“Failure processes of old manufacturing firms in different European countries”

This study aims to detect failure processes on the example of old bankrupted European manufacturing firms. Two study designs are applied, namely the original six variables from Laitinen’s (1991) model and an extended dataset with eleven variables for a five-year timespan before declared bankruptcy. On both occasions, two different failure processes are detected which indicate elements of either quickly or gradually failing firms. Clear contingencies between detected processes and firms’ countries of origin exist. There is some evidence that firms of different sizes follow varying failure processes, but this does not apply when discriminating between exporters and non-exporters.


Introduction ©
In recent years, research about firm failure has flourished.This is especially observable in the domain of failure prediction studies, where dozens of papers appear annually, both applying classical statistical and novel machine learning techniques.When a myriad of different prediction models have been created since the piloting multivariate study by Altman (1968), another close research domain -firm failure processes -has received episodic attention since the piloting monograph by Argenti (1976).Studies focusing on firm failure processes offer an insight in which different ways firms collapse.Both causal and symptomatic domains have been elaborated in this research stream.In a few past years, the topic has reemerged and multiple recent studies are available (e.g., Laitinen and Lukason, 2014;Laitinen et al., 2014;du Jardin, 2015), some of which also provide first evidence of inter-country comparison of failure processes.Recent studies also carry several limitations like small and multi-sector datasets.Therefore, this study aims to provide single-sectoral evidence from a large set of European countries relying on multiple different study designs.
From a more theoretical perspective, this study contributes to the liability of obsolescence concept explaining the failure of old firms.The main foundation of this approach is that firms become inert to changes in environment and therefore start to decline (Barron et al., 1994;Henderson, 1999;Thornhill and Amit, 2003).Still, the timing of such a decline and its characteristics depicted through different financial variables remain understudied in the available research focusing on the liability of obsolescence concept.The topic of failure processes has also been neglected in bankruptcy prediction studies and only a few of them (e.g., Pompe and Bilderbeek, 2005) have outlined the large dispersion in the pre-bankruptcy values of firms' financial ratios, therefore, offering some indication of the possible presence of different failure processes.
The general objective of the paper is to create a taxonomy of failure processes on the example of old European manufacturing firms.This will be achieved by applying two different study designs on a dataset of manufacturing firms from all size categories from 15 European countries.The paper is focused on manufacturing firms that were at least 10 years old by the moment of bankruptcy declaration.
The paper is structured as follows.The literature review focuses on past studies about failure processes.In the literature review part, also key terminology is outlined and hypotheses are set to guide the research.The overview of past studies is followed by the section describing the sample, variables and methods of the analysis.After that, the results of the study are presented and discussed.The paper ends with a conclusion also encompassing implications.
1.A review of literature 1.1.Context of failure and failure process.Failure has been defined differently through studies, but researchers focusing on failure prediction and failure processes have almost univocally used courtdeclared permanent insolvency (i.e., bankruptcy) as the definition of a failed firm (Mellahi and Wilkinson, 2004;Balcaen and Ooghe, 2006;Laitinen et al., 2014).The popularity of bankruptcy as a definition is obviously connected to the fact that such firms are deleted from business registers and are therefore "dead".In case of other possible definitions (see, e.g., Mellahi and Wilkinson, 2004, p. 22) it is always arguable, what situation to consider as failure.For instance, during economic downturn, a majority of firms can witness temporary liquidity problems or even temporary payment defaults.As another example, a firm can deliberately earn losses when applying a strategy of out-marketing competitors.
There is more terminological and content-related variation when considering firm failure processes (Laitinen and Lukason, 2014, p. 811).When several studies follow the example of Laitinen's (1991) study by applying the term "failure process" (e.g., Laitinen and Lukason, 2014;du Jardin, 2015), then, others apply completely different terms (e.g., "trajectories" in Argenti, 1976 and "patterns" in D'Aveni, 1989) for a very similar concept.Using different terms is often merely a linguistic issue.More variety arises when considering the variables different authors have applied for modelling the failure process.Such diversity can be aggregated as whether studies apply "why", "how" or both of them for depicting the failure process.Thus, there are studies focusing only on failure causes (e.g.Crutzen and van Caillie, 2010), financial indicators (e.g., Laitinen, 1991) and on both of them (Laitinen and Lukason, 2014).This study focuses solely on financial variables to model the failure process and failure is defined as bankruptcy (i.e., permanent insolvency declared at court).

