Does location matter to firms? Some locations or geographic areas have increasingly been recognized as being more favorable for firm growth than others (Lee, 2018; 2009; Storey, 1994; Davidsson et al., 2002). Such is the case of firms located in an industrial district, or a cluster, like Silicon Valley in the United States or Zhongguancun Science Park in China (Tarfasa et al., 2016; Pyke et al., 1990). Firms are motivated to locate close to each other because of Marshallian agglomeration externalities (Folta et al., 2006). These spillovers, which constitute the essence of location advantage, are associated with labor market pooling, specialized input suppliers, and knowledge spillovers (Lee, 2018; Porter, 2000). Audretsch (1998) said that the marginal cost of transmitting knowledge, especially tacit knowledge, rises with distance. This suggests that being located in a cluster is more conducive to firm growth because, there, firms are granted access to location resources, which does not occur in other regions (Pe’er, Vertinsky 2016; Wonglimpiyarat, 2016; Cuervo-Cazurra et al., 2014; Audretsch & Dohse, 2007; Barringer et al., 2005; Acs & Armington, 2004; Audretsch & Dohse, 2004; Altenburg & Meyer-Stamer, 1999).

Although establishing a comprehensive theory about the effect of geographic clustering on firm growth is difficult, the empirical literature has offered diverse but conflicting results. Concerning the effect of geographic clustering, some studies have supported the conventional wisdom: the concentration of economic activity in a geographic space has a significant impact on firm growth (Harabi, 2007; Liedholm, 2002; Storey, 1994). Hoogstra and van Dijk (2004) concluded that ‘location matters’ but that the effect differs by type of economic activity. McPherson (1996), in an analysis of five southern African countries, provided some evidence of the existence of agglomeration externalities and found that urban-based firms grow faster than small and micro firms do in rural areas. Other studies, however, have stated that being located in a cluster is statistically insignificant (Glancey, 1998; Kolvereid, 1992). Lee (2018) found that location per se has no positive effect on firm growth in an empirical analysis of firms in nine industries across six countries, and Van Geenhuizen and Reyes-Gonzalez (2007) reported that clustered firms have no significant influence on innovation and speed of growth. To summarize, the effects on a firm’s growth of being surrounded by firms belonging to the same industry are controversial. With this in mind, we have posited three hypotheses regarding firm location in different ever-widening geographic areas:

H_1a – Being located in a cluster has a positive effect on firm growth rate.

Gibrat (1931) presented the first formal model that relates the dynamics of firm size to the structure of the industry, known in the literature as the “Law of proportional effect” (Sutton, 1997). Gibrat's Law states that the average growth rate is the same for all firms at any given time (Parker, 2009). Consequently, the expected growth rate of a firm is independent of its size, and the probability of a given growth rate during a specific time interval is the same for all firms in the same industry (Becchetti & Trovato 2002; Sutton 1997). Within this context, the issue of whether firm size has a systematic influence on the growth rate of a firm has been the subject of extensive research in empirical studies (Mukhopadhyay & Amirkhalkhali 2010). Despite the apparent power of Gibrat’s Law, most empirical analyses have rejected the hypothesis of independence of growth in relation to size: the size presents an inverse relationship to the firm's growth (Fowowe, 2017, Tarfasa et al., 2016, Brenner & Schimke, 2015, Maças Nunes et al., 2013; Mateev & Anastasov, 2010, Harabi, 2007; Oliveira & Fortunato, 2006b; Almus & Nerlinger, 2000; 1999; McPherson, 1996; Hall, 1987; Mansfield, 1962). Small firms grow faster (Coad & Tamvada 2012; Davidsson et al. 2002; Liedholm 2002) than their larger counterparts (Maças Nunes et al., 2013; Evans 1987). Hence, it is expected that the smaller the firm, the higher its growth rate:

H₂ – The growth rate of firms is independent of their size.

Jovanovic (1982) and Evans (1987) theoretically address the age of firms as a determinant of firm growth. While Evans found that firm growth decreases at a diminishing rate with firm size, Jovanovic concluded that, on average, older firms grow more slowly than younger ones within an industry (i.e. firm growth decreases with firm age). This may happen because young firms need to grow to be able to reduce uncertainty and accumulate sufficient resources that allow them to withstand unpredictable external shocks (Dobbs & Hamilton 2007). Concerning empirical research, most analyses have concluded that age and firm growth are inversely related to each other (Nichter & Goldmark, 2009; Liedholm, 2002; McPherson, 1996): unlike mature firms, young firms exhibit high growth rates (Coad et al., 2018; Fowowe, 2017; Coad et al., 2016; Hampel-Milagrosa et al., 2015; Maças Nunes et al., 2013; Lotti et al., 2009; Morone & Testa 2008; Harabi, 2007; Davidsson et al., 2002; Almus & Nerlinger, 1999, Storey, 1994, Variyam & Kraybill, 1992). Hence, it is expected that young firms have high growth rates:

H₃ – The older a firm is, the more negatively affected its growth is.

