1990) finds that managers exercise discretion over the timing of loan loss provisions to meet the regulatory capital. As past research (e.g., Dechow & Shakespeare, 2009; Moyer, 1990) documents managers opportunistically time loan loss provisions and securitization transactions, an independent investigation of managers' preference over the two instruments is important. 16 A dollar decrease in loan loss provisions increases net income after tax by a dollar minus tax rate. 17 Federal Deposit Insurance Corporation Improvement Act of 1991introduced the implementation of new capital standards in 1991. Total risk-weighted assets are defined as the sum of balance sheet assets and off-balance-sheet activities, weighted according to designed risk levels. 18 Regulatory capital is the sum of shareholders' equity, equity-debt hybrid instruments, and loan loss allowances, less goodwill and other intangibles. Tier I capital is defined as the sum of book value of equity, non-cumulative perpetual preferred stock, minority interest in equity minus goodwill and other intangible assets. Tier II capital is the sum of loan loss reserves, hybrid capital instruments, perpetual debt, and convertible debts.

25

in the form of additional loan loss provisions.” Implicit in the reasoning is that an increase in loan loss provisions is a way to convey financial strength, even though increasing in loan loss provisions reduces current period earnings. Consistent with this view, a reduction in loan loss provisions may signal bad news about future prospects. Taken together, the conflicting directions arising from multiple motivations restrict bank managers' ability to use loan loss provisions, thus it is possible to have an upper bound on the total amount of discretionary loan loss provisions that a manager can exercise. Although both loan loss provisions and gains from structuring loan transfers can be used for meeting earnings benchmarks, banks may prefer to pick the instrument. I argue such decision-making depends on the respective costs and benefits of each instrument. Managing earnings through loan loss provisions costs much less, but the potential impact on earnings is limited to some extent. Therefore, managers are more likely to use loan loss provisions to meet or beat earnings targets if the magnitude of earnings manipulation is close to the limit of loan loss provisions. Based on the line of reasoning, I predict that bank managers prefer to use loan loss provisions to avoid small earnings decreases or to just meet or beat analysts' forecasts, before resorting to costlier methods, such as structuring transactions. This argument leads to the first hypothesis, stated in alternative form: H1. Banks manage earnings through loan loss provisions before structuring transactions to avoid small earnings decreases or to just meet or beat analysts' forecasts. 2.4. Market reactions to the disclosure of loan loss provisions and gains from loan sale and securitization Signaling theory suggests that discretionary loan loss provisions reveal the private information of bank managers. The primary motivation for managers to use their discretion in reporting unexpected loan loss provisions is to signal future performance, thus reducing information asymmetry (Akerlof, 1970). The credibility of a signal is related to its cost. Poor performing bank managers have little incentives to engage in false signaling through increasing loan loss provisions, given that the cost of a false signal is greater for banks that are not expected to perform well. Banks with poor performance will report even lower earnings if they increase loan loss provisions, which in turn increases the chance being audited by regulatory agencies (Kanagaretnam et al., 2003). Several studies (Barth, Beaver, & Stinson, 1991; Liu & Ryan, 1995; Wahlen, 1994) investigate market reactions to unexpected loan loss provisions. Beaver et al. (1989) finds that investors interpret an increase in loan loss provisions as a sign of strength, since better financial positions let managers make a generous allowance for loan losses. Similarly, Wahlen (1994) documents a positive relation between discretionary loan loss provisions and stock returns and future cash flows, suggesting that discretionary loan loss provisions predict future cash flows. Liu and Ryan (1995) show market responses to discretionary and nondiscretionary components of loan loss provision are positive and negative, respectively. The reputation effects (Milgrom & Roberts, 1982) also explain why discretionary loan loss provisions are positively priced. Healy and Palepu (1993) develop an analytical model to show that firms with better reputation are more likely to disclose higher loan loss provisions. Consistent with signaling and reputation theories, I expect lower than expected loan loss provisions to be priced negatively. In the case of structuring transactions, investors view the reliability of reported gains from securitizations as being low. There are several reasons. First, transactions structured for earnings benchmarks may lack economic substance. Earnings with greater components from gains of structured transactions without economic substance will be poor predictors of future earnings. Consequently, Niu and Richardson (2004) find that gains from securitizations are less

26

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

informative than other components of earnings and thus receive a lower pricing multiple. Second, as noted by Dechow et al. (2008), managers can make legitimate “errors” in their assumptions, such as discount rate and prepayment rate, to determine securitization gains, and it is difficult for outside users to evaluate the reasonableness of such assumption either ex ante or ex post. Moreover, Karaoglu (2005) finds that securitization gains are attributed both to cherry-picking of loans whose market values exceed their book values and to overvaluation of the retained interests on the balance sheet.19 These results suggest that managers use discretion to determine reported securitization gains in an opportunistic manner. A recent study (Michalak & Uhde, 2009) also highlights the existence of market discipline, as documented by a downgrade of asset-backed security and an increase of systematic risk at issuing banks. Loan sales differ from securitizations in that loan sales are whole asset sales without retained interest. Banks preferentially sell loans of low quality borrowers based on negative private information that is unobservable to investors, thus resulting in moral hazard and adverse selection problems (Berndt & Gupta, 2009).20 Based on the line of reasoning, I predict that higher than expected loan sales and securitization gains are viewed negatively by investors. Taken together, a joint use of loan loss provisions and securitization gains to achieve earnings benchmarks is priced more negatively by investors, compared to only loan loss provisions or gains manipulated from loan sales and securitizations. This argument leads to the second hypothesis: H2. Earnings with lower discretionary loan loss provisions and higher discretionary gains from loan sales and securitizations are priced more negatively. 3. Research design 3.1. Measures of discretionary loan loss provision I use the following regression model, similar to that of Beatty et al. (1995), to estimate the nondiscretionary portion of the loan loss provisions. Specifically, LLP

¼ α þ β1 SIZE þ β2 Δ NP LOAN þ β3 LOAN IND þ β4 LOAN FG þ B5 LOAN DEP þ β6 LOAN COM þ β7 LOAN AGR þ β8 LOAN RST þ β9 LLR þ β10 LCO þ ε

LOAN_COM commercial and industrial loans scaled by total assets at quarter q-4; LOAN_AGR agricultural loans scaled by total assets at quarter q-4; LOAN_RST loans secured by real estate scaled by total assets at quarter q-4; LLR loan loss reserve scaled by total assets at quarter q-4; LCO net loan charge-offs scaled by total assets at quarter q-4. The expectation of the nondiscretionary loan loss provisions is regressed over the pooled time-series cross-sectional sample. The amount of net loan charge-offs (LCO) is related to loan loss provisions by construction. Loans are charged off when they are deemed uncollectible. Net loan charge-offs (uncollectible loans net of any expected recovery) reflect the expectations of the collectibility of current loans. Change of non-performing loans (ΔNP_LOAN) is a proxy for change of default risk. LLR is the beginning balance of loan loss reserve. Following prior research, I also control for loan characteristics and regional effects, such as loans to individuals (LOAN_IND), loans to foreign governments (LOAN_FG), loans to depository institutions (LOAN_DEP), commercial and industrial loans (LOAN_COM), agricultural loans (LOAN_AGR), and loans secured by real estate (LOAN_RST). The coefficients on net loan charge-offs and change in non-performing loans are expected to be positive because loan loss provisions are larger when the expectations of the uncollectibility of current loans is higher and when there is an increase in nonperforming loans. I do not make the predictions on the coefficient of loan loss reserve due to the mixed results documented by prior research. Beatty et al. (1995) finds that loan loss provision is positively associated with the amount of the loan loss reserve at the beginning of the year. Conversely, Kanagaretnam et al. (2003) shows that loan loss reserve has a negative effect on loan loss provision, because a large initial reserve will result in a lower provision in the current period.21 The residuals from Eq. (1) represent the discretional components of loan loss provisions. 3.2. Measures of discretionary gains from loan sales and securitizations Consistent with Beatty et al. (2002), I adopt a simple approach to estimate the nondiscretionary portion of gains from loans sales and securitizations. GAIN ¼ α þ β1 SIZE þ β2 UN GL þ ε

ð2Þ

ð1Þ where: LLP

loan loss provisions in quarter q scaled by total assets at q-4; SIZE natural log of total assets in $000; ΔNP_LOAN change of non-performing loans in quarter q relative to q-4, scaled by total assets at q-4; LOAN_IND loans to individuals scaled by total assets at quarter q-4; LOAN_FG loans to foreign governments scaled by total assets at quarter q-4; LOAN_DEP loans to depository institutions and acceptances of other banks scaled by total assets at quarter q-4;

19 The book value consists of two parts: the carry value of components sold and the retained interest kept by firms. In a typical securitization, firms retain some interest in the receivables. The accounting rule SFAS140 requires the retained interest to be recorded at fair value. However, limited guidance is provided on how to calculate fair value and what assumptions (such as discount rate, prepayment rates, and the likelihood of default) should be used (Dechow et al., 2008). 20 Because of the negative signal sent by loan sales, Berndt and Gupta (2009) proposes to impose restrictions on loan sales. In particular, banks need to disclose more details about the loans being traded in the secondary market and retain a proportion of loans on balance sheet in order to reduce the adverse selection problem.

Where: GAIN SIZE UN_GL

gains (or) losses from loan sales and securitizations in quarter q divided by total assets at q-4; natural log of total assets at quarter q; reported unrealized security gains and losses at the beginning of the quarter divided by total assets at quarter q-4.

