Geographical factors and business failure: An empirical study from the Madrid metropolitan area

Economic Modelling xxx (2018) 1–9

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.journals.elsevier.com/economic-modelling

Geographical factors and business failure: An empirical study from the Madrid metropolitan area
pez-Hernandez b, Mariluz Mat
e-S
anchez-Val a, *, Fernando Lo c Christian Camilo Rodriguez Fuentes a b c

Department of Financial Economics and Accounting at the Technical University of Cartagena, Post-mail: Calle Real 3, 30201, Cartagena, Murcia, Spain Department of Quatitative Methods and Computing at the Technical University of Cartagena, Post mail: Calle Real 3, 30201, Cartagena, Murcia, Spain Technical University of Cartagena, Post-mail: Calle Real 3, 30201, Cartagena, Murcia, Spain

A R T I C L E I N F O

A B S T R A C T

JEL Code: G30 M21 R12

Geography has been considered a decisive factor in different fields of business-related research. This paper provides some evidence concerning the role of geography on business failure in urban environments. The paper use spatial econometric methodology to evaluate the impact of the geographical location of external economic agents on the probability of business failure. In addition, it is shown that probabilities of business failure for geographically close firms are correlated. A firm-level empirical application based on 3125 industrial small and medium firms (SMEs) located on the Madrid metropolitan area (Spain) confirms that the geographical proximity between firms, external economic agents and transport facilities has a determinant impact on business failure among these companies. This study contributes to gaining a greater understanding of the factors that determine SMEs business failure, highlighting the importance of geographical factors.

Keywords: Business failure Spatial probit Geographical proximity External economic agents Transport facilities

1. Introduction Business failure has been paid a good deal of attention in the scholarly literature (Van Gelder et al., 2007). These studies have mainly focused on internal financial variables, and have aimed to produce predictable models of business failure. Owing to the unavailability of financial information, small-sized enterprises are a relatively neglected subject as far as business failure is concerned. The limited number of studies that focus on Small and Medium Enterprises (SMEs) find that some financial features may play a significant role in predicting financial business failure (some studies in Altman et al., 2010; Sohn and Kim, 2013; Andreeva et al., 2016 or Calabrese et al., 2017). However, failure is also caused by external factors over which entrepreneurs have little or no control (Everett and Watson, 1998). Therefore, companies fail not only as a result of the decisions adopted by their executives, but also of unavoidable environmental factors, which can vary substantially from place to place (Raspe and van Oort, 2011). In this context, the location of businesses and their proximity to one another and to external agents has scarcely been considered. Some results can be found in Fernandes and Artes (2016), who developed a credit scoring model which includes a variable

that represents spatial dependence between Brazilian SMEs. The spatial model proposed by Fernandes and Artes (2016) yields better results than the models which do not include this variable. Calabrese et al. (2017) analyse the effects of including spatial dependence between London small businesses into standard scoring models. They also find that spatial interdependence is a significant variable. In addition, the inclusion of this variable improves the ability to predict business defaults. The present study builds on the above to analyse the effects on business failure of interdependence between geographically close companies, and of the geographical proximity to external agents, by imposing additional research questions to the current literature. In particular, we apply a spatial econometric analysis to a specific metropolitan area in Spain in order to answer the following questions: are there significant spatial co-localised patterns in the local distribution of business failure in the case under consideration? Does geographical distance between companies and both external agents (such as suppliers and providers) and transportation facilities have an influence on the probability of business failure? The answers to these questions may be relevant for researches and policy makers in order to understand the impact of accessibility and geographical proximity on the business failure. External information

* Corresponding author. E-mail addresses: [email protected] (M. Mat
e-S
anchez-Val), [email protected] (F. L
opez-Hernandez), [email protected] (C.C. Rodriguez Fuentes). https://doi.org/10.1016/j.econmod.2018.05.022 Received 7 September 2017; Received in revised form 16 April 2018; Accepted 24 May 2018 Available online xxxx 0264-9993/© 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/bync-nd/4.0/).

Please cite this article in press as: Mat
e-S
anchez-Val, M., et al., Geographical factors and business failure: An empirical study from the Madrid metropolitan area, Economic Modelling (2018), https://doi.org/10.1016/j.econmod.2018.05.022

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

flows from external agents provides firms with a privileged awareness of market changes, and allow them to adopt the necessary measures to adapt to changing conditions. Therefore, geographical proximity between firms and external economic agents in their local environments could be considered to improve the firms’ chances of survival. This study is based on detailed empirical data collected among industrial small and medium firms (SME) located in the Madrid metropolitan area (Spain). The territory of Madrid is an interesting scenario to develop our study given the prominent place held by the region in terms of business failure. Madrid accounts for approximately 10% of all failed companies in Spain in the period 2011–2015 (Official Spanish Statistical Institute, www.ine.es). Our study geo-locate different transportation facilities and economic agents in the Madrid area whose geographical proximity to SMEs could have a positive impact on their probabilities of survival. We have applied this information in an exploratory analysis based on Join-Count tests, with the goal of contrasting spatial co-localized patterns in SMEs business survival. Afterwards, relying on the spatial conditional probit model, we estimate the role played by the spatial dependence between SMEs and the geographical proximity to external economic agents and the probability of failure. Therefore, in contrast with previous studies, we consider other geographical variables apart from spatial interdependencies between SMEs in order to examine the relevance of business environment on the prediction of business failure. The results indicate that geography is a determinant factor on business failure among SMEs in the analysed sample. In this regard, significant spatial co-localized patterns concerning the distribution of failed/healthy companies in Madrid can be identified. In addition, the examination of other factors, such as geographical interdependence and proximity between firms and transportation facilities and external economic agents also yielded significant results. Our analysis measures both the extension and the intensity of the areas of influence of external economic agents stressing the substantial role played by logistic centres, industrial estates and transport facilities in reducing business failure in the territory of Madrid. After this introduction, the rest of the paper is organized as follows. Section 2 reviews the literature that the present paper draws upon. Section 3 describes the data set, the variables and the methodology used. Section 4 shows our empirical results. Finally, Section 5 includes the discussion and our main conclusions.