Growth can be seriously hampered when firms are subject to financial restrictions (Musso & Schiavo, 2008) . Of course, access to finance does not directly cause growth, but credit constraints may affect it. Therefore, without adequate access to financing, the staying power of the business and its growth potential is jeopardized (Ullah, 2019; Rahaman, 2011). Research on the role of financing in the firm growth process highlights that access to finance impacts firm behavior (investment, production, innovation, and exporting decisions) (Rostamkalaei & Freel 2016). Consequently, financial capital is essential for the growth of firms because it can be easily converted into other types of resources (Tarfasa et al., 2016). It must be pointed out that there are two sources of financial resources: internal and external. The former stems from the injections of capital provided by the business proprietor and the profits of the firm; the latter is provided from financial institutions, suppliers, and the capital market (Wang et al., 2022; Serrasqueiro et al. 2021). Firms that can finance themselves with their profits are less exposed to external financing sources (Kunt-Demirgüç & Maksimovic 1998). Internal sources of financing are typically the first option for a business proprietor; however, this form of funding is likely to be limited, which may constrain the growth of the firm (Rostamkalaei & Freel 2016). According to Rahaman (2011), as the level of external financial constraints decreases, the tendency for firms is to transition from relying on internal funds to seeking external sources of financing to support their growth. It should also be mentioned that internal financial resources are related to the idiosyncratic characteristics of the business proprietor (motivation, number of founders, networks, and personal and family resources) and internal factors of the firm (vision and mission). The external financial resources are related to the owner’s idiosyncratic characteristics (such as age and experience) and factors that are external to the firm (political and economic) (Vaz, 2021 ), which portray the features of the financial system in which the firm is situated and determine external financing options (Dobbs & Hamilton 2007).

According to the empirical literature, financial capital is one of the determining factors of firm growth (Serrasqueiro & Maças Nunes, 2016; Coad et al., 2013; Guariglia et al., 2011; Serrasqueiro et al., 2010; Segarra & Teruel, 2009; Zhou & Wit, 2009; Hermelo & Vassolo, 2007; Oliveira & Fortunato, 2006a; Carpenter & Petersen, 2002; Cooper et al., 1994). Based on the above arguments, we have formulated the following research hypothesis.

H₄ – There is a positive relationship between internal and external finance and the growth of a firm.

The data was obtained from the SABI database, which offers exhaustive information from balance sheets and financial sources, both on public and private firms belonging to Portugal. The data included all variables found in annual reports e.g., number of employees, sales, assets, financial ratios, date of establishment and industry classification codes. Using this information, we chose different criteria by first selecting Portuguese firms classified according to Rev. 3 on divisions 13 and 14. Secondly, we restricted our analysis to active firms (i = 1, …, n) throughout all sample periods (2011–2019 inclusive). Thirdly, given that statistical tests can only be computed based on raw data without missing values in the data matrix, we excluded firms with missing values for the number of employees, age, assets, liability, and location. Finally, we only considered firms having at least one employee throughout the period, and we only considered firms having a turnover greater than zero throughout the period. The final sample was a balanced panel dataset and comprised a total of 2.487 firms. During the 9 years covered in this study, these firms survived, were not absorbed by other firms and did not go bankrupt.

To measure dependent variable growth, we used the growth rate given by the difference between the logarithm of a size-related variable in the current and immediately previous period. The literature has traditionally studied the growth of firms by analyzing the evolution of the series of employment, sales or assets. More specifically, we measured firm growth by taking the log-differences to minimize the effect of heteroscedasticity in statistical analysis (Coad & Holz, 2010), the usual procedure for calculating growth rates (Coad et al., 2018, Serrasqueiro & Maças Nunes, 2016, Brenner & Schimke, 2015, Gopinath, 2012, Rahaman, 2011, Serrasqueiro et al., 2010); that is,

(1) Growthi,t=log(Sizei,t)−log(Sizei,t−1),

where Size_i,t is measured by any of the following variables identified in the literature: employment (Coad et al., 2016; Delmar & Wiklund, 2008), sales (Guariglia et al., 2011), and assets (Mateev & Anastasov, 2010; Heshmati, 2001). Accordingly, we used three different measures for growth that reflect different aspects of the growth process. Several measures in the analysis of firm growth allowed us to provide a more complete picture and robustness of the effect of size on the growth of firms (Dobbs & Hamilton, 2007; Delmar et al., 2003).