Similar to Eq. (1), the above model is implemented over the pooled time-series cross-sectional sample. The coefficient on reported unrealized security gains and losses is expected to be positive, given that realized security gains and losses are higher when there are higher unrealized security gains and losses. The residuals from Eq. (2) are the proxy of discretionary gains from loan sales and securitizations. 3.3. Testing Hypothesis 1 Hypothesis 1 tests whether banks have a preference for managing loan loss provisions over structuring from loan sales and securitizations for a gain. Next, I create two scenarios in which I expect the incentives of managing loan loss provisions vs. structuring transactions 21

Following Beatty et al. (1995), I scale all variables by beginning total assets.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

would differ – those relying on discretionary gains from loan sales and securitizations to achieve earnings benchmarks and those not. 3.3.1. Scenario 1 The first scenario tests whether bank managers use loan loss provisions and securitization gains to avoid small earnings decreases or just meet/beat analysts' forecasts when small earnings changes are larger than gains from loan sales and securitizations. In this scenario, bank managers have little incentives to structure loan sales and securitizations because gains from loan sales and securitizations become quite insignificant. In other words, earnings benchmarks have been met before the gains from loan sales and securitizations are used. Following Degeorge et al. (1999), I define small earnings surprise as current quarter earnings greater than earnings in quarter q-4 by less than 0.2% of total assets in quarter q-4. I use earnings from the same quarter in the prior year as a proxy for earnings benchmark.22 Just meeting or beating earnings benchmarks is a commonly used proxy for earnings management.23 Following prior research (Harris, Shi, & Xie, 2010; McVay, 2006), I define just meeting/beating analysts' forecasts as earnings meeting or beating analysts' forecasts by zero or one penny. MEET1 and FMEET1 are the threshold variables under this scenario. The following equations are used to test H1 in this scenario. AB LLP ¼ α þ β11 Threshold þ β12 ΔROA þ β13 HOM þ β14 ROA þ β15 CAP þ β16 GDP þ ε

ð3aÞ

AB GAIN ¼ α þ λ11 Threshold þ λ12 ΔROA þ λ13 HOM þ λ14 ROA þ λ15 CAP þ λ16 GDP þ ε

ð3bÞ

27

the motivations of using loan loss provisions and gains from securitizations and loan sales to satisfy regulatory capital, signal future prospects, and smooth earnings, respectively. 24 Homogenous loans (HOM) is included, as banks holding more homogenous loans manage loan loss provisions to smooth reported earnings (Liu & Ryan, 2006). In addition, HOM is used to control for the characteristics of securitizations, given that homogenous loans are the first and the most important type of securitized financial assets (Chen, Liu, & Ryan, 2007). I use growth in GDP to control for the effect of pro-cyclicality on loan loss provisions, as banks increase their loan loss provisions when an economy faces a downturn while decrease their provisions when an economy falls in upswings (Borio, Furfine, & Lowe, 2001). Variable GDP is also included in Eq. (3b) to control for the macroeconomic effects on securitizations. Given that managers have little incentives to inflate earnings through loan sales and securitizations under this scenario, the estimate on λ11 in Eq. (3b) is predicted to be insignificant or negative.25 3.3.2. Scenario 2 The second scenario is that earnings exceed the prior year's level or meet/beat analysts' target, but small earnings surprises are lower than gains from loan sales and securitizations. In this scenario, gains from loan sales and securitizations become significant. Therefore, banks are motivated to structure loan sales and securitizations because they will be unable to report small positive earnings growth without the gains from loan sales and securitizations. In the case of analysts' forecasts, earnings per share meet or beat analysts' forecasts by zero or one penny, but earnings per share before gains from loan sales and securitizations miss analysts' forecasts. The threshold variables under this scenario are defined as follows: 1 if 0 b = INq-INq-4 b 0.2%*ASSETSq-4 and INq − INq-4 b GAINq, and 0 otherwise; GAIN represents gains from loan sales and securitizations; FMEET2 1 if earnings meet or beat analysts' forecast by zero or one penny and earnings before discretionary gains from securitization and loan sales miss analysts forecast, and 0 otherwise.

MEET2 Where: AB_LLP discretionary loan loss provisions estimated from Eq. (1); AB_GAIN discretionary gains from loan sales and securitizations estimated from Eq. (2); MEET1 1 if 0 b = INq-INq-4 b 0.2%*ASSETSq-4 and INq − INq-4 > GAINq, and 0 otherwise; GAIN represents gains from loan sales and securitizations; FMEET1 1 if earnings meet or beat analysts' forecast by zero or one penny and earnings before securitization gains meet or beat analysts forecast, and 0 otherwise; ΔROA one-year ahead changes in earnings before loan loss provisions and taxes, scaled by total assets at quarter q-4; HOM % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP total capital ratio; GDP growth in GDP per capita (annual %).

It is expected that the coefficients on MEET2 and FMEET2 in Eq. (3b) are positive, suggesting that bank managers use loan sales and securitizations to report higher earnings. Moreover, the coefficients on MEET2 and FMEET2 in Eq. (3a) are expected to be negative, which means that loan loss provisions are used to manage earnings for benchmarks even when incentives for managing gains from loan sales and securitizations are present. H1 is considered to be supported if loan loss provisions are managed in both of the two scenarios, while gains from loan sales and securitizations are managed in Scenario 2 when the incentives to structure transactions are present. 3.4. Testing Hypothesis 2

Following prior research (Ahmed et al., 1999; Collins et al., 1995, Karaoglu, 2005), I use total capital ratio (CAP), one-year ahead changes in earnings before loan loss provisions and taxes (ΔROA), and earnings before loan loss provisions and taxes (ROA) to proxy

I use Wahlen (1994)'s model to test market reactions to the disclosure of loan loss and securitization gains information. Quarterly earnings (EARN) and changes in quarterly earnings (ΔEARN) are included to capture earnings levels and earnings surprises that are released during the event window. The variables unexpected change in non-performing loans (UΔNP_LOAN) and unexpected loan chargeoffs (ULCO) are used as proxies for the other new information in loan

22 Graham et al. (2005)'s survey results indicate that managers believe last year earnings are an important benchmark. 23 Persuasive evidence has been documented by prior research that earnings are likely managed when firms just meet or beat analyst forecast. McVay (2006) finds that firms use accounting choices such as manipulating tax expense and exercising discretion over income statement classification to meet analyst forecast.

24 Costly regulatory monitoring motivates managers to use loan sales and securitization influencing earnings. Karaoglu (2005) finds that gains from loan sales and securitizations are negatively associated with regulatory capital. 25 The negative coefficients are predicted, given that banks may smooth earnings through securitizations by recognizing gains (losses) when pre-securitization earnings are lower (higher).

28

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

loss disclosure. 26 The coefficients on UΔNP_LOAN and ULCO are expected to be negative, as UΔNP_LOAN and ULCO reflect bad news in the disclosure of loan loss. AB RET ¼ λ0 þ λ1 EARN þ λ2 ΔEARN þ λ3 DLLP þ λ4 DGAIN þ λ5 BOTH þ λ6 UΔNPLOAN þ λ7 ULCO þ ε

ð6Þ

Where: AB_RET

three day (− 1, 0, and 1) cumulative abnormal returns where day 0 is the quarterly earnings announcement date; EARN earnings before loan loss provisions, gains from loan sales and securitizations, and taxes, scaled by total assets at the beginning of quarter q; ΔEARN changes in earnings before loan loss provisions, gains from loan sales and securitizations, and taxes, scaled by total assets at the beginning of quarter q; DLLP 1 if discretionary loan loss provisions at quarter q is less than the median, and 0 otherwise; DGAIN 1 if discretionary gains from loan sales and securitizations at quarter q is greater than the median, and 0 otherwise; BOTH 1 if DGAIN = 1 and DLLP = 1, 0 otherwise; UΔNP_LOAN unexpected change in non-performing loans; ULCO unexpected loan charge-offs. The dependent variable (AB_RET) is the residual from Fama–French three factor model estimated over the three-day period (one day before, the day of, and the days after each of the bank's quarterly earnings announcement date).27 H2 predicts that the coefficient on BOTH (λ5) is more significantly negative than the coefficients on DLLP (λ3) and DGAIN (λ4). 4. Sample and descriptive statistics 4.1. Sample selection Loan loss provisions, gains from loan sales and securitizations, and other quarterly financial data are obtained from FR Y-9C forms. These forms have been compiled by the Federal Reserve Bank of Chicago since 1986. Data of fiscal year-end is obtained from the Compustat quarterly file. Analysts' consensus (mean) earnings forecasts and common shares outstanding are retrieved from I/B/E/S unadjusted summary files, and daily stock returns are derived from CRSP. Table 1 shows the sample selection process. After merging FR Y-9C with Compustat, there are 10,670 bank-quarter observations from 2001 to 2007. I select this time period because this is when SFAS No. 140 became effective. I delete observations without data in the same quarter of the previous year, observations with missing values, and the top and bottom 1% of loan loss provisions and gains from loan sales and securitizations. These screenings leave 7612 bank26 Following Beatty et al. (1995), I measure the nondiscretionary component of loan charge-offs using non-performing loans (NPL) and loan loss reserve (LLR) at the beginning of each quarter. Similarly, the expected change in non-performing loans is estimated using the change in non-performing loans (NPL) in the previous period. The residuals from each model are the proxies for unexpected components of loan charge-offs and change in non-performing loans, respectively. All of the variables are scaled by beginning total assets.

LCOt ¼ γ þ η1 NPLt−1 þ η2 LLRt−1 þ o ΔNP LOANt ¼ α þ β 1 ΔNP LOANt−1 þ γ 27

ð4Þ ð5Þ

I estimate the cumulative abnormal return using a three day window. The expected return is estimated using a Fama–French Three Factor over the period as 290 to 40 days prior to the earnings announcement date. The residual over the three-day period (−1, 1) around earnings announcement dates is used to calculate cumulative abnormal returns.

Table 1 Sample selection process. # of Observations FR Y-9C Merged with Compustat Quarterly Data Less: without observations of quarter q-4 Less: missing values Less: top and bottom 1% of LLP and GAIN # of observations used in loss avoidance test # of observations used in loss avoidance test Less: observations not in I/B/E/S # of observations used in meeting or beating analysts forecast test # of observations used in meeting or beating analysts forecast test Less: observations without return Data from CRSP Less: without observations in quarter q-1 and missing values Less: top and bottom 1% of AB_RET # of observations used in market reaction test

10,670 (2715) (82) (261) 7612 7612 (3026) 4586 4586 (680) (402) (82) 3422

quarter observations, representing 382 unique banks in the analysis of avoidance of earnings decreases. In the analysis of meeting or beating analysts' forecast, I further merge the data in the analysis of the avoidance of earnings decreases with the I/B/E/S, and this leaves 4586 bank-quarter observations, representing 274 unique banks. In the market reaction analysis, I start from 4586 observations in the analysis of meeting or beating analysts' forecast. 680 and 402 observations are excluded due to lacking returns data and missing values respectively. In addition, I delete the top and bottom 1% of abnormal returns and earnings surprise to avoid the effect of extreme values. These screenings leave 3422 bank-year observations in the analysis of market reaction, representing 208 bank holding companies. 4.2. Estimations of discretionary loan loss provisions and gains from loan transfers and securitizations Table 2 reports the results of OLS regressions to estimate the discretionary loan loss provisions and gains from loan sales and securitizations. All regression models are significant at p b 0.001. The adjusted R2 is 0.286 in estimating the discretionary loan loss provisions and is 0.092 in estimating discretionary gains from loan sales and securitizations. The overall goodness of fit in the two equations is consistent to prior research (Beatty et al., 2002). 28 The significant coefficients in loan loss provisions regression indicate that loan loss provisions are positively correlated with the change of non-performing loans (ΔNP_LOAN), the proportion of individual loans (LOAN_IND), commercial loans (LOAN_COM), agricultural loans (LOAN_AGR), and loans secured by real estate (LOAN_RST). In addition, loan loss provisions increase with the amount of the loan loss reserve at the beginning of the year (LLR) and decrease in loans to depository institutions (LOAN_DEP). In the case of gains from loan sales and securitizations regression, the significant positive coefficient on UN_GL suggests that realized securitization gains increase with the magnitude of unrealized gains and losses. The results from Table 2 are similar to those reported in Beatty and Harris (1999). Discretionary loan loss provisions and discretionary gains from loan sales and securitizations are estimated as the residuals from Eqs. (1) and (2), respectively. 29 28 In my test, I separate discretionary loan loss provisions and securitization gains from non-discretionary components. Given an adjusted R2 of just 9% in the model of estimating securitization gains, I take Dechow et al. (2008)'s approach and use an alternative measure in the sensitivity test by assuming that the entire gain from securitization is discretionary. Unreported results show that my major results are not affected by this alternative measure. 29 Unreported statistics show that the means (medians) loan loss provisions and gains from loan sales and securitizations are 0.12% (0.08%) and 0.07% (0.01%) of total assets respectively. I use the standard score of gains from loan sales and securitizations in the regression tests, because this variable is highly skewed. The standard score is also called z-score. It is calculated as the difference between the raw score (gains from loan sales and securitizations) and the mean of population, divided by the standard deviation of the population.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37 Table 2 Estimation of discretionary loan loss provisions and discretionary gains. LLP

Intercept SIZE NP_LOAN LOAN_IND LOAN_FG LOAN_DEP LOAN_COM LOAN_AGR LOAN_RST LLR LCO UN_GL Adjusted R2 Total # of obs.