the opposite conclusion. These studies state that the presence of a large number of firms operating in the same industrial sector may increase competition, reducing the probabilities of survival (Khelil, 2016). Da Silva and McComb (2012) conclude that a high density of firms operating in the same industry located within a one-mile radius results in a higher failure rate than that found among firms located further away. Folta et al. (2006) reach similar conclusions, and claim that the probability of failure increases in areas where a relatively large number of akin companies operate. However, in general terms, it is assumed that the advantages derived from inter-industrial specialization offset the limitations and therefore, that companies located in these environments have a higher probability of surviving (Weterings and Marsili, 2015). In addition, the presence of intra-industry specialization, often attracts a diversity of actors, including industrial agents and organizations, universities, industrial research laboratories, trade associations and other knowledge-generating organizations. Diversity encourages the production and assimilation of knowledge through cross-sector spillovers, which is a stimulus for growth at the regional level (Harrison et al., 1997). In this regard, intra-industry specialization will result in lower operational costs and will thus increase the firms’ chances of survival (Pe~ na, 2002). This theoretical framework also highlights the local labour market as a source of economic benefits. Industry specialization tends to provide a pool of specialized labour to which companies have easy access. 2.2. Knowledge spillover Knowledge spillover occurs when there is a flow of information between agents working in the same area. Knowledge is more likely to spill over between geographically closer firms. This proximity facilitates the formation and transmission of social capital, enhancing trust and the ability to share vital information (Karlsson et al., 2015). Managers working in the same environment normally have the opportunity to build face-to-face relationships, exchange ideas and learn from one another's’ experience. As a result, positive network externalities will ensue and companies will be able to learn from the failure and success of other firms sooner than they would if no direct contact between was possible (Maskell, 2001). Transportation facilities also foster knowledge spillovers between companies and external economic agents. Proximity to major highways, seaports, rail stations and airports strengthens the firms' interaction with their environments. Previous studies suggest that the proximity of transport infrastructures is a major factor in the choice of location of new firms (Chatman et al., 2016). Therefore, it is expected that companies near transport infrastructures have easier access to external knowledge, resulting in a lower probability of failure.

2. Business failure: spatial considerations and implications Geography has been considered a decisive factor in different fields of business performance. Regarding previous literature, two different theoretical perspectives may be identified, which examine the impact of geography on business results: these theoretical perspectives deal respectively in transportation costs and external economies. The former is underpinned on the hypothesis that companies close to other economic agents have easy access to external resources, such as suppliers and financial providers, and therefore, minimize transportation costs (Weber, 1909). The external economies perspective states that business location triggers different forms of interaction between firms and between firms and their environment (Marshall, 1920). In particular, industry specialization and knowledge spillovers can potentially strengthen interaction between companies and external agents.

3. Methodology and database In order to analyse the effect of geography on business failure, two tests have been undertaken. Initially, the Join-Count tests are used to identify spatial pattern structures in binary variables (failure versus nofailure companies). Second, a spatial probit model is estimated in order to determine which factors play a greater role in business failure, including the instances of spatial interdependence identified with the Joint-Count tests. The following subsections provide a brief description of the methodologies applied.

2.1. Industry specialization 3.1. Spatial autocorrelation tests for qualitative data: the Joint-Count tests The benefits to be derived from industrial specialization are related to the exchange of knowledge between companies working in similar or different sectors. Empirical studies suggest that both inter-industry and intra-industry specialization exerts a positive effect on business performance. This is due to the proximity between firms and between firms and specialized suppliers, which lowers transaction costs (Fujita and Thisse, 2002). Despite this general understanding, the literature on business survival is not conclusive about the effects of industrial specialization. Although in general terms results are positive, some studies have reached

The Joint-Count tests compare spatial co-localized patterns in dichotomy variables (Cliff and Ord, 1981 p.36). In this paper two values of this variable must be distinguished: failed business (F) versus healthy business (H).1 Based on these categories, three different connections are

1 Traditionally the letters B (Black) and W (White) are used to denote the two possible categories.

2

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9



defined: FF, FH and HH. FF and HH represent close companies which fall in the same category while FH represents the number of close (connected) companies which fall in different categories. In order to connect companies, we selected a binary weight matrix W to establish a connectivity criterion. In particular, the elements of W, wij ði; j ¼ 1; …; nÞ have a value of 1 if companies i and j are neighbours, and 0 if otherwise. By definition, the elements in the main diagonal are equal to 0. Based on geographical distance, we consider that each company i is connected with its k nearest neighbors. The row-standardization of the k-nearest neighbour W matrix adopted here was based on several definitions proposed in the existing literature. As is well-known, using this approach, the W matrix contains elements of either 1/k or 0. This is an exogenous criterion, which prevents endogeneity. From this connectivity criterion the Join-Count statistics (JFF, JFH, JHH) are defined as follows in equations (5)–(6): JFF ¼

n X n 1X wij FFij 2 i¼1 j¼1

(1)

JFH ¼

n X n 1X wij FHij 2 i¼1 j¼1

(2)

n X n 1X wij  ðJFF þ JFH Þ 2 i¼1 j¼1

(4)

y* ¼ ρWy* þ Xβ þ ε; ε  Nð0; In Þ

(5)

1 if 0 if

where y is the observed value of the limited-dependent variable, y * is the unobserved latent dependent variable, and X is a matrix of explanatory variables; W is an n  n spatial weight matrix defining the neighborhood structure; ρ is the spatial autoregressive parameter, if ρ ¼ 0 the spatial probit model collapses to the standard binary probit model, otherwise, if ρ6¼0, the (nx1) vector Wy* consisting of an average of the k neighbouring companies (average of failed companies), and it creates a mechanism for modelling interdependence in business failure. Finally, ε is the error term. Several alternatives have been proposed to estimate the parameters of a spatial probit model: The Generalised Method of Moment (Pinkse and Slade, 1998); Maximum Likelihood using the Expectation-Maximization algorithm (McMillen, 1992); and, Bayesian Gibbs sample approach (LeSage, 2000). In this paper we use the Maximum Likelihood procedure recently developed by Martinetti and Geniaux (2017). This proposal is efficient and reliable since conditional estimators outperform the respective full-likelihood estimators. The estimates were calculated with the R package ProbitSpatial.

