The curious case of Level 3 instruments

ARTICLE IN PRESS Research in Accounting Regulation ■■ (2017) ■■–■■

Contents lists available at ScienceDirect

Research in Accounting Regulation j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / r a c r e g

Regular Paper

The curious case of Level 3 instruments Robson Glasscock a,*, David W. Harless b, Jack Dorminey c a

University of Wyoming, USA Virginia Commonwealth University, USA c West Virginia University, USA b

A R T I C L E

I N F O

Article history: Available online Keywords: Fair value Level 3 Earnings management

A B S T R A C T

Standard setters and regulators face an ever-present concern over the discretionary influence firms have in financial reporting. For information to have enhanced relevance, some level of discretion in financial reporting is often necessary. Prior work suggests that firms may be opportunistic in exercising choice and influence where discretion is available to advantageously affect reported results. This study examines if aggressive firms take the opportunity afforded by the wide discretion in Level 3 valuations under the original Accounting Standards Codification (ASC) 820 standard to manipulate financial reporting. Minimal evidence is found to support an association between Level 3 valuations and other metrics reflecting earnings management. The findings may be driven by the high-level, and typically limited, disclosures that firms are required to make under the originally promulgated ASC 820. This primary finding suggests that FASB’s move to increase the disclosures required under the standard was warranted. © 2017 Published by Elsevier Ltd.

Introduction This study examines if the discretion afforded under Accounting Standards Codification (ASC) 820 – Fair Value Measurements and Disclosures (ASC 820) for valuations that rely on significant unobservable inputs is used by firms to alter financial reporting. Standard setters and regulators worry about the impact of managerial discretion on reported results. Among the transactions giving rise to such concerns is fair value accounting and especially Level 3 valuations. As the representation of fair valuation expands within the Codification, the effect of subjective inputs and discretionary aspects of reporting will increase in importance.1

The accounting framework, by its very design, involves a tension between relevance and reliability (Laux & Leuz, 2009). In the interest of relevance, some level of discretionary reporting is often necessary.2 Such is the case when fair values are reported for assets or liabilities where no observable or comparable market price exists and are therefore based on significant unobservable inputs (identified as ‘Level 3’ under ASC 820). The expanded discretion afforded may amplify the opportunity for biased reporting (i.e., a threat to faithful representation). Yet, this leeway may be necessary in the interest of relevant reporting. The objective of this study is not to determine to what degree the trade-off is appropriate, but rather to assess if this standard results in opportunistic reporting due to the level of discretion.

Data Availability: XBRL data used in this study are available from Calcbench, Inc. All other data are from public sources identified in the study. * Corresponding author. Fax: 307-766-4028 E-mail address: [email protected] (R. Glasscock). 1 Efforts surrounding convergence with IFRS have been associated with an increased role of fair value metrics in financial reporting.

2 Prior to 2010, much of the fair value literature deals with a tradeoff between relevance and reliability. Reliability refers to a measure that can be validated. In its update of the Conceptual Framework (FASB, 2010a, BC3.20-3.25), FASB replaced reliability with faithful representation. Faithful representation, in the context of the Conceptual Framework, is used to indicate that the information is complete, neutral, and free from error.

http://dx.doi.org/10.1016/j.racreg.2017.04.006 1052-0457/© 2017 Published by Elsevier Ltd.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS 2

R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

Assuming aggressively reporting firms seek additional opportunity, the discretion allowed under the current guidance found in ASC 820 may provide another avenue for advantageous reporting. In this study, the association between firm aggressiveness and Level 3 incomes is evaluated. Aggressiveness in financial reporting is assessed in three ways, all of which have been used extensively in past studies. Aggressiveness is measured as (1) the absolute value of prior period discretionary accruals3,4 (Dechow, Sloan, & Sweeney, 1995; Kim, Park, & Wier, 2012; Kothari, Leone, & Wasley, 2005), (2) composite real activities manipulation5 (Cohen & Zarowin, 2010; Roychowdhury, 2006), and (3) “Street” earnings that are greater than or equal to analysts’ consensus estimates (Matsumoto, 2002; Skinner & Sloan, 2002). Further, both discretionary accruals and real earnings management are estimated using multiple empirical model specifications. A positive association between the aggressiveness measures and changes in Level 3 valuations recognized in earnings would support the notion that aggressive firms may implement ASC 820 in a way that biases reported financial results. However, the results do not provide conclusive evidence that otherwise aggressive firms report biased (i.e., overstated) gains/losses on Level 3 instruments. Various analyses and robustness tests are conducted: (1) univariate tests, (2) multivariate tests, (3) constraining the analyses to “markto-market” adjustments which are recognized in earnings, and (4) a variety of “suspect firm” analyses. Results of the analyses generally do not support the conjecture that earnings are manipulated via the allowable discretion in Level 3 estimates. The apparent lack of an association is unexpected, particularly in light of the Public Company Accounting Oversight Board’s (PCAOB) recommendation of expanded audit evidence in fair valuation situations,6 suggesting an expectation that these valuations represent an increased audit risk. One potential explanation for the lack of an associative finding is that the standard does not provide sufficiently precise reporting disclosures that are necessary to differentiate between normal and aggressive changes in Level 3 valuations. This assertion is consistent with FASB’s ongoing attention to the reporting and disclosure requirements for Level 3 valuations.7 Specifically, FASB has made nontrivial expansions in the level of valuation disclosures required by

3 Discretionary accruals refer to the use of excessively employed accruals with the objective of altering reported results. 4 A total of seven measures are used because discretionary accruals, while a single category of measure, is assessed in five alternative representations. 5 Roychowdhury (2006, p.336) defines real activities management as “. . . management actions that deviate from normal business practices, undertaken with the primary objective of meeting certain earnings thresholds.” 6 The AICPA’s issuance of the Clarified Statements on Auditing Standards regarding the evaluation of misstatements identified during an audit, AU-C 450 (AICPA, 2014), also suggests that the existing audit practice surrounding these valuations needed to be enhanced for nonissuers, as well. 7 See Deloitte’s summary of the March 4, 2015 FASB discussion related to the fair value measurement guidance in ASC 820. http://www.iasplus .com/en/publications/us/aje/2015/0305.

management under ASC 820 since its initial release. This suggests that FASB may have realized that the disclosures under the initial ASC 820 were insufficient to provide market participants with adequately useful information. The findings herein indicate FASB’s actions were necessary.

Context and motivation On May 12, 2011, the Financial Accounting Standards Board (FASB) issued a press release stating that the FASB and the International Accounting Standards Board (IASB) completed a significant milestone in the process of moving towards a single, global set of high-quality financial accounting standards via the adoption of common standards governing fair value accounting techniques and disclosures (FASB, 2011a). Despite the world-wide importance of fair value accounting, relatively little empirical evidence exists regarding modern fair value standards. This sentiment is expressed by DeFond (2010): “Going forward, I think there are accounting developments on the horizon about which we know relatively little, and hence are logical prospects for future research. One example is fair value accounting, which represents a potentially sea-changing development in the accounting environment.” Additionally, the augmented disclosures required by ASC 8208 (FASB, 2010b) are thought by some to provide information which may be used to construct more direct tests of managerial discretion in fair value estimates than were possible previously (Barth & Taylor, 2010). This study examines disclosures under ASC 820 to determine if aggressive firms use the discretion over the inputs used in the Level 3 category9 asset valuations to report biased (overstated) gains/losses10 for those instruments. Specific examples of Level 3 items include (1) auction rate securities tied to collateralized student loan debt, (2) investments in hedge funds, (3) investments in private equity firms, (4) collateralized debt obligations, (5) credit default swaps, and (6) derivatives relating to commodity basis differentials. Biased gains/losses may be either unrealized or realized. The unrealized gains/losses used in this study are attributable to Level 3 financial instruments which are not designated as cash flow hedges or represent the ineffective portion of cash flow hedges. These unrealized gains/losses are booked to income statement accounts and affect the current period’s earnings. The realized gains/losses attributable to Level 3 holdings are recognized when the item is sold. While it

8 FASB’s ASC 820: Fair value Measurements and Disclosures subsumes the pre-codification standard SFAS 157 (FASB, 2006). 9 ASC 820 identifies fair value measurement as falling into one of three categories (levels): quoted prices in active markets for identical assets or liabilities (Level 1), (2), significant other observable inputs (Level 2), and (3) significant unobservable inputs (Level 3). 10 For example, a loss is overstated when a company objectively should have reported a loss of 50 but instead they recognize a loss of 30. Negative 30 is the larger number (i.e., [−30] – [−50] = 20).