Table 3 Descriptive statistics.

GAIN

Variables

Estimate

p value

Estimate

p value

− 0.0006 0.0001 0.0242 0.0041 0.0031 − 0.0035 0.0013 0.0015 0.0004 0.0721 0.1820

b 0.0001 0.0001 0.0002 b 0.0001 0.728 0.012 b 0.0001 0.024 0.0005 0.0001 0.0001

0.0001 0.0001

b0.0001 b0.0001

0.0056 0.092

b0.0001

0.2860 7612

29

Variable definitions: LLP = loan loss provisions in quarter q scaled by total assets at q4; GAIN = gains from loan sales and securitization in quarter q scaled by total assets at q-4; SIZE = natural log of total assets in $000; ΔNP_LOAN = change of non-performing loans in quarter q relative to q-4, scaled by total assets at q-4; LOAN_IND = loans to individuals scaled by total assets at quarter q-4; LOAN_FG = loans to foreign governments scaled by total assets at quarter q-4; LOAN_DEP = loans to depository institutions and acceptances of other banks scaled by total assets at quarter q-4; LOAN_COM = commercial and industrial loans scaled by total assets at quarter q-4; LOAN_AGR = agricultural loans scaled by total assets at quarter q-4; LOAN_RST = loans secured by real estate scaled by total assets at quarter q-4; LLR = loan loss reserve scaled by total assets at quarter q-4; LCO = net loan charge-offs scaled by total assets at quarter q-4; UN_GL = reported unrealized security gains and losses at the beginning of the quarter, divided by total assets at quarter q-4.

4.3. Descriptive statistics The descriptive statistics of discretionary loan loss provisions, discretionary gains, and other variables are shown in Table 3. Panel A presents the descriptive statistics of the sample in the analysis of avoiding earnings decreases. Although both the means of discretionary loan loss provisions and discretionary gains from loan sales and securitizations are close to 0 as constructed, the cross-sectional variation is substantial: discretionary loan loss provision (discretionary gains from loan sales and securitizations) is understated by 0.03% (0.05%) at the 25th percentile but overstated by 0.02% (0.01)% at the 75th percentile. MEET1 and MEET2 measure the proportion of banks with earnings meeting or exceeding that of the same quarter of prior fiscal year under different scenarios. Specifically, about 44.28% banks report small earnings increase when pre-securitization income is above the prior year's levels (mean MEET1). 11.23% banks rely on gains from loan sales and securitizations to report small earnings increase (mean MEET2). The mean of one-year ahead changes in earnings before loan loss provisions and taxes (ΔROA) is 0.38%, indicating banks are profitable. The mean of total capital ratio (CAP) is 12.86%. This suggests that, on average, banks in the sample are well capitalized. The descriptive statistics of percentage of homogenous loans (HOM) and earnings before loan loss provisions and taxes (ROA) are comparable to those reported in Liu and Ryan (2006). Table 3 Panel B reports the descriptive statistics of the variables used in meeting or beating analysts' forecasts analysis. Overall, about 40.54% banks meet or beat analysts' consensus (mean FMEET1) forecasts by zero or one penny when pre-securitization income exceeds the target. 20.12% banks rely on gains from loan sales and securitizations to meet or beat analysts' forecast (mean FMEET2). The descriptive statistics of other variables are similar to those reported in Panel A. Prior studies also document a “kink” in the distribution of earnings around targets as an indication of earnings management (Burgstahler & Dichev, 1997; Degeorge et al., 1999). 30 The literature, however, is 30 Burgstahler and Dichev (1997) show too few firms just below the earnings threshold and too many firms just above. Similarly, Degeorge et al. (1999) observes a “kink” in the distribution of forecast errors.

Mean

Std dev

25th pctl

Median

75th pctl

− 0.0001 − 0.0002 0.0000 0.0000 0.0036 0.3485 0.0089 0.1244 0.0270

0.0002 0.0001 1.0000 0.0000 0.0051 0.4785 0.0135 0.1415 0.0310

Panel B: meeting or beating analysts' forecasts test (N = 4586) AB_LLP 0.0000 0.0008 − 0.0003 − 0.0001 AB_GAIN 0.0000 0.0012 − 0.0005 − 0.0003 FMEET1 0.4054 04528 0.0000 0.0000 FMEET2 0.2012 0.3742 0.0000 0.0000 ΔROA 0.0042 0.0035 − 0.0013 0.0038 HOM 0.3354 0.1614 0.2092 0.3450 ROA 0.0105 0.0065 0.0056 0.0110 CAP 0.1279 0.0386 0.1088 0.1209 GDP 0.0259 0.0075 0.0160 0.0230

0.0002 0.0002 1.0000 0.0000 0.0051 0.4786 0.0157 0.1367 0.0280

Panel A: earnings decrease avoidance test (N = 7612) AB_LLP 0.0000 0.0007 − 0.0003 AB_GAIN 0.0000 0.0011 − 0.0005 MEET1 0.4428 0.4982 0.0000 MEET2 0.1123 0.2825 0.0000 ΔROA 0.0038 0.0029 − 0.0009 HOM 0.3562 0.1628 0.2162 ROA 0.0101 0.0074 0.0049 CAP 0.1286 0.0435 0.1112 GDP 0.0263 0.0078 0.0180

Variable definitions: AB_LLP =discretionary loan loss provisions; AB_GAIN= discretionary gains from loan sales and securitizations; MEET1 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq – GAINq > INq-4, and 0 otherwise; MEET2 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq – GAINq b INq-4, and 0 otherwise; FMEET1 = 1 if earnings before gains from securitization and loan sales meet or beat analysts' forecasts by zero or one penny, and 0 otherwise; FMEET2= 1 if earnings meet or beat analysts' forecasts by zero or one penny but earnings before gains from securitization and loan sales miss analysts forecast, and 0 otherwise; ΔROA= one-year ahead changes in earnings before loan loss provisions and taxes, scaled by total assets in quarter q-4; HOM= % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA= earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP= total capital ratio; GDP = GDP growth is real growth in GDP per capita (annual %).

inconclusive on the validity of earnings benchmarks. Dechow, Richardson, and Tuna (2003) finds that discretionary accruals in small profit firms are no different from small loss firms, suggesting the “kink” driven by earnings management is unsubstantiated. 31 In this study I provide evidence whether there is a “kink” around the thresholds in banking industry. Fig. 1 plots the distribution of earnings surprises (current quarter earnings minus earnings in quarter q-4, scaled by beginning total assets). This panel shows that roughly 54% of the observations fall in bin 0 and bin 1(firms reporting small positive earnings growth), which is much higher than 18% of the observations in bin −1 and bin − 2. Fig. 2 presents the distribution of the forecast error (actual earnings per share minus the analysts' consensus forecast) in 1-penny bins in a range around zeros. The percentage of observations meeting or beating analysts' forecasts (bin 0 and bin 1 combined) is much higher than that missing analyst forecasts (bin −1 and bin −2 combined). The findings from Figs. 1 and 2 indicate there is an upward kink in the earnings distribution from bin − 1 to bin 0, consistent to the common interpretation that firms intentionally manage earnings to just meet/beat analysts' forecasts and to avoid small earnings decreases.

31 Other researchers provide alternative explanations of the “kink” in the earnings distributions. Durtschi and Easton (2005) document that the discontinuities in the distribution of earnings around zero may be induced by the scaling effect. Beaver, McNichols, and Nelson (2007) show that the asymmetric effects of income taxes and special items contribute to a large portion of the break in the distribution.

30

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

Fig. 1. Histogram of earnings surprises: exploring the threshold of avoiding earnings decreases. Earnings surprises (ES)= (the current quarter earnings− earnings in quarter q-4)/beginning total assets Observations are assigned into bins of 0.1% width. For example, observations with 0%≤ ES b 0.1% are assigned into “bin 0” and those with 0.2% ≤ ES b 0.1% into “bin 1”. Observations with −0.1%≤ ES b 0% are assigned into “bin −1” and those with −0.2% ≤ ES b −0.1% into “bin −2”.

5. Empirical results 5.1. Analysis of avoidance of earnings decreases 5.1.1. Tests of discretionary loan loss provisions In Panel A of Table 4, I report the multivariate analysis of the association between discretionary loan loss provisions and avoidance of small earnings decreases. H1 predicts that banks manage loan loss provisions more pervasively than structure loan transfers for earnings benchmarks. Model 1 tests whether banks use loan loss provisions to avoid small earnings decreases when pre-securitization earnings are greater than earnings in the same quarter of prior year. Consistent to the prediction, the coefficient on MEET1 is −0.0002 with a p value b0.0001 (All p values are two-tailed unless indicated otherwise). In other words, banks with positive earnings growth report lower discretionary loan loss provisions when earnings before the gains have already exceeded the earnings at quarter q-4 (Scenario 1). Unreported statistics show that the median quarterly net income (scaled by total assets at the end of quarter q-4) of the sample in testing loss avoidance is 0.0069. This means banks with small earnings surprises recognize lower discretionary loan loss provisions equivalent to 3% (−0.0002/0.0069) of income. Although the economic magnitude of smaller discretionary loan loss provisions is immaterial under the conventional 5% rule-of-thumb for materiality, the effects on earnings per share can be more pronounced because meeting or beating earnings benchmarks by even a small amount has great impact on stock prices (Barth et al., 1999; Degeorge et al., 1999; Matsumoto, 2002).