where wij represent the elements of the weight matrix; FFij ¼ 1 if the units i and j both belong to the category “F”, and FFij ¼ 0 if otherwise; FHij ¼ 1 if the units i and j belong to different categories and FHij ¼ 0 if otherwise. n is the total number of companies under analysis. The statistic JHH can be computed from the two previous statistics (5) and (6) as JHH ¼

y0 y<0

y* ¼

3.3. Database and variables 3.3.1. Region under study Only companies located in the municipality of Madrid (Spain) were taken into consideration. The territory of Madrid is an interesting scenario to develop our study owing to the prominent role played by the region in nation-wide business-default statistics. Madrid also offers a good case study for a regional analysis because it possesses a complex and dynamic infrastructure of public and private transport, logistic facilities, universities and R&D centres, which encourages industrial development and attract new SMEs. In addition, the availability of detailed micro-data on industrial firms and of the geographical coordinates of different economic agents offers excellent conditions for undertaking this study. Furthermore, the SME is an ideal unit of analysis due to its significant weight in the current economic system; SMEs account for over 99.8% of the productive system in Spain (Directorio Central de Empresas, 2017; www.ine.es). In addition, given the particular characteristics of these companies, they are highly dependent from environmental conditions (Carreira and Silva, 2010).

(3)

Details about asymptotic distributions in Join-Count tests can be found in Cliff and Ord (1981). Spatial co-localised patterns which result from the application of Join-Count tests can be positive or negative. A positive co-localised pattern indicates a spatial structure in which there is a high probability of finding companies that belong to the category F or H surrounded by companies which fall in the same category (e.g. Fig. 1A), while a negative result reveals the spatial interconnection of companies which fall in different categories (e.g. Fig. 1B). When the spatial distribution is random, no spatial co-localized pattern can be attested (e.g. Fig. 1. C). The R package spdep was used to get the Joint-Count statistics.

3.3.2. Database The financial and accounting data applied in our empirical application as well as the geographical coordinates of each company were obtained from the SABI (Iberian Balance Analysis System) database, which provides a wide range of information about business characteristics of Spanish firms (i.e. size; age or internationalization). We selected industrial firms based on the criterion established in the National Classification of Economics Activities (Nomenclature of Economic Activities, 2007). The sample is made up of 3125 SMEs over the period 2013–2015. In addition, localization (latitude and longitude coordinates) of external economic agents which could have an impact on the chances of survival of these business probability of survival (such as industrial estates; R&D and universities or logistic centres) is available as Open Data at the Madrid City Council website (http://datos.madrid.es/). The road coordinates were obtained from Google Maps. Information concerning logistic centres was collected from the Railway Infrastructure Administrator (http://www.adif.es/), the location of thirteen freight terminals have also been taken into consideration in our analysis.

3.2. Spatial probit model Models with limited dependent variables have received significant attention in connection with business failure (Altman et al., 2010; Calabrese et al., 2017) but in the presence of spatial autocorrelation in the dependent variable, estimates become inconsistent and inefficient (McMillen, 1992). In this case, it is necessary to apply the spatial-probit model with interdependence in the latent-variable (LeSage and Pace, 2009), i.e. in the unobserved argument to the probit-modelled probability of a binary outcome. The spatial probit model is represented as follows,

3.3.3. Variables 3.3.3.1. Dependent variable: business failure. Following previous literature, we base our definition of business failure on financial distress or

Fig. 1. Spatial co-localised patterns. 3

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

research centres and universities (DMinR&D). Proximity to these centres facilitates knowledge flows, reducing the probability of failure, and therefore we expect failure rates in the proximity to R&D to be lower (Audretsch and Feldman, 1996). Close interaction with R&D centres does not only contribute to regional development, but also benefits those firms which have a better profile in terms of innovation (Romijn and Albu, 2002). The location of Madrid's major universities (i.e. “Aut
onoma”, “Complutense”, “Polit
ecnica”, “Alcala de Henares”, “Carlos III” and “UNED”) and other research institutes of R&D have also been taken into account. Finally, we consider the distance between SMEs and logistic centres (DMinLC), which are defined as companies that chiefly provide transportation, distribution or storage services. 3.3.3.2.2. Transport facilities. In this category, we consider the distance between firms to 45 junctions between three major roads “M-30”, “M-40” and “M-45” and other national roads (DMinHWJ). Little is known about the potential effect of this variable on the probability of failure. However, some studies suggest that the proximity of transport infrastructures has a positive effect in the chances of survival of new firms (Chatman et al., 2016). Therefore, we expect that companies that are closer to road junctions have easier access to external agents and therefore, lower probabilities of failure. Finally, we also include a representative variable of the economic characteristics of the area where each of the subject companies is located. In particular, we take into consideration income per capita in the censual area within the municipality of Madrid in which each subject company is located. 3.3.3.2.3. Control variables. Three ratios were chosen to represent the financial characteristics of firms. These ratios are related to the liquidity, profitability and financial leverage of companies. The Liquidity ratio measures how well a company pays off its short-term liabilities, in cash or current assets. Among different possibilities, we have chosen to measure liquidity on the basis of the current ratio, that is, the ratio between current assets and current liabilities. Profitability ratios measure the organization's overall financial performance by evaluating its ability to generate revenues. We measure Profitability as the ratio between net operating income and total assets. Finally, the Debt ratio measures the extent to which the company is financed by its debt holders (as opposed to its owners). This ratio is computed as total liabilities to total assets. In addition, we include representative variables for those among the firms' characteristics which have a greatest bearing on the firms' survival. The size of the company is a relevant factor in this regard. Previous studies suggest that there is a positive relationship between the size of firms and their chances of survival (Back, 2005). Larger companies enjoy such advantages as economies of scale and greater market presence, which result in enhanced probabilities of survival. Following the criteria established by the European Commission on 6 May 2003, we consider three categories in this regard: micro-companies (<2 million of total assets), small firms (2–10 million of total assets) and medium size companies (10–43 million of total assets). The technological intensity of firms is also taken into consideration in this study. According to previous studies, companies engage in high-tech are less likely to fail (Chen and Williams, 1999). This variable is based on the criteria set forth in NACE (Nomenclature of Economic Activities) codes. In this way, companies are defined as Low Technological intensity firms (LT), Low Medium Technological Intensity firms (LMT), High-Medium Technological Intensity firms (HMT) and High Technological Intensity firms (HT). We also consider the firms' internationalization. Theoretically, companies with greater access to international markets are less likely to fail (Esteve et al., 2004). These companies engage in riskier operations and, therefore, tend to adopt relatively conservative financial policies. In addition, the international character of these companies allows them a greater degree of market diversification, reducing the probability of failure. The SABI database includes a specific entry with information about the international projection of firms. Finally, the Age of the company has also been included in our analysis. Following Berger and Udell (1998) we define variable Age as the logarithm of the number of years that have passed since the constitution of the company. This variable is related to the nature of the