The independent variables included internal factors related to the essence and characteristics of the firm and external factors related to economic ones (Vaz, 2021; Fadahunsi, 2012). We considered five variables: Size, Age, Internal and External Finance, and Location. Unless mentioned, all but the dummy and proxy variables were subjected to logarithmic transformation (natural log). The variable Size (Size_i,t-1) is a measure of the number of firm resources. To be consistent with the growth measure chosen, Size was measured by (the logarithm of) the number of employees, sales, and assets in the previous periods. The variable Age (Age_i,t-1) was measured by (the log of) the number of years a firm had been active in the business sector in the previous periods. Concerning the variables of financing sources, we followed the methodology of Serrasqueiro et al. (2010). The variable Internal Finance (Internali,t−1Finance) is a proxy, and we used cash flow, given by the ratio between earnings after tax plus depreciation and total assets in the previous period; The variable External Finance (Externali,t−1Finance) is also a proxy, and we used the level of debt given by the ratio between total liabilities of total assets in the previous period.

Finally, concerning Location, the literature has proposed different measures for this independent variable. Folta et al. (2006), using data on 806 private and public U.S. biotechnology firms, asserted that methods for determining clusters are imprecise even though they considered the Metropolitan Statistical Area to define clusters. Glancey (1998) simply made a dichotomic distinction between firms located in rural areas and those located in urban ones. In our body of work, we analyzed the Portuguese Textile Cluster, a geographically proximate group of interconnected firms mainly located at four poles. To measure the local levels of agglomeration and the local market concentration, we needed to determine the boundaries within which these forces operate; therefore, we constructed series that measures three different dimensions. A first series, denoted as LCluster_i, simply indicates whether a firm pertains (or not) to one of the four poles of the Portuguese textile and clothing industry. Like other previous studies (Lee, 2018; Tarfasa et al., 2016; Van Geenhuizen & Reyes-Gonzalez, 2007), we created a dummy variable that takes the value of 1 if the firm belonged to one of the above-mentioned poles, or 0 otherwise. A second series, denoted as LMunicipality_i, accounted for all firms belonging to each firm’s municipality. Here the cluster and the resulting spillovers were measured as concentrated within a municipality. The greater the number of firms in a municipality, the greater the effect of location on firm growth. Finally, we considered that location spillovers are measured not only by the firms within a municipality but also by the conglomerate “surrounding” that firm: that is, firms belonging to the municipality as well as those neighboring the municipality account for the location externality of each firm. Accordingly, we followed the methodology of Pe’er et al. (2016), based on concentric rings with various radii around the geographic centroid, defining a third variable denoted as LWideCluster_i This variable considered the total number of firms belonging to their respective municipalities plus those belonging to the neighboring municipalities.

The distribution of firms by Size_i,t-1 (measured by the number of employees in 2011) and Age_i,t-1 (the firm’s number of years from foundation to 2011) is presented in Figures 1-2, which show that most firms are small and young. The size distribution of firms is J-shaped: there are a few large firms and many small firms. The 79 smaller firms have 1 employee, while the largest has 761 employees. In 2011, the 86 youngest firms in the sample were 1 year old, while the oldest was 106.

Table 1 displays basic statistics for the measures Growth_i,t, Size_i,t-1, Age_i,t-1, Internali,t−1Finance and (Externali,t−1Finance). Throughout the period of analysis, firms on average grew more in assets followed by sales and, finally, employees. In the analysis, it can be seen that the average in our sample of the firms in the Portuguese textile and clothing industry was about 28 employees, whereas the median, a measure that is less susceptible to outliers, was 12 employees, in 2011. The mean and median confirmed the usual definition of small and medium-sized businesses adopted by the European Union. In terms of sales, in 2011, firms sold on average around 1.662.912 euros and had approximately 1.735.375 euros in assets. On average, the age of firms was approximately 16 years old, whereas the median was 12 years old, which means that most of the firms were young. Concerning the analysis of the (Internali,t−1Finance) and (Externali,t−1Finance) variables, the high average debt of firms stood out.