Fig. 2. Histogram of forecast error: exploring the threshold of meeting or beating analysts' forecasts.Analysts' forecast error (F_error) = the actual EPS − consensus analyst forecast in the month immediately prior to earnings announcement.Observations are assigned into bins of $0.01 width. For example, observations with F_error= 0.00 are assigned into “bin 0” and those with F_error= 0.01 into “bin 1”.

Model 2 (Table 4 Panel A) tests whether bank managers manage loan loss provisions when pre-securitization income is below the prior year's levels and income after securitization exceeds the prior year's level (scenario 2). The coefficient on MEET2 is −0.0001 with p value b0.0001, indicating that banks relying on gains from loan sales and securitizations to report small earnings growth recognize lower loan loss provisions. 32 Overall, the results in model 1 and model 2 indicate that loan loss provisions are managed under these two scenarios. Next, I use model 3 (Table 4 Panel A) to test whether the level of discretionary loan loss provisions stays the same for banks reporting small positive earnings growth with or without the incentives of managing gains from loan sales and securitizations. Model 3 includes both of MEET1 and MEET2 and it is predicted that the coefficients on MEET1 and MEET2 will not be statistically different. Consistent to my expectation, the coefficients on MEET1 and MEET2 are − 0.0002. A Chi-square test fails to reject the null hypothesis that the coefficients on MEET1 and MEET2 are the same. As expected, ROA (earnings before loan loss provisions and taxes) and ΔROA (one-year ahead changes in earnings before loan loss provisions and taxes) have significantly positive coefficients, suggesting that bank managers exercise discretion over loan loss provisions to signal future prospects and smooth earnings. In addition, I include two interaction terms: ROA* ΔROA and ROA*CAP in model 4 (Table 4 Panel A) to further examine whether the conflicting directions arising from the multiple motivations limit the use of loan loss provisions. Consistent to past research (Kanagaretnam, Lobo, & Yang, 2004), the coefficient on the interaction term ΔROA*ROA is significantly negative. The results indicate that the joint impact of ΔROA*ROA on discretionary loan loss provisions decreases as ΔROA*ROA increases, supporting my supposition that the use of discretionary loan loss provisions is limited to some extent because signaling purpose reduces management's flexibility to utilize discretionary loan loss provisions for smoothing income. In sum, the findings from model 3 and model 4 imply that it is likely to have an upper bound on the level of discretionary loan loss provisions that a manager can exercise. In addition, I find that the coefficients on CAP and ROA*CAP are insignificant, suggesting that the impact of loan loss provisions on capital is restricted to some extent. Finally, consistent to the prediction, I find that the coefficients on GDP are negative and significant in all models. The negative coefficients show that bank managers tend to increase loan loss provisions when the external economy experiences a decline, Together, the results in Table 4 Panel A support the notion that managers use loan loss provisions to achieve a small positive earnings increase without relying on gains from loan sales and securitizations. 5.1.2. Tests of discretionary gains from loan sales and securitizations Table 4 Panel B presents the results in testing the association between discretionary gains from loan sales and securitizations and the avoidance of small earnings decreases. All models are significant at p b 0.0001. In model 1, the coefficient on MEET1 is −0.0004 and highly significant, indicating that when banks have already reported small earnings growth (earnings before loan sales and securitizations are positive), the discretionary gains are significant low (Scenario 1). In model 2 (Table 4 Panel B), however, the coefficient on MEET2 is 0.0009 with p b 0.0001, indicating that banks report higher discretionary gains when they rely on gains from loan sales and securitizations to report small positive earnings growth (Scenario 2). In model 3(Table 4 Panel B), I compare the levels of discretionary gains for banks reporting small positive earnings growth under these two scenarios. The coefficient on MEET1 is −0.0003, which indicates 32 The magnitude of the coefficient is smaller than that in model 1. It is because banks reporting small positive earnings growth without relying on gains from loan loss provisions and securitizations become part of the default group (MEET2 = 0).

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

31

Table 4 Multivariate analysis of earnings decrease avoidance, loan loss provision, and gains from securitization and loan sales. Panel A: dependent variable – AB_LLP

Intercept MEET1 MEET2 ΔROA HOM ROA CAP GDP ROA*ΔROA ROA*CAP Adjusted R2 N

Model 1 (Scenario 1)

Model 2 (Scenario2)

Estimate

p value

Estimate

p value

Estimate

p value

Estimate

p value

0.0001 − 0.0002

0.02 b 0.0001

0.0001

0.03

0.0025 − 0.0001 0.0032 0.0001 − 0.044

0.02 0.27 0.01 0.87 b 0.0001

− 0.0001 0.0028 − 0.0001 0.0043 0.0001 − 0.043

b0.0001 0.01 0.32 0.02 0.42 0.01

0.0001 − 0.0002 − 0.0002 0.0027 − 0.0001 0.0041 0.0001 − 0.044

0.01 b 0.0001 b 0.0001 0.02 0.44 0.01 0.81 b 0.0001

0.0003 − 0.0002 − 0.0002 0.0025 − 0.0001 0.0037 0.0001 − 0.04 − 0.0148 0.0006 0.053 7612

0.01 b0.0001 b0.0001 0.05 0.35 0.10 0.75 b0.0001 b0.0001 0.12

0.041 7612

Model 3

0.04 7612

0.042 7612

Model 4

Note: p values are calculated based White (1980) heteroscadisticity-robust t-values. Variable definitions AB_LLP = discretionary loan loss provisions; MEET1 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq − GAINq > INq-4, and 0 otherwise; MEET2 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq − GAINq b INq-4, and 0 otherwise; ΔROA = one-year ahead changes in earnings before loan loss provisions and taxes, scaled by total assets at quarter q-4; HOM = % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA = earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP = total capital ratio; GDP = GDP growth is real growth in GDP per capita (annual %). Panel B: Dependent variable – AB_GAIN Model 1 (Scenario 1)

Intercept MEET1 MEET2 ΔROA HOM ROA CAP GDP Adjusted R2 N

Model 2 (Scenario2)

Model 3

Estimate

p value

Estimate

p value

Estimate

p value

0.0002 − 0.0004

0.001 b0.0001

− 0.0001

0.48

− 0.0180 0.0002 − 0.0426 − 0.0001 − 0.0371 0.152 7612

0.08 0.05 b0.0001 0.01 0.01

0.0009 − 0.0132 0.0001 − 0.0324 − 0.0001 − 0.037 0.283 7612

b0.0001 0.06 0.22 0.01 0.01 0.01

0.0003 − 0.0003 0.0008 − 0.0124 0.0002 − 0.0411 − 0.0001 − 0.039 0.304 7612

0.08 b0.0001 b0.0001 0.05 0.12 b0.0001 0.01 0.01

Note: p values are calculated based White (1980) heteroscadisticity-robust t-values. Variable definitions: AB_GAIN = discretionary gains from loan sales and securitizations; MEET1 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq – GAINq > INq-4, and 0 otherwise; MEET2 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq – GAINq b INq-4, and 0 otherwise; ΔROA = one-year ahead changes in earnings before loan loss provisions and taxes, scaled by totalassets at quarter q-4; HOM = % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA = earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP = total capital ratio;GDP = GDP growth is real growth in GDP per capita (annual %).

that banks report lower discretionary gains absent of the incentive to avoid small earnings decreases. In comparison, the coefficient on MEET2 is 0.0008, suggesting that banks with incentives to manage gains from loan sales and securitizations report higher discretionary gains. Most notably, the coefficients on earnings before loan loss provisions and taxes (ROA) and capital ratio (CAP) in all models are significantly negative, consistent to Karaoglu (2005)'s findings that bank managers use the discretion in reported gains to manage regulatory capital and earnings. 33 Overall, the results in Table 4 Panel A indicate that banks manage loan loss provisions with or without incentives to manage the gains from loan sales and securitizations. In comparison, the results in Table 4 Panel B show that gains from loan sales and securitizations are used only when such incentives are present. Taken together, the findings in Table 4 are

33 Table 4 Panel A shows insignificant coefficients on CAP, while Table 4 Panel B indicates significantly negative coefficients on CAP. A possible explanation is that the riskbased regulations set an upper bound for loan loss provisions to be included as Tier II capital. So the impact of loan loss provisions on regulatory capital is limited to some extent. Unlike loan loss provisions, loan sales and securitizations do not have such restrictions. Managers may increase their capital by selling or securitizing loans.

consistent with the prediction in H1 that banks manage earnings through loan loss provisions, before through structuring transactions, to achieve small earnings surprises. 5.2. Analysis of meeting or beating analysts' forecasts 5.2.1. Tests of discretionary loan loss provisions Table 5 shows the results from OLS regressions that examine whether banks manage loan loss provisions and gains from loan sales and securitizations to just meet or beat analysts' forecasts. In Table 5 Panel A, all models are significant at p b 0.001. In model 1, the coefficient on FMEET1 is − 0.0001, significant at the 0.001 level, indicating that banks meeting or beating analysts' forecasts by zero or one penny report lower discretionary loan loss provisions when pre-securitization gains exceed prior year level (Scenario 1). Unreported statistics show that the median net income of the sample in testing meeting or beating analysts' forecasts is 0.0059. Thus, on average, banks that just meet or beat analysts' forecasts report lower discretionary loan loss provisions equivalent to 1.61% (−0.0001/0.0062) of net income. Similarly in model 2 (Table 5 Panel A), the coefficient on FMEET2 is − 0.0001 and highly significant, indicating that banks relying on gains from loan transfers to just meet or beat analysts' forecasts also report lower discretionary loan loss

