Key variable definitions and descriptive statistics

In the following we discuss the key measures used throughout the paper. We show descriptive statistics in Table B.2, and summarize variable definitions in Table B.11 (appendix).

The variable selection process is based on prior research on the characteristics of earnings management firms and on covariates associated with cost of debt. Obviously, we cannot include in our analysis market-based variables.

3.3.1 Owner-managers

As outlined above we limit the sample to owner-managers. We identify owner-managers in the following way: First, we use the data on ownership that we obtain from the Danish Business Authority. We create ultimate ownership percentages through direct links (individual owns company) and indirect links (individual owns company X that owns company Y that owns company Z…). From these data, we identify an owner-manager as a person that owns at least

Table B.2: Descriptive statistics All observations (unconditional)SDEM=1 (conditional) (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12) NMeanS.D.P25P50P75NMeanS.D.P25P50P75 Earnings management and earnings benchmarks SDEM98,5050.0820.2740.0000.0000.0008,0651.0000.0001.0001.0001.000 LossAvoid98,5050.0050.0710.0000.0000.0008,0650.0630.2420.0000.0000.000 Decreaseavoid98,5050.0060.0750.0000.0000.0008,0650.0700.2550.0000.0000.000 SalaryTA98,5050.1160.1020.0430.0860.1578,0650.1160.1020.0440.0850.155 ΔSalaryTA98,5050.0020.026-0.0030.0000.0078,065-0.0180.020-0.022-0.011-0.005 SmallLoss98,5050.0630.2430.0000.0000.0008,0650.0700.2550.0000.0000.000 SmallDecrease98,5050.1110.3150.0000.0000.0008,0650.1070.3090.0000.0000.000 Cost of debt and leverage CostDebtt+198,5050.0440.0380.0190.0370.0578,0650.0420.0390.0170.0340.053 DebtTA98,5050.4820.2150.3150.4880.6478,0650.4680.2080.3080.4730.624 TLTA98,5050.6170.2190.4660.6480.7928,0650.6010.2110.4510.6270.766 NewDebtt+198,5050.3280.4700.0000.0001.0008,0650.3720.4830.0000.0001.000 ROA and its components ROA98,5050.0750.1230.0050.0530.1308,0650.1070.1300.0230.0790.164 OPCF98,5050.0530.221-0.0580.0460.1668,0650.0920.229-0.0310.0780.211 OPACC98,5050.0220.197-0.0800.0080.1128,0650.0150.201-0.0880.0060.111 DACCGP80,699-0.0000.071-0.037-0.0020.0366,7550.0120.073-0.0270.0080.049 DACCEG80,006-0.0000.085-0.047-0.0020.0446,7110.0100.088-0.0400.0060.057 ABS_DACCGP80,6990.0510.0480.0160.0370.0716,7550.0540.0500.0170.0390.074 ABS_DACCEG80,0060.0620.0570.0200.0450.0876,7110.0660.0600.0220.0480.092 ΔROA98,5050.0050.116-0.0500.0010.0548,0650.0220.124-0.0400.0140.076 Other controls StdROA98,5050.0910.0920.0380.0660.1128,0650.0940.0940.0400.0680.116 PPE98,5050.2580.2490.0560.1650.4108,0650.2470.2410.0570.1590.385 CashTA98,5050.1440.1850.0040.0620.2268,0650.1580.1880.0060.0840.251 TA (mDKK)98,5058.92811.1442.5144.75510.1558,0659.99511.6862.9835.61311.802 Personals PersEquityTA98,5050.2700.5480.0000.1010.3698,0650.2760.5400.0040.1110.375 Age98,37149.6859.30743.00050.00057.0008,05150.3349.47943.00050.00058.000 Criminal98,5050.2130.4090.0000.0000.0008,0650.1990.3990.0000.0000.000 Female98,5050.0710.2570.0000.0000.0008,0650.0630.2420.0000.0000.000 HighEduc98,5050.0780.2680.0000.0000.0008,0650.0940.2920.0000.0000.000

102

103 95% of the company¹⁵ and is the CEO. For the firm-year observations where ownership data are missing¹⁶, the current CEO is identified as an owner-manager if she was also a CEO on the date the firm was founded.

3.3.2 Salary-dividend earnings management (SDEM)

We define the event of earnings management as an indicator variable SDEM taking the value one if (1) the owner-manager decreases her salary significantly (at least by 5 percent and at least by DKK 10t¹⁷ (EUR 1.3t)), (2) the salary decrease does not shift the owner-manager’s marginal labor income to a lower tax bracket¹⁸, and (3) the owner-manager contemporarily increases dividends to at least offset the after-tax salary decrease, and zero otherwise. Marginal tax rates for labor income vs. capital income along with a numerical example are presented in appendix.