Concerning the distribution of firms among location, we found that most of them were concentrated in four main sub-regions (the textile cluster): Cávado (24% of the total firms); Ave (33%); Área Metropolitana do Porto (21%), and Tâmega e Sousa (8%). Most firms (n = 2.130 [86%]) were located inside the textile cluster, while only 357 (14%) were based outside the cluster (Table 3). This reinforces the idea that the textile sector (and the sector’s main infrastructures) is geographically concentrated in the “Norte” region of Portugal (see Figure 3). According to ATP (2019), the Portuguese textile and clothing industry is mainly located in the “Norte” region of Portugal which represents 87% of the turnover and 85% of employment. The rationale of this distribution may be due to there being a positive location externality, where there is a higher average growth rate for firms within a cluster than the growth rate of firms located outside it (Table 3). We provide further analysis on this topic in the following section.

Figure 1. Source: SABI

Figure 2. Source: SABI

Table 1.

Variable Name	No of Obs.	Mean	Median	Standard Deviation	Min.	Max.
Growthi,tEmployees	2.487	0.0073	0.0000	0.09719	-1.51	1.95
Growthi,tSales		0.0131	0.0125	0.11913	-1.85	2.04
Growthi,tAssets		0.0159	0.0113	0.10419	-1.33	1.17
Sizei,t−1Employees		1.0931	1.0792	0.52752	0.00	2.88
Sizei,t−1Sales		5.5723	5.4702	0.70179	3.43	7.88
Sizei,t−1Assets		5.4785	5.4201	0.73658	3.70	8.45
Agei,t−1		1.0587	1.0792	0.38710	0.00	2.03
Internali,t−1Finance		0.0535	0.0559	0.27910	-5.85	6.20
Externali,t−1Finance		0.8201	0.6771	0.90717	0.02	19.59

Source: Own computations from SABI.

Table 2.

Variable Name	p5	p10	p25	p50	p75	p90	p95
Growthi,tEmployees	-0.1091	-0.0621	-0.0067	0.0000	0.0296	0.0792	0.1249
Growthi,tSales	-0.1393	-0.0909	-0.0347	0.0125	0.0591	0.1168	0.1670
Growthi,tAssets	-0.1266	-0.0817	-0.0299	0.0113	0.0593	0.1233	0.1753
No of Obs.	2.487

Source: Own computations from SABI.

Table 2 displays the average firm growth at each percentile for the Portuguese textile and clothing industry. Concerning growth, it was reported that: (1) Whatever the growth measurement used, the descriptive statistics indicate that until the 25^th percentile, growth was negative. For the modal value of the distribution (p50) and later (p75, p90, and p95) percentiles, growth was positive. This seems to support the idea that most firms do not maintain their initial size. Firms exhibiting an increase in the number of employees, sales or assets—suggesting an upsizing (positive) growth—coexist with others that shrink—suggesting a downsizing (negative) growth; and (2), firms grew more in assets, followed by sales and, finally, employees; the same happened for the firms that shrank.

Figure 3. Source: https://www.ccdr-n.pt/storage/app/media/files/ficheiros_ccdrn/institucional/mapa_norte.jpg

Table 3.

	Variable name	No. of obs.	Mean	Median	Standard deviation	Min.	Max.
Out of the cluster	Growthi,tEmployees	357	-0.0012	0.0000	0.10571	-1.13	1.26
	Growthi,tSales		0.0036	0.0073	0.13374	-1.74	1.48
	Growthi,tAssets		0.0066	0.0049	0.10568	-1.07	1.17
In the cluster	Growthi,tEmployees	2.130	0.0087	0.0000	0.09562	-1.51	1.95
	Growthi,tSales		0.0146	0.0132	0.11643	-1.85	2.04
	Growthi,tAssets		0.0175	0.0126	0.10385	-1.33	1.09

Source: Own computations from SABI.

Table 4 reports the Pearson Correlation of the variables used in this study on the Portuguese textile and clothing industry. The correlation between the measures of firm growth (Growthi,tEmployees, Growthi,tSales, and Growthi,tAssets) is moderate but significant at the 0.01 level. These results provide us with some confidence relative to the econometric analysis performed in the next section, given that the measures of growth were not all continuous, which might be problematic in the regression analysis (see Coad et al. 2018, ft.3). We found a low, positive correlation which was significant at the 0.01 level, coinciding with common definitions of growth for Size_i,t-1. There was also a low, negative correlation but also significant at the 0.01 level with Age_i,t-1 (Internali,t−1Finance, and Externali,t−1Finance variables reporting a low, positive correlation that was significant at the 0.01 and 0.05 level, matching common definitions of growth. Finally, the variables LCluster_i, LMunicipality_i, and LWideCluster_i had a positive correlation and significant at the 0.01 and 0.05 levels for all of the measures of the growth variable, also being highly correlated among themselves.