32

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

provisions (Scenario 2). 34 In addition, as reported in model 3 (Table 5 Panel A), the coefficients on FMEET1 and FMEET1 are both −0.0001 with p values b 0.0001 and 0.025, respectively. Most notably, when I compare the two tables (Table 4 Panel A vs. Table 5 Panel A), I find the magnitude of the coefficients on MEET1 (Table 4 Panel A) is larger than the coefficients on FMEET1 (Table 5 Panel A), implying that managers understate loan loss provision to a greater extent in avoiding small earnings decreases relative to meeting/beating analysts' forecasts.35 Finally, model 4 (Table 5 Panel A) shows that the interaction term ΔROA*ROA is significantly negative, similar to the previous results that the conflicting motivations limit bank managers' ability to use loan loss provisions. 5.2.2. Tests of discretionary gains from loan sales and securitizations In Table 5 Panel B, I present the results from OLS regressions using discretionary gains from loan sales and securitizations as the dependent variable. In model 1, the coefficient on FMEET1 is −0.0003 and highly significant, which means that banks just meeting or beating analysts' forecasts report lower discretionary gains when presecuritization earnings have already met or beaten analysts' forecasts (Scenario 1). This is the circumstance that banks even intentionally manage gains from loan sales and securitizations downward. In contrast, the coefficient on FMEET2 in model 2 is 0.0003 with a p value b 0.0001 (Scenario 2). This means, on average, that banks report higher discretionary gains from loan sales and securitizations when pre-securitization earnings miss the target, which is equivalent to 4.8% (0.0003/0.0062) of median net income. In model 3 (Table 5 Panel B), the coefficients on FMEET1 and FMEET2 are −0.0003 and 0.0003 respectively, both highly significant. The results provide evidence that banks manage gains from loan sales and securitizations to meet or beat analysts' forecasts only when incentives are present.36 Overall, H1 is supported by the results in Tables 4 and 5 that gains from loan sales and securitizations are used as a secondary instrument to meet or beat earnings benchmarks. 5.3. Market reaction analysis The results of testing H2 are reported in Table 6. Model 1 tests whether lower discretionary loan loss provisions (lower than the 34 The results in table 5 (panel A) complement previous research in supporting the argument that managers exercise discretion over loan loss provisions to achieve earnings benchmarks. A similar approach to my paper is taken by Robb (1998) who find that banks manage earnings through discretionary loan loss provisions to meet market expectations when an analyst consensus exists. This paper extends the literature by investigating how bank managers trade off loan loss provisions and securitizations gains to just meet or beat analysts' forecasts. My supplementary analysis demonstrates that the mean dispersion in analysts' forecasts in those just meeting or beating firms (meet forecast by zero or one penny) is 0.0228, which is much lower than the mean dispersion 0.0531 in other meeting firms (meet forecast by more than one penny). Dispersion in analysts' forecasts is measured using standard deviation of forecast divided by the absolute value of mean forecast. Consistent to Robb (1998)'s findings, the results suggest a consensus in earnings forecasts also drives managers to manipulate earnings to achieve market expectations. 35 The differential magnitude of coefficients on meeting/beating analysts' forecasts vs. avoiding earnings declines may provide insight into the hierarchy of earnings thresholds. As evidence shown in Table 3, about 55.51% (sum of MEET1 and MEET2) banks report small positive earnings growth and 60.66% (sum of FMEET1 and FMEET2) banks just meet or beat analyst expectation. The results indicate that bank managers pay more attention to meeting/ beating analysts' forecast than to reporting positive earnings surprises. It is possible because meeting analyst expectation is easier to achieve, as documented by a lower coefficient on meeting/beating analysts' forecast. Further evidence shows that conditioned on meeting analyst expectation, about 84% banks report positive earnings surprises and that conditioned on reporting small positive earnings growth, about 75% banks meet analyst target. In short, these results suggest that meeting/beating analysts' forecast is the threshold bank managers most attempt to achieve, which is consistent to Brown and Caylor (2005)'s findings. This study provides an appealing rationale for this threshold hierarchy by documenting the differential magnitude of earnings management in achieving each earnings benchmark. 36 Again, qualitatively similar results are found when comparing the magnitude of the coefficients on MEET2 (Table 4 Panel B) vs. FMEET2 (Table 5 Panel B), thus supporting the argument that managers exercise less discretion in just meeting/beating analysts' forecasts.

median) are negatively associated with stock returns. Consistent to the expectation, the coefficient on DLLP is − 0.0075 with a p value of 0.08. In Model 2 (Table 6), the coefficient on DGAIN tests whether banks reporting higher discretionary gains from loan sales and securitizations are priced negatively. As predicted, the coefficient on DGAIN is − 0.0102 and highly significant. Model 3 (Table 6) compares the magnitude of market reactions among those banks reporting both a lower level of discretionary loan loss provisions and a higher level of discretionary gains from loan sales and securitizations (BOTH), banks reporting only a lower level of discretionary loan loss provisions (DLLP), and banks reporting a higher level of discretionary gains from loan sales and securitizations (DGAIN). As predicted, the coefficient on BOTH is −0.0208 with a p value of 0.01. In comparison, the coefficient on DLLP is −0.0077 but insignificant at the conventional level. The coefficient on DGAIN is −0.0115 with a p value of 0.06. In addition, a Chi-square test shows that the magnitude of coefficient on BOTH is greater than that on DGAIN at p b 0.10. The magnitude and significance of coefficient on BOTH imply that investors discipline the joint use of loan loss provisions and securitization gains in meeting earnings benchmarks to a greater extent. Furthermore, the results on Table 6 also show that the coefficients on EARN in all models are significantly positive. In contrast, the coefficients on ULCO are significantly negative, suggesting that unexpected charge-offs provide additional loan loss information to investors. As a sensitivity check, first Eq. (6) is re-estimated using changes in variables of interest. Variables ΔLLP and ΔGAIN are defined as follows: ΔLLP=1 if changes in discretionary loan loss provisions in quarter q are negative, and 0 otherwise; ΔGAIN =1 if changes in discretionary gains from loan sales and securitizations in quarter q are positive, and 0 otherwise. Qualitatively similar results (not reported) are found when these additional proxies are used.37 Second, as a check on influence of the extreme observations on the results, I divide the sample into three thirds based on the magnitudes of discretionary loan loss provisions and gains from loan sales and securitizations. I repeat the analyses (Eqs. (3a) and (3b)) using a subsample that includes only the top and bottom thirds. This procedure increases the likelihood that earnings management can be detected by investors. Untabulated reports show that the coefficient on discretionary loan loss provisions is significantly positive, suggesting that investors interpret the higher (lower) level of discretionary loan loss provisions as good (bad) news. Moreover, the negative coefficient on discretionary securitization gains indicates that market reacts negatively to firms that inflate earnings with securitizations. Taken together, the additional findings support H2 and they are robust to the extent of earnings management.

5.4. Sensitivity analyses 5.4.1. Controlling for fixed effects I perform additional analyses to check the robustness of my results. First, I control for firm fixed effects. My estimations of nondiscretionary loan loss provisions and gains from loan sales and securitizations are based on the regression in the pooled time-series cross-sectional models. The pooled models assume that the parameters are stationary across the observation period and equal across banks, and as a result, the reported significance level could be overstated because of the cross-sectional and time-series dependence in the residuals. Given the importance of firm and year effects, I use the fixed-effects

37 In addition, I reexamined equation 6 using higher discretionary loan loss provisions (equal to 1 if discretionary loan loss provisions is greater than the median, and 0 otherwise) to show the consistency of my results with previous research. Unreported results indicate that the market reacts positively to higher levels of discretionary loan loss provisions, supporting the argument that higher discretionary loan loss provisions is viewed by investors as credible signal of firm growth.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

33

Table 5 Multivariate analysis of meeting or beating analysts' forecasts, loan loss provisions, and gains from securitization and loan sales. Panel A: Dependent variable – AB_LLP

Intercept FMEET1 FMEET2 ΔROA HOM ROA CAP GDP ROA*ΔROA ROA*CAP Adjusted R2 N

Model 1 (Scenario 1)

Model 2 (Scenario2)

Estimate

p value

Estimate

p value

Estimate

p value

Estimate

p value

0.0001 − 0.0001

0.05 b 0.0001

0.0001

0.04

0.0022 − 0.0001 0.004 0.0001 − 0.041

0.01 0.18 0.01 0.58 b 0.0001

− 0.0001 0.0025 − 0.0001 0.0047 0.0001 − 0.042

b0.0001 0.01 0.15 0.01 0.64 b0.0001

0.0001 − 0.0001 − 0.0001 0.0024 − 0.0001 0.0042 0.0001 − 0.041

0.05 b0.0001 0.01 0.01 0.22 0.01 0.46 b0.0001

0.0002 − 0.0001 − 0.0001 0.0022 − 0.0001 0.0038 0.0001 − 0.038 − 0.0126 0.0004 0.051 4586

0.05 b 0.0001 b 0.0001 0.08 0.16 0.05 0.75 b 0.0001 b 0.0001 0.24

0.039 4586

Model 3

0.038 4586

0.042 4586

Model 4

Note: p values are calculated based White (1980) heteroscadisticity-robust t-values. Variable definitions AB_LLP = discretionary loan loss provisions; FMEET1 = 1 if earnings before gains from securitization and loan sales meet or beat analysts' forecasts by zero or one penny, and 0 otherwise; FMEET2 = 1 if earnings meet or beat analysts' forecast by zero or one penny but earnings before gains from securitization and loan sales miss analysts forecast, and 0 otherwise; ΔROA = one-year ahead changes in earnings before loan loss provisions and taxes, scaled by total assets at quarter q-4; HOM = % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA = earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP = total capital ratio; GDP = GDP growth is real growth in GDP per capita (annual %). Panel B: Dependent variable – AB_GAIN Model 1 (Scenario 1)

Intercept FMEET1 FMEET2 ΔROA HOM ROA CAP GDP Adjusted R2 N

Model 2 (Scenario2)

Model 3

Estimate

p value

Estimate

p value

Estimate

p value

0.0002 − 0.0003

0.001 b0.0001

0.0001

0.34

− 0.0150 0.0004 − 0.0378 − 0.0001 − 0.0320 0.145 4586

0.05 b0.0001 b0.0001 0.27 0.01

0.0003 − 0.015 0.0005 − 0.0372 − 0.0001 − 0.031 0.182 4586

b 0.0001 0.05 b 0.0001 b 0.0001 0.34 0.01

0.0001 − 0.0003 0.0003 − 0.016 0.0004 − 0.037 − 0.0002 − 0.0034 0.253 4586

0.12 b 0.0001 b 0.0001 0.06 b 0.0001 b 0.0001 0.15 0.01

Note: p values are calculated based White (1980) heteroscadisticity-robust t-values. Variable definitions: AB_GAIN = discretionary gains from loan sales and securitizations; FMEET1 = 1 if earnings before gains from securitization and loan sales meet or beat analysts' forecasts by zero or one penny, and 0 otherwise; FMEET2 = 1 if earnings meet or beat analysts' forecasts by zero or one penny but earnings before gains from securitization and loan sales miss analysts forecast, and 0 otherwise; ΔROA = one-year ahead changes in earnings before loan loss provisions and taxes, scaled by total assets at quarter q-4; HOM = % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA = earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP = total capital ratio; GDP = GDP growth is real growth in GDP per capita (annual %).

model to reexamine the test equations. My results are not sensitive to this alternative expectation model. 5.4.2. Controlling for equity incentives Consistent with Cheng and Warifield (2005) I control for the impact of CEO equity incentives on earnings management. 38 CEO equity incentive is measured as the ratio of equity shares to total outstanding shares. CEO equity shares include current period option grants, unexercisable options, exercisable options, restricted stock grants, and stock ownership. I include CEO equity incentive variable into the main regression models. Unreported statistics indicate that the hierarchy of managing earnings to avoid small earnings declines and to just meet or beat analysts' forecasts still holds after controlling for equity incentives.