Failure processes detected in previous
studies.Empirical detection of failure processes is a rather scarce topic in available literature.Most of the available studies have focused on the example of old firms, whereas study designs have been diverse.Studies can be divided between those providing case study evidence (e.g., Argenti, 1976; Sheppard and Chowdhury, 2005; Ooghe and de Prijcker, 2008), and those applying statistical analysis on samples (e.g., D 'Aveni, 1989;Laitinen, 1991;Laitinen et al., 2014).Sample sizes have varied, being below 100 firms in earlier studies and remarkably larger in recent ones (e.g., 558 firms in Laitinen et al., 2014).Available research has almost exclusively applied multi-sector approach and only a scant amount of studies (e.g., Lukason, 2014) are focusing on a single sector.In multi-sector studies, the manufacturing sector has mostly been included, although analysis results might not include sectoral breakdown.Process extraction methods have mostly included classical statistical tools like factor and/or cluster analysis (e.g., D'Aveni, 1989; Laitinen, 1991; Laitinen et al., 2014), but there are examples of more sophisticated novel tools as well (e.g., du Jardin, 2015).Thus, the set of applied methods has been rather constrained, unlike in the connected research stream of bankruptcy prediction, where novel methods are introduced almost annually (see e.g., Ravi Kumar and Ravi, 2007; Kirkos, 2012).In case of old firms, the modelling focus has been set on the last years of existence, which can for instance include only two years (e.g., Laitinen and Lukason, 2014), four to six years (e.g., D 'Aveni, 1989;Laitinen, 1991;Laitinen et al., 2014) or even ten years (e.g., Hambrick and D'Aveni, 1988) before bankruptcy.The considered time span has some connections to firm size, as, for instance, in Hambrick and D'Aveni (1988), the focus was on large US corporations, whereas in Laitinen and Lukason (2014) mostly on micro-firms.
Most of the studies have detected a small number (three or four) of different processes, which represent the chronically unsuccessful firm (i.e., financial ratios indicate poor performance long before bankruptcy is declared), gradually deteriorating firm (i.e., performance becomes worse step by step until bankruptcy is declared) and suddenly collapsing firm (i.e., forthcoming bankruptcy might not be indicated through financial ratios not earlier than just before bankruptcy) (e.g., D 'Aveni, 1989;Laitinen, 1991;Laitinen et al., 2014).In Laitinen's (1991) study, process extraction was based on a theoretical model outlining which variables and how could be interconnected in the failure process.These variables were (Laitinen, 1991, p. 656): return on investment ratio, rate of growth in total assets, net sales to total assets ratio, traditional cash flow to net sales ratio, capital assets ratio and current ratio.In Laitinen et al. (2014) study that applied similar variables as Laitinen (1991), manufacturing firms were pooled together with mining and quarrying firms, accounting for 30.7% of the applied sample.The representation of processes in Laitinen et al. ( 2014) study was as follows: a) acute failure 23% (two different acute failure sub-processes detected), b) gradual failure 62%, and c) chronic failure 15%.In Laitinen's (1991) model, liquidity and solvency were captured by three variables: traditional cash flow ratio (dynamic liquidity, but this variable also portrays cash flow based profitability), current ratio (static liquidity) and capital assets ratio (i.e., 1 -equity ratio; static solvency).In a model specifically focusing on firm liquidity and solidity (see Laitinen, 1995), a larger set of variables was used to portray the static and dynamic concepts of liquidity and solidity.Namely, dynamic liquidity was measured by two variables (traditional cash flow and operating cash flow ratios), static liquidity by two variables (quick ratio and average payment period of accounts receivables), dynamic and static solvency were both measured by one variable (respectively shareholder's capital to total capital and cash flow to total debt ratios) (Laitinen, 1995, p. 441).In Laitinen (1995), it was demonstrated that failing firms can face various combinations of liquidity and solvency problems.
In the model proposed for old firm failure by Ooghe and de Prijcker (2008, p. 233), some important contributors to failure have been noted to be declining sales, high expenses (operational costs) and inappropriate capital expenditures (excess leverage).In Moulton et al. (1996), different failure processes have been described with the development in total assets, total debt and total sales.All different failure processes are characterized by debt growth exceeding assets growth, whereas sales growth has varied from negative to positive (Moulton et al., 1996, p. 585).Also, changes in firms' total assets and costs during downfall have been well documented in another stream of literature, namely turnaround studies (e.g., Robbins and Pearce, 1992; Barker and Duhaime, 1997; Smith and Graves, 2005).
Thus, a failure process can be modelled in a narrower classical approach, but also in an extended framework by incorporating additional variables.This has been depicted on Figure 1.The development of total debt, sales and operating costs links well to the variables applied in Laitinen (1991), as these variables are statically included in the denominators and nominators of different ratios applied in that study.Although past studies have found that (at least three) different failure processes exist reflecting the chronically, gradually and acutely failing firms, this finding needs confirmation in case of old manufacturing firms.Therefore, we set the first general hypothesis as follows.
Hypothesis 1. Different failure processes exist for old manufacturing firms.
We will incorporate two stages in the empirical analysis, namely, in the first stage, we will use the variables from the classical Laitinen (1991) model and, thereafter, we will apply the extended approach portrayed on Figure 1.