Table 4.

Variable name	1	2	3	4	5	6	7	8	9	10	11	12
1 Growthi,tEmployees	1.000
2 Growthi,tSales	0.313**	1.000
3 Growthi,tAssets	0.164**	0.357**	1.000
4 Sizei,t−1Employees	0.129**	0.059**	0.048**	1.000
5 Sizei,t−1Sales	0.103**	0.123**	0.090**	0.820**	1.000
6 Sizei,t−1Assets	0.055**	0.028**	0.103**	0.721**	0.916**	1.000
7 Agei,t−1	-0.106**	-0.124**	-0.122**	0.298**	0.337**	0.425**	1.000
8 Internali,t−1Finance	0.029**	0.157**	0.242**	0.028**	0.070**	0.055**	-0.058**	1,000
9 Externali,t−1Finance	0.014*	0.072**	0.186**	-0.060**	-0.100**	-0.147**	-0.087**	-0.518**	1.000
10 LClusteri	0.036**	0.033**	0.037**	0.059**	0.048**	-0.008	-0.099**	0.063**	-0.055**	1.000
11 LMunicipalityi	0.019**	0.017*	0.022**	0.004	0.013	-0.016*	-0.050**	0.041**	-0.036**	0.409**	1.000
12 LWideClusteri	0.028**	0.024**	0.023**	0.025**	0.033**	-0.006	-0.078**	0.053**	-0.046**	0.548**	0.701**	1.000

Source: Own computations from SABI. **

The correlation is significant at the 0.01 level.

To perform the studies, we followed the tradition of modeling firm growth as a stochastic process (Coad et al., 2018). To model the dynamics of firm growth, we began by opting for a multiple linear regression. To avoid endogeneity, we regressed each measure of growth with non-correspondent measures of Size. The multiple multivariate regression specification was.

(2) Growthi,t=β0+β1Sizei,t−1Sales+β2Agei,t−1+β3Internali,t−1Finance+β4Externali,t−1Finance+β5LClusteri+β6LMunicipalityi+β7LWideClusteri+εi,t,

where Growth_i,t was the growth rate experienced by firm i for the period 2011-2019, and ε_i,t represented the statistical residual. The variable Sizei,t−1Sales represented firms’ Size in 2011; Age_i,t-1 represented the number of years of the firm’s existence until 2011; Internali,t−1Finance represented the cash flow of the firms in the previous period; Externali,t−1Finance represented the level of debt of the firms in the previous period; LCluster_i indicated whether a firm pertained (or not) to one of the four poles of the Portuguese textile and clothing industry; LMunicipality_i accounted for all firms belonging to each firm’s municipality; and LWideCluster_i accounted for the total number of firms belonging to the firm’s municipality plus those belonging to the neighboring ones.

The results are presented in Table 5. The empirical evidence obtained in this study, for size, age, and financial resources, shows that the findings were similar to the other empirical studies (Coad et al., 2018, Fowowe, 2017, Serrasqueiro & Maças Nunes, 2016, Tarfasa et al., 2016). Gibrat (1931) prediction that size is only determined by random influences and the null hypothesis that every firm has the same probability to grow can be rejected, so H₂ was rejected. Jovanovic (1982) and Evans (1987) showed that young firms on average, exhibit higher growth rates than mature firms, so our analysis could not reject hypothesis H₃. The financial resources, internal or external, are determinants stimulating the growth of firms in the Portuguese textile and clothing industry. Similar results were found in the Musso and Schiavo (2008) study, meaning that we could not reject the previous formulated hypothesis, H₄.

Regarding location, each definition exhibited a positive and small coefficient when regressed with the different measures of growth. The standalone effect on the growth of firms located in an agglomeration was not statistically significant, so our results reject hypotheses H_1abc. This suggests that firms can grow in any geographic area as the location externalities are weak. These results were in line with Lee (2018), concerning location in a cluster per se having no positive effect on firm growth, and contrasted with other empirical studies claiming that concentration of economic activity in a geographic space has a significant impact on firm growth (Harabi, 2007; Hoogstra & Van Dijk, 2004; Liedholm, 2002; McPherson, 1996; Storey, 1994). However, this finding is puzzling: if there is no location externality, how can it be explained that 87% of turnover and 85% of employment of the sector are located in four main sub-regions in the “Norte” region of Portugal? We will call this the “location puzzle” for the Portuguese textile and clothing industry.