5.4.3. Controlling for other risk factors in affecting securitizations I also add risk measurement variables on Eq. (3b) (model of gains from loan sales and securitizations as a function of earnings smoothing,

38 Cheng and Warifield (2005) find that managers with high equity incentives are more likely to report earnings that just meet or just beat analysts' forecasts.

communication of private information, and regulatory management) based on prior studies. Demsetz (2000) and Karaoglu (2005) document that securitizations can be used to manage interest rate risk and credit risk. Following prior research, I use the standard deviation of asset returns and the changes in three-month Treasury bill rates over the quarter prior to the securitizations to control for such incentives that motivate bank managers engaging earnings management through securitizations. My results are not affected by the inclusions of these risk variables.

5.4.4. Controlling for the simultaneity of discretion choices I also take account of simultaneity of discretion choices. Beatty et al. (1995) finds that loan-loss provisions, loan charge-offs, and the decision to issue securities are jointly determined. The joint determination of these variables is based on the assumption that managers trade off discretionary transactions to achieve conflicting earnings and capital management objectives. To address the potential for simultaneity, I take a three stage least squares (3 SLS) approach. In the first stage, each endogenous variable is regressed on its instrumental variables. The predicted values of endogenous variables are independent of equation errors. In the second stage, the predicted values replace the right side of endogenous variables. The 3 SLS

34

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

Table 6 The joint effect of loan loss provision and gains from securitization and loan sales on earnings. Dependent variable – AB_RET Model 1

Intercept EARN ΔEARN DLLP DGAIN BOTH UΔNP_LOAN ULCO Adjusted R2 N

Model 2

Model 3

Estimate

p value

Estimate

p value

Estimate

p value

0.017 0.0425 0.0221 − 0.0075

0.01 0.01 0.12 0.08

0.025 0.0417 0.0204

0.03 0.01 0.14

− 0.0102

0.02

0.0074 − 0.147 0.058 3422

0.65 b 0.0001

0.0312 0.0478 0.0234 − 0.0077 − 0.0115 − 0.0208 0.0075 − 0.118 0.062 3422

0.01 0.01 0.10 0.15 0.06 0.01 0.58 b0.0001

0.0082 − 0.125 0.051 3422

0.76 b0.0001

Variable definitions: AB_RET = three day (− 1, 0, and 1) cumulative abnormal returns where day 0 is the quarterly earnings announcement date; EARN = earnings before loan loss provisions, securitization gains, and taxes, scaled by total assets at the beginning of quarter q; ΔEARN = changes in earnings before loan loss provisions, securitization gains, and taxes, scaled by total assets at the beginning of quarter q; DLLP = 1 if discretionary loan loss provisions at quarter q is less than the median, and 0 otherwise; DGAIN = 1 if discretionary gains from loan sales and securitization at quarter q is greater than the median, and 0 otherwise; BOTH = 1 if DGAIN = 1 and DLLP = 1, 0 otherwise; UΔNP_LOAN = unexpected change in non-performing loans; ULCO = unexpected loan charge offs.

approach enables me to control for cross-equation correlation. 39 The system of four simultaneous equations is expressed as follows: LLP ¼ α þ β 11 GAIN þ β12 CHFUND þ β13 LCO þ β 14 Threshold þ β15 ΔROA þ β 16 HOM þ β 17 ROA þ β 18 CAP þ β 19 GDP þ β20 LLP þ ε

ð7aÞ GAIN ¼ α þ λ11 LLP þ λ12 CHFUND þ λ13 LCO þ λ14 Threshold þ λ15 ΔROA þ λ16 HOM þ λ17 ROA þ λ18 CAP þ λ19 GDP þ λ20 GAIN þ ε

ð7bÞ LCO ¼ α þ η11 LLP þ η12 CHFUND þ η13 GAIN þ η17 CAP þ η18 GDP þ η19 LCO þ ε

ð7cÞ

CHFUND ¼ α þ γ 11 LLP þ γ12 LCO þ γ 13 GAIN þ γ17 CAP þ γ 18 GDP ð7dÞ þ γ 19 CHFUND þ ε In the main testing models, I construct two thresholds (MEET1/ FMEET1 and MEET2/FMEET2) that affect managerial behavior in reporting earnings. Threshold variables in Eqs. (7a) and (7b) represent MEET1/FMEET1 and MEET2/FMEET2 under each scenario. The determinants of loan loss provisions and securitization gains in Eqs. (7a), (7b) are similar to Eqs. (3a), (3b) except for the system 39 The nondiscretionary portions of the loan loss provisions and gains from loan sales and securitization are estimated using Eqs. (1) and (2), respectively. Following Beatty et al.(1995), I measure the nondiscretionary component of loan charge-offs using prior quarter-end non-performing loans (NPL) and prior quarter-end loan loss reserve (LLR). Similarly, the nondiscretionary part of issuing securities depends on the sum of prior quarter-end capital notes and preferred stock (NOTE) and prior quarter-end common equity (EQUITY). All of the variables are scaled by beginning total assets. To summarize, the nondiscretionary portions of loan charge-offs and issuing securities can be expressed as follows:

LCOt ¼ NPLt−1 þ LLRt−1 þ η

5.4.5. Imposing other constraints on variables of interest In the main testing models, I use gains from loan sales and securitizations to set the constraints for MEET variables (INq − GAINq > INq-4 or INq − GAINq b INq-4). In the supplementary analysis, I also impose similar constraints using loan loss provisions. First, I create a scenario in which banks report small positive earnings surprises and earnings before discretionary loan loss provisions meet the benchmarks. This is the situation where managers have little incentives to inflate earnings through loan loss provisions. Variables MEET3 and FMEET3 are used to proxy the thresholds under this scenario. 40 Unreported tables show that the coefficients on MEET3 and FMEET3 in Eq. (3a) (Eq. (3b)) are significantly positive (negative), 41 suggesting that bank managers employ income-decreasing tools to smooth earnings when net income before discretionary loan loss provisions exceeds earnings benchmark. Next, I create a scenario in which banks report small positive earnings surprises but earnings before discretionary loan loss provisions are below the targets. This is the scenario that bank managers have incentives to understate loan loss provisions. In order to further test managers' preference over the tools, I spit this scenario into two subparts: those relying on gains from securitizations to report small positive growth (MEET4/FMEET4) and those not (MEET5/FMEET5).42 Variables MEET4 40 Specifically, variables MEET3 and FMEET3 are defined as follows: MEET3 = 1 if 0 b = INq-INq-4 b 0.2%*ASSETSq-4 and INq + AB_LLPq > INq-4, and 0 otherwise; FMEET3 = 1 if earnings just meet or beat analysts' forecast and earnings before discretionary loan loss provisions meet analysts' forecast, and 0 otherwise. 41 As mentioned in Section 3.3.1, Eq. (3) (Eq. (4)) defines loan loss provisions (loan sales and securitization gains) as a function of earnings smoothing, communication of private information, and regulatory management. The positive (negative) coefficients on MEET3/FMEET3 in Eqs. (3a), (3b) indicate bank managers downward earnings by overstating loan loss provisions (recognizing securitization losses). 42 The threshold variables are defined as follows:

•

ð6Þ

Where: LCO = net loan charge-offs; NPL = prior year-end non-performing loans; CHFUND = the sum of prior year-end common stock, preferred stock, and capital notes issued; NOTE = prior year-end capital notes and preferred stock; EQUITY = prior yearend common equity.

In the case of reporting small positive earnings growth

MEET4 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4, and INq + AB_LLPq b INq-4, and INq − AB_GAINq b INq-4, and 0 otherwise; MEET5 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4, and INq + AB_LLPq b INq-4, and INq − AB_GAINq > INq-4, and 0 otherwise;

•

ð5Þ

CHFUNDt ¼ NOTEt−1 þ EQUITYt−1 þ o

approach. Eqs. (7c) and (7d) are constructed based on Beatty et al.'s (1995) findings that loan charge-offs and the decision to issue stock are managed to achieve uncertain capital goal. GDP is included in the system to control for the macroeconomic effects. Table 7 presents the results of three-stage least squares estimation of system under each scenario. The main results show: (i) negative coefficients on MEET1, MEET2, FMEET1, and FMEET2 in LLP equation (β14 = −0.0002 in Panel A and Panel B; β14 = − 0.0001 in Panel C and Panel D), consistent with the prediction that loan loss provisions are managed to avoid small earnings decreases without relying on gains from loan sales and securitizations; and (ii) negative coefficients on MEET1 and FMEET1 (λ14 = − 0.005 in Panel A and λ14 = − 0.003 in Panel C) and positive coefficients on MEET2 and FMEET2 (λ14 = 0.009 in Panel B and λ14 = 0.004 in Panel D) in GAIN equation, consistent with the prediction that bank managers report higher discretionary gains from loan sales and securitizations only when incentives are present. The coefficients on other control variables are consistent to the findings as reported by Beatty et al. (1995). Overall, the results in Table 7 provide additional support for H1.

In the case of meeting or beating analysts' forecasts

FMEET4 = 1 if earnings just meet or beat analysts' forecasts by zero or one penny, but earnings before discretionary loan loss provisions miss analysts' forecasts and earnings before discretionary securitization gains miss analysts' forecasts as well, and 0 otherwise; FMEET5 = 1 if earnings just meet or beat analysts' forecast by zero or one penny and earnings before discretionary securitization gains meet analysts' forecasts, but earnings before discretionary loan loss provisions miss analysts' forecasts; and 0 otherwise.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

35

Table 7 Results of three-stage least squares estimation. LLP Coeff.

GAIN p-value

Coeff.

LCO p-value

Coeff.