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

is not likely that management can substantially influence the sales price of a Level 3 instrument in an arm’s length transaction in period t, management still has discretion over realized gains/losses in period t because management determined the fair value of the instrument in all periods between inception and sale. During the interim period, firms are able to build reserves or borrow from future performance. The practical significance of this research question is supported by recent PCAOB inspection reports and auditing standards. For instance, Acuitas, Inc. (2012) summarized PCAOB inspection reports between 2008 and 2010. The percentage of audits with deficiencies more than doubled between 2008 and 2010, and over half of the audit deficiencies were related to inadequate substantive testing of fair value and impairment. This is evidence of the difficulty that auditors face when auditing fair value estimates derived from significant subjective inputs. The PCAOB makes clear the difficulty in assessing proper valuation and, in response, recommends greater granularity in the verifications associated with audit. If the PCAOB has concluded that auditors may not be gathering sufficient appropriate audit evidence for fair value, this may also indicate that fair value estimates are a viable earnings management tool. This notion is supported by the Clarified Statements on Auditing Standards regarding the evaluation of misstatements identified during an audit, AU-C 450 (AICPA, 2014). AU-C 450 requires auditors to consider lowering materiality for misstatements that, for example, alter the current period’s financial metrics to be in accordance with “previous communications.” AU-C 450 contains specific examples of misstatements which may require a reduced materiality threshold. These misstatements include items which increase equity compensation for managers or bring earnings closer to previously issued earnings forecasts. Bartov, Givoly, and Hayn (2002) conclude that meeting-or-beating results in abnormal returns. Therefore, misstatements that allow firms to narrowly meet or beat expected earnings are one example of errors that auditors should consider for materiality reductions. It is clear that both the PCAOB and AICPA have concerns over the discretion afforded to managers under the current standard and are recommending increased audit vigilance as a result. This is taken as prima facie evidence that the current audit practice surrounding this type of valuation may be insufficient given the perceived risk. Simply stated: how does one audit or attest to what cannot be verified? This paper extends the post-SFAS 157/ASC 820 fair value accounting literature in two primary ways. First, Valencia (2011) and Fiechter and Meyer (2009) provide evidence that fair value accounting is used as an earnings management tool conditional upon the properties of the firm’s earnings. Both Valencia (2011) and Fiechter and Meyer (2009) conduct analyses where the primary explanatory variables are related to earnings, changes in earnings, or net income. Traditional financial reporting aggressiveness measures (e.g., discretionary accruals, meeting-or-beating, and real activities manipulation) have not been previously considered by researchers. Second, all of the modern (i.e., post-

3

SFAS 157) fair value studies have examined fair value accounting using samples consisting solely of financial services firms. Financial services firms have unique operating and accounting environments, and findings supporting the notion that managers willfully overstate fair value estimates may not be generalizable to nonfinancial services firms. Relevant literature Relevance vs. reliability The debate surrounding fair value accounting centers on two desirable characteristics of financial reporting, relevance and reliability,11 and the tradeoff between them. Laux and Leuz (2009) write that accounting standards setters have debated the tradeoff between relevance and reliability for decades. Laux and Leuz (2009) focus on banks using fair value accounting in times of financial turmoil, the viability of historical cost accounting as the de facto alternative to fair value accounting, implementation problems with fair value, and litigation risks managers face when deviating from observable market prices during financial crises. The authors also cite empirical evidence that managers fail to take writedowns on assets that are materially impaired. Laux and Leuz (2009, 827–828) write, “Few dispute that transparency is important. But the controversy rests on whether fair value accounting is indeed helpful in providing transparency and whether it leads to undesirable actions on the parts of banks and firms.” This study provides direct evidence regarding one such undesirable action (e.g., a firm intentionally manipulating fair value estimates to manage earnings). The ability of managers to influence Level 3 fair value estimates may be substantial. Federal Reserve researchers report that fair value estimates for bank loans can vary between 200 and 500 basis points depending on the particular valuation inputs and model used (Bell & Griffin, 2012; Bies, 2004). Additionally, PCAOB inspections in 2010 reveal that auditors fail to appropriately evaluate the reasonableness of significant assumptions used by management and also rely on inquiry alone rather than a combination of inquiry and other supporting evidence (Bell & Griffin, 2012). This casts doubt on the argument that management cannot use fair value estimates as an earnings management tool due to constraints imposed by auditors. Some researchers argue that fair value standards applied to more discretionary (e.g., Level 3) items are a “boon” to aggressive executives (Aboody, Barth, & Kasznik, 2006; Bartov, Mohanram, & Nissim, 2007; Bell & Griffin, 2012). Similar to Bell and Griffin (2012), Benston (2008) is also critical of the ability of auditors to constrain opportunistic and overly optimistic managers. He goes on to provide several examples where fair value accounting may be used opportunistically and cautions that dishonest managers may easily use fair value accounting to manipulate earnings. This study attempts to provide direct empirical evidence of the valid-

11 Reliability was replaced with faithful representation in the Conceptual Framework in 2010.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

4

ity of Benston’s (2008) concerns by studying the fair value estimates of firms that appear to be relatively more aggressive using established and academically accepted measures. Fair value relevance and opportunistic use of fair value estimates Existing ASC 820 papers fall into two broad categories. One group examines the value-relevance of fair value estimates and disclosures while the second attempts to determine if managers use their discretion over fair value in an opportunistic manner. The value-relevance studies (Du, Li, & Xu, 2014; Goh, Ng, & Yong, 2009; Kolev, 2009; Lu & Mande, 2014; Song, Thomas, & Yi, 2010) often employ a modified version of the Ohlson (1995) model. The Ohlson (1995) model expresses share price as a function of book value and the present value of cumulative expected abnormal earnings. Value-relevance studies decompose the book value of the firm into assets/liabilities carried at historical cost and the ASC 820 framework for assets/liabilities carried at fair value (e.g., Level 1, Level 2, and Level 3). The relationship between each asset/liability level and share price shows that, while even the most subjective fair value estimates are still value-relevant, investors place less reliance on fair value adjustments that are subject to larger amounts of managerial discretion. Valencia (2011) and Fiechter and Meyer (2009) explore whether financial services institutions use Level 3 fair value estimates opportunistically. Valencia (2011) concludes that banks which would have reported losses had they not been able to recognize the Level 3 valuation changes in earnings, report unrealized gains that are higher than those banks that would have reported positive earnings even without recognized Level 3 valuation changes. Fiechter and Meyer (2009) examine whether banks use the discretionary nature of Level 3 fair value estimates to smooth earnings. Fiechter and Meyer (2009) find a negative association between unrealized gains/losses on Level 3 instruments and net income before unrealized gains/losses on Level 3 instruments. They conclude that this is evidence consistent with firms intentionally using Level 3 valuations to smooth earnings. Earnings announcements Findings from prior research are also consistent with firms being rewarded for reporting earnings consistent with the market’s expectations (Bartov et al., 2002; Skinner & Sloan, 2002). Skinner and Sloan (2002) explore the curious phenomenon of portfolios of growth stocks underperforming portfolios of value stocks. They find that this underperformance is primarily explained by large and asymmetric price reactions when growth stocks report earnings which are considered negative surprises by the market. The prices of value stocks do not decline as significantly if these firms report earnings that are less than the market’s expectations. Bartov et al. (2002) examine the market’s reactions to firms meeting-or-beating analysts’ consensus estimates. Primary conclusions of the study are that while firms are rewarded for both meeting and beating earnings expectations, the premium for beating expectations is larger than the premium for simply meeting expectations. This

reward continues to hold in cases where firms likely engaged in either expectations management or earnings management. Firms that appeared to use earnings management to report earnings in accordance with analysts’ expectations received a slight but economically insignificant discount to the premium. Taken together, these findings provide support for an expectation that firms may intentionally report biased estimates of fair value. Whether or not aggressive managers abuse the leeway they have in determining the fair value of Level 3 items remains an open empirical question. Conclusions from the earnings announcement literature, fair value relevance literature, and opportunistic use of fair value studies cited above provide support for the expectation that aggressiveness in financial reporting is positively associated with realized and unrealized gains/ losses on Level 3 instruments which are recognized in earnings. Empirical model The research design used in this paper differs from both Valencia (2011) and Fiechter and Meyer (2009) in that management’s use of ASC 820 Level 3 fair value estimates are examined conditional upon the characteristics of the firm’s financial reporting environment (discretionary accruals and meeting-or-beating) or operating decisions (real earnings management) and not solely upon the characteristics of the firm’s reported earnings. The empirical model attempts to isolate changes in Level 3 instruments based upon equity market volatility, the risk-free rate, firm size, firm profitability, the “fear index,” and the type of financial instrument held before estimating the impact of aggressive financial reporting on Level 3 valuations. See Appendix A-1, Empirical Model, which contains the estimation equation, variable definitions, and support for the inclusion of each measure. Data collection and classification Data sources Analyst forecast data to calculate meeting-or-beating (MBE) is obtained from IBES via the Thomson Reuters Spreadsheet Link (TRSL) interface. Financial accounting data is obtained from COMPUSTAT. The closing price of the Volatility Index (VIX) for the S&P 500 is downloaded from the Chicago Board Options Exchange (CBOE) via WRDS. The price of the S&P 500 Composite Index is obtained from CRSP’s Daily Stock file, and the three-month treasury yield is downloaded using the Federal Reserve Economic Data Microsoft Excel add-in.12 Level 3 data collection The United States Securities and Exchange Commission (SEC) requires that large accelerated filers (i.e., firms with common equity market capitalization of at least $700 million) prepare their financial statements in an interac-

12

https://fred.stlouisfed.org/.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

5

Table 1 Initial sample of XBRL and hand-collected data. Panel A: Reconciliation of XBRL data to Hand-Collected Sample Observations XBRL Data Obtained from Calcbench, Inc. Less: Missing Value for Realized and Unrealized Gain/Loss Tag Duplicate Observation Financial Services Firms Cumulative 6-month and 9-month Data Annual Realized and Unrealized Data Add: Interpolated Observations Initial XBRL Sample Less: No Level 3 Roll-forward Table in 10-Q/10-K Only Cumulative 6- or 9-Month Reported Total Realized and Unrealized Gains/Losses Not Reported OCI vs. Earnings Not Disclosed Foreign Filer Without Any 10-Q’s/10-K’s No 10-Q or 10-K Filed Interpolation Not Possible Due to Missing Observations in Panel Initial Hand-Collected Sample

1,409 (1) (1) (613) (282) (212) 192 492 (6) (33) (8) (11) (6) (5) (45) 378

Panel B: Reconciliation of Hand-Collected Sample to Total Observations in Each FVE Model

Initial Hand-Collected Sample Less: Missing Values- PPEGTQ Missing Values- XSGAQ Missing Values- Other Total FVE Observations

Discretionary

Real Activities

Accruals Models

Manipulation Models

MBE

378 (40) (14) 324

378 (159) (15) 204

378 (5) 373

FVE is realized and unrealized gains and losses in period t scaled by total assets in period t-1. PPEGTQ is gross property, plant, and equipment. XSGAQ is selling, general, and administrative expenses.