Contingencies of failure processes.
There is very little evidence whether failure processes vary through different types of manufacturing firms.From Laitinen (1991, p. 661), it can be only concluded that different industries are characterized by varying failure processes.Still, the sample in Laitinen (1991) was very small, thus, from some sectors, only one or a few observations were available.In the follow-up study by Laitinen et al. (2014) applying the similar variables as in Laitinen (1991) on a larger dataset from multiple European countries, clear evidence was found about the dissimilarity of failure processes through countries.In that study, the general and paired comparison of countries revealed abundance of inter-country differences (Laitinen et al., 2014, pp. 218-219).Namely, although the gradual failure process was dominating in all studied countries, the shares in specific countries could vary from 44%-82%, whereas in case of other processes, the dispersion was even higher (e.g.some detected processes were practically non-existent in specific countries) (Laitinen et al., 2014, p. 218).This will ground the following hypothesis concerning the representation of different failure processes through countries.

Hypothesis 2a. Old manufacturing firms in different countries are characterized by different failure processes.
Past studies detecting failure processes have mostly relied on either specific size categories while in some others, the sizes of included firms have been dispersed without providing further evidence about the interaction of size and processes.For instance, in Hambrick and D'Aveni (1988), only large firms were studied, whereas Laitinen (1991) and Laitinen et al. ( 2014) focused on small-and medium-sized firms.Although there can be small differences in the formulas of ratios, then when comparing the return on assets and equity ratios through the samples of failed firms in Hambrick and D'Aveni (1988) and Laitinen (1991), it can be concluded that they behave differently in the pre-bankruptcy years.Only in a study by Laitinen and Lukason (2014), it can be observed that some processes are specifically characteristic to firms in a certain size category.From a more theoretical perspective, the theories about the liability of size would suggest that larger firms are more likely to survive, but that might not apply when the industry has reached a mature stage (Mellahi and Wilkinson, 2004).Smaller firms have been noted to have more problems with raising capital and higher interest rates, but also being more reluctant or incapable of making necessary operational or strategic changes (Aldrich and Auster, 1986).Thus, there is some ground to assume that manufacturing firms in different size categories (especially when comparing very small and large firms) can follow different failure processes and therefore we set the following hypothesis.