In the following sections, we conjecture different explanations for this puzzle, namely, the existence of intra-division spillovers (section 4.3), the existence of inter-division spillovers (section 4.4) and the existence of other benefits for belonging to a cluster (section 4.5).

Table 5.

Variable name	Growthi,tEmployees	Growthi,tSales	Growthi,tAssets
Sizei,t−1Sales	0.021 (0.000***)		0.019 (0.000***)
Sizei,t−1Employees		0.022 (0.000***)
Agei,t-1	-0.052 (0.000***)	-0.050 (0.000***)	-0.037 (0.000***)
Internali,t−1Finance	0.007 (0.004***)	0.093 (0.000***)	0.146 (0.000***)
Externali,t−1Finance	0.002 (0.000***)	0.019 (0.000***)	0.036 (0.000***)
LClusteri	0.003 (0.222)	0.003 (0.326)	0.005 (0.032**)
LMunicipalityi	8.219-7 (0.871)	2.052-6 (0.736)	7.644-6 (0.120)
LWideClusteri	1.130-6 (0.630)	4.994-7 (0.859)	-2.668-6 (0.240)
Constant	-0.054 (0.000***)	0.025 (0.000***)	-0.090 (0.000***)
R2	0.034	0.074	0.211
Observations	2.487

Source: Own computations from SABI.

Note: The value in parentheses is significant at the t-test.

The poor results in the multivariate analysis concerning location effects are puzzling. In this section, we conjecture that the origin of this puzzle is the dissimilarity between the activities performed by firms belonging to the Portuguese textile cluster. Our hypothesis is that location spillovers exist among textile-related firms, but only among those developing close textile activities. Specifically, we might adapt hypothesis 1c as follows: “Being located in an “extended” or “wide” municipality with firms developing similar textile activities has a positive effect on firm growth rate."

To check our hypothesis, we adopted the regression strategy of separating firms into two groups regarding their particular textile activities. The first group comprise firms involved in the preparation and spinning of textile fibers, textile weaving, textile finishing, and manufacture of other textiles, classified according to Rev. 3 on division 13 “textile manufacture”; the second group is made up of firms involved in clothing as their main output and classified according to the Rev. 3 on division 14 “clothing industry”. To be consistent with this new analysis, we created two new location variables, LwideClusteri13 and LwideClusteri14 that accounted for the total number of firms belonging to their respective municipalities plus those belonging to the neighborhood municipalities for each respective group.¹

Next, to address the location effect on both groups of firms in the Portuguese textile and clothing industry, we regressed equation (2) for these subsamples, Table 6 showing our results.²

The relationship between growth and size was significantly positive, while the variable Age was significantly negative for both industries (similar results were obtained in studies such as Coad et al., 2018, Fowowe, 2017, and Tarfasa et al., 2016). The variables of Internal and External finance continue positive and significant for both groups. The coefficient of the variable LwideClusteri13 for division 13 (textile manufacture) was negative and not significant, while the value attained in LwideClusteri14 for division 14 (clothing industry) was positive and significant, albeit with a small β₆ coefficient.

Table 6.

	13	14	13A	13B
Variable name	Growthi,tAssets
Sizei,t−1Sales	0.017 (0.000***)	0.021 (0.000***)	0.015 (0.000***)	0.018 (0.000***)
Agei,t-1	-0.033 (0.000***)	-0.040 (0.000***)	-0.035 (0.000***)	-0.031 (0.000***)
Internali,t−1Finance	0.133 (0.000***)	0.167 (0.000***)	0.088 (0.000***)	0.170 (0.000***)
Externali,t−1Finance	0.025 (0.000***)	0.043 (0.000***)	0.020 (0.000***)	0.030 (0.000***)
LwideClusteri13	-6.992-6 (0.230)
LwideClusteri14		8.765-6 (0.003***)
LwideClusteri13A			3.657-5 (0.259)
LwideClusteri13B				-1.903-5 (0.080*)
Constant	-0.070 (0.000***)	-0.104 (0.000***)	-0.056 (0.000***)	-0.082 (0.000***)
R2	0.184	0.237	0.162	0.209
Observations	910	1.577	269	641

Source: Own computations from SABI.