tive data format using eXtensible Business Markup Language (XBRL) for reports filed on or after June 15, 2009. The SEC intended the XBRL format to be more useful to investors because XBRL “tagged” documents enable investors to quickly and easily download data directly into spreadsheets. The SEC stated that this should help investors analyze the data in a variety of ways using commercially available software (SEC, 2009). Calcbench Inc. uses cloud-based computing to process and store data from all eXtensible Business Reporting Language (XBRL) tagged financial reports filed with the United States Securities and Exchange Commission. The population of publicly traded firms with realized or unrealized gains/losses on Level 3 assets was obtained from Calcbench Inc. XBRL prepared documents are mandated beginning the second calendar quarter of 2009, but interpolation based on data in the 10-Q’s is possible. Therefore, the sample period in this study includes the first calendar quarter in 2009 through the second calendar quarter of 2012. After interpolation there are 175 firms in the sample and 902 firmquarters. Financial services firms (i.e., SIC codes 6000– 6799) are excluded due to the extant research on financial services firms and fair value accounting. As shown in Panel A of Table 1, the remaining XBRL sample includes 105 firms and 492 firm-quarters of which 410 firm-quarters report non-zero realized and unrealized gains/losses on Level 3 assets. A recent white paper from Columbia University’s Center for Excellence in Accounting & Security Analysis outlines

a series of problems in the current XBRL reporting environment. Among the problems discussed by Harris and Morsfield (2012) are low-quality XBRL document tagging due to limited liability of filers for errors combined with the fact that XBRL tagging is unaudited, filers utilizing incorrect XBRL tags, and errors causing the tagged data to not reconcile with the EDGAR filling. In fact, the first recommendation made by Harris and Morsfield (2012) is that the error rates of XBRL data be significantly reduced. Based on the findings of Harris and Morsfield (2012) data is hand-collected from EDGAR fillings for the 492 firmquarters discussed above. As shown in Tables 1 and 2, the hand-collected data contains 86 firms and 378 firmquarters of which 333 firm-quarters report non-zero realized and unrealized gains/losses on Level 3 assets. The XBRL data does not agree with the EDGAR fillings in nearly one quarter of cases, and the hand-collected data is used for all of the following descriptive statistics and empirical analyses. Panel B of Table 1 reconciles the hand-collected observations to the observations used in each FVE model. Level 3 classifications The average amount of total realized and unrealized gains/losses (unscaled) and unrealized gains/losses (unscaled), respectively, in the sample is $5.35 million and $5.39 million, respectively. The average number of common shares outstanding and the average net income reported is examined for all firms available in COMPUSTAT

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

6

Table 2 Sample industry membership and Level 3 instrument types. SIC

Description

Firms

Obs

Most Common Instrument

10 12 13 20 26 28 29 34 35 36 37 38 39 42 45 47 49 50 53 56 58 73 99

Metal Mining Coal Mining Oil and Gas Extraction Food Paper Chemicals Petroleum Refining Fabricated Metals Excluding Machinery Industrial and Commercial Machinery Electric Excluding Computers Transportation Equipment Measuring, Analyzing, and Controlling Miscellaneous Manufacturing Motor Freight Transportation Transportation by Air Transportation Services Electric, Gas, and Sanitary Services Wholesale Trade- Durable Goods General Merchandise Stores Apparel and Accessory Stores Eating and Drinking Places Business Services Nonclassifiable Establishments

1 2 9 1 1 9 2 1 6 6 1 3 2 1 2 1 24 1 2 1 1 8 1 86

2 14 36 3 4 33 3 1 25 29 4 17 3 10 3 2 146 2 5 1 2 25 8 378

Non-Specific Derivatives Collars, Power, and Physical Commodity Derivatives Collars, Power, and Physical Commodity Derivatives Other Investments, Contingent Consideration Auction Rate Securities Mortgage-Backed Instruments Other Investments, Contingent Consideration Auction Rate Securities Private Equity, Corporate Debt, Venture Capital Auction Rate Securities Asset-Backed Securities Auction Rate Securities Non-Specific Derivatives Auction Rate Securities Energy Derivatives- Oil and Natural Gas Other Investments, Contingent Consideration Non-Specific Derivatives Other Investments, Contingent Consideration Auction Rate Securities Auction Rate Securities Auction Rate Securities Non-Specific Derivatives Private Equity, Corporate Debt, Venture Capital

between 2000 and 2014. Approximately $2.94 million dollars in earnings is needed for the average firm in COMPUSTAT to alter EPS by one cent, and this provides support for the practical significance of the dependent variable in this study. Table 2 presents the sample by industry. Specifically, it details the number of unique firms and firm-quarters within each two-digit SIC code. The most frequent Level 3 instrument type within each industry is also shown in Table 2. The industries with the heaviest representation are Electric, Gas, Sanitary, Oil and Gas Extraction, and Chemicals. Three industries (Electric, Gas, and Sanitary firms) provide 56.87 percent of total firm-quarters in the sample. The least represented industries are Apparel and Accessory Stores and Fabricated Metals Excluding Machinery. Auction rate securities and non-specific derivatives are the most frequently occurring assets in the hand-collected sample. The majority of the auction rate securities are described in EDGAR fillings as being associated with collateralized student loan debt. Firms typically provide only a few sentences describing their Level 3 holdings. High-level descriptions of Level 3s and a valuation reconciliation are disclosed, but specific information (e.g., term structures, counterparties, credit risk, use of valuation specialists, maturities, etc.) is not currently required to be disclosed by ASC 820. For example, in the June 30, 2012 10-Q, American Electric Power defines Financial Transmission Rights (FTRs) as, “A financial instrument that entitles the holder to receive compensation for certain congestion-related transmission charges that arise when the power grid is congested resulting in differences in locational prices.” In Note 8, the company describes the FTRs as, “Certain OTC and bilaterally executed derivative instruments are executed in less active markets with a lower availability of pricing information. Long-dated and

illiquid complex or structured transactions and FTRs can introduce the need for internally developed modeling inputs based upon extrapolations and assumptions of observable market data to estimate fair value. When such inputs have a significant impact on the measurement of fair value, the instrument is categorized as Level 3” (American Electric Power Company, 2012, 58). The American Electric Power disclosure is an example of the discussion and detail commonly included in Level 3 disclosures. It is clear that users of the financial statements are left wondering what assumptions and other inputs were used in determining the reported values. 13 The disclosure provides facts surrounding the instrument, but very little information regarding its value. Attempting to control for valuation changes in instruments which differ considerably across the sample is difficult. This difficulty is magnified by brief corporate disclosures that do not provide enough detail for specific model inputs. In addition to those measures common to other fair value accounting studies, macroeconomic variables anticipated to impact a wide variety of financial instruments are also included. Prior studies (e.g., Fiechter & Meyer, 2009; Valencia, 2011) have included industry indicator variables based on twodigit SIC codes as aggregate measures of Level 3 instrument type. The rationale is that firms within the same industry will likely hold the same Level 3 instruments. This is a strong assumption given the variety of firms that are collapsed into each two-digit SIC code. Specifically, collapsing firms into groups based on the first two SIC digits assigns chocolate

13 In all of the Level 3 disclosure statements examined for this study, it is impossible to know what factors and what valuation techniques are employed in estimating the values that are reported in the financial statements.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

producers, chewing gum manufacturers, rice millers, and meat packing plants to the same group. It is reasonable to believe that meat packing plants may use very different derivative financial instruments than chocolate producers or chewing gum manufacturers. Indicator variables at the industry level also assume that every firm within the same industry will have the same investment risk tolerance and financial sophistication. Accordingly, binary variables are assigned based on hand-collected descriptions of Level 3 instruments. Results Descriptive statistics and univariate models See Panel A of Table 3 which is included in Appendix A-2 and contains summary statistics for all variables contained in the statistical models. The summary statistics reported are generally consistent with previously published research. Panel B of Table 3, which is also included in Appendix A-2, reports the results of tests of differences in means of each of the seven financial reporting measures based upon whether firms reported positive, zero, or negative values for realized and unrealized gains and losses on Level 3 instruments. No clear pattern emerges from the results in Panel B of Table 3. Table 4, which is included in Appendix A-3, contains univariate correlations among the variables of interest. Statistically, it becomes difficult to measure the “holding other things constant” impact of key variables when most of the variation between said variables is shared. The correlations reported in Table 4 do not indicate that the models herein will suffer from this problem. Table 5, which is included in Appendix A-4, represents univariate statistical tests of association between each aggressive financial reporting measures and Level 3 changes. Panel A of Table 5 examines total realized and unrealized gains and losses, while Panel B of Table 5 constrains the analysis to only “mark-to-market” adjustments. The reported results indicate that the aggressiveness measures are not strongly related to changes in Level 3 valuations. Multivariate models Results from multivariate statistical tests are presented in Table 6 of Appendix A-5. The surprising results show that neither measures of financial reporting aggressiveness nor underlying financial markets variables explain a meaningful portion of the variation in Level 3 instruments. Alternative model specifications and various measures taken from the hand-collection of EDGAR fillings are discussed at length in Appendix A-5. The primary conclusion that follows from the results in Appendix A-5 is that investors are simply not able to use disclosed information in SEC fillings to come up with objective, arm’s length valuation estimates for Level 3 instruments. Additional analyses and robustness checks Appendix A-6 contains Table 7 and re-estimates the empirical models using dependent variables only consisting of