economic failure. Previous studies suggest that this the most accurate definition of business failure, because some financially distressed firms never file for bankruptcy while we financially healthy companies may file for bankruptcy owing to strategic reasons unrelated to financial distress. As such, sticking to a legal definition of failure could provide an unrealistic sample of failed and non-failed companies (Balcaen and Ooghe, 2006). From this perspective, we define financial distressed firms as those companies whose accountancy records attest to three straight years of negative shareholders' equity or two straight years of negative shareholders’ equity and one additional year for which no information is available (Correa et al., 2003; Rubio-Misas, 2008, Tascon and Casta~ no, 2012). 3.3.3.2. Explanatory variables 3.3.3.2.1. Environmental variables. In order to design the density and environmental variables, we follow Da Silva and Mc Comb (2012), who compute the geographical variables considering different buffers or concentric rings ri of different radii around each economic agent. Later, they select which radii provide the best fit for the analysed model. From this perspective, space is considered a continuous dimension, and the measure of interdependence between different economic agents is considered as being independent from any arbitrary jurisdictional division. The iterative procedure employed in order to determine the different radii ri will be described in subsection 4.2. Density variables deal with the number of firms inside of each firm's buffer. There are two such variables, and they evaluate the intensity of inter and intra industry specialization. Sectorial Density (DensSec) establishes the number of firms operating in the same sector inside of the buffer; Density (Dens) evaluates the diversification of the environment by calculating the density of companies operating in different industrial sectors within the buffer. According to previous studies, these variables should reduce the probability of business failure. As the density of the environment increases, intra and inter specialization effects will be more intense, having an increasingly significant effect on the chances of business survival. In general, these effects are positive, for increased density results in economies of scale and improved informational flows, decreasing the chances of business failure (Weterings and Marsili, 2015). However, in order to take into account the potential trade-off between competition and clustering effects, as discussed in section 2, sectorial density is included in squared terms.2 Additional environmental factors based on distances are defined as dichotomy variables. These variables have a value of 1 if the company is inside the buffer of the external economic agent under evaluation (R&D centre, large company, etc.) and a value of 0 if otherwise. We prefer using dichotomy variables to reduce multicolinearity in the estimation of probit models because in an urban environment all distance-based variables present moderate to high correlations. Moreover, this procedure allows considering non-linearity in the variables associated with geographical proximity between companies and the different external economic agents. In this way, we build a variable to evaluate the distance from the company to larger companies (DMinLC), which are here regarded to be as leader in their sectors. They suffer less from asymmetric information, and react to market changes faster than other companies. Therefore, the proximity to larger companies will have a positive spillover effect on their neighbours, which will receive more information and can imitate management practices implemented by larger companies (Pirinsky and Wang, 2010). The distance to industrial estates (DMinIP) is also taken into consideration. This variable evaluates the industrial inter-territorial relationships; the proximity of other companies leads to benefits associated with the diffusion of knowledge and easier access to different modes of transport (Mota and Castro, 2004). According to the census of industrial estates, there are twenty-two such facilities in the municipality of Madrid. We also take into consideration the distance to

2

Authors thank the anonymous referee for this comment. 4

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

firms' financial needs and their available resources. In general, younger firms face more constrained market conditions and find more financial difficulties than older firms, and the chances of failure increase accordingly (Rubio-Misas, 2008). Table 1 gives an overview of the list of

variables that have been taken into account in our model as well as descriptive information concerning the percentage of firms to which these variables apply. 4. Results

Table 1 List of dependent and independent variables included in the model. Variable

Dependent variable BF 1 if firm fails, 0 otherwise (sane firm). Independent variables

Income Current ratio Debt Equity ratio Liquidity ratio Age

a

Micro Small

Medium LTa LMT HMT HT Local

a

Imp Exp ImpExp

Dens DensSub DMinLC

DMinIP

DMinR&D

DMinHWJ

DMinLS

4.1. Spatial co-localised patterns in business failure

Description

The Join-Count tests were used to assess the spatial co-localised patterns for the variable business failure in our sample. This involves taking into account the connectivity criterion defined on the basis of different k-values (k ¼ 4,6,8,10,12). Given a significance level of 5%, the null hypothesis (H0: random co-localized pattern) and the alternative hypothesis (HA: existence of a spatial co-localized pattern). The JoinCount results (Table 2) indicate the existence of a positive spatial colocalised pattern for the variable business failure in our sample. This involves taking into account the connectivity criterion defined on the basis of different k-values. The Join-Count results (Table 2) indicate the existence of a positive spatial co-localized pattern and significance among failed companies for all k values (first columns in Table 2). In all these cases, the number of pairs of connected failed companies (JFF) is higher than the expected values (J
FF). Therefore, these results reveal that failed companies are highly likely to be surrounded by other failed companies. Moreover, the spatial co-localized pattern of companies that belong to different category yields significant results (Table 2). Therefore, the spatial structure healthy-failed can also be found in our sample. In the case of healthy companies (last columns in Table 2), the number of healthy companies located in the vicinity of other healthy companies founded is higher than expected. These results indicate that the failure of a company increases the probability that neighbouring companies will also fail and, therefore, this information must be included in the probit model.

15.74%

Control variables

Mean (std)

Log income (pc) of population in the census track where the firm is localized. Short Term Assets to Short Term Liabilities. Total Liabilities to Total Assets.

1.86 (0.51)

Current Assets to Current Liabilities

3.40 (9.84)

Age of company (years from foundation).

