Corporate default prediction with payment disturbances in managers’ earlier entrepreneurial practices

Figures & data

Table 1. Theoretical framework of the study.

General theoretical explanations
Concepts from the literature: upper echelons, (non-)learning from failure, serial entrepreneurship Key integrational findings: Firm performance is dependent of the actions of managers, while it is especially topical in the case of SMEs, where different corporate governance layers overlap and firms can be subject to one-man rule. Previously failed entrepreneurs are likely to start new companies, while entrepreneurial failure can be recurrent. Therefore, in the case of SMEs past entrepreneurial behaviour is likely to reoccur in the new firms.
Specific theoretical explanations
Previous payment disturbances as predictors of corporate default	Limitations of financial ratios as predictors of corporate default
Concepts from the literature: Poor financial management skills of entrepreneurs, recurrent poor liquidity management of a firm, recurrent serial non-compliance with financial reporting regulations associated with recurrent financial distress. Key integrational findings: Poor financial management skills have been found to be persistent in time. In the same firm, this has been contextualized as poor liquidity management (including the occurrence of payment disturbances) being persistent in time. In addition, not submitting financial reports on time or not at all has been found to be characteristic to serial entrepreneurship, while the latter has also been found to be linked with poor financial performance of the same firms. Therefore, in the case of serial entrepreneurs running SMEs, it is likely that payment difficulties occur in different companies ran by them.	Concepts from the literature: Existence of different failure processes, modest accuracies of the failure prediction models of SMEs, decrease in the prediction accuracies for less severe failure definitions, delays of financial reports in the case of distressed firms. Key integrational findings: In the SME sector, financial ratios have often been found to have modest prediction accuracies. The latter is more characteristic to less severe failure definitions, such as the initial emergence of payment difficulties. Such findings can be caused by proportionally more SMEs following a very quick failure process, while the delay of financial reports can also aggravate the respective information asymmetry.
Theoretical expectations of the study
Payment disturbances in the earlier entrepreneurial practices can predict corporate default.	Payment disturbances in the earlier entrepreneurial practices can be more accurate predictors of corporate default than financial ratios.

Figure 1. A conceptual scheme of events and information availability in the failure process. Note: The lighter green colour indicates a reduction in the predictive value of the earlier payment behaviour variables.

Table 2. Variables of the study.

Variable	Code	Formula
Dependent
Permanent payment default	DEFAULT	1 – Occurrence of a permanent payment default (by using its start date), 0 – firm having no default at the end of year 2019
Independents (earlier payment behaviour)
Frequency tendency of payment disturbances	COUNTPD	Number of month-ends with payment disturbances for 60 months
Average tendency of payment disturbances	MEANPD	Natural logarithm of the mean value of payment disturbances for 60 month-ends
Maximum tendency of payment disturbances	MAXPD	Natural logarithm of the maximum value of payment disturbances for 60 month-ends
Independents (financial ratios)
Profitability	ROA	Net income/total assets
Productivity	STA	Operating revenue/total assets
Solvency (and capital structure)	EA	Total equity/total assets
Liquidity	WCTA	(Current assets – current liabilities)/total assets

Notes: Earlier payment behaviour variables are calculated for all Estonian firms where the individual has been a board member by accounting for the exact membership start and end dates. Financial ratios are calculated from the last annual report (for defaulted firms, the pre-default report) for the (non-)defaulted firms included in the analysis. For defaulted firms, the three payment behaviour variables focus on the five full years before the onset of default, while in the case of non-defaulted firms, the five full years 2014–2018 are applied. For MEANPD and MAXPD, the original values of ≤1 Euro were replaced with 0 without applying the natural logarithm.

Table 3. Descriptive statistics by the status of firms.

Status	Statistic	COUNTPD	MEANPD	MAXPD	ROA	STA	EA	WCTA
Non-defaulted (N = 42,803)	Mean	7.4	2.0	3.3	0.1	2.3	0.6	0.4
	Std. deviation	14.2	2.8	3.9	0.3	2.3	0.3	0.4
	Median	1.0	0.0	0.0	0.1	1.6	0.7	0.4
Defaulted (N = 1380)	Mean	25.8	5.0	6.9	0.0	2.7	0.4	0.3
	Std. deviation	22.2	3.1	3.6	0.5	2.8	0.5	0.5
	Median	21.0	5.6	8.0	0.0	1.7	0.5	0.3
All firms (N = 44,183)	Mean	8.0	2.1	3.4	0.1	2.3	0.6	0.4
	Std. deviation	14.8	2.8	3.9	0.3	2.3	0.4	0.4
	Median	1.0	0.0	0.0	0.1	1.6	0.7	0.4
Welch ANOVA statistic and p-value		934; <0.001	1318; <0.001	1348; <0.001	47; <0.001	29; <0.001	20; <0.001	160; <0.001

Table 4. Average marginal effects from the logistic regression analysis by single payment behaviour variables for different firm populations.

(Sub-)population	Average marginal effects for three variables
	COUNTPD	MEANPD	MAXPD
All firms	0.0013684	0.0089776	0.0068433
Smaller firms	0.0013589	0.0090237	0.0067359
Larger firms	0.0013929	0.0092558	0.0072014
Younger firms	0.0012473	0.0081539	0.0062043
Older firms	0.0014828	0.0098126	0.0075278
Non-B2B sector	0.0013708	0.0087481	0.0064922
B2B sector	0.0013669	0.0090852	0.0069862
Non-exporter	0.0014421	0.0094122	0.0071722
Exporter	0.0012284	0.0082125	0.0062782
Background from one firm	0.0007602	0.0043615	0.0031354
Background from many firms	0.0017631	0.0120685	0.0093268

Note: p-values for marginal effects for all instances <0.001.