7

“mark-to-market” adjustments to Level 3 items which are reported in current earnings. Therefore, these valuation changes are based solely on estimates and are thus most impacted by managerial discretion. The results of the analysis in Appendix A-6 are consistent with the results presented in Appendix A-5, and no additional evidence supporting the notion that aggressive managers manage earnings by overstating opaque financial instrument valuations is found. Final tests for differences in means of the dependent variables are included in Appendix 7, Table 8. These tests are based on classifying firms as “suspect” or “non-suspect” based on prior research. Panel A examines total realized and unrealized gains and losses, while Panel B examines only mark-to-market adjustments. The results remain mixed, and no clear evidence is found supporting earnings management through Level 3 instruments. Conclusions This study examines the relationship between established measures of aggressiveness in financial reporting and the opportunistic use of ASC 820 fair value estimates. Based on prior literature, a positive relationship between aggressiveness and realized and unrealized gains/losses on Level 3 instruments is expected. Evidence for such an association is not found. Regardless of whether macroeconomic variables or descriptions of Level 3 holdings taken directly from EDGAR fillings are used, the models are unable to explain a significant portion of the variation in Level 3 valuation changes or gains/losses from sales of securities. The lack of individual or joint significance of most control variables illustrates the substantial level of discretion management has over Level 3 items. ASC 820 requires roll-forwards and specific disclosures about the Level 3 instruments, but firms typically only disclose a few highlevel sentences about their Level 3 holdings. It appears that these disclosures are not sufficient for parties outside the firm to derive their own estimates of fair value. External parties must therefore blindly rely on the firms’ own valuations. For example, firms do not disclose if they have net “long” or “short” positions relating to congestion on power grids or what sectors the hedge funds they have invested in operate within. If the sample includes some firms whose Level 3 items increase in value when the VIX increases and other firms whose Level 3 items decrease in value when the VIX increases, finding a relationship between aggressiveness and Level 3 valuation would be burdensome. In summary, there does not appear to be evidence that management opportunistically uses discretion over Level 3 items to manage earnings, but it may be that the original SFAS 157/ASC 820 disclosure environment often precludes effective research designs of this empirical question. This is not the first study to conclude that the originally promulgated fair value accounting standards require subsequent modifications to convey relevant and reliable information to financial statement users. Cochran, Coffman, and Harless (2004) report that the fair value of mortgage loan servicing rights is related to non-servicing character-

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

8

DA_KPW = Lagged absolute value of discretionary accruals as in Kim et al. (2012). RAM_CZ = Composite RAM measure calculated as abnormal production minus abnormal cash flow from operations minus abnormal discretionary expenses and estimated as in Cohen and Zarowin (2010). RAM_R = Composite RAM measure calculated as abnormal production minus abnormal cash flow from operations minus abnormal discretionary expenses and estimated as in Roychowdhury (2006). MBE = Indicator variable equal to one if the firm’s “Street” earnings per share is greater than or equal to the most recent analysts’ consensus estimate, else 0. FVE = Total realized and unrealized gains/losses included in earnings related to changes in Level 3 valuations per ASC 820, scaled by total assets at the beginning of the quarter. LEV = debt-to-assets ratio defined as LTQ/ATQ in period t-1 LTQ = Total liabilities- Quarterly. ATQ = Total assets- Quarterly. LMVE = Natural log of the firm’s market capitalization (e.g., Ln(MKVALTQ)) MKVALTQ = Sum of all issue-level market values, including trading and non-trading issues. MTM = Unrealized gains/losses included in earnings related to Level 3 instruments per ASC 820, scaled by total assets at the beginning of the quarter. REAL_UNREAL = Total realized and unrealized gains/ losses included in earnings related to Level 3 instruments per ASC 820, unscaled. ROA = Return-on-assets defined as NIQ/Average ATQ NIQ = Net Income- Quarterly. SP = Quarterly standard deviation of the S&P 500 Index. TREAS = Three-month U.S. Treasury yield. TYPE = Indicator variable for each Level 3 instrument type, shown as D1-D5 in empirical results. UNREAL = Unrealized gains/losses included in earnings related to Level 3 instruments per ASC 820, unscaled. VIX = Closing price of the VIX on the date of the 10-K/ 10-Q issuance.

istics of firms. Similarly, Dorminey and Apostolou (2012) provide evidence that the information conveyed to market participants about derivatives and hedging activities under current accounting standards causes both confusion and the inability of investors to appropriately gauge risk. Calls for increased disclosure have also come from regulators. For example, the SEC (2010, 155) concludes, “The results of these (studies) also provide support for the view expressed by investors that additional disclosures would be most useful in the case of fair values derived from illiquid markets or model estimates.” This paper provides evidence that the FASB’s recent amendments to ASC 820 which require more disclosure regarding Level 3 items were warranted. Specifically, ASU 2011-04 (FASB, 2011b) is effective for fiscal periods beginning after December 15, 2011 and requires firms to disclose more information about the valuation process and the Level 3 items that are held. For example, firms must now disclose quantitative information about significant unobservable inputs, the sensitivity of fair value measurements to changes in unobservable inputs, and interrelationships between unobservable inputs if dependence between inputs potentially changes fair value estimates. The expanded disclosure environment may allow researchers to reexamine the relationship between aggressive financial reporting and Level 3 instruments. This is worthy of consideration in future research.

Aacknowledgments We thank Michael McKenzie (Ernst and Young), Bryan Turner (KPMG), Myung Seok Park, Benson Wier, Alisa Brink, Frederic Sterbenz, Ben Gilbert, Nicole Choi, Suman Banerjee, Sridhar Gogineni, Valentina Zamora, Eric Johnson, Joy Embree, and seminar participants at the University of New Mexico, Seattle University, and Virginia Commonwealth University for valuable comments.

Appendix A-1:

Empirical model The generic representation of the primary analysis model is as follows.

FVEit = βˆ 0 + βˆ1 AGGRit + βˆ 2TREASt + βˆ 3VIX t + βˆ 4 SPt (1) 12 + βˆ 5LMVEit + βˆ 6 LEVit + βˆ 7 ROAit + ∑ i=8 βˆ i TYPEit + εit , AGGR = Aggressiveness as measured by: DA_KLW, DA_MJCONS, DA_DSS, DA_KPW, RAM_CZ, RAM_R, or MBE. DA_KLW = Lagged absolute value of discretionary accruals as in Kothari et al. (2005). DA_MJCONS = Lagged absolute value of discretionary accruals as in Dechow et al. (1995) but augmented with an unscaled intercept. DA_DSS = Lagged absolute value of discretionary accruals as in Dechow et al. (1995).

The main interest is in the association between the aggressiveness measures (AGGR) and changes in Level 3 valuations recognized in earnings (FVE). The model includes the placeholder AGGR to one of represent seven financial aggressiveness measures.14 The primary analysis uses seven measures of aggressiveness: four discretionary accruals measures, two real earnings management measures, and meeting-or-beating.

14

DA_MJCONS, DA_DSS, DA_KPW, RAM_CZ, RAM_R, or MBE.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

Macro-economic variables The empirical models used in the multivariate tests include macroeconomic variables to control for valuation changes in the investments, derivatives, and financial instruments that make up the Level 3 instruments in the sample. The objective in this study is to identify if FVE is augmented through discretionary valuations. Normal fluctuations in the valuations of the instruments under consideration will result in FVE with or without aggressive reporting. Therefore, the association between FVE and AGGR cannot be isolated unless the natural valuation characteristics of the instruments giving rise to the income is considered. Aggressive valuations, and the attendant income, can only be detected if the normal valuations are identified. This is problematic as firms do not disclose sufficient information to derive estimated parameters for the Black– Scholes model on a case by case basis. However, the factors that are directly related to the valuation of these instruments are commonly understood and can be adequately measured. While the disclosures surrounding Level 3 valuations are insufficient to permit computation of the specific incomes generated, the underlying economic forces that give rise to changes in valuations can be reasonably measured and included in the model. 15 Accordingly, macro-economic variables generally applicable to the valuation models used for Level 3 items are included. Specifically, the three-month treasury yield (TREAS) is included as a measure of the risk-free rate, the Chicago Board Options Exchange (CBOE) Volatility Index (VIX) price as a proxy for options volatility, and the standard deviation of the S&P 500 (SP) as a measure of volatility in equities. Both measures of volatility are necessary and included as each pertains and relates to different classes of financial instruments included in the sample. These controls are similar to those used by Hutchinson, Lo, and Poggio (1994).

9

variables based on instrument type (e.g., auction rate securities, nonspecific derivatives, collars, power, and physical commodity derivatives, oil and natural gas derivatives, and investments in private equity, venture capital, or corporate debt) are included due to the diverse nature of the Level 3 holdings. While there is no strong a priori expectation that different instruments will interact in the model in a unique way, the inclusion of these indicators serves to provide an unrestricted model, thus mitigating that possibility as influential to the results. Fiechter and Meyer (2009) demonstrate that financial services firms use fair value estimates to smooth earnings. Accordingly, return on assets (ROA) is included as a measure of profitability. Because each firm in the sample is likely to implement alternative investments in different ways, standard errors are clustered at the firm level. This specification permits the use of those macroeconomic variables that influence all financial valuations of interest while simultaneously controlling for the unique implementation of those instruments for each firm in the sample. Appendix A-2: Descriptive statistics and tests of differences in means

Control variables Prior studies (e.g., Valencia, 2011; Fiechter & Meyer, 2009) also include measures of firm size and leverage when estimating unrealized gains/losses for Level 3 assets. The log of the firm’s market capitalization (LMVE) and the debt-toassets ratio (LEV) at the beginning of the quarter are included to control for size and leverage, respectively. Indicator

Panel A of Table 3 contains descriptive statistics. The means of both total realized and unrealized gains/losses on Level 3 (FVE) and unrealized gains/losses on Level 3 instruments (MTM) are positive.16 Each of the discretionary accruals measures have means that are generally consistent with previous studies (e.g., Cohen, Dey, & Lys, 2008). The means of the real activities manipulation (RAM) variables are negative. As these measures have been constructed to be positive as RAM increases, this suggests that, on average, firms in the sample are not heavy users of operational forms of earnings management. Meetingor-beating frequencies reported here are similar to those reported by Matsumoto (2002). Panel B of Table 3 separates firms based on whether FVE is positive, negative, or break-even. Results of tests of differences in means of the aggressiveness measures across the FVE groups are mixed. The strongest evidence of differences is between the firms that reported positive FVE and firms that reported negative FVE.

15 All financial valuations (including those of Level 3) are ultimately centered on the risk-to-return tradeoff (e.g., CAPM, Black-Scholes, KVM). Additional unique factors also affect valuations for different instruments, but all share the inclusion of the risk-to-return tradeoff. Therefore, those common factors are applied here to proxy for ‘normal’ valuation levels.