17.69 (12.54)

Firm
s characteristic factorsa

n (%)

1 if firm size is small (<2 millions Total Assets), 0 otherwise. 1 if firm size is small (2–10 millions Total Assets), 0 otherwise. 1 if firm size is medium (10–43 millions Total Assets), 0 otherwise. 1 if firm has Low intensity technologic, 0 otherwise. 1 if firm has Medium-Low intensity technologic, 0 otherwise. 1 if firm has Medium-High intensity technologic, 0 otherwise. 1 if firm has High intensity technologic, 0 otherwise. 1 if firm without internationalisation (not import and/or export), 0 otherwise. 1 if firm only import, 0 otherwise. 1 if firm only export, 0 otherwise. 1 if firm import and export, 0 otherwise.

2622 (83.9%) 353 (11.3%)

Firm
s geographic factors

Mean (std) or %

Number of total firms of all sectors within a radius r1. Number of firms of the same subsector (NACE2007, 2 digits) within a radius r2. Minimum distance to a large firm within a radius r3 ¼ 1. Minimum distance to a large firm outside a radius r3 ¼ 0. Minimum distance to industrial parks within a radius r4 ¼ 1. Minimum distance to industrial parks outside a radius r4 ¼ 0. Minimum distance to R&D institutes or universities within a radius r5 ¼ 1. Minimum distance to R&D institutes or universities outside a radius r5 ¼ 0. Network of road junctions “M-30”, “M-40” and “M45” with national roads. Minimum distance to network of road highway junction within a radius r6 ¼ 1. Minimum distance to network of road highway junction outside a radius r6 ¼ 0. Minimum distance to logistic services within a radius r7 ¼ 1. Minimum distance to logistic services outside a radius r7 ¼ 0.

17.47 (24.19)b 3.86 (6.31)b

0.11 (0.45) 1.13 (2.38)

150 (4.8%) 1798 (57.5%) 764 (24.25%) 410 (13.1%) 143 (4.9%)

4.2. Probit model of business failure

2518 (80.58%) 118 (3.8%) 344 (11.0%) 145 (4.6%)

In this subsection, we propose the estimation of probit and spatial probit models to explain business failure with reference to location. With this aim, we estimate the probability of business failure on a set of representative variables which evaluates the distance between firms and external agents which could the firms could interact. The baseline model for the expected probit of firm i is then given by,   P BF ¼ 1jXs ¼ β0 þ β1 Income PC þ β2 Liquidity þ β3 Profitability þ β4 Debt þβ5 Age þ β6 Small þ β7 Medium þ β8 LMT þ β9 HMT þ β10 HT þ β11 Imp

791 (25.3%)

β12 Exp þ β13 ImpExp þ β14 Dens þ β15 DensSub þ β16 DensSub2 þ β17 DMinLC þ β18 DMinIP þ β19 DMinR&D þ β20 DMinHWJþ

436 (13.9%)

β21 DMinLS þ ε (6) In order to avoid endogeneity in (7), financial ratios are instrumentalised by applying a two-year temporal lag for each of them. In addition, the impact of geographical factors on business failure depends on the considered cut-off points (see ri values in Table 1). In order to identify these cut-off points, the probit model (7) has been estimated by selecting ri values that maximized the likelihood. Da Silva and Mc Comb (2012) and Rosenthal and Strange (2003) follow a similar procedure, selecting the radii which provide the highest significance for the coefficients of the model. The ri values that maximize this function are: r1 ¼ 0.05; r2 ¼ 0.24; r3 ¼ 0.29; r4 ¼ 0.88; r5 ¼ 2.39; r6 ¼ 2.59; r7 ¼ 2.99. Fig. 2 illustrates the locations of firms in the metropolitan area of Madrid and the buffers around strategic centres. The radius ri defines the influence areas of each strategic centre.3

12.7%

24.4%

75.5%

ri (with i ¼ 1, …,7) represents the different cut-off points to define each variable (see subsection 4.2). a The reference category in the probit model is a micro firm with low technological intensity and without internationalisation. b Mean of number firms around each one in sample.

3 To evaluate the sensitivity of cut-points the model is re-estimate increasing (resp. decreasing) the radii a 10% defining ri’ ¼ 1.1  ri (resp. ri’’ ¼ 0.9  ri).

5

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

Table 2 Join-Count tests for business failure (BF). k-nearest neighbours

4 6 8 10 12

Join-Count FF test (Failed-Failed) a

JFF

J
FF

185** 270** 352** 438** 517**

154.66 231.98 309.31 386 463

z-value

Join-Count FH test (Failed-Healthy) b

2.09 3.00 2.90 3.10 2.51

JFH

J
FHa

z-value

1588** 2371** 3180** 3998** 4824**

1658.69 2488.03 3317.38 4146.72 4976.07

2.98 3.93 3.92 3.74 3.41

Join-Count HH test (Healthy-Healthy) b

JHH

J
HHa

z-valueb

4477** 6733.5** 8968** 11189** 13408**

4436.66 6654.98 8873.31 11091.64 13309.97

2.09 3.20 3.19 2.85 2.51

*** significant at 1%, ** significant at 5% and * significant at 10%. a J
FF is the expected value of JFF; E [JFH] ¼ J
FH; E [JHH] ¼ J
HH. b p-value of the Z statistic.

the local level (cut-off points less than 1 Km); for some variables (e.g. Dens or DensSub), the cut-off points can be placed a few hundred meters’ distance, which suggests that business density is a significant local factor on business failure for these companies. Second, the cut-off points for

Some comments are necessary in the light of these results. Two groups of geographic factors can be considered based on the ri values obtained. First, the variables Dens; DensSub; DMinLC and DMinIP are shown to have a high potential to have an effect on the probability of business failure at