Table 5. Accuracies from the logistic regression analysis with single payment behaviour variables for different firm populations.

(Sub-)population	Accuracies by variables
	COUNTPD			MEANPD			MAXPD
	ND (%)	D (%)	T (%)	ND (%)	D (%)	T (%)	ND (%)	D (%)	T (%)
All firms	83.5	56.5	70.0	71.8	71.0	71.4	63.5	77.9	70.7
Smaller firms	84.5	54.3	69.4	72.6	68.3	70.4	63.7	74.8	69.2
Larger firms	83.1	58.6	70.9	70.8	74.5	72.6	63.6	81.3	72.4
Younger firms	82.9	52.2	67.6	70.7	69.0	69.8	62.5	75.7	69.1
Older firms	84.7	59.5	72.1	73.0	73.2	73.1	64.7	79.8	72.3
Background from one firm	87.8	43.7	65.8	78.2	57.5	67.8	72.2	63.8	68.0
Background from many firms	78.8	60.9	69.9	66.2	72.7	69.5	57.0	79.5	68.2
Non-B2B sector	82.8	55.6	69.2	71.5	66.8	69.2	62.1	73.0	67.5
B2B sector	83.7	56.9	70.3	71.9	72.0	72.0	64.0	79.4	71.7
Non-exporter	82.9	55.1	69.0	71.2	70.5	70.9	62.6	77.0	69.8
Exporter	85.2	58.3	71.7	72.9	72.8	72.9	65.3	79.5	72.4

Note: Abbreviations for accuracy (%): ND: non-defaulted; D: defaulted; T: total.

Table 6. Accuracies by two types of variables and both combined with three different prediction methods.

Method	Past payment behaviour			Financial ratios			Both included
	ND (%)	D (%)	T (%)	ND (%)	D (%)	T (%)	ND (%)	D (%)	T (%)
Logistic regression	75.5	66.2	70.9	66.0	53.0	59.5	76.5	68.6	72.5
Neural networks	75.9	69.8	72.9	71.5	52.9	62.1	75.5	74.9	75.2
Decision tree	73.1	74.3	73.7	75.5	56.8	66.1	80.8	70.6	75.7

Note: Abbreviations for accuracy (%): ND, non-defaulted; D, defaulted; T, total. In the case of neural networks and decision tree, the highest accuracy out of five runs is presented, while it is important to note that the difference between the highest and lowest accuracy was constantly below one percentage point. In the case of logistic regression, the accuracy of a single model is presented, as it remains constant for different runs.

Table 7. Study’s results and their theoretical essence.

Hypotheses	Main results	Theoretical essence
H1a. The frequency of payment disturbances in the earlier entrepreneurial practices of managers increases the likelihood of corporate default. H1b. The size of payment disturbances in the earlier entrepreneurial practices of managers increases the likelihood of corporate default.	Hypotheses H1a and H1b were accepted. The latter result is valid in the case of the single frequency proxy applied and both size proxies (reflecting average and maximum tendencies) applied by using the marginal effects from the logistic regression. Firm context (size, age, business-to-business, international orientation, earlier entrepreneurship) plays a role, meaning that the values of marginal effects vary for different firm types. Still, except for the earlier entrepreneurship context (i.e. whether the manager has run additional firms during earlier five years), the differences in the marginal effects for the other firm contexts cannot be considered large.	The general theoretical meaning of the results is that failures (linking to the non-learning from failure literature) in the earlier careers of managers (linking to the serial entrepreneurship literature) affect a company’s performance (linking to the upper echelons literature). The specific theoretical meaning of the results is that poor (financial) management and specifically liquidity management skills remain persistent in serial entrepreneurship (linking to the liquidity and financial management literature). In addition, as the out of the sample tests on the population of defaulted firms not submitting any financial reports indicate the existence of predictive value of payment disturbances, bad payment behaviour is associated with accounting law violations (linking to the non-compliance literature).
H2. Payment disturbances occurring in managers’ earlier entrepreneurial practices are better predictors of a firm’s default than financial ratios.	Hypothesis H2 is accepted, as payment disturbances variables together and also individually are more accurate than financial ratios. The highest accuracy achieved with three variables reflecting payment disturbances is 73.7%, while the same for financial ratios is 66.1%. The latter results are obtained with the decision tree, while in the case of other two methods the differences in the accuracies for the two variables’ pools are even larger. The variable reflecting the average tendency is individually the most accurate, followed by the maximum and frequency tendencies, although the differences in the individual accuracies are minimal (the maximum difference being 1.4 percentage points) and all exceed 70%.	Financial ratios have poor predictive value in signaling future corporate default (linking to the literature of SME failure prediction), potentially indicating their inability to detect the early stages of the quick failure process (linking to the literature on failure processes and their stages). In addition, in the case of the financial reports’ absence, payment disturbances remain an important available tool to predict corporate default (linking to the non-compliance literature).

Data availability statement

Data not available due to legal restrictions.