16 See Appendix B for a discussion on the simultaneity between FVE, MTM, and Discretionary Accruals Measures.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS 10

R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

Table 3 Descriptive statistics and tests of differences in financial reporting aggressiveness measures across FVE conditions. Panel A: Summary statistics Variable

Mean

Sd

P25

P75

REAL_UNREAL FVE UNREAL MTM DA_KLW DA_MJCONS DA_DSS DA_KPW RAM_R RAM_CZ MBE TREAS VIX SP LMVE LEV ROA

5.35 0.0001 5.39 0.0002 0.204 0.231 0.115 0.094 −0.021 −0.076 0.729 0.102 23.85 37.31 9.15 0.55 0.02

49.68 0.0022 34.33 0.0016 0.527 0.541 0.264 0.218 0.863 0.305 0.445 0.06 8.74 14.66 1.39 0.18 0.02

−1.90 −0.0001 −0.123 −0.0000 0.011 0.014 0.01 0.009 −0.123 −0.163 0 0.04 17.59 27.91 8.06 0.43 0.01

2.10 0.0002 1.717 0.0001 0.135 0.183 0.088 0.079 0.106 0.013 1 0.16 25.92 42.60 9.94 0.69 0.03

Panel B: t-tests of differences in mean aggressiveness conditional on total realized and unrealized gains/losses.

Variable Mean When: FVE >0 FVE=0 FVE <0 t-Tests: Difference in Means FVE >0 vs. FVE=0 Difference in Means FVE >0 vs. FVE<0 Difference in Means FVE=0 vs. FVE <0

DA_KLW

DA_MJCONS

DA_DSS

DA_KPW

RAM_R

RAM_CZ

MBE

0.25 .20 0.17 p-value 0.22 0.09 0.27

0.29 .20 0.21 p-value 0.11 0.13 0.55

0.12 0.14 0.08 p-value 0.65 0.08 0.04

.10 0.12 0.05 p-value 0.75 0.01 0.003

0.07 −0.03 −0.16 p-value 0.24 0.04 0.15

−0.08 −0.14 −.10 p-value 0.15 0.38 0.76

0.73 0.82 0.67 p-value 0.91 0.12 0.02

The t-tests allow for unequal variances between the groups. REAL_UNREAL is unrealized gains/losses in period t, unscaled. FVE is realized and unrealized gains/losses in period t scaled by total assets in period t-1. UNREAL is unrealized gains/losses in period t scaled by total assets in period t-1, unscaled. MTM is unrealized gains/losses in period t scaled by total assets in period t-1. DA_KLW is the absolute value of discretionary accruals in period t-1 from Kothari et al. (2005). DA_MJCONS is the absolute value of discretionary accruals in period t-1 using the Modified Jones Model with the inclusion of an unscaled intercept. DA_DSS is the absolute value of discretionary accruals in period t-1 from Dechow et al. (1995). DA_KPW is the absolute value of discretionary accruals in period t-1 from Kim et al. (2012). RAM_R is real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated with the inclusion of intercept terms as in Roychowdhury (2006). RAM _CZ is composite real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated as in Cohen and Zarowin (2010). All discretionary accruals and real earnings management variables are Winsorized at the 1% level. MBE is an indicator variable equal to one if the firm’s “Street” earnings are greater than or equal to the most recent analysts’ consensus estimates. TREAS is the three-month yield on US treasuries. VIX is the closing price of the CBOE Volatility Index. SP is the quarterly standard deviation of the S&P 500 Index. LMVE is the natural log of the market value of equity. LEV is the debt to assets ratio in period t-1. ROA is the return on assets ratio in period t.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

11

Appendix A-3: Univariate correlations Table 4 Univariate correlations. FVE FVE MTM DA_KLW DA_MJCONS DA_DSS DA_KPW RAM_R RAM_CZ MBE TREAS VIX SP LMVE LEV ROA

RAM_CZ MBE TREAS VIX SP LMVE LEV ROA

MTM

DA_KLW

DA_MJCONS

DA_DSS

DA_KPW

RAM_R

1.000 0.6763 (0.0000) 0.0568 (0.3081) 0.0315 (0.5724) 0.0815 (0.1431) 0.0958 (0.0851) 0.0115 (0.8706) 0.0406 (0.5646) 0.0532 (0.3057) 0.1145 (0.0270) 0.0718 (0.1663) 0.0613 (0.2377) 0.0207 (0.6990) 0.0052 (0.9205) 0.0736 (0.1558)

(0.0028) (0.9648) (0.0041) (0.9495) (0.0733) (0.2519) 0.0305 (0.6342) 0.0192 (0.8154) 0.0156 (0.8490) (0.0338) (0.5763) 0.1313 (0.0295) 0.0693 (0.2519) 0.0606 (0.3165) (0.0046) (0.9420 0.0586 (0.3326) 0.0187 (0.7582)

0.8807 (0.0000) 0.3400 (0.0000) 0.5380 (0.0000) (0.0398) (0.5501) (0.0460) (0.4893) 0.0088 (0.8619) 0.0918 (0.0688) 0.0930 (0.0651) 0.0852 (0.0913) 0.1063 (0.0401) (0.0246) (0.6268) 0.0349 (0.4899)

0.3812 (0.0000) 0.4284 (0.0000) (0.0159) (0.8116) (0.0666) (0.3170) 0.0204 (0.6870) 0.0758 (0.1330) 0.0895 (0.0760) 0.0630 (0.2124) 0.0966 (0.0625) (0.0221) (0.6616) 0.0309 (0.5412)

0.6032 (0.0000) (0.0195) (0.7699) 0.0873 (0.1893) 0.0876 (0.0825) 0.0914 (0.0700) (0.0524) (0.2996) (0.0208) (0.6813) 0.0481 (0.3539) (0.0670) (0.1844) 0.0002 (0.9970)

(0.0898) (0.1766) 0.0567 (0.3943) 0.0347 (0.4918) 0.0297 (0.5569) 0.0338 (0.5031) 0.0552 (0.2743) 0.0637 (0.2199) (0.0811) (0.1080) 0.0115 (0.8196)

0.4011 (0.0000) (0.0189) (0.7481) (0.0264) (0.6542) 0.0928 (0.1147) 0.1036 (0.0781) 0.0672 (0.2562) (0.0206) (0.7268) 0.0310 (0.5992)

RAM_CZ

MBE

TREAS

VIX

SP

LMVE

LEV

1.0000 0.0187 (0.7517) (0.0113) (0.8483) 0.0528 (0.3702) 0.0544 (0.3559) (0.1319) (0.0254) 0.1670 (0.0043) (0.1724) (0.0032)

1.0000 1.0000 1.0000 1.0000 1.0000 1.0000

1.0000 0.1506 (0.0009) 0.0224 (0.6227) (0.0343) (0.4498) 0.2640 (0.0000) (0.0911) (0.0455) 0.2015 (0.0000)

1.0000 0.0874 (0.0540) 0.0195 (0.6678) 0.2012 (0.0000) 0.0070 (0.8777) 0.0349 (0.4457)

1.0000 0.8538 (0.0000) (0.0440) 0.3488 (0.0294) 0.5185 0.0523 0.2523

1.0000 (0.0973) (0.0379) (0.0169) (0.7109) (0.0007) (0.9874)

1.0000 0.0864 (0.0652) 0.2864 (0.0000)

1.0000 (0.3198) (0.0000)

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

12

Appendix A-4: Univariate regressions Table 5 Univariate regressions of total realized and unrealized gains/losses or unrealized gains/losses on each aggressiveness measure. Panel A: Total realized and unrealized gains and losses FVE Intercept DA_KLW

0.0001 (0.816) 0.0002** (1.701)

DA_MJCONS

FVE 0.0001 (0.889)

FVE 0.0001 (0.668)

FVE

FVE

0.0001 (0.525)

0.0003 (1.643)

FVE 0.0002 (1.670)

−0.0001 (−0.181)

0.0001 (0.960)

DA_DSS

0.0006 (0.806)

DA_KPW

0.0011* (1.494)

RAM_CZ

0.0003 (1.079)

RAM_R

0.0000 (0.508)

MBE Obs. R2

FVE

324 0.003

324 0.001

324 0.007

324 0.009

204 0.002

204 0.000

0.0003 (0.719) 373 0.003

MTM

MTM

MTM

MTM

MTM

MTM

Panel B: Unrealized gains and losses MTM Intercept DA_KLW

0.0002 (1.871) −0.0000 (−0.054)

0.0002 (1.842)

0.0002 (2.103)

0.0002 (1.561)

0.0002 (1.582)

0.0002 (1.495)

−0.0000 (−0.084)

DA_MJCONS

−0.0004* (−1.538)

DA_DSS DA_KPW

0.0002 (0.449)

RAM_CZ

0.0001 (0.620)

RAM_R

0.0000 (0.676)

MBE Obs. R2

0.0003 (1.353)

246 0.000

246 0.000

246 0.005

246 0.001

151 0.000

151 0.000

−0.0001 (−0.507) 275 0.001

** and * represent statistical significance at the one-tailed 5% and 10% level, respectively. t-Statistics are reported in parentheses. Standard errors are clustered by firm. FVE is realized and unrealized gains/losses in period t scaled by total assets in period t-1. MTM is unrealized gains/losses in period t scaled by total assets in period t-1. DA_KLW is the absolute value of discretionary accruals in period t-1 from Kothari et al. (2005). DA_MJCONS is the absolute value of discretionary accruals in period t-1 using the Modified Jones Model with the inclusion of an unscaled intercept. DA_DSS is the absolute value of discretionary accruals in period t-1 from Dechow et al. (1995). DA_KPW is the absolute value of discretionary accruals in period t-1 from Kim et al. (2012). RAM_R is real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated with the inclusion of intercept terms as in Roychowdhury (2006). RAM _CZ is composite real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated as in Cohen and Zarowin (2010). All discretionary accruals and real earnings management variables are Winsorized at the 1% level. MBE is an indicator variable equal to one if the firm’s “Street” earnings are greater than or equal to the most recent analysts’ consensus estimates.