Fig. 2. Location of firms and buffers around strategic points. 6

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

distances DMinR&D; DMinHWJ and DMinSL range between 2 km and 3 km. These geographical factors do not have local incidence, which suggests these to be less critical issues with regard to the probability of business failure. Table 3 shows the spatial probit maximum likelihood estimations. Regarding the first model, the results are as expected concerning the impact of firm characteristics on the probability of failure. First, we find that the Income per capita has a significant positive effect, decreasing the probability of business failure. This result highlights the need to take the economic conditions of the area where a firm is going to be situated into consideration (Altman et al., 2010). The financial ratios also are useful indicators. Although all of them bear significantly on the firms’ chances of survival, profitability may be regarded as the most significant (it is related to the highest coefficient). This agrees with previous studies, which highlight the relevance of profitability ratios in predicting business failure (Khelil, 2016). Regarding firms’ characteristics, we find that there is a negative relationship between the size of firms and the probability of business failure. This is consistent with previous studies, which stress the benefits of economies of scale and the greater market presence that larger companies enjoy. The probability of business failure recedes as the firms grow larger (Rubio-Misas, 2008). Also, technological intensity also contributes to reduce the probability of business failure, which also match previous studies (Chen and Williams, 1999). Likewise, international projection (Imp, Exp and ImpExp) helps to reduce the probability of business failure. This result also matches previous studies, which stress that internationalized companies tend to engage in high-risk operations, while implementing relatively conservative financial and economic policies, which reduces the probability of failure (Esteve et al., 2004). Finally, younger companies face more financial difficulties than older ones, and they are accordingly more likely to fail (Berger and Udell, 1998). Representative variables of the business environment also play a significant role in our model. Density is evaluated through two variables. Dens represents the number of companies located in a firm's environment. This variable is negative and non-significant, and it provides evidence about the positive effects of concentration in reducing business failure (Weterings and Marsili, 2015). A similar result is obtained when the variable that reflects density within the same subsector (DensSub) is examined. In addition, our results corroborate a trade-off between clustering versus competition when density variables are examined. In this sense, a significant and positive value for sectorial density in square terms indicates that density plays a positive effect in decreasing the probability of failure, but this effect becomes negative when the density value is too high (Khelil, 2016). The distance to strategic centres, such as industrial estates, R&D centres and universities are also significant and negative. Therefore, the model suggests that proximity to these points increases a firm's chances of survival for these companies. Finally, proximity to transport facilities plays a relevant role in business survival, decreasing the probability of business failure.4

Table 3 Probit and Spatial probit model estimationsa. Variable

Probit

Income per capita Liquidity ratio Profitability ratio Indebtedness ratio Age Microa Small Medium LTa LMT HMT HT Locala Imp Exp ImpExp Dens DensSub DensSub^2 DMinLC DMinIP DMinR&D DMinHWJ DMinLS Spatial effect (ρ) Log Likelihood LR probit vs spatial probit

Spatial probit

Coef.

p-value

Coef.

p-value

0.1833*** 0.0727*** 0.3234*** 0.0746*** 0.0021 – 0.4346*** 0.5073** – 0.1212* 0.2778** 0.5271*** – 0.3300 0.4105*** 0.3159** 0.0026** 0.0438*** 0.0007*** 0.0941 0.1886* 0.1706*** 0.1973*** 0.2291*** – 1152.04

(0.000) (0.000) (0.000) (0.000) (0.397) – (0.001) (0.028) – (0.083) (0.018) (0.001) – (0.132) (0.001) (0.061) (0.110) (0.001) (0.019) (0.283) (0.074) (0.013) (0.002) (0.002) –

0.1246*** 0.0756*** 0.3258*** 0.0710*** 0.0011** – 0.4492*** 0.4989** – 0.0990** 0.2012** 0.4950*** – 0.3500 0.3816*** 0.3209** 0.0020* 0.0360*** 0.0006** 0.0162 0.1316 0.1172*** 0.1269** 0.1754*** 0.2232*** 1149.17 5.74**

(0.000) (0.000) (0.000) (0.000) (0.052) – (0.000) (0.028) – (0.149) (0.039) (0.003) – (0.108) (0.002) (0.056) (0.172) (0.003) (0.045) (0.827) (0.123) (0.036) (0.018) (0.003) (0.007)

(*) significant at 10% (**) significant at 5% (***) significant at 1%. a The reference category in probit model is a micro firm with low technological intensity and without internationalisation. Table 4 Marginal effects. Spatial Probit model.

4.3. Spatial probit models of business failure

Income per capita

0.0333**

Liquidity Profitability Indebtedness Age Small Medium LMT HMT HT Imp Exp ImpExp Dens DensSub DensSub^2 DMinLC DMinIP DMinR&D DMinHWJ DMinLS

0.0195** 0.0668** 0.0146** 0.0003 0.1164** 0.1292** 0.0264** 0.0529** 0.1280** 0.0913** 0.0999** 0.0833** 0.0005 0.0094** 0.0001** 0.0054 0.0347* 0.0245** 0.0340** 0.0460**

** Significant at 5% * Significant at 10%.

The identification of spatial patterns in the dependent variable (see Table 2) could lead to spatial dependence in the residual of the probit model. If this were the case, previous estimations would be inconsistent (McMillen, 1992). To overcome this limitation, we extended the conditional probit model by introducing a spatial lag in the dependent variable

and testing the influence of spillovers. The spatial probit model with 6 nearest neighbourhoods (Wk¼6) is represented in the second column of Table 3. We have selected the weight matrix Wk¼6 because it maximizes the log-likelihood of the model to the alternative matrices.5 Thus, this is the weight matrix that best fits the model (Stakhovych and Bijmolt, 2009). In general, we observe that the results yielded by the extended spatial model closely match those yielded by the non–spatial one. Yet, now we find a significant and positive spatial interaction coefficient

4 In order to give robustness to our results, we also estimate this model considering a legal definition of business failure considering default companies those firms which have been declared bankrupt, put into receivership or dissolved either through voluntary or involuntary liquidation in any of the years of the period 2013–2015. For these companies we assign the value of one and zero in otherwise. Estimation results for the model (7) with this dependent variable were significant and with the expected signs for the geographical variables.

5

7

Similar results for k ¼ 4 and 8 available under request.

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

Table 5 Areas of influence and marginal effects of density and strategic centres.