Hypothesis 2b.
Old manufacturing firms in different size categories are characterized by different failure processes.
The interconnection of firm failure processes and firms' international commitment by means of exporting is a neglected topic in the available literature.There is a multitude of evidence about the process of withdrawal from foreign markets (e.g., Pauwels and Mathyssens, 1999;Turcan, 2003;Crick, 2004), but such studies do not rely emphasis on failure.Still, de-internationalization cannot be automatically considered a failure (Vissak et al., 2012;Vissak and Francioni, 2013) and it has been shown that nonlinear internationalization (i.e., multiple consecutive enters and full or partial retreats from foreign markets) can be a highly frequent phenomenon among firms (Vissak and Masso, 2015).Vissak and Masso (2015) showed on the whole population of Estonian firms that full deinternationalization is a rare process among all exporters, thus it can be deduced that deinternationalization ending with bankruptcy is even more rare among the population of all firms.Although without general consensus, there is more evidence that exporting firms have higher likelihood of survival (Wagner, 2013).Still, the multitude of evidence concerning the interconnection of survival opportunities and exporting does not answer the question whether exporting and non-exporting old firms arrive to bankruptcy similarly or not.The literature review by Wagner (2012) demonstrated that when exporting firms are generally more productive, then the relevant studies about profitability do not indicate homogenous results.As there is no univocal basis to hypothesize about the relationship between the involvement in exporting and failure processes, we simply make use of the fact of probable discrepancies in the values of financial ratios between exporting and non-exporting firms, therefore, setting the last hypothesis as follows.
Hypothesis 2c.Old exporting and non-exporting manufacturing firms are characterized by different failure processes.
The testing of Hypotheses 2a, 2b and 2c is evidently dependent of the fact that different failure processes emerge from the preceding analysis.Still, based on the review of past literature, this is obviously a reasonable expectation to be set.

Data of bankrupted firms.
In line with past studies, financial data are applied for extracting failure processes in this research.Financial data are obtained from Amadeus database incorporating the balance sheets and income statements of European firms.The following criteria will be applied for all firms to be included in the analysis.Firstly, all firms must have all relevant information listed in Amadeus.This means not only financial information needed to calculate variables, but also firm foundation and bankruptcy dates.Secondly, all included firms should be at least ten years old to be classified as "old".A similar age criterion has been applied, for instance, in Moulton et al. (1996).The age of a firm is determined by deducting exact foundation time (year, month, day) from bankruptcy time (year, month, day).Thirdly, each included firm should have six consecutive pre-bankruptcy annual reports available.This enables calculation of five consecutive variables, as for some variables information from two consecutive years has to be applied.Fourthly, the main industry of all firms included should be NACE Rev. 2 section C, i.e., the manufacturing industry.Fifthly, the last annual report of each firm should be in the interval of 0.75 to 1.25 years before bankruptcy declaration.This makes the dataset homogenous, as in case of using just the last available annual report, it is not known whether some annual reports are missing from the in-between period.It is impossible to obtain sufficient number of cases having data exactly one year before bankruptcy declaration.Still, it must be noted that the mean and median interval between bankruptcy declaration and the last available report is exactly one year in the applied sample.
The application of previous restrictions results in 1235 valid cases from 15 different European countries (see Appendix 1).Still, many countries are represented only with a few cases, but they will not be excluded from analysis.The five countries with the largest number of cases are: Italy (340 cases), France (290), Romania (198), Spain (166), and Hungary (129).In the analysis, another taxonomy will be also applied, which indicates whether the country belongs to a former socialist or capitalist countries' group.The regional dispersion can be followed in Appendixes 1 and 2. Based on median operating revenue of five last years of action and using the EU official size criteria, firms are grouped as follows: 797 micro firms, 346 small firms, 92 medium-and large-sized firms.Medium-and largesized firms are not distinguished herewith, as there are only a few large firms having median turnover over 50 million euros.In the sample, for 337 firms export revenue has been noted for last five years and out of them, 43% are non-exporters and 57% had been exporting during at least one year out of five.