First, the relationship between aggressiveness in financial reporting and subjective fair value accounting estimates is explored using univariate regressions. Only two of the seven aggressiveness measures are significantly associated with FVE. One of the seven aggressiveness measures is associated with MTM, and counter to expectations, the parameter estimate is negative.

Appendix A-5: Multivariate results Equation (1) is estimated using panel data with standard errors clustered at the firm level. Standard errors are not clustered on both firm and time because the number

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

of time periods in the data precludes consistent estimation of time-based serial correlation (e.g., Peterson, 2009; Thompson, 2011). However, inferences remain the same when two-way clustering is used. Estimation results of equation (1) are shown in Table 6. The null hypothesis that all of the parameter estimates are jointly equal to zero is not rejected, and very few of the control variables are individually significant.17 A variety of additional tests were conducted including alternative estimation samples and model specifications. Specifically, the following additional analyses were conducted: (i) estimated a loglinear model to allow for a nonlinear relationship between aggressiveness and FVE/MTM, (ii) estimated the model in levels but included a quadratic term for aggressiveness to allow for a nonlinear relationship between aggressiveness and FVE/MTM, (iii) eliminated firms that reported no valuation changes in FVE between periods, (iv) split the estimation sample up into one sample for firms that reported positive FVE and a second sample of firms that reported negative FVE, (v) scaled FVE/MTM by Level 3 assets instead of total assets, (vi) estimated the model without the indicator variables for Level 3 instrument type, (vii) included quarterly indicator variables instead of the macroeconomic variables, and (viii) estimated the model using differenced macroeconomic variables. Regardless of the above, inferences remain quantitatively similar to the results presented in Table 6. Perhaps most surprisingly, regressing FVE and MTM (untabulated) only on the indicator variables for Level 3 instrument type also results in an inability to reject the null hypothesis that all of the indicator variables are jointly equal to zero. This suggests that valuation changes and proceeds from sales for auction rate securities tied to collateralized student loan debt are statistically indistinguishable from energy derivatives, commodity derivatives, nonspecific derivatives, investments in hedge funds, investments in private equity firms, corporate debt, and mortgage backed securities. It is unreasonable to believe that groups of instruments this diverse would all trade for, or be sold for, similar amounts. Rather than taking these results at face value, it may be that the disclosure requirements of ASC 820 simply do not allow parties external to the firm to come up with their own valuations for Level 3 holdings. That is, joint consideration of macroeconomic variables and firm-level disclosures results in imprecise estimates of fair values; sufficiently imprecise so as to preclude support for the hypothesized association. The relatively small adjusted R-squared measures reported in Table 6 warrant further investigation and discussion. Research concerning the post-ASC 820 fair value accounting rules is relevantly recent as the new requirements have been in effect for less than a decade. Value-relevance studies typically contain price or returns as the dependent variable with the fair value hierarchy measurements as independent variables (e.g., Goh et al., 2009; Kolev, 2009; Song et al., 2010). A second stream of fair value accounting studies examines Level 3 classifications and transfers between categories descriptively or with limited dependent variable models (e.g., Altamuro

17 The variance inflation factors (VIF) do not suggest that the model estimates suffer from multicollinearity.

13

& Zhang, 2013; Webinger, Comer, & Bloom, 2013). Finally, researchers have explored the relationships between fair value and audit fees (Ettredge, Xu, & Yi, 2010), fair value and information asymmetry (Liao, Kang, Morris, & Tang, 2013), and adjusted stock market betas and fair value (Riedl & Serafeim, 2011). The goodness of fit statistics from these studies is not directly comparable to the current study since the dependent variables are not FVE or MTM. Prior research using the same, or similar, dependent variables herein are Fiechter and Meyer (2009), Valencia (2011), and He, Wong, and Young (2012). Valencia (2011) does not report R-squared or adjusted R-squared, and Fiechter and Meyer (2009) do not report adjusted R-squared. The adjusted R-squareds from He et al. (2012) range from 0.040 to 0.072, and unadjusted R-squareds from Fiechter and Meyer (2009) are between 0.112 and 0.153. Unadjusted R-squared measures (untabulated) for the models in Table 6 are bounded by 0.047 and 0.050. These are smaller than those of Fiechter and Meyer (2009), but this may be attributed to the fact that Fiechter and Meyer (2009) use a sample consisting of only financial services firms, their empirical model contains many bank-specific control variables (e.g., loan loss provisions and nonperforming assets), and they have three times more observations than the current study. The dependent variable from He et al. (2012) is realized gains and losses on sales of available for sale securities. While this is a component of FVE, this measure differs from the current study. Additionally, the sample from He et al. (2012) consists solely of Chinese firms and empirical models include two independent variables based on the Chinese province the firm operates within and two independent corporate governance variables. These are major differences which may result in the discrepancy of goodness of fit measures between the studies. Nonetheless, the adjusted R-squared measures reported in Table 6 indicate that the empirical models do not explain a meaningful portion of the variation in FVE. One example of a Level 3 disclosure is the American Electric Power case discussed in the previous section. The only information disclosed by American Electric Power is that the instrument’s fair value changes as a result of congestion on a power grid and that the instrument is long-dated, complex, and illiquid. The firm does not make further disclosures. Readers of the financials do not know if the instrument increases or decreases in value as a result of power grid congestion, the revenues or costs associated with various levels of congestion are not disclosed, and neither is the geographic area the instrument pertains to. Similarly, in 3M’s 2009 10-K, Level 3 instruments are simply described as, “ . . . auction rate securities that represented interests in investment grade credit default swaps”(3M Company, 2010, 78). 3M goes on to state that the securities failed to auction since 2007. No further information is provided about either specific qualitative aspects of the swaps (e.g., the identity of the counterparty or the industry the counterparty operates within) or the actual inputs used for valuation. The estimation models related to Table 6 attempt to deal with these vague disclosures by grouping the Level 3 instruments into categories based on hand-collected data obtained from reading the 10-Q or 10-K for each observation in the sample. For example, the 3M instrument is categorized via an

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

14

Table 6 Total realized and unrealized gains/losses and aggressiveness in financial reporting.

Intercept DA_KLW

FVE

FVE

FVE

FVE

−0.0003 (−0.221) 0.0002* (1.657)

−0.0003 (−0.238)

−0.0004 (−0.329)

−0.0004 (−0.341)

DA_MJCONS

FVE 0.0012 (1.105)

FVE 0.0012 (1.114)

0.0014** (1.999)

DA_DSS

0.0007 (0.886) −0.0000 (−0.029)

RAM_R RAM_CZ

0.0002 (0.571)

MBE

VIX SP LMVE LEV ROA D1 D2 D3 D4 D5 Obs. Adj. R2 F- Statistic

−0.0008 (−0.686)

0.0001 (1.002)

DA_KPW

TREAS

FVE

0.0025* (1.487) 0.0000 (0.484) −0.0000 (−0.137) −0.0000 (−0.137) −0.0000 (−0.052) 0.0018 (0.322) −0.0002 (−0.330) −0.0002 (−0.353) 0.0008 (0.945) 0.0001 (0.163) −0.0004 (−0.669) 305 0.000 1.196

0.0025* (1.525) 0.0000 (0.491) −0.0000 (−0.119) −0.0000 (−0.090) −0.0001 (−0.105) 0.0019 (0.337) −0.0002 (−0.359) −0.0002 (−0.341) 0.0008 (0.929) 0.0001 (0.152) −0.0003 (−0.663) 305 −0.002 1.162

0.0026* (1.590) 0.0000 (0.624) −0.0000 (−0.204) −0.0000 (−0.267) −0.0000 (−0.002) 0.0028 (0.526) −0.0001 (−0.249) −0.0001 (−0.146) 0.0009 (1.077) 0.0002 (0.367) −0.0003 (−0.636) 305 0.011 1.154

0.0022* (1.571) 0.0000 (0.645) −0.0000 (−0.181) −0.0000 (−0.127) −0.0001 (−0.132) 0.0020 (0.352) −0.0001 (−0.266) −0.0001 (−0.134) 0.0009 (1.057) 0.0001 (0.260) −0.0003 (−0.606) 305 0.005 0.805

0.0024 (1.139) −0.0000 (−0.991) 0.0000 (1.159) −0.0002* (−1.636) 0.0005 (0.983) 0.0039 (0.616) −0.0005 (−1.191) 0.0004 (0.469) −0.0005 (−0.863) −0.0004 (−0.760) −0.0005 (−1.066) 203 −0.025 0.881

0.0024 (1.135) −0.0000 (−1.015) 0.0000 (1.166) −0.0002* (−1.640) 0.0005 (0.911) 0.0042 (0.686) −0.0005 (−1.194) 0.0003 (0.450) −0.0005 (−0.867) −0.0004 (−0.767) −0.0006 (−1.064) 203 −0.025 0.933

0.0002 (0.373) 0.0039** (2.269) −0.0000 (−0.016) 0.0000 (0.589) −0.0000 (−0.050) 0.0002 (0.319) 0.0104 (1.284) −0.0003 (−0.701) −0.0005 (−0.777) 0.0006 (0.811) 0.0000 (0.048) −0.0004 (−0.912) 352 0.013 0.892