Note: The value in parentheses is significant at the t-test.

These results partially solve the location puzzle: while firms in the clothing industry consider it important to be in the cluster, it is not relevant to textile manufacture. This finding leads to the creation of a new puzzle, as, in our database, 83% of firms in the textile manufacture industry (division 13) were also located in the four main sub-regions in the “Norte” region of Portugal. So, we next wonder whether further disaggregation of the textile manufacture (division 13), would be able to solve the location puzzle. Accordingly, we separated firms belonging to division 13 into two groups of firms. The first one, which we denoted as division 13A, includes firms devoted to activities such as preparation, spinning, weaving and textile finishing activities, and are classified according to Rev. 3 on groups 131, 132, and 133 (see Table 1). The second group, denoted division 13B, includes firms’ developing activities related to the manufacture of home textiles and other textiles, classified in group 139 (see Table 1). To be consistent with this new analysis, we created two new location variables, LwideClusteri13A and LwideClusteri13B, and we regressed equation (2) for these new subsamples. Table 6 displays our results. The relationship between growth and size, age, and internal and external finance remained unchanged. The coefficient of the variable LwideClusteri13A was not significant and for the LwideClusteri13B a significantly negative coefficient was reported, suggesting that location is a restrictive factor of growth.

These results show that the Marshallian agglomeration externalities, associated with labor market pooling, specialized input suppliers, and knowledge spillovers (Lee, 2018; Folta et al., 2006; Porter, 2000) did not result in greater growth of firms belonging to the textile manufacture. Hence, our “location puzzle” has been solved at least for division 14 but it still unexplained for firms belonging to division 13.

In our database, 83% of firms in textile manufacture (division 13) were located in the “Norte” region of Portugal. Nonetheless, the previous section shows that the location decision of business proprietor is unusual as no external effect was reported for firms in division 13.

Table 7.

	14	13A	13B
Variable name	Growthi,tAssets
Sizei,t−1Sales	0.021 (0.000***)	0.015 (0.000***)	0.018 (0.000***)
Agei,t-1	-0.040 (0.000***)	-0.036 (0.000***)	-0.031 (0.000***)
Internali,t−1Finance	0.167 (0.000***)	0.088 (0.000***)	0.169 (0.000***)
Externali,t−1Finance	0.043 (0.000***)	0.020 (0.000***)	0.030 (0.000***)
LwideClusteri13A	0.000 (0.002***)	0.000 (0.226)	0.000 (0.065*)
LwideClusteri13B	0.000 (0.002***)	0.000 (0.171)	0.000 (0.042**)
LwideClusteri14	2.385-5 (0.000***)	1.143-5 (0.475)	-7.391-6 (0.502)
Constant	-0.108 (0.000***)	-0.052 (0.002***)	-0.085 (0.000***)
R2	0.238	0.164	0.210
Observations	1.577	269	641

Source: Own computations from SABI.

Note: The value in parentheses is significant at the t-test

In this section, we conjecture that the origin of the unsolved location puzzle for division 13 stemmed from the existence of location spillovers among firms developing different, but complementary activities. Our hypothesis is that location spillovers exist for firms in division 13, but only among those developing complementary textile activities, because complementarities exist when firms of some industry are located close to firms of other industries. Thus, we needed to adapt Hypothesis 1c again to consider this location complementarity for division 13.

To check our hypothesis, we adopted the close regression strategy that was followed in section 4.3. Therefore, we assessed the cross-relationship betweenLwideClusteri13A, LwideClusteri13B and LwideClusteri14 and their respective groups. Table 7 presents our results.

The results for growth and size, age, and internal and external finance were still unchanged. Concerning location, a cross-division effect was not reported. The coefficients for location-related variables were not found to be significant for subdivisions 13A and 13B, while the value taken for division 14 was positive and significant, albeit with a β₇ coefficient close to zero. These results imply that only the firms in the clothing industry (division 14) enjoyed inter-industry location externalities, because they are located close to textile-related firms.