Financial variables.
The financial variables are chosen for the analysis exactly based on the model brought out on Figure 1.For extracting the First Set of Processes (also referred to as Analysis Stage 1), six variables from Laitinen's (1991) study will be applied.The Second Set of Processes (also referred to as Analysis Stage 2) is extracted by using five variables in addition to those given in Laitinen's (1991) study as depicted on Figure 1.Two additional variables (traditional cash flow to total debt, i.e., CFTD, and cash and equivalents to current liabilities, i.e., CCL) reflect dynamic solidity and static liquidity as given in Laitinen's (1995) study.In this study, terms "debt" and "liabilities" are applied synonymously, e.g., "total debt" includes all short-term (i.e., current) and longterm liabilities no matter of their type.Two variables (DeltaTD and DeltaS) reflect the development of total debt and total sales as given in Moulton et al. (1996) study and one variable (DeltaOC) reflects the development of operating costs as given in Ooghe and de Prijcker's (2008) study.In this study, terms "sales" and "operating revenue" are used as synonyms reflecting all revenue obtained from operating activities.The operating costs' variable does not include amortization and depreciation, as the latter can be subject to different accounting practices and therefore not portray the incurred costs correctly.The formulas of all variables are shown in Table 1.Some variables have been presented in a form of a ratio and others in a %, the latter to make it easier to follow results.All variables have been calculated for five consecutive pre-bankruptcy years, whereas the number behind the variable in Tables 3 and 4 denotes the respective pre-bankruptcy year (e.g., 1 denotes one year before the bankruptcy was declared).The specific processes detected in the Analysis Stage 1 will be denoted as 1.n (e.g., first process among the First Set of Processes is denoted 1.1) and among the Second Set of Processes as 2.m.In Laitinen's (1991) study, four periods (t -1, t -2, t -4, t -6) were applied, but herewith the aim is to portray the exact step-by-step financial development, thus five consecutive years have been chosen.The results indicate two different failure processes (see Table 3 What concerns exporting and non-exporting firms, then in both groups Process 1.1 accounts for about two thirds of cases, thus, there is clearly no contingency between exporting behavior and detected failure processes (see Appendix 4).All aforementioned results of Chi-Square Tests can be followed in Table 5.

Conclusion
This paper aimed to detect the presence of different failure processes for bankrupted manufacturing firms.
For that purpose, factor and cluster analyses were applied on a dataset of 1235 bankrupted companies from 15 different European countries, although only five countries had a representation of over 100 cases.
The focus was on five consecutive pre-bankruptcy years.Different financial variables (financial ratios and changes in balance sheet and profit statement accounts) were applied.The study incorporated two different study designs: namely, an initial model incorporating six variables from Laitinen's (1991) study and an extended model with five additional variables.
Both study designs resulted in two different failure processes.The processes extracted in case of both study designs have elements of acute and gradual failure (see D'Aveni, 1989; Laitinen, 1991), but unlike in past studies on the same topic (e.g.Laitinen, 1991;Laitinen et al., 2014), the chronic failure process does not emerge among the extracted processes.The two extracted processes can be very differently associated with firms' country of origin: namely, in some countries one of the extracted processes is clearly dominating and in others, their frequencies can be very similar.There is no strong association between firm size and the process it follows.Also, there is no difference in the representation of processes for exporting and non-exporting firms.
There are several limitations in this study that could be taken into account in the future research.Namely, the dataset included a smaller number of medium-sized firms than in two size categories for smaller firms.Also, there are only a few large-sized firms in the analysis, which have therefore been consolidated with the group of medium-sized firms.
As past studies (e.g., Hambrick and D'Aveni, 1988) have noted the failure process for large corporation to span over a lengthy time, it could be interesting to pay specific attention to large-sized firms.Also, currently the internationalization context of firms is captured by a single variable, namely whether a firm has been engaged in exporting during the last five years of existence.As numerous different internationalization patterns exist (e.g., Vissak and Masso, 2015), such a diversity could be taken into account in further analysis.Still, as in the current analysis, there is a relatively small number of exporting firms (namely 192; see Appendix 4), the dataset is too small to take such internationalization differences into account.The study aimed to extract the main processes existing in the data, but as the main processes could be disaggregated into sub-processes, future studies can be focused on them as well.As different process detection methods can have their specific peculiarities, future studies can be extended by using a larger set of extraction methods: for instance, novel machine learning techniques.Finally, although this study encompassed more than 1000 applicable observations, the further analysis would benefit from a larger sample, in which, for instance, the countries currently represented with only a few cases would have a proportional amount of observations.
Several implications for practice can be drawn from this study.The median values of financial ratios indicate that the collapse of old manufacturing firms could be forecastable with bankruptcy prediction models.Namely, for all failure processes detected in this study, it can be observed that for instance profitability, capital structure and liquidity ratios obtain low values for the last years before collapse.Those dimensions have been common predictors of collapse (see, e.g., Dimitras et al. (1996) overview of prediction studies), thus, bankruptcy prediction models could easily capture such failure signals.Such a downturn could also be observable for managers, who can thus implement different turnaround strategies for vitalizing firms.As the majority of firms are following processes in case of which failure signals are observable years ahead of failure, policy measures could be implemented to address the low values of financial ratios of failing firms.For instance, many bankrupted firms witness very low total equity to total assets ratios, which could lead to more strict capital requirements.