CHFUND p-value

Coeff.

p-value

Panel A: results of three-stage least squares estimation under Scenario 1: income before gains from loan sales and securitization is above the prior year's level (threshold= MEET1) Intercept − 0.0016 0.00 − 0.0010 0.95 − 0.0010 0.37 − 0.0030 0.06 GAIN 0.1755 0.28 1.0386 0.01 − 0.2790 0.28 LLP 0.2904 0.18 0.7710 b 0.0001 − 0.1550 0.08 CHFUND − 0.0004 0.09 − 0.0005 0.49 0.0003 0.04 LCO 0.1861 b0.0001 0.1085 0.08 0.5170 0.04 MEET1 − 0.0002 b0.0001 − 0.0005 b 0.0001 ΔROA 0.0040 0.06 − 0.0240 0.05 HOM − 0.0002 0.84 0.0006 b 0.0001 ROA 0.0241 0.01 − 0.1030 b 0.0001 CAP 0.0001 0.26 − 0.0002 0.01 0.0001 0.03 − 0.0004 0.94 GDP − 0.1076 b0.0001 − 0.0421 0.01 − 0.0693 b 0.0001 0.0260 0.20 Instruments Yes Yes Yes Yes System-weighted R2 = .44 Panel B: results of three-stage least squares estimation under Scenario 2: income before gains from loan sales and securitization is below the prior year's level, but income after gains from loan sales and securitizations exceeds the prior year's level. (threshold = MEET2) Intercept − 0.0010 0.00 − 0.0016 0.23 − 0.0010 0.34 − 0.0023 0.11 GAIN 0.1770 0.29 1.0564 0.01 − 0.2355 0.36 LLP 0.2327 0.15 0.9720 b 0.0001 − 0.1650 0.12 CHFUND − 0.0003 0.09 − 0.0005 0.51 0.0350 0.04 LCO 0.1860 b0.0001 0.1076 0.08 0.5380 0.04 MEET2 − 0.0002 b0.0001 0.0009 b 0.0001 ΔROA 0.0030 0.07 − 0.0260 0.02 HOM − 0.0002 0.44 0.0005 b 0.0001 ROA 0.0240 0.01 − 0.0940 b 0.0001 CAP 0.0001 0.26 − 0.0002 0.03 0.0001 0.04 − 0.0004 0.94 GDP − 0.1076 b0.0001 − 0.0414 0.01 − 0.0693 b 0.0001 0.0260 0.20 Instruments Yes Yes Yes Yes System-weighted R2 = 0.46 Panel C: results of three-stage least squares estimation under Scenario 1: earnings meet or beat analysts' forecast by zero or one penny and pre-securitization gains meets or beat analysts' forecast. (threshold = FMEET1) Intercept − 0.0050 0.00 − 0.0028 0.01 − 0.0080 0.95 0.0023 0.11 GAIN 0.2533 0.12 0.9180 0.19 0.2355 0.36 LLP 0.3795 0.18 0.7918 b 0.0001 − 0.1650 0.12 CHFUND − 0.0002 0.69 − 0.0004 0.82 0.0003 0.04 LCO 0.1630 b0.0001 0.1074 0.01 0.5380 0.04 FMEET1 − 0.0001 b0.0001 − 0.0003 b 0.0001 ΔROA 0.0025 0.06 − 0.0260 0.01 HOM − 0.0001 0.28 0.0005 b 0.0001 ROA 0.0330 b0.0001 − 0.0420 b 0.0001 CAP 0.0002 0.32 − 0.0002 0.03 0.0001 0.04 − 0.0003 0.76 GDP − 0.1479 b0.0001 − 0.0354 0.04 − 0.0930 b 0.0001 0.0240 0.20 Instruments Yes Yes Yes Yes 2 System-weighted R = 0.45 Panel D: results of three-stage least squares estimation under Scenario 2: earnings meet or beat analysts' forecast by zero or one penny but pre-securitization gains miss analysts' forecast (threshold = FMEET2) Intercept − 0.0028 0.00 − 0.0150 b 0.0001 − 0.0080 0.95 0.0020 0.13 GAIN 0.1680 0.12 0.9240 0.13 0.1724 0.45 LLP 0.4730 0.11 0.9710 b 0.0001 − 0.1771 0.10 CHFUND − 0.0002 0.66 − 0.0004 0.66 0.0003 0.04 LCO 0.1960 b0.0001 0.1375 b 0.0001 0.5610 0.77 FMEET2 − 0.0001 b0.0001 0.0004 b 0.0001 ΔROA 0.0030 0.01 − 0.0250 0.03 HOM − 0.0001 0.53 0.0005 b 0.00001 ROA 0.0560 b0.0001 − 0.0540 b 0.0001 CAP 0.0001 0.45 − 0.0002 0.03 0.0001 0.04 − 0.0003 0.72 GDP − 0.1672 b0.0001 − 0.0514 0.05 − 0.0977 b 0.0001 0.0220 0.21 Instruments Yes Yes Yes Yes 2 System-weighted R = .43 Variable definitions: LLP = loan loss provisions scaled by total assets at q-4; GAIN = gains from loan sales and securitization scaled by total assets at q-4; CHFUND = the sum of prior year-end common stock, preferred stock, and capital notes issued scaled by total assets at q-4; LCO = net loan charge-offs scaled by total assets at q-4; MEET1 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq – GAINq > INq-4, and 0 otherwise; MEET2 = 1 if 0 b = INq − INq-4 b 0.2%*ASSETSq-4 and INq – GAINq b INq-4, and 0 otherwise; FMEET1 = 1 if earnings before gains from securitization and loan sales meet or beat analysts' forecasts by zero or one penny, and 0 otherwise; FMEET2 = 1 if earnings meet or beat analysts' forecast by zero or one penny but earnings before gains from securitization and loan sales miss analysts forecast, and 0 otherwise; ΔROA = one-year ahead changes in earnings before loan loss provisions and taxes, scaled by total assets at quarter q-4; HOM = % of homogenous loans (1–4 family residential mortgage, consumer loans, loans to financial institutions, or acceptances to other banks) to total loans; ROA = earnings before loan loss provisions and taxes, scales by total assets at quarter q-4; CAP = total capital ratio; GDP = GDP growth is real growth in GDP per capita (annual %); Instruments in LLP equation = SIZE; ΔNP_LOAN; LOAN_IND; LOAN_FG; LOAN_DEP; LOAN_COM; LOAN_AGR; LOAN_ LLR; and LCO (defined in Table 2); Instruments in GAIN equation = SIZE; UN_GL (defined in Table 2); Instruments in LCO equation = NPL (prior year-end non-performing loans scaled by total assets at q-4); LLR; Instruments in CHFUND equation = NOTE (prior year-end capital notes and preferred stock scaled by total assets at q-4); EQUITY (prior year-end common equity scaled by total assets at q-4).

36

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37

and FMEET4 describe the scenario where earnings before discretionary loan loss provisions miss the targets and pre-securitization earnings miss earnings targets as well. Under this scenario I expect the coefficients on MEET4/FMEET4 in Eq. (3b) to be positive given the incentives in managing gains from loan sales and securitizations. Conversely, variables MEET5 and FMEET5 describe the scenario where pre-securitization earnings meet earnings targets. Due to lack of incentives under this scenario, I expect the coefficients on MEET5/FMEET5 in Eq. (3b) to be insignificant or negative. Untabulated results indicate that consistent to the expectation, I find that the coefficients on MEET4/FMEET4 (MEET5/ FMEET5) in Eq. (3b) are significantly positive (negative), supporting the argument that bank managers securitize loans to increase earnings when the incentives are present. In addition, I find that the coefficients on MEET4/FMEET4 and MEET5/FMEET5 in Eq. (3a) are all negative, significant at the 0.0001 level, suggesting that loan loss provisions are used to manage earnings with or without the incentives to manage gains from loan sales and securitizations. Taken as a whole, these results provide further substantiation consistent with the supposition that banks manage earnings through loan loss provisions, ahead of structuring transactions, to avoid small earnings decrease and to meet/beat analysts' forecasts. 6. Summary and conclusions This study examines the relationship between managing specific accruals and structuring transactions in meeting or beating earnings target and the financial reporting consequences when both of these two instruments are used. The objective of this study is to gain a better understanding of how managers use specific accruals and structure transactions to influence reported earnings. My results indicate that managers have a preference to pick the instruments. Specifically, managers use loan loss provisions to achieve earnings benchmarks without relying on gains from loan sales and securitizations. The findings suggest that banks manage loan loss provisions more pervasively than, and before, structuring loan sales and securitizations. In addition, I discover that earnings of banks reporting both lower discretionary loan loss provisions and higher discretionary gains from loan sales and securitizations are priced more negatively. The findings suggest that investors impose an incremental penalty when bank managers simultaneously manipulating specific accruals and structuring transactions. This study gives insight into the rationale for managers' choices of earnings management instruments. My results support the notion that mangers' choice of each instrument is a function of the firms' ability to use the instrument and the costs of doing so. This study adds to a stream of research that documents earnings management through managing special accruals and structuring transactions (Dechow et al., 2008; Karaoglu, 2005; Niu & Richardson, 2004). Moreover, evidence from the market reaction to the joint use of these two instruments has implications in making investment decisions. Acknowledgments I would like to acknowledge the extensive help provided by Dechun Wang and David Smith. I also thank two anonymous referees for their useful comments. Appendix A Effective in 1997, SFAS 125 “Accounting for Transfers and Servicing of Financial Assets and Extinguishment of Liabilities,” was introduced to standardize the accounting treatment for securitizations. Using a fair value perspective, SFAS 125 created incentives for managers to manipulate earnings through securitizations. However, very few firms were voluntarily disclosing the details of securitizations after

the release of SFAS 125. 43 As a result, SFAS 140, effective in 2000, required firms to disclose the details of securitization gains. The accounting standard SFAS140 uses the approach “surrender or control” to determine subsequent treatment. Securitization transactions can be classified as sales or secured borrowings. The transfer of receivables is viewed as a sale if the transferor surrenders control over assets. SFAS 140, paragraph 9 specifies the transfer of financial assets as a sale when the following three conditions have been met: (1) the transferred assets have been isolated from the transferor; (2) the transferee has the right to pledge or exchange the assets; and (3) the transferor does not maintain effective control over the assets. In a sale, derecognition is required. Thus, the firm has no rights to claim future cash flows from the receivables. Under secured borrowings, the accounts receivables are left on the books until the customers pay, and any cash receipts from securitizations are recorded as loans. The secured borrowings treatment has no ability to affect income. As compared to secured borrowings, the sales treatment has several accounting benefits. First, leverage is lower given that securitization is viewed as off-balance sheet financing. Second, efficiency ratios such as days-sales-outstanding are improved due to the removal of receivables on balance sheet. Finally and most importantly, the sales treatment offers the opportunity for managers to overvalue the retained interest and determine the gains recorded in the income statement. In the discussions with securitizations specialists, Dechow et al. (2008) document that managers have little incentives to structure a transaction as a secured borrowing, because a typical purpose for structuring transactions is to achieve gains with the use of sale accounting.44 The recent financial crisis has triggered the criticism on the use of fair value accounting rule SFAS140. As a consequence, the FASB issued SFAS 166, Accounting for Transfers of Financial Assets and No. 167, Amendments to FASB Interpretation No. 46(R), effective in 2010. The new standards require additional disclosures on securitization transactions. Robert Herz, chairman of the FASB, said: “These changes were proposed and considered to improve existing standards and to address concerns about companies who were stretching the use of off-balance sheet entities to the detriment of investors. The new standards eliminate existing exceptions, strengthen the standards relating to securitizations and special-purpose entities, and enhance disclosure requirements. They'll provide better transparency for investors about a company's activities and risks in these areas.” References Adiel, R. (1996). Reinsurance and management of regulatory ratios and taxes in the property-casualty industry. Journal of Accounting and Economics, 22(1–3), 207–240. Akerlof, G. (1970). The market for ‘lemons’: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 89, 488–500 (August). Ahmed, A., Takeda, C., & Thomas, S. (1999). Bank loan loss provision: A reexamination of capital management, earnings management and signaling effects. Journal of Accounting and Economics, 28, 1–25 (November). Barth, M., Beaver, W., & Stinson, C. (1991). Supplemental data and the structure of the thrift share prices. The Accounting Review, 66, 56–66 (January). Barth, M., Elliott, J., & Finn, M. (1999). Market rewards associated with patterns of increasing earnings. Journal of Accounting Research, 37, 387–413 (Autumn). Barth, M., & Landsman, W. (2010). How did financial reporting contribute to the financial crisis? The European Accounting Review, 19(3), 399–423. Bartov, E., Givoly, D., & Hayn, C. (2002). The rewards to meeting or beating earnings expectations. Journal of Accounting and Economics, 33, 173–204. Barua, A., & Cready, W. (2008). Classification Shifting and Special Items: Evidence of earnings management or a research design consequence? Working paper. Beatty, A., Chamberlain, S., & Magliolo, J. (1995). Managing financial reports of commercial banks: The influence of taxes, regulatory capital, and earnings. Journal of Accounting Research, 33, 231–261 (Autumn). Beatty, A., Ke, B., & Petroni, K. (2002). Earnings management to avoid earnings declines across publicly and privately held banks. The Accounting Review, 77(3), 547–570. 43 Karaoglu (2005) provides empirical evidence that banks use gains from loan sales and securitizations under SFAS125 to influence both reported earnings and regulatory capital. In a response to the fair value perspective, the new accounting standard SFAS140 was passed to address the limitations of SFAS125. 44 Following Dechow et al. (2008), I treat all securitization transactions as sales.