Radius Marginal effect (%)

Dens

DensSub

DMinLC

DMinIP

DMinR&D

DMinHWJ

DMinLS

0.05 0.05

0.24 0.94

0.29 0.54

0.88 3.47

2.39 2.45

2.59 3.4

2.99 4.6

ρ ¼ 0.2232, which suggests that the probability of a firm failing strongly

Table 5). Moreover, the areas of influence of these economic agents are larger (ri ¼ 2.99 and 0.88 respectively) than those related to other types of economic agent. Therefore, companies do not have to be located particularly close to these centres to see their chances of survival increased. Our findings confirm that geographical proximity to other companies is beneficial, owing to better information flows and easier access to different modes of transportation (Mota and Castro, 2004). These results also highlight how beneficial it is for firms to be located in the proximity (less than 2.39 Km) of Research Centres and Universities. The benefits include best access to information. In addition, proximity favours synergies between R&D centres and companies, which fosters innovation activities and reduces their cost (Romijn and Albu, 2002). These effects tend to make companies less likely to fail. Finally, local density of firms also enhances the probability of firm survival. Higher local density decreases the probability of business failure, but only when other firms are located in close proximity to the subject companies. The density variable, however, has very low value (0.05%). This could be motivated by the inclusion of other variables which tend to offset the potential effect of density indicators. The location of companies and their relationship with geographical factors is an attractive research topic. This area of study is an interesting future avenue of research since many of the geographic factors concerning business survival remain underexplored. Our results indicate that companies located in the vicinity to certain external agents have lower probability of business failure. This positive effect could be explained by the reduction in activity costs as well as by the advantages linked to external information flows (Karlsson et al., 2015). Finally, we have to admit that the approach followed in this paper may have at least three fundamental shortcomings that must be noted. First, it is necessary to stress that our study was undertaken during a critical economic period. SMEs are particularly sensitive to economic cycles, and analyses carried out in different conditions should be put forward. Secondly, the method applied to select the areas of influence around each strategic centre (selection of ri) is based on our own information and assumptions, and the regression process may be overparameterized. Therefore, alternative procedures to determine areas of influence should be proposed. Our study is based on geographical distance between companies and external agents; alternative relationships of proximity should be examined in this context. In addition, we include the most popular factors in relation to business failure, but it needs to be taken into account, that the omission of relevant variables in the model may have a significant impact on results. Third, our results are significant for the sample under examination, but further studies, which consider other scenarios and geographical contexts need to be carried before our conclusions can be extrapolated. A final reflexion is necessary before end the paper.6 The methodology applied in this work does not show a causal relationship between firm localization and the probability of business failure. It is possible that solvent firms with a priori low bankruptcy probability choose the localization close to a strategic point and therefore the probability to found failure firms around this strategic point is low. Much more sophisticated methodologies would need to be brought to bear before we could be comfortable concluding about the causal effect of firm failures on neighbouring firm failures. Further work on the exact causal mechanisms behind these relationships would be required before policy implications for future business location could be drawn.

depends on the failure of neighbouring companies. 4.4. Marginal effects The coefficients from the spatial probit model are not directly interpretable. A further modification is needed for their marginal effects to be interpretable. LeSage and Pace (2009) show that these effects can be calculated by using the inverse matrix (I - ρW)1 in the following equation:   ∂E½yjxr  ¼ Φ ½I  ρW1 βr xr ∘ ½I  ρW1 βr ∂xr

(7)

where xr is the mean of the rth variable, Φ is a standard normal distribution, and ∘ is element-by-element multiplication. Table 4 shows the marginal effects for the coefficients of the spatial probit model. The examination of these results can offer a useful guide to avoid business failure in this urban environment. With this method, it is possible to assign values (in terms of %) to each of the elements considered on the probability of business failure. These results show that the most important factors for business survival are related to the characteristics of the firms, especially size and degree of internationalization. In addition, companies which operate in highly technological subsectors are nearly 10% less likely to fail. Financial features, notably profitability and the economic characteristics of the area within which the company is located also play a relevant role in determining the probability of business failure. Regarding environmental factors, our results suggest that proximity to logistic centres and industrial estates are the most relevant factors concerning a firm's chances of survival. In this sense, if the company is within the area of influence of a logistic centre, the probability of the firm failing will drop by 4.60%, and by 3.47% if the firm is within the area of influence of an industrial estate. Companies also benefit from being within the area of influence of road junctions, R&D centres, universities and large companies, which reduce the probability of their neighbours failing by approximately by a 2.45%. 5. Discussion and conclusions The overall aim of our study was to test the impact of geographical factors on business failure in an urban context, specifically in the Madrid metropolitan area. In order to achieve this purpose, we analyzed the relevance of geographical proximity to external economic agents and transport facilities, which could have a bearing on the firms’ chances of survival. In addition, we evaluated if the interaction of nearby peer companies also has an influence on their chances of survival. The data used includes a sample of industrial SMEs located in the municipality of Madrid. This data was subject to a join-count test in order to detect possible spatial co-localized spatial patterns among pairs of connected peer companies. This analysis yielded significant results concerning the spatial distribution of healthy and failed companies. Subsequently, we used a spatial probit model to analyze the effect of geographical proximity variables on business failure. Our results confirm spatial co-localized patterns in pairs of failed companies. Therefore, failed companies are likely to be surrounded by other failed companies. In addition, proximity to external economic agents reduces the probability of failure, especially concerning Logistic Centres (DMinLS) and Industrial estates (DMinIP) (see

6

8

We appreciate this subjection of an anonymous referee.

