4. RESULTS 1 Descriptive statistics
3.2 Key variables
3.2.1Criminal executives and criminal workforces
On the person level, I define an indicator CRIME that takes the value one if a person has a prior criminal record, excluding crimes related to traffic offences such as parking tickets, speeding tickets, etc. (similar to Kallunki et al. 2018), and zero otherwise21. Then, I aggregate the CRIME information to the firm-year level, and construct the following variables:
%CrimEXEC denotes the percentage of executives with a criminal record (percentage of executives where CRIME=1) within a firm-year, and %CrimEMPL denotes the percentage of employees with a criminal record (percentage of employees where CRIME=1) within a firm- year. %CrimEXEC is used to test H1 and %CrimEMPL is used to test H2.
used for accrual estimation (for example, observations for the year 2014 include cash-flow information for the years 2012-2016).
20 The number of person-firm-year observations is slightly higher than the number of person-year observations,
because one person can be employed at more than one firm at the same time.
21 In untabulated tests I repeat all analyses using only traffic related offences and find insignificant results,
157 Then, to test H3, I divide the observations into four groups based on both the proportion of executives with a criminal record and the proportion of employees with a criminal record. First, I define when the executive team is classified as criminal and when the workforce is classified as criminal. The variable CrimEXEC takes the value one if the majority of the executive team members has a criminal record (%CrimEXEC>0.5), and zero otherwise. The variable
CrimEMPL takes the value one if the proportion of employees with a criminal background is above the within-year median (%CrimEMPL>within-year median of %CrimEMPL), and zero otherwise. I use the within-year median of %CrimEMPL to define CrimEMPL to overcome fluctuations of criminal employee distributions over time, and to measure the βcriminalityβ of the workforce relative to other firms. I use a cutoff of 0.5 (i.e. the βmajorityβ) of executives to define CrimEXEC, because only one executive is identified for 82% of the sample observations, and thus using the within-year median is not feasible. From these definitions, I categorize the observations into four groups, as depicted in Table C.2:
Table C.2: Four groups based on CrimEXEC and CrimEMPL
Groups based on CrimEXEC
and CrimEMPL
CrimEXEC=1 CrimEXEC=0
CrimEMPL=1 1/1 0/1
CrimEMPL=0 1/0 0/0
Where the 1/1 group (both executives and the workforce are relatively criminal) is the group relevant for testing H3.
3.2.2Discretionary accruals
I estimate discretionary accruals (DACC) as the residuals of the following estimation:
πππ΄πΆπΆπ,π‘ = πΌ0+ π½1πΈπππΏπΊπ π,π‘+ π½2πΈπππΏπΊπ π,π‘+1+ π½3πΈπππΏπΊπ π,π‘β πππ΄π,π‘β1+ π½4πππΆπΉπ,π‘β2+ π½5πππΆπΉπ,π‘β1+ π½6πππΆπΉπ,π‘+ π½7πππΆπΉπ,π‘+1+ π½8πππΆπΉπ,π‘+2+ π½9π·π’ππππΆπΉπ,π‘+ π½10π·π’ππππΆπΉπ,π‘β πππΆπΉπ,π‘+ π½11π ππ΄π,π‘β1+
β πΌππ·ππππ π + β ππΈπ΄π + ππ,π‘
(1)
where i indexes firms and t indexes time (year). OPACC is comprehensive operating accruals, including both working capital accruals and non-current operating accruals. Following Larson et al. (2018) I control for current comprehensive operating cash flows (OPCF), two leads and lags of OPCF, growth in employees (EMPLGR22), and an interaction of EMPLGR and lagged net operating assets scaled by assets (EMPLGRt*NOAt-1).
22 Revenue data, as used in conventional research when estimating discretionary accruals, is not available for the
158
Additionally, I control for negative cash flows (dumOPCF) and an interaction between negative cash flow and cash flow (DumOPCF*OPCF) to allow a piecewise linear relation between current OPCF and OPACC (Ball and Shivakumar 2006)23. Further, I complement Larson et al.βs (2018) model and include lagged return on assets (ROAt-1) to control for
performance (Kothari et al. 2005). I control for lagged ROA and not current ROA because current ROA and current OPCF would perfectly explain OPACC. I also control for future employee growth (EMPLGRt+1) because firms invest based on expectations to future growth
(Collins et al. 2017)24. All continuous variables are winsorized at the 1% and 99% level to accommodate for outliers, and all variables are defined in appendix. I point out that in all but descriptive analyses I estimate discretionary accruals in a one-step procedure (Chen et al. 2018).
3.2.3New finance
In my identification of an opportunistic setting in which the firm has incentive to manage earnings, I use events where the firm raises new financing (either debt or equity financing). In my identification I follow Godsell et al. (2017), who likewise base their analysis on ORBIS data, and use a similar method to identify an opportunistic setting when using accruals as proxy for earnings management25.
First, I calculate the difference between long-term bank debt in year t+1 and long-term bank debt in year t-1, and scale the difference by assets in year t-1. I define DEBT_ISSUE as an indicator variable taking the value one if the change in debt scaled by assets is larger than 0.05, and zero otherwise. Second, I calculate the difference between shareholdersβ equity in year t+1
and shareholdersβ equity in year t-1, and further deduct the sum of net income in year t and net income in year t+1, and scale this number by assets in year t-1. I define EQUITY_ISSUE as an indicator variable taking the value one if the change in equity (controlling for concurrent income) scaled by lagged assets is larger than 0.05, and zero otherwise. Finally, I define the
revenue. Instead, I use employee growth β a growth measure not subject to manipulation (as for example revenue) β similar to Allen et al. (2013) and Larson et al. (2018).
23 I point out that Larson et al. (2018) use a piecewise version of MTB (market-to-book ratio) to model
conditionally conservative accruals, but use negative cash flows (DumCF and DumCF*CF) in robustness tests. Market values are naturally not available for private firms.
24 I note that Collins et al. (2017) use MTB to proxy for growth opportunities, which is not available for my sample.
Therefore, I use realized employee growth for year t+1 instead.
25
159 variable NEW_FIN as an indicator variable taking the value one if either DEBT_ISSUE or
EQUITY_ISSUE equals one, and zero otherwise.
The variable captures firms raising new finance in year t or t+1. Due to the lack of cash flow statements I am not able to directly observe cash flows originating from financing activities; hence I proxy those using ORBISβ standardized balance sheet items.