Empirical Design

The empirical approaches employed throughout the dissertation are primarily econometric, and include multiple regressions estimated with ordinary least squares, panel data estimation with firm fixed effects estimated with ordinary least squares, multiple logistic regressions

40

estimated with maximum likelihood, propensity score matching, and instrumental variable regressions. Further, the results in Chapter 2 are complemented with interview insights about banks’ lending practices. In all standard multiple OLS regressions and logistic regressions standard errors are clustered by firm and year to correct for cross-sectional and time-series dependence in the error term (Gow et al. 2010).

In the first paper, I rely solely on quantitative estimations. I estimate discretionary accruals using recent model enhancements (Collins et al. 2017; Larson et al. 2018) and use the residuals from the estimation model as a proxy for accrual discretion exercised in the financial reporting.

Then, I estimate a bankruptcy probability model following Beaver et al. (2005), and use the predicted values of this estimation as a proxy for financial distress. Then, to test the research question of this paper, I estimate standard earnings persistence regressions with both future return on assets and future operating cash flows as dependent variables, respectively, and investigate the influence of discretionary accruals on earnings persistence/cash flow prediction.

In the second paper, my co-authors and I base the main analysis on quantitative estimations and complement the results with interview insights with a number of large Danish banks. We estimate the propensity to shift salary to dividends (salary-dividend earnings management:

SDEM, hereinafter) using logistic regression. Then, to test if the use of SDEM has implications for the firm’s cost of debt, we use multiple regressions and estimate future cost of debt as a function of current SDEM and controls, and complement with propensity score matching and IV regressions. To extend our understanding of how banks use financial reporting information in the lending decision, we conduct semi-structured interviews with four large Danish banks. To avoid blurred answers we initially tell the interviewees that the research project explores earnings management in private firms, but not the specific channel through which we investigate earnings management (SDEM).

In the third paper, I rely on quantitative estimations. I estimate accruals in a one-step procedure (Chen et al. 2018) based on Larson et al.’s (2018) accrual estimation model, with several additional controls improving my ability to determine innate (“normal”) accruals, and hence discretionary accruals. In the main analysis I rely on the slope on an interaction term between the variable of interest and an indicator of new finance issuance. The interaction coefficient thus captures the incremental effect of the variable of interest on accruals, given that the firm issues new finance. This econometric procedure is similar to that of several related research papers (Ayers et al. 2006 Table 1, Panel D; Balsam et al. 2002 Table 3; Call et al. 2014 Table 5; Frankel et al. 2016 Table 5; Gul et al. 2003 Table 4; Doukakis et al. 2019 equation 1).

41 3.5 Limitations and future research

Investigating financial reporting in Danish private firms comes with benefits in the form of large sample sizes, access to proprietary data difficult to obtain elsewhere, and special agency settings. Despite those important benefits researching earnings management in private firms comes with several major limitations: First, market prices are naturally not available. Therefore, this dissertation is limited from linking financial reporting to stock returns to test the “value” of financial reporting (Allen et al. 2013; Richardson et al. 2006; Richardson and Sloan 2005;

Dechow and Dichev 2002; Xie 2001; Subramanyam 1996), using market based proxies for growth opportunities (Collins et al. 2017), and using market based variables in the estimation of the probability of default (Hillegeist et al. 2004; Shumway 2001).

Second, a wide range of conventional variables used by prior research (Biggerstaff et al.

2015; Ali and Hirshleifer 2017; Liu 2016; Davidson et al. 2015; Kallunki et al. 2018; Dhaliwal et al. 2011) to capture earnings management, or outcomes of earnings management, are not available for private firms, such as restatements, meeting or beating analyst forecasts, SEC enforcements, and internal control deficiencies, as well as other proxies for opportunistic firm behavior, such as option backdating, insider trading, and shareholder litigations. Future research could benefit from obtaining data on other opportunistic firm outcomes of private firms than those employed in this dissertation. For example, researchers could obtain data on the auditors’

adjustments to managers’ submitted accounting data (see e.g. Lennox et al. 2018), or enforcement actions of the business authorities, if such data is available anywhere.

Third, detailed data on loan characteristics are not available and thus this dissertation relies on proxies of cost of debt using financial expenses scaled by liabilities net of trade payables.

Other contracting terms that may influence the total cost of debt include collateral, distance to the bank, covenants, and length of bank relationship (Cassar et al. 2015; Agarwal and Hauswald 2010; Granja et al. 2019). If such data are available elsewhere, future research could benefit greatly from including those factors in the analysis. Although not currently available, I point out that the Danish Central Bank has just recently mandated Danish banks to file with the Central Bank detailed loan-level data (instead of aggregated data as off now)⁹. The data will be available for researchers through Statistics Denmark’s researcher access later this year. I believe such data can contribute greatly to further refine the research questions asked in this dissertation.

9See http://www.nationalbanken.dk/da/statistik/FIONA/Sider/Banker,-realkreditinstitutter-mv.aspx. Look for

“kreditregister”. Additionally, I have been in contact with the Central Bank, who plans to launch the dataset in the third quarter of 2019. Which data that will become available to researchers is not yet determined.

42

I point out that the institutional setting of using Danish firms may impair the generalizability of the findings of this dissertation. For example, as with most European countries all Danish limited liability firms are mandated to disclose financial statements publicly¹⁰, and the data can be easily extracted from a central database by firm stakeholders, and therefore capital providers’

information acquisition costs are likely lower than for example in the US or Canada (Minnis and Shroff 2017), which may influence how firm stakeholders use and rely on reported statements.

Further, the Danish setting is characterized by high legal and enforcement quality and a low level of alignment between financial accounts for external reporting and tax purposes (Burgstahler et al. 2006)¹¹. It would be interesting to see how lenders rely on financial statement data (and adjust for managers’ salary) in jurisdictions where public financial disclosure is not mandatory, or in jurisdictions without approximate tax neutrality. In the light of those caveats, however, most of the findings are based on theories of information asymmetries – issues that are present in most countries around the world.

10 And as in other European countries the information content of the financial report increases with firm size.

11 See also Leuz et al. (2003) and Blaylock et al. (2015) for additional factors characterizing Denmark.

43