Where do firms manage earnings?

Indeed, research on the related phenomenon of accounting method choice dates back to studies such as Watts and Zimmerman (1978), Hagerman and Zmijewski (1979), Bowen et al. (1981), and Holthausen (1981).

For reviews see Healy and Wahlen (1999), Schipper (1989), Dechow and Skinner (2000), Fields et al. (2001), and Dechow et al. (2010).

The subsidiary may also be required to prepare financial statements in accordance with local GAAP for a variety of reasons. Our point here is that when the earnings are consolidated and reported for the entire company, all the earnings are reported using U.S. GAAP and are subject to U.S. securities laws.

December 7, 2009, speech before the American Institute of Certified Public Accountants (AICPA) National Conference on Current SEC and PCAOB Developments, Washington, DC (Goelzer 2009).

There are other important papers that are somewhat related to our research but less directly. For example, Duru and Reeb (2002) report evidence that analysts have lower forecast accuracy when firms have greater international diversification. Looking across firms—thus, more similar in spirit to Leuz et al. (2003), Pincus et al. (2007) document the occurrence of the accrual anomaly in foreign countries, and DeFond et al. (2007) provide evidence that earnings announcements are more informative in countries with strong investor protections. In addition, there are other papers beyond Erickson et al. (2004) that study the relation between taxes and earnings management. For example, Badertscher et al. (2009) (discussed below), Frank et al. (2009), and Rego (2003). Frank et al. (2009) find a positive relation between aggressive tax reporting and aggressive financial reporting, and Rego (2003) finds economies of scale in tax planning, such that multinational firms are better able to avoid tax. We do not study tax planning in our paper.

In an analysis of fraudulent financial reporting over the period 1998–2007, the Committee of Sponsoring Organizations of the Treadway Commission (COSO) states: “The SEC named the CEO and/or CFO for some level of involvement in 89 % of the fraud cases, up from 83 % of cases in 1987–1997. Within 2 years of the completion of the SEC’s investigation, about 20 % of CEOs/CFOs had been indicted and over 60 % of those indicted were convicted” (Beasley et al. 2010). Thus, many cases of financial manipulation name central managers as the guilty party. However, some cases are done by “rogue” managers, as the two SEC AAER cases in Sect. 2 indicate. Top management at Boston Scientific and Bristow did not appear to be, and were not accused of being, involved in the fraud.

Many tax havens have a corporate tax rate of zero, but conceptually for this paper we only need havens to impose lower taxes on income than does the U.S.

Our premise assumes that the earnings from the subsidiary are not repatriated and the U.S. tax is not incurred. The Badertscher et al. (2009) study assumes all the income is taxable in the U.S. and that managers lessen the tax via nonconforming earnings management.

Similarly, companies can tax plan for a variety of other reasons, for example, to maximize foreign tax credits. While firms in our sample may be doing such tax planning, to our knowledge such tax planning will not affect our tests. We have constructed the tests using pre-tax measures to avoid incorrect inferences due to tax planning.

Studies examining the absolute value of discretionary accruals include Dechow and Dichev (2002), Frankel et al. (2002), Klein (2002), Chung and Kallapur (2003), Myers et al. (2003), Leuz et al. (2003), and Bergstresser and Philippon (2006). As discussed later, we control for operating volatility in our analyses, following the recommendations of Hribar and Nichols (2007).

See http://​info.​worldbank.​org/​governance/​wgi/​pdf/​rl.​pdf for a detailed discussion and listing of factors.

We use the lists in Miedema (2008) as per Dyreng and Lindsey (2009).

See http://​www.​oecd.​org/​document/​23/​0,3343,en_​2649_​33745_​30575447_​1_​1_​1_​1,00.​html for further details.

This definition captures noncorporate entities, which may not all meet the technical definition of an SPE. We follow prior convention for labeling purposes but recognize that there is measurement error in this variable.

The financials and utilities industries are dropped because we eliminate regulated industries from the sample. Of the 14 firms in Compustat in the tobacco industry, none fulfill all of our sample criteria. Thus, we include a breakdown of 27 (not 30) industries in Table 2.

The mean is not zero because the model is estimated over the larger 69,819 firm-year sample, which is prior to imposing other data requirements to arrive at the final sample.

We use robust regression to control for outliers. In the regressions, all continuous variables are mean centered at zero for ease of interpretation of the interaction effects (Aiken and West 1991). We multiply the dependent variable by 100 to facilitate interpretation of the coefficients as percentages. The standard errors in all regressions are computed after clustering observations by firm and year to mitigate the effects of cross-sectional and intra-firm correlation in the residuals (Petersen 2009). For all regressions we present one-tailed p values for t statistics where we have a prediction and two-tailed p values otherwise.

The breakdown of pre-tax income into pre-tax domestic income and pre-tax foreign income is required by the SEC to be included in the tax footnote of firm’s financial statements to correspond with the breakdown of tax expense into domestic and foreign components.

We use the superscript j to indicate a vector of controls that includes all of the controls in the vector k and also includes RULE OF LAW and HAVEN INTENSITY.

Calculated as −18.633 × 0.487 = −9.1. Note that 0.487 is the standard deviation of RULE OF LAW from Table 3.

Many studies use the absolute value of discretionary accruals, including Dechow and Dichev (2002), Frankel et al. (2002), Klein (2002), Chung and Kallapur (2003), Myers et al. (2003), Leuz et al. (2003), and Bergstresser and Philippon (2006).

We thank an anonymous referee for this suggestion.

A corrupt country was defined to be any country in the most corrupt quartile of the World Bank’s Corruption Index.

The additional control variables include change in receivables, change in inventory, change in cash sales, change in return-on-assets, change in the number of employees, the level of “soft” assets, an indicator for whether the firm issued debt or equity in the period, an indicator for whether the firm has outstanding leases, a measure of ex ante financing needs, Altman’s Z, and industry fixed effects. See Dechow et al. (2011) for detailed definitions of these variables.

Not every treatment firm will match with a control firm, and the propensity score approach involves a trade-off. If the matching process is relaxed so that more firms match, then the resulting matches will be less precise. Conversely, if the matching is required to be very precise, then there will be fewer successful matches.