* and ** represents statistical significance at the one-tailed 10% and 5% level, respectively. Standard errors are clustered by firm and t-statistics are reported in parentheses. FVE is realized and unrealized gains/losses in period t scaled by total assets in period t-1. DA_KLW is the absolute value of discretionary accruals in period t-1 from Kothari et al. (2005). DA_MJCONS is the absolute value of discretionary accruals in period t-1 using the Modified Jones Model with the inclusion of an unscaled intercept. DA_DSS is the absolute value of discretionary accruals in period t-1 from Dechow et al. (1995). DA_KPW is the absolute value of discretionary accruals in period t-1 from Kim et al. (2012). RAM_R is real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated with the inclusion of intercept terms as in Roychowdhury (2006). RAM _CZ is composite real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated as in Cohen and Zarowin (2010). All discretionary accruals and real earnings management variables are Winsorized at the 1% level. MBE is an indicator variable equal to one if the firm’s “Street” earnings are greater than or equal to the most recent analysts’ consensus estimates. TREAS is the three-month yield on US treasuries. VIX is the closing price of the CBOE Volatility Index. SP is the quarterly standard deviation of the S&P 500 Index. LMVE is the natural log of the market value of equity. LEV is the debt to assets ratio in period t-1. ROA is the return on assets ratio in period t. D1-D5 are indicator variables for Level 3 instrument type.

indicator variable equal to one for auction rate securities, and this indicator variable is also equal to one for any other firm that describes the Level 3 instruments as auction rate securities. This is deemed to be an improvement over prior studies which have simply assigned indicator variables based on the two-digit SIC code of the firm, which has little to do with the actual underlying financial instrument. However, after the hand-collection of data, the assignment of indicator variables based upon the disclosed information about the Level 3 items in SEC filings, including control variables common across prior SFAS 157/ASC 820 fair value studies (e.g., measures of size, leverage, and profitability), and exploring a variety of model specifications discussed above does little to improve R-squared. This in-

dicates that these instruments are so complex, and the disclosures are so paltry, that investors are left without useful information from ASC 820 in its original form to develop objective, arm’s-length estimates of valuation changes. This conclusion is echoed by He et al. (2012) who write, “Specifically, we find that the new (fair value accounting) for trading securities fails to generate the intended benefits of improved transparency. . .18”

18 While He et al. (2012) study the adoption of IFRS-based standards in China, the IFRS fair value standard is essentially the same as SFAS 157/ASC 820.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

Appendix A-6: Unrealized gains and losses estimation Unlike total realized and unrealized gains/losses, unrealized gains/losses are based entirely on estimates (e.g., there is no counterparty agreeing on the price of a Level 3 item sale in period t). Therefore, these adjustments are impacted the most by managerial discretion. The first additional analysis is restricted to mark-to-market entries which are recognized in current earnings. These results are shown in Table 7. Similar to the results reported for

15

tests of the initial hypothesis, the null hypothesis that all of the parameter estimates are jointly equal to zero is unable to be rejected at conventional levels, and very few of the control variables are individually significant. All of the additional tests, model specifications, and samples discussed above were repeated for the unrealized gains/ losses sample. For the sake of brevity, these are not repeated here. However, inferences remain quantitatively similar to the results in Table 6 regardless of the additional procedures employed. There is no support found for the hypothesized association in an examination of only unrealized gains/losses.

Table 7 Unrealized gains/losses and aggressiveness in financial reporting.

Intercept DA_KLW

MTM

MTM

MTM

MTM

−0.0003 (−0.424) 0.0000 (0.024)

−0.0003 (−0.437)

−0.0003 (−0.567)

−0.0002 (−0.331)

DA_MJCONS

MTM 0.0007 (1.045)

MTM 0.0006 (0.930)

0.0006 (1.280) −0.0003 (−1.051)

DA_DSS

−0.0000 (−0.375)

RAM_R

−0.0002 (−0.461)

RAM_CZ MBE

VIX SP LMVE LEV ROA D1 D2 D3 D4 D5 Obs. Adj. R2 F- Statistic

−0.0006 (−0.843)

0.0000 (0.191)

DA_KPW

TREAS

MTM

0.0023* (1.561) −0.0000 (−0.033) 0.0000 (0.450) −0.0000 (−0.707) 0.0005 (1.049) 0.0001 (0.017) 0.0001 (0.584) 0.0002 (0.822) 0.0006 (1.263) 0.0000 (0.409) −0.0002** (−1.738) 230 −0.009 0.795

0.0024* (1.589) −0.0000 (−0.053) 0.0000 (0.462) −0.0000 (−0.712) 0.0005 (1.037) 0.0000 (0.013) 0.0001 (0.567) 0.0002 (0.816) 0.0006 (1.265) 0.0000 (0.445) −0.0002** (−1.771) 230 −0.009 0.806

0.0025* (1.665) −0.0000 (−0.070) 0.0000 (0.517) −0.0000 (−0.865) 0.0005 (1.118) 0.0005 (0.142) 0.0001 (1.020) 0.0003 (0.986) 0.0007* (1.378) 0.0001 (0.975) −0.0002* (−1.375) 230 −0.004 1.139

0.0025* (1.605) −0.0000 (−0.167) 0.0000 (0.476) −0.0000 (−0.606) 0.0005 (0.976) 0.0001 (0.020) 0.0000 (0.151) 0.0001 (0.541) 0.0006 (1.209) 0.0000 (0.056) −0.0003** (−1.807) 230 −0.006 0.824

0.0021 (1.102) −0.0000 (−0.706) 0.0000 (0.916) −0.0001 (−1.307) 0.0002 (0.713) 0.0050 (0.931) −0.0000 (−0.317) 0.0009 (1.033) 0.0007 (0.968) −0.0000 (−0.102) −0.0001 (−0.585) 150 −0.037 0.476

0.0022 (1.063) −0.0000 (−0.712) 0.0000 (0.912) −0.0001 (−1.295) 0.0003 (0.783) 0.0046 (0.970) −0.0000 (−0.250) 0.0009 (1.048) 0.0007 (0.943) 0.0000 (0.029) −0.0001 (−0.567) 150 −0.037 0.526

−0.0001 (−0.286) 0.0036** (2.204) −0.0000 (−0.389) 0.0000 (1.044) −0.0000 (−0.390) 0.0005 (1.271) 0.0046 (0.929) 0.0000 (0.276) 0.0002 (0.714) 0.0008** (1.599) 0.0001 (0.546) −0.0003*** (−2.393) 257 0.016 0.858

* , **, and *** represents statistical significance at the one-tailed 10%, 5%, and 1% level, respectively. t-Statistics are reported in parentheses. Standard errors are clustered by firm. MTM is unrealized gains/losses in period t scaled by total assets in period t-1. DA_KLW is the absolute value of discretionary accruals in period t-1 from Kothari et al. (2005). DA_MJCONS is the absolute value of discretionary accruals in period t-1 using the Modified Jones Model with the inclusion of an intercept. DA_DSS is the absolute value of discretionary accruals in period t-1 from Dechow et al. (1995). DA_KPW is the absolute value of discretionary accruals in period t-1 from Kim et al. (2012). RAM_R is real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated with the inclusion of intercept terms as in Roychowdhury (2006). RAM _CZ is composite real activities manipulation defined as Abnormal Production minus Abnormal Cash Flows from Operations minus Abnormal Discretionary Expenditures estimated as in Cohen and Zarowin (2010). All discretionary accruals and real earnings management variables are Winsorized at the 1% level. MBE is an indicator variable equal to one if the firm’s “Street” earnings are greater than or equal to the most recent analysts’ consensus estimates. D1- D5 are indicator variables for Level 3 instrument type.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

16

Appendix A-7: Suspect firm analyses Table 8 Suspect firm analyses of differences in means for suspect vs. non-suspect firms. Panel A: Analysis of FVE Suspect Firm Criteria Mean FVE for Suspect Firms Mean FVE for Non-Suspect Firms Difference

IBQt/ATQt ∈ [0, .005] −0.0004 0.0002 −0.0006

(IBQt - IBQt-1)/ATQt ∈ [0, .005] 0.0003 0.0000 0.0003*

(IBQt - IBQt-4)/ATQt ∈ [0, .005] −0.0000 0.0002 −0.0002

MBE ≤ $.01 0.0008 .0000 0.0008**

IBQt/ATQt ∈ [0, .005] −0.0001 −0.0002 0.0001

(IBQt - IBQt-1)/ATQt ∈ [0, .005] 0.0004 −0.0004 0.0008**

(IBQt - IBQt-4)/ATQt ∈ [0, .005] 0.0001 −0.0003 0.0004

MBE ≤ $.01 0.0002 −0.0002 0.0004*

Panel B: Analysis of MTM Suspect Firm Criteria Mean MTM for Suspect Firms Mean MTM for Non-Suspect Firms Difference

* and ** represent statistical significance at the .10 and .05 level, respectively. Reported p-values are for the alternate hypothesis that the mean of the suspect firms is greater than the mean of the non-suspect firms. The t-tests allow for unequal variances between the groups. FVE is realized and unrealized gains/losses in period t scaled by total assets in period t-1. MTM is unrealized gains/losses in period t scaled by total assets in period t-1. IBQ is income before extraordinary items, quarterly. ATQ is total assets, quarterly. MBE is an indicator variable equal to one if the firm’s “Street” earnings are greater than or equal to the most recent analysts’ consensus estimates.