In the previous section, we did not empirically find external spillovers that explain the reasons that resulted in the present location of firms in division 13, the textile manufacture, which represents 37% of the total firms. This suggests that other motivations or benefits play a role. The usual intuition of a business holder’s location choice is usually related to cultural (entrepreneurial culture), behavioral, historical and institutional factors, that influence your location decisions (Musolino et al., 2020). For example, modern environmental regulation would find it less costly to move the firms together with sites endowed with abatement technologies. This would save fixed costs, then increase profits and trigger firm growth. Yet, our previous finding has not reported location effects for firms in the textile manufacture (division 13). Therefore, an alternative explanation is that firms obtain some benefit for locating at a cluster. Next, we explored the relationship between locations in clusters and one of the possible benefits: having access to external financial resources.

We regressed equation (3) with the variable LCluster_i, filtered by selecting two groups. One group includes the 2.130 firms belonging to the four poles of the Portuguese textile and clothing industry and a second group comprised 357 firms that were not in any of the poles. We also undertook the same analysis for divisions 13A, 13B, and 14. The linear multivariate regression specification then became:

(3) Growthi,t=β0+β1Sizei,t−1Sales+β2Agei,t−1+β3Internali,t−1Finance+β4Externali,t−1Finance+εi,t,

All variables are the same as in equation (2). The results are presented in Table 8 and Table 9.

The relationship between growth and size and age remained unchanged. The results show that the effect of external financial resources on the growth of firms was greater for firms located inside the cluster. The coefficient to the variables of the financial resources (internal or external) is positive and significant, but the value of the coefficient β₄ is higher for firms that are located in one of the four poles, in all the division's portions of the sample.

Table 8.

	Inside the cluster			Outside the cluster
Variable name	Growthi,tEmployees	Growthi,tSales	Growthi,tAssets	Growthi,tEmployees	Growthi,tSales	Growthi,tAssets
Sizei,t−1Sales	0.020 (0.000***)		0.016 (0.000***)	0.023 (0.000***)		0.025 (0.000***)
Sizei,t−1Employees		0.020 (0.000***)			0.020 (0.000***)
Agei,t-1	-0.052 (0.000***)	-0.048 (0.000***)	-0.030 (0.000***)	-0.045 (0.000***)	-0.037 (0.000***)	-0.029 (0.000***)
Internali,t−1Finance	0.018 (0.000***)	0.156 (0.000***)	0.229 (0.000***)	-0.005 (0.144)	0.038 (0.000***)	0.068 (0.000***)
Externali,t−1Finance	0.004 (0.000***)	0.023 (0.000***)	0.044 (0.000***)	-0.001 (0.414)	0.008 (0.000***)	0.019 (0.000***)
Constant	-0.049 (0.000***)	0.019 (0.000***)	-0.090 (0.000***)	-0.070 (0.000***)	0.022 (0.043**)	-0.112 (0.000***)
R2	0.034	0.102	0.267	0.035	0.034	0.156
Observations	2.130			357

Source: Own computations from SABI.

Note: The value in parentheses was significant at the t-test.

Table 9.

	Inside the cluster			Outside the cluster
	13A	13B	14	13A	13B	14
Variable name	Growthi,tAssets
Sizei,t−1Sales	0.012 (0.000***)	0.018 (0.000***)	0.016 (0.000***)	0.013 (0.204)	0.017 (0.000***)	0.028 (0.000***)
Agei,t-1	0.026 (0.000***)	-0.032 (0.000***)	-0.033 (0.000***)	-0.070 (0.001***)	-0.016 (0.070**)	-0.024 (0.003***)
Internali,t−1Finance	0.386 (0.000***)	0.158 (0.000***)	0.261 (0.000***)	0.053 (0.000***)	0.221 (0.000***)	0.078 (0.000***)
Externali,t−1Finance	0.113 (0.000***)	0.033 (0.000***)	0.046 (0.000***)	0.010 (0.000***)	0.025 (0.000***)	0.023 (0.000***)
Constant	-0.198 (0.000***)	-0.086 (0.000***)	-0.089 (0.000***)	0.022 (0.725)	-0.092 (0.000***)	-0.146 (0.000***)
R2	2	0.211	0.298	0.376	0.214	0.150
Observations	250	506	1.374	19	135	203

Source: Own computations from SABI.

Note: The value in parentheses was significant at the t-test.

This result supports the conventional wisdom that firms can gain substantial advantages if they are located in a cluster (Lee, 2018; Porter, 2000). It provides empirical evidence that the growth of firms located in the textile and clothing cluster is positively correlated with external finance, and this helps to explain why textile manufacture firms are located in the cluster. This result implies that business proprietors strategically (and rationally) decide to locate their firms in the cluster because it helps mitigate information asymmetry problems in accessing external finance. In other words, the location itself becomes a solvency signal for firms.