Dividends can be distributed as ordinary dividends when the annual report is approved for publication (i.e. dividends related to the income of year t are distributed in year t+1 when the annual report is published) or as extraordinary dividends (i.e. dividends related to the income of year t are distributed during year t). Therefore, we allow the manager to pay out the dividends either during the fiscal year of the salary decrease or in the following year. We estimate dividends from balance sheet and income statement items.

We observe SDEM in around 8 percent of the firm-year observations, and find that SDEM on average increases earnings scaled by assets by 1.8 percentage points. Further, we observe that managers partly reverse their salary decreases from SDEM years in the following year: For all SDEM observations we observe an average salary change of -16% in the SDEM year followed by an average salary increase by 8% in the following year. Further, in Figure B.1 we show the development of salary/TA and dividend/TA for SDEM firms and their propensity score matched

15 We use 95% rather than 100% because of potential rounding of ownership stakes. Further, we assess this identification as conservative, because other ownership structures which assimilate the owner-manager structure are excluded. These quasi owner-manager ownership structures include married couples owning a business together, family members owning a business together, or even close friends owning a business together.

16 In December 2014 new regulation was enforced which required firm owners to file ownership data with the Danish Business Authority, with a retrospective effect, meaning that managers had to disclose the starting date of their ownership. Hence, the ownership data which we acquire from the Danish Business Authority is limited in coverage back in time.

17 We impose 5% and a monetary amount to define a significant salary decrease. Comparable approaches (as well as a similar cutoff of 5%) are used in the literature to define significant R&D jumps (Eberhart et al. 2004; Dube 2019).

18 We have data on (1) the managers’ income from the firm, and (2) the manager’s total taxable labor income from Denmark. The cases where the owner-manager lowers her salary from the firm so that the marginal labor income (of total taxable labor income) is shifted to a lower labor income tax bracket are probably due to tax optimization rather than earnings management, and therefore we do not include those cases in the SDEM definition.

104

Figure B.1: Salary and dividend changes (scaled by total assets) by SDEM firms and their matched peers

This figure shows the level of salary (salary/TA) and dividend (dividend/TA) preceding, during, and following the SDEM year for SDEM firms and the matching year for matching firms. Year t=0 (x-axis) refers to the SDEM year for SDEM firms, and the matching year for control firms.

Control firms are matched with propensity score matching, as described later in the paper. Descriptive statistics of the two propensity score matched samples are presented in Table B.9.

peers¹⁹. In this figure, we observe partly reversal of both salary and dividends following the SDEM year²⁰. However both salary and dividends remain on a higher level following the SDEM year than in the years before the SDEM year.

3.3.3 Magnitude of debt and Cost of debt

Data coverage on actual interest bearing debt (or bank debt) is very limited in our dataset, and therefore we proxy it by calculating total liabilities net of trade payables. We scale by assets and term this measure DebtTA. Similarly, the actual interest rate on debt is not provided in the dataset, so we proxy it as financial expenses divided by average total liabilities net of trade payables, and term this measure CostDebt. The procedure of estimating the cost of debt as financial expenses scaled by debt is comparable to that in related studies (Francis et al. 2005;

Minnis 2011; Gassen and Fülbier 2015; Vander Bauwhede et al. 2015). However, we differ from those studies in that we scale financial expenses by debt net of trade payables and not bank debt due to data limitations. We acknowledge that our approach contains significant noise and truncate the CostDebt construct at 0 percent and 30 percent to avoid extreme observations from

19 We discuss how we propensity score match in section 3, and show the results of the matching in Table B.9.

20 Recall that dividend increases are allowed both for year t and year t+1 for the SDEM definition, and therefore the dividend reversal happens in year t+2.

105 a noisy measure blurring our results (see e.g. Gassen and Fülbier 2015)²¹. We observe an average CostDebt of 0.044, which is lower than observed in comparable studies²². The difference is likely due to our definition of debt and thus CostDebt is probably understated.

However, empirical estimations in section 4 give confidence that CostDebt, although a noisy measure, is a valid proxy of the variations in the true (and for us unobservable) cost of debt.

3.3.4 Earnings measures and earnings benchmarks:

We make two overall adjustments to earnings and cash flows. First, we compute earnings and cash flows net of salary changes (variable names are added with the term “netsalary”), which proxy the performance signal the manager receives before making the decision to use SDEM.

We use these measures to estimate the propensity to use SDEM. Second, we generate pre-managed earnings and cash flows: these measures adjust for salary changes (variable names are added with the term “premanaged”), but only when observations are identified as SDEM observations. We use these measures when we (i) investigate the impact of SDEM on future cost of debt, controlling for the underlying performance (i.e. pre-managed), and (ii) to identify the incidences in which the owner-manager uses SDEM to transform losses (earnings decreases) into profits (earnings increases).