The final series of tests classifies firms as “suspect” or “non-suspect” based upon financial statement data. Small changes in scaled earnings between financial reporting periods, no change in scaled earnings between financial reporting periods, and “just” meeting or beating analysts’ consensus earnings by one cent or less have all been used in prior research to classify firms into the “suspect” groups (e.g., Roychowdhury, 2006; Cohen et al., 2008). One “suspect” group based on differences in seasonal earnings is added. Many firms exhibit an element of seasonality in earnings. For example, retailers often have their busiest quarter of the year followed by one of their slowest quarters of the year. Defining suspect firms based only on changes in scaled earnings between adjacent quarters ignores this. The means of FVE are larger in two of the four suspect firm groups, and these differences are statistically significant. The means of MTM are larger in all four of the suspect firm groups, and two of the four are statistically significant. All of the statistically significant differences are from suspect firm classifications based upon differenced earnings and meeting-or-beating. The results of these tests provide some support that fair value discretion is used to increase earnings for aggressive firms, but the results are still quite mixed. Appendix B Simultaneity between FVE, MTM, and discretionary accruals measures In certain circumstances the dependent variables in this study are necessarily correlated with total accruals due to the equation used to derive total accruals. This is problematic econometrically because in these situations the dependent variable will be included as a part of discretionary accruals to the extent that the regressors in the discretionary accruals models are unable to explain varia-

tion in total accruals. The inclusion of FVE or MTM into the discretionary accruals measures induces simultaneity bias in OLS and leads to biased and inconsistent parameter estimates (Wooldridge, 2002). The following series of equations illustrate the simultaneity problem:

Total Accruals = Net Income Before Extraordinary Items − Operating Activities Net Cash Flow (B.1)

Operating Activities Net Cash Flow = Net Income + Total Adjustments,

(B.2)

where Total Adjustments are all the reconciling items contained in the operating section of the statement of cash flows. By substituting Eq. (B.2) into Eq. (B.1)

Total Accruals = Net Income Before Extraordinary Items − (Net Income + Total Adjustments) Assuming firms do not have extraordinary items, Net Income Before Extraordinary Items equals Net Income

Total Accruals = Net Income Before Extraordinary Items − Net Income Before Extraordinary Items − Total Adjustments Total Accruals = ( −1) × ( Total Adjustments )

(B.3)

Therefore, total accruals equal the negative of total reconciling items on the operating statement of cash flows. Any realized or unrealized gain/loss that is included in FVE or MTM and is also included as a reconciling item in the operating section of the statement of cash flows induces simultaneity bias in the OLS parameter estimates. Accordingly, the lag of each discretionary accruals measure is used to avoid simultaneity bias. Using lags to avoid simultaneity is both econometrically sound and has precedence in the accounting literature (e.g., Matsumoto, 2002).

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006

ARTICLE IN PRESS R. Glasscock et al. / Research in Accounting Regulation ■■ (2017) ■■–■■

References 3M Company. (2010). Form 10-K, December 31, 2009. Retrieved from: https://www.sec.gov. (Accessed 22 April). Aboody, D., Barth, M. E., & Kasznik, R. (2006). Do firms understate stock option-based compensation expense disclosed under FAS 123? Review of Accounting Studies, 11(4), 429–461. Acuitas, Inc. (2012). Survey of fair value audit deficiencies. Retrieved from: http://www.acuitasinc.com/documents/Survey_of_Big_4_Fair_Value _Audit_Deficiencies_as_Reported_in_the_Wall_Street_Journal.pdf. (Accessed 22 April). Altamuro, J., & Zhang, H. (2013). The financial reporting of fair value based on managerial inputs versus market prices: Evidence from mortgage servicing rights. Review of Accounting Studies, 18, 833–858. American Electric Power Company. (2012). Form 10-Q, June 30, 2012. Retrieved from: https://www.sec.gov. (Accessed 22 April). American Institute of Certified Public Accountants (AICPA) (2014). Evaluation of misstatements identified during the audit. New York: AICPA. AU-C Section 450. Barth, M., & Taylor, D. (2010). In defense of fair value: Weighting the evidence on earnings management and asset securitizations. Journal of Accounting and Economics, 49(1–2), 26–33. Bartov, E., Givoly, D., & Hayn, C. (2002). The rewards to meeting-or-beating earnings expectations. Journal of Accounting and Economics, 33(2), 173–204. Bartov, E., Mohanram, P., & Nissim, D. (2007). Managerial discretion and the economic determinations of the disclosed volatility parameter for valuing ESOs. Review of Accounting Studies, 12(1), 155–179. Bell, T. B., & Griffin, J. B. (2012). Commentary on auditing high-uncertainty fair value estimates. Auditing: A Journal of Practice and Theory, 3(1), 147–155. Benston, G. J. (2008). The shortcomings of fair-value accounting described in SFAS 157. Journal of Accounting and Public Policy, 27(2), 101–114. Bies, S. (2004). Remarks to the International Association of Credit Portfolio Managers General Meeting. New York, NY. November 18, 2004. Retrieved from: http://www.federalreserve.gov/newsevents/speech/2004speech .htm. (Accessed 22 April). Cochran, R., Coffman, E., & Harless, D. (2004). Fair value capitalization of mortgage loan servicing rights. Research in Accounting Regulation, 17, 153–165. Cohen, D., Dey, A., & Lys, T. (2008). Real and accruals-based earnings management in the pre- and post-Sarbanes-Oxley periods. The Accounting Review, 83(3), 757–787. Cohen, D., & Zarowin, P. (2010). Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics, 50(1), 2–19. Dechow, P., Sloan, R., & Sweeney, A. (1995). Detecting earnings management. The Accounting Review, 70(2), 193–225. DeFond, M. (2010). Earnings quality research: Advances, challenges and future research. Journal of Accounting and Economics, 50(2–3), 402–409. Dorminey, J., & Apostolou, B. (2012). Hedging derivatives in the banking industry. Research in Accounting Regulation, 24(2), 65–73. Du, H., Li, S., & Xu, R. (2014). Adjustment of valuation inputs and its effect on value relevance of fair value measurement. Research in Accounting Regulation, 26(1), 54–66. Ettredge, M., Xu, Y., & Yi, H. (2010). Fair value measurements, auditor industry expertise, and audit fees: Evidence from the banking industry. Auditing: A Journal of Practice and Theory, 33(3), 33–58. Fiechter, P., & Meyer, C. (2009). Discretionary measurement of financial assets during the financial crisis. Working paper, University of Zurich. Financial Accounting Standards Board (FASB) (2006). Fair value measurements. Statement of Financial Accounting Standards No. 157. Stanford, CT. Financial Accounting Standards Board (FASB) (2010a). Conceptual Framework for Financial Reporting – Chapter 1, The Objective of General Purpose Financial Reporting, and Chapter 3, Qualitative Characteristics of Useful Financial Information (a replacement of FASB Concepts Statements No. 1 and No. 2) Concepts Statement No. 8. Stanford, CT. Financial Accounting Standards Board (FASB) (2010b). Fair value measurement. ASC 820. Stanford, CT.

17

Financial Accounting Standards Board (FASB) (2011a). Update No. 201104—Fair value measurement (topic 820): Amendments to achieve common fair value measurement and disclosure requirements in U.S. GAAP and IFRSs. ASC 820. Stanford, CT. Financial Accounting Standards Board (FASB) (2011b). IASB and FASB issue common fair value measurement and disclosure requirements. News Release. Stanford, CT. Goh, B., Ng, J., & Yong, K. (2009). Market pricing of banks’ fair value assets reported under SFAS 157 during the 2008 economic crisis. Working paper, Singapore Management University. Harris, T., & Morsfield, S. (2012). An evaluation of the current state and future of XBRL and interactive data for investors and analysts. Working paper, Columbia University. He, X., Wong, T. J., & Young, D. (2012). Challenges for implementation of fair value accounting in emerging markets: Evidence from China. Contemporary Accounting Research, 29(2), 538–562. Hutchinson, J., Lo, A., & Poggio, T. (1994). A nonparametric approach to pricing and hedging derivative securities via learning networks. The Journal of Finance, 49(3), 851–889. Kim, Y., Park, M. S., & Wier, B. (2012). Is earnings quality associated with corporate social responsibility? The Accounting Review, 87(3), 761– 796. Kolev, K. (2009). Do investors perceive mark-to-model as marking-to-myth? Early evidence from FAS No. 157 disclosure. Working paper, New York University. Kothari, S. P., Leone, A., & Wasley, C. (2005). Performance matched discretionary accrual measures. Journal of Accounting and Economics, 39(1), 163–197. Laux, C., & Leuz, C. (2009). The crisis of fair-value accounting: Making sense of the recent debate. Accounting Organizations and Society, 34(6–7), 826–834. Liao, L., Kang, H., Morris, R., & Tang, Q. (2013). Information asymmetry of fair value accounting during the financial crisis. Journal of Contemporary Accounting & Economics, 9, 221–226. Lu, H.-Y., & Mande, V. (2014). Does disaggregation of fair value information increase the value relevance of the fair value hierarchy? Research in Accounting Regulation, 26(1), 90–97. Matsumoto, D. (2002). Management’s incentives to avoid negative earnings surprises. The Accounting Review, 77(3), 483–514. Ohlson, J. (1995). Earnings, book values, and dividends in equity valuation. Contemporary Accounting Research, 11(2), 661–687. Peterson, M. A. (2009). Estimating standard errors in finance panel datasets: Comparing approaches. The Review of Financial Studies, 22(1), 435–480. Riedl, E., & Serafeim, G. (2011). Information risk and fair values: An examination of equity betas. Journal of Accounting Research, 49(4), 1083–1122. Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42(3), 335–370. Securities and Exchange Commission (SEC) (2009). Interactive data to improve financial reporting. Washington, DC. Securities and Exchange Commission (SEC) (2010). Report and recommendations pursuant to Section 133 of the Emergency Economic Stabilization Act of 2008: Study on mark to market accounting. Retrieved from: http://www.sec.gov/news/studies/2008/marktomarket123008 .pdf. (Accessed 22 April). Skinner, D., & 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(2–3), 289–312. Song, C., Thomas, W., & Yi, H. (2010). Value relevance of FAS No. 157 fair value hierarchy information and the impact on corporate governance mechanisms. The Accounting Review, 85(4), 1375–1410. Thompson, S. B. (2011). Simple formulas for standard errors that cluster by both firm and time. Journal of Financial Economics, 99(1), 1–10. Valencia, A. (2011). Opportunistic behavior using Level 3 fair-values under SFAS 157. Working paper, Florida State University. Webinger, M., Comer, M., & Bloom, R. (2013). The effect of additional guidance on fair value measurement and disclosure in illiquid or inactive markets. Research in Accounting Regulation, 25, 220–229. Wooldridge, J. (2002). Econometric analysis of cross section and panel data. Boston, MA: The MIT Press.

Please cite this article in press as: Robson Glasscock, David Harless, Jack Dorminey, The curious case of Level 3 instruments, Research in Accounting Regulation (2017), doi: 10.1016/j.racreg.2017.04.006