X. Cheng / Advances in Accounting, incorporating Advances in International Accounting 28 (2012) 22–37 Beatty, A., & Harris, D. (1999). The effects of taxes, agency costs and information asymmetry on earnings management: A comparison of public and private firms. Review of Accounting Studies, 4(3&4), 299–326. Beaver, W., Eder, C., Ryan, S., & Wolfson, M. (1989). Financial reporting, supplemental disclosures, and bank share prices. Journal of Accounting Research, 27, 157–178 (Autumn). Beaver, W., McNichols, M., & Nelson, K. (2003). Management of the loss reserve accrual and the distribution of earnings in the property-casualty insurance industry. Journal of Accounting and Economics, 35(3), 347–376. Beaver, W., McNichols, M., & Nelson, K. (2007). An alternative interpretation of the discontinuity in earnings distributions. Review of Accounting Studies, 12(4), 525–556 (December). Berndt, A., & Gupta, A. (2009). Moral hazard and adverse selection in the originate-todistribute model of bank credit. Working paper. Borio, C., Furfine, C., & Lowe, P. (2001). Procyclicality of the financial system and financial stability: Issues and policy options. BIS Paper(No. 1), 1–57. Brown, L., & Caylor, M. (2005). A temporal analysis of quarterly earnings thresholds: Propensities and valuation consequences. The Accounting Review, 80(2), 423–440. Burgstahler, D., & Dichev, I. (1997). Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics, 24(December), 99–126. Chen, W., Liu, C., & Ryan, S. (2007). Characteristics of Securitizations that Determine Issuers' Retention of the Risks of the Securitized Assets. Working paper. Cheng, Q., Warfield, T., & Ye, M. (2011). Equity incentives and earnings management: Evidence from the banking industry. Journal of Accounting, Auditing and Finance, 26, 317–349. Cheng, Q., & Warifield, T. (2005). Equity incentives and earnings management. The Accounting Review, 80(2), 441–476. Cohen, D., & Zarowin, P. (2010). Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics, 50, 2–19. Collins, J., Shackelford, D., & Wahlen, J. (1995). Bank differences in the coordination of regulatory capital, earnings and taxes. Journal of Accounting Research, 33, 263–291. Cox, J., Faucette, J., & Lickstein, C. (2010). Why Did the Credit Crisis Spread to Global Markets? http://www.uiowa.edu/ifdebook/ebook2/contents/part5-II.shtml Dechow, P. M., Myers, L., & Shakespeare, C. (2008). Fair Value Accounting and Gains from Asset Securitizations: A Convenient Earnings Management Tool with Compensation Side-Benefits. Working paper, University of Michigan, University of Illinois and University of Michigan. Dechow, P. M., & Shakespeare, C. (2009). Do managers time securitization transactions to obtain accounting benefits? The Accounting Review, 84, 99–132. Dechow, P., Richardson, S., & Tuna, I. (2003). Why are earnings kinky? An examination of earnings management explanation. Review of Accounting Studies, 8, 355–384. Degeorge, F., Patel, J., & Zeckhauser, R. (1999). Earnings management to exceed thresholds. Journal of Business, 27, 1–33 (January). Demsetz, R. (2000). Bank Loan Sales: A New Look at the Motivations for Secondary Market Activity. The Journal of Financial Research, 23, 197–222. Dickson, G. (2011). Securitization Litigation & the Foreclosure Crisis: What's on the Horizon? http://www.theshortsaleguide.com/profiles/blogs/securitization-litigationamp Durtschi, C., & Easton, P. (2005). Earnings management? The shape of the frequency distribution of earnings metrics are not evidence ipso facto. Journal of Accounting Research, 43, 557–592. FASB, Financial Accounting Standard Board (1975). Accounting for Contingencies. Statement of Financial Accounting Standards No. 5. FASB, Financial Accounting Standard Board (1996). Accounting for Transfers and Servicing of Financial Assets and Extinguishments of Liabilities. Statement of Financial Accounting Standards No. 125. FASB, Financial Accounting Standard Board (2000). Accounting for Transfers and Servicing of Financial Assets and Extinguishments of Liabilities-a replacement of FASB Statement No. 125. Statement of Financial Accounting Standards No. 140. Fan, Y., & Thomas, W. (2011). Managing Earnings Using Classification Shifting: Evidence from Quarterly Special Items. The Accounting Review, 85(4), 1303–1323. Flannery, M. (2011). The Faces of Market Discipline. Journal of Financial Services Research, 20(2), 107–119. Graham, J. R., Harvey, C., & Rajgopal, S. (2005). The economic implications of corporate financial reporting. Journal of Accounting and Economics, 40, 3–73. Harris, D., Shi, L., & Xie, H. (2010). Is Meeting and Beating Earnings Benchmarks Associated with Accounting Fraud? Working paper. : Syracuse University.

37

Healy, P., & Palepu, K. (1993). The effect of firms' financial disclosure strategies on stock prices. Accounting Horizons, 7(1), 1–11. Healy, P., & Wahlen, J. (1999). A review of the earnings management literature and its implications for standard setting. Accounting Horizons, 365–383 (December). Kanagaretnam, K., Lobo, G., & Mathieu, R. (2003). Managerial incentives for income smoothing through bank loan loss provisions. Review of Quantitative Finance and Accounting, 20, 63–80. Kanagaretnam, K., Lobo, G., & Yang, D. (2004). Joint Tests of Signaling and Income smoothing through Bank Loan Loss Provisions. Contemporary Accounting Research, 21(4), 843–884. Karaoglu, E. (2005). Regulatory capital and earnings management in banks: the case of loan sales and securitization. Working paper. : University of Southern California. Kim, M., & Kross, A. (1998). The Impact of the 1989 Change in Bank Capital Standards on loan loss provisions and loan write-offs. Journal of Accounting and Economics, 25, 69–99. Koh, K., Matsumoto, D., & Rajgopal, S. (2008). Meeting or beating analyst expectations in the post-scandals world: Changes in stock market rewards and managerial actions. Contemporary Accounting Research, 25, 1067–1098. Lee (2007). Earnings Management to just meet analysts' forecast. Working paper. Li, F., Rider, E., & Moore, E. (2009). Accrual Based Earnings Management, Real Transactions Manipulation and Expectations Management: U.S. and International Evidence. Working paper. Liu, C., & Ryan, S. (1995). The effect of bank loan portfolio composition on the market reaction to and anticipation of loan loss provisions. Journal of Accounting Research, 33, 77–94. Liu, C., & Ryan, S. (2006). Income smoothing over business cycle: changes in banks' coordinated management of provisions for loan losses and loan charge-offs from the pre-1990 bust to the 1990s boom. The Accounting Review, 81(2), 421–441. Marquardt, C., & Wiedman, C. (2004). How are earnings managed? An examination of specific accruals. Contemporary Accounting Research, 21, 461–491. Marquardt, C., & Wiedman, C. (2005). Earnings management through transaction structuring: contingent convertible debt and diluted EPS. Journal of Accounting Research, 43(2), 205–243. Matsumoto, D. (2002). Management's incentives to avoid negative earnings surprises. The Accounting Review, 77(3), 483–514. Matsunaga, S., & Park, C. (2001). The effect of missing a quarterly earnings benchmark on the CEO's annual bonus. The Accounting Review, 76, 313–332. McNichols, M., & Wilson, P. (1988). Evidence of earnings management from the provision of bad debts. Journal of Accounting Research, 12, 123–154. McVay, S. (2006). Earnings management using classification shifting: An examination of core earnings and special items. The Accounting Review, 81, 501–531 (October). Michalak, T., & Uhde, A. (2009). Credit risk securitization and banking stability: Evidence from the micro-level for Europe. Working paper. Milgrom, P., & Roberts, J. (1982). Prediction, reputation, and entry deterrence. Journal of Economic Theory, 27, 280–312. Minton, B., Opler, T., & Stanton, S. (1997). Asset Securitization among Industrial Firms. Working Paper. : Ohio State University. Moyer, S. (1990). Capital adequacy ratio regulations and accounting choices in commercial banks. Journal of Accounting and Economics, 13(issue 2), 123–154. Niu, F., & Richardson, G. (2004). Earnings quality, off-balance sheet risk, and the financial components approach to accounting for transfers of financial assets. Working paper. Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42(3), 335–370. Robb, S. (1998). The effect of analysts' forecasts on earnings management in financial institutions. Journal of Financial Research, 21(3), 315–332. Skinner, D. J., & Sloan, R. (2002). Earnings surprises, growth expectations and stock returns or don't let an earnings torpedo sink your portfolio. Review of Accounting Studies, 7, 289–312. Wahlen, J. (1994). The nature of information in commercial bank loan loss disclosures. The Accounting Review, 69, 455–478. White, H. (1980). A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica, 48, 817–838. Zang, A. (2007). Evidence on the tradeoff between real manipulation and accrual manipulation. Working Paper. : University of Rochester.