M. Mat
e-S
anchez-Val et al.

Economic Modelling xxx (2018) 1–9

Acknowledgements

Fujita, M., Thisse, J., 2002. Economics of Agglomeration: Cities, Industrial Location and Regional Growth. Cambridge University Press, Cambridge. Harrison, B., Kelley, M., Gant, J., 1997. Innovative firm behavior and local milieu: exploring the intersection of agglomeration, firm effects, and technological change. Econ. Geogr. 72 (3), 233–258. € € 2015. Regional characteristics and the survival of new Karlsson, C., Klaesson, J., Oner, O., firms. In: ERSA Conference Papers (No. Ersa15p6). European Regional Science Association. Khelil, N., 2016. The many faces of entrepreneurial failure: insights from an empirical taxonomy. J. Bus. Ventur. 31 (1), 72–94. LeSage, J.P., Pace, R.K., 2009. Introduction to Spatial Econometrics. CRC Press. LeSage, James P., 2000. Bayesian estimation of limited dependent variable spatial autoregressive model. Geogr. Anal. 32, 19–35. Marshall, A., 1920. Principles of Economics. Macmillan, London. Martinetti, D., Geniaux, G., 2017. Approximate likelihood estimation of spatial probit models. Reg. Sci. Urban Econ. 64, 30–45. Maskell, P., 2001. Towards a knowledge based theory of the geographical cluster. Ind. Corp. Change 10 (4), 921–943. McMillen, D., 1992. Probit with spatial autocorrelation. J. Reg. Sci. 32 (3), 335–348. Mota, J., Castro, L.M., 2004. A capabilities perspective on the evolution of firm boundaries: a comparative case example from the Portuguese moulds industry. J. Manag. Stud. 41, 295–316. NACE: Nomenclature of Economic Activities, 2007. http://ec.europa.eu/eurostat/ documents/3859598/5902521/KS-RA-07-015-EN.PDF. Pe~ na, I., 2002. Intellectual capital and business start-up success. J. Intellect. Cap. 3 (2), 180–198. Pirinsky, C., Wang, Q., 2010. Geographic Location and Corporate Finance: a Review. Handbook of Emerging Issues in Corporate Governance. Pinkse, J., Slade, M., 1998. Contracting in space: An application of spatial statistics to discrete-choice models. J. Econom. 85 (1), 125–154. Raspe, O., van Oort, F., 2011. Growth of new firms and spatially bounded knowledge externalities. Ann. Reg. Sci. 46 (3), 495–518. Romijn, H., Albu, M., 2002. Innovation, networking and proximity: lessons from small high technology firms in the UK. Reg. Stud. 36, 81–86. Rosenthal, S.S., Strange, W.C., 2003. Geography, industrial organization, and agglomeration. Rev. Econ. Stat. 85 (2), 377–393. Rubio-Misas, M., 2008. An
alisis del fracaso empresarial en Andalucía. Especial referencia a la edad de la empresa. Cuadernos de Ciencias Econ
omicas y Empresariales 1 (54), 35–56. Sohn, S.Y., Kim, Y.S., 2013. Behavioral credit scoring model for technology-based firms that considers uncertain financial ratios obtained from relationship banking. Small Bus. Econ. 41 (4), 931–943. Stakhovych, S., Bijmolt, T.H.A., 2009. Specification of spatial models: a simulation study on weights matrices. Pap. Reg. Sci. 88 (2), 389–408. Tasc
on, M., Casta~ no, F., 2012. Variables y modelos para la identificaci
on y predicci
on del fracaso empresarial: revisi
on de la investigaci
on empírica reciente. Span. Account. Rev. 15 (1), 7–58. Van Gelder, J., De Vries, R., Frese, M., Goutbeek, J., 2007. Differences in psychological strategies of failed and operational business owners in the Fiji Islands. J. Small Bus. Manag. 45 (3), 388–400. Weber, A., 1909. Uber den standort der industrien. Mohr Verlag, Turingen, Germany. Weterings, A., Marsili, O., 2015. Spatial concentration of industries and new firm exits: does this relationship differ between exits by closure and by M&A? Reg. Stud. 49 (1), 44–58.

The authors gratefully acknowledge financial support from the Fundaci
on S
eneca number 19884/GERM/15. Appendix A. Supplementary data Supplementary data related to this article can be found at https://doi. org/10.1016/j.econmod.2018.05.022. References Altman, E., Sabato, G., Wilson, N., 2010. The value of non-financial information in small and medium-sized enterprise risk management. J. Credit Risk 6, 1–33. Andreeva, G., Calabrese, R., Osmetti, A.S., 2016. A comparative analysis of the UK and Italian small businesses using generalised extreme value models. Eur. J. Oper. Res. 249 (2), 506–516. Audretsch, D., Feldman, P., 1996. R&D spillovers and the geography of innovation and production. Am. Econ. Rev. 86 (4), 253–273. Back, P., 2005. Explaining financial difficulties based on previous payment behavior, management background variables and financial ratios. Eur. Account. Rev. 14 (4), 839–868. Balcaen, S., Ooghe, H., 2006. 35 years of studies on business failure: an overview of the classic statistical methodologies and their related problems. Br. Account. Rev. 38, 63–93. Berger, A., Udell, G., 1998. The economics of small business finance: the roles of private equity and debt markets in the financial growth cycle. J. Bank. Finance 22 (6–8), 613–673. Calabrese, R., Andreeva, G., Ansell, J., 2017. ‘Birds of a feather’ fail together: exploring the nature of dependency in SME defaults. Risk Anal.: Int. J. 1–14 https://doi.org/ 10.1111/risa.1286. Carreira, C., Silva, F., 2010. No deep pockets: some stylised empirical results on firms' financial constraints. J. Econ. Surv. 24 (4), 731–753. Chatman, D., Noland, R., Klein, N., 2016. Firm births, access to transit, and agglomeration in Portland, Oregon, and Dallas, Texas. Transport. Res. Rec.: J. Transport. Res. Board 2598, 1–10. Chen, J., Williams, M., 1999. The determinants of business failures in the US lowtechnology and high-technology industries. Appl. Econ. 31 (12), 1551–1563. Cliff, A.D., Ord, J.K., 1981. Spatial Processes: Models and Applications. Pion, London. Correa, A., Acosta, M., Gonz
alez, A.L., 2003. La insolvencia empresarial: un an
alisis empírico para la peque~ na y mediana empresa. Span. Account. Rev. 6 (12), 47–79. Da Silva, D., McComb, R., 2012. Geographic concentration and high tech firm survival. Reg. Sci. Urban Econ. 42 (4), 691–701. DIRCE: Directorio Central de Empresas, 2017. http://www.ine.es. Esteve, S., Sanchis, A., Sanchis, J., 2004. The determinants of survival Spanish manufacturing firms. Rev. Ind. Organ. 25, 251–273. Everett, J., Watson, J., 1998. Small business failure and external risk factors. Small Bus. Econ. 11 (4), 371–390. Fernandes, G.B., Artes, R., 2016. Spatial dependence in credit risk and its improvement in credit scoring. Eur. J. Oper. Res. 249 (2), 517–524. Folta, T., Cooper, A., Baik, Y., 2006. Geographic cluster size and firm performance. J. Bus. Ventur. 21 (2), 217–242.

9