Fig. 1 .
Fig. 1.Classical and extended framework for modelling failure processes

3. Process extraction method. The
(Costello and Osborne, 2005)udy are detected with a twostage method, namely the consecutive application of factor and cluster analysis.This "factor-cluster analysis" method has been applied in multiple areas, also outside the field of social sciences.It has proven to be very efficient for the extraction of failure processes (see, e.g.,Laitinen et al., 2014).In this study, unweighted least squares extraction with Varimax rotation is applied and the number of factors is chosen by using the Eigen value exceeding one rule.The Eigen value exceeding one rule is the most classical approach, although like other methods for the choice of the number of factors, it has limitations as well(Costello and Osborne, 2005).
3. Results and discussion3.1.Analysis Stage 1 -the First Set of Processes.In Analysis Stage 1, the variables from Laitinen's (1991) study have been applied to extract failure

Table 2 .
Characteristics of factor and cluster solutions in Analysis Stages 1 and 2

Table 3 .
Median values of variables for First Set of Processes Process 2.1 firms have also higher CCL figures.Still, the difference in the median values of CACL is not as large as in case of comparing Processes 1.1 and 1.2.Also, the CACL t -1 median value is lower for Process 2.1 than for 2.2.In case of both processes, the dynamic solidity reflected by total debt coverage with traditional cash flow (CFTD) is very poor throughout the five years and turns negative during one or two last years of result is similarly to Analysis Stage 1 checked up to k = 20, but none of the cluster solutions obtains higher pseudo-F value, thus, the k = 2 solution in Analysis Stage 2 could also hold the maximum pseudo-F value over all possible values of k.

Table 4 .
Median values of variables for Second Set of Processes

Table 5 .
Contingencies of processes with firms' country of origin, size and export behavior

Table 6 .
Contingency of First and Second Set of Processes

Table 7 .
Results of hypotheses testingThe findings of this study provide some important new evidence to failure literature.They have been summarized in Table7.Namely, in case of both study designs, two different failure processes emerged.These different processes are mostly discriminated by the speed of failure, Note: the two sets of failure processes referred to are: a) the First set is based only on variables from Laitinen's (1991) model, b) the Second set is based on refined Laitinen's (1991) model by supplementing additional five variables.For Hypotheses 2a, 2b and 2c the Chi-Square Test p-level 0.01 has been applied to accept or reject each of the specific hypotheses.3.4.Summary of Analysis Stages.