I use discretionary accruals as a measure of discretion exercised in financial reporting. Larson et al. (2018) argue that barring some compelling reason to focus on working capital accruals alone, researchers should focus on comprehensive operating accruals, and therefore I use
8 All Danish bankruptcies are made publicly available by the Danish Official Gazette (statstidende.dk). Konkurs.dk
draws information from this information source. Data from konkurs.dk is cross-checked with data from Statistics Denmark. Konkurs.dk provides bankruptcy data on firm level, and Statistics Denmark provides summarized monthly bankruptcy data for the full economy.
9 The auditing exemption requirements are as follows: For two consecutive years the company cannot exceed two
of the following three thresholds: (1) Total assets of DKK 4m, (2) operating revenue of DKK 8m, and (3) number of full time equivalent employees of 12. However, revenue and employee data are not available for the full sample, and hence I use a conservative (higher than the actual threshold) total asset restriction criteria.
Formerly, the thresholds were even lower. The thresholds were increased in 2006, 2011, and 2013, respectively.
10 An international definition of SMEs does not exist. I use total asset constraint to define my SME sample, and use
the thresholds set forward by the European Commission. DKK 323m approximately equals EUR 43m. I use a DKK/EUR rate of 7.5.
61
Table A.1: Sample identification
Note Screen applied Observations
dropped
Sample size Decrease in
sample size, %
1 All firm-year observations, fiscal years 2003-2015 2,635,218
Keep financial reports with 12 months 106,237 2,528,981 4% Remove observations with missing data on total assets 192,364 2,336,617 8% 2 Keep firm-years with ta≥10m & ta≤323m 1,859,598 477,019 80% Remove observations with missing data on net income 10 477,009 0% 3 Remove certain industries 179,468 297,541 38% 4 Remove subsidiaries 12,460 285,081 4% Remove listed firms 436 284,645 0% Keep observations with data available for estimation 118,972 165,673 42% This table shows the sample selection procedure. Notes: (1): The period 2003-2016 are the years for which bankruptcy data are available, and to allow one year’s lag between the fiscal year end and the bankruptcy filing, I restrict the period to include 2003-2015. (2) The lower cap aims to assure that all financial statements are audited, and the upper cap conforms to the SME definition of the European Commission. I note that the total asset criterion is only one of three to define a company as an SME, but – similar to the audit requirement discussion stated above – revenue and employee data are not available for all observations. (3): Consistent with prior accounting and finance research I exclude certain regulated industries (financials and utilities), and further exclude state-owned companies. (4) To avoid double counting I exclude subsidiaries.
comprehensive operating accruals as my measure for accruals. In robustness tests I estimate discretionary working capital accruals, and results remain unchanged.
Inspired by Collins et al. (2017) I model normal and discretionary accruals as a non-linear function of growth and current profitability, and a linear function of lagged accruals and size. Further, I complement the model with current and lagged cash flows, because Allen et al. (2013) show that the component of normal accruals predicted by cash flows11 is the most persistent component of accruals, indicating that controlling for cash flows is important when dividing accruals into normal accruals and discretionary accruals. I do not include leaded cash flows in the model, because this would induce a mechanical bias between modelled accruals and future profitability/cash flows (Allen et al. 2013). If future cash flows are already controlled for when estimating discretionary accruals, discretionary accruals cannot contain information value about future cash flows. Because future cash flows are correlated with future profitability, a similar argument can be raised regarding future profitability.
I estimate Eq. (1) for each industry-year, classify growth and profitability measures into quintiles by industry-year (Collins et al. 2017), and require at least 30 observations per industry- year. Discretionary accruals (DACC, hereinafter) are the residuals from estimating Eq. (1).
11 In Table 5 of Allen et al., they term this component “MDDMATCH” and find that it to a very high extend maps
62 𝑂𝑃𝐴𝐶𝐶𝑖,𝑡 = 𝛼0,𝑖,𝑡+ 𝛼1 1 𝑇𝐴𝑖,𝑡+ ∑ 𝑅𝑂𝐴_𝐼𝑁𝐷𝑘,𝑖,𝑡 5 𝑘 + ∑ ∆𝐺𝑃_𝐼𝑁𝐷5𝑘 𝑘,𝑖,𝑡 +𝛽1𝑂𝑃𝐴𝐶𝐶𝑖,𝑡−1+ 𝛽2𝑂𝑃𝐶𝐹𝑖,𝑡+ 𝛽3𝑂𝑃𝐶𝐹𝑖,𝑡−1+ 𝜀𝑖,𝑡 (1)
Where OPACC is comprehensive operating accruals estimated using the balance sheet approach12 scaled by lagged assets, ROA is net income scaled by lagged assets, ΔGP is the change in gross profit scaled by lagged assets, and OPCF is comprehensive operating cash flows scaled by lagged assets, for firm i in year t. ROA_IND is an indicator variable that takes the value one if ROA in the industry-year belongs to the kth quintile, and zero otherwise. A similar procedure is used to define ΔGP_IND (see Collins et al. 2017). For completeness, in Table A.2 I report the regression coefficients using a pooled regression, where quintile indicators are generated per industry-year.
Data availability is constrained by exemption rules, as disclosure of revenue and cash flow statements is voluntary for most firms in the sample. Therefore I use gross profit growth as proxy for economic activity when estimating discretionary accruals. In untabulated analysis I estimate discretionary accruals for the 134,693 (28,089) firm-years with employee (revenue) data available, and find a high correlation of 0.93 (0.80) between the discretionary accruals estimated with gross profit growth and employee growth (revenue growth) respectively. In robustness tests I repeat all analyses using revenue growth and employee growth, respectively, for the subsamples in which the data are available.