The Brazilian Automotive Industry in the Nineties

Munich Personal RePEc Archive Brazilian automotive industry in the nineties Kerlyng Cecchini and Joaquim José Martins Guilhoto and Geoffrey J.D. Hewings and Dridi Chokri Tecpar – Instituto de Tecnologia do Paraná, Brazil, University of São Paulo, University of Illinois, University of Alberta 2007 Online at http://mpra.ub.uni-muenchen.de/41197/ MPRA Paper No. 41197, posted 11. September 2012 11:34 UTC BRAZILIAN AUTOMOTIVE INDUSTRY IN THE NINETIES Kerlyng Cecchini1 Joaquim José Martins Guilhoto2 Geoffrey J.D. Hewings3 Chokri Dridi4 ABSTRACT This paper aims to carry out an analysis of fuzzy clusters in the Brazilian automotive industry to contribute to the analysis of the relative importance of these economic activities in the national productive structure and in their regional contexts. The intention is to assess whether, once they have been established in the structure of a determined region, the economic activities of the industry establish productive relationships similar to other industries to the point of leading an industrial group in the regions or in the national economy. INTRODUCTION In the nineties, the automotive industry stood out in the international scene due to investments made in emerging markets such as India, China and Brazil. In the case of Brazil in particular, the macroeconomic and political ambient that was aimed towards modernization and increased capacity for production and greater economic integration of the industry in Mercosur, became extra stimuli for new investments in the sector. With the Regime Automotivo (system of tax benefits for car manufacturers) that was set up in 1995, the automobile manufacturers benefited from special import tariffs for products from other Mercosur countries. From that period onward, state governments that hoped to capture future investments in the industry, broadened incentives. The result was a fiscal war in the Brazilian economy in the late nineteen nineties. Tecpar – Instituto de Tecnologia do Paraná, Brazil Department of Economics, University of São Paulo, Brazil; REAL, University of Illinois; and CNPq Scholar 3 REAL, University of Illinois 4 University of Alberta, Canada; REAL, University of Illinois 1 2 2 The adoption of these incentives by the Brazilian government to encourage development of the automobile industry and stimulate a much needed debate on the efficiency of recent public policies, raised questions concerning the relative importance of this industry in the development of regional economies and their role in the national productive structure. In economic theory, the input-output analysis has been widely explored in studies that intend to identify and analyze the productive profile of different economies. Within the analyses derived from the application of this model, we may highlight the analyses that concentrate on the behavior of block industries, identifying clusters or groupings of industries according to the similarity of their goods and services. This paper aims to carry out an analysis of fuzzy clusters in the Brazilian automotive industry to contribute to the analysis of the relative importance of these economic activities in the national productive structure and in their regional contexts. The intention is to assess whether, once they have been established in the structure of a determined region, the economic activities of the industry establish productive relationships similar to other industries to the point of leading an industrial group in the regions or in the national economy. 2. LITERATURE REVIEW 2.2 The automotive industry in Brazil 2.2.1 The role of the state: from implantation to the recent phases of consolidation and modernization The active role of the state was decisive for the development of the Brazilian automobile industry, both at the early stages and consolidation, in the fifties and sixties, and more recently in the phases of modernization and expansion to other regions in the nineties. Even with the first assembly line of vehicles established in Brazil in 1919, the initial efforts to set up the automobile industry in Brazil were not made until the nineteen fifties during the Getúlio Vargas administration with the strengthening of the steel industry and the creation of the National Steel Company and other important industries (Ferro, 1992) In the late fifties, there was an effective commitment on the part of the government to develop the sector as a result of the policy of then president Juscelino Kubitschek. His policy program, called the Plan of Goals, covered five areas: energy, transport, supply of foodstuff, education and base industry. The Plan of Goals included a specific development program for the automobile industry, organized by the mediation of the National Bank for Economic and Social Development (BNDES), which eased restrictions on the import of equipment, raw materials and components for a certain period of time (Baer, 1995; Orenstein & Sochaczewski, 1990). This program was run by the Executive Group of the Automobile Industry (GEIA). According to Santos & Burity (2002), within the responsibilities of the GEIA were the definition of rules of installation, production targets and plans to nationalize the industry, with a priority for 3 the production of cargo vehicles and attenuating the deterioration of the balance of payments resulting from increased imports of cars and spare parts. As a result, at the end of the Kubstischek administration, around half of car production consisted of passenger vehicles and the rest was made up of utility vehicles and trucks (Baer, 1995). According to Santos & Burity (2002), the concession of quotas for the import of spare parts that were not produced in the country, the exchange rate favored importing equipment and tax exemptions for the importation of components for automobiles, were some of the exchange rate and fiscal incentives adopted by the government at this time. In the seventies, the BNDES was responsible for the financial support and numerous restructuring programs for industries without access to affordable and more long-term credit by way of specific programs, one of them for the spare automobile parts industry. The National Development Plan (II PND) also deserves to be mentioned. Nevertheless, after the petroleum crises and the II PND, divergences arose between the public and private sectors (Bedê, 1997). In the early nineties, the state became involved in making policies to strengthen the industry once again. As highlighted by Bonelli & Veiga (2003), in no other industry “was incentive so extensive and industrial policy so explicit” during this time. Policies were adopted from the start of the decade, as shown in Table 1 below. Table 1: Policies for the automobile industry from 1990-1995 Government Fernando Collor (1990-1992) Policies/Measures Program of tariff reductions (80% a 35%) from 1990 to 1994. Financial incentives suspended, nontariff barriers eliminated. Fiscal incentives (reduction of industrialized products tax) for small cars. Mercosur initiative with Argentina, Paraguay and Uruguay. 1992: Industrial Arbitration Process. Targets set for prices, production and export, jobs and salaries negotiated with industry, suppliers, trade unions and government. Reduction of taxes (industrialized product tax and ICMS) enforced to increase demand Itamar Franco (1993-1994) February, 1993: Arbitration process of the industry is renegotiated. New goals are established. April, 1993: fiscal incentives are given to cars in the lower price range; industrial product tax falls to 0.1%, ICMS is reduced and exceptions are made for COFINS. October, 1994: Government lowers tariffs to 20% (which was the goal only for 2001) 4 Fernando Henrique Cardoso (1994-2002) February, 1995: new meeting for industrial arbitration process (tariffs again increased to 32%, industrialized product tax for low price cars is raised to 8%) March, 1995: tariffs raised to 70% June, 1995: New policy established by the government (import quotas, reduction of tariffs to 2% for equipment and components associated with exporters, incentives for investments are given – accelerated depreciation). Import of cars from Argentina is exception to the new measures. Source: Laplane & Sarti (1997) With a view to restarting investments and the promotion of exports via increased competitiveness, the Brazilian government adopted the Regime Automotivo in 1995. The program established increased protection for the industry and was reformulated in 1997 to involved less developed states. Besides a fiscal incentive package for companies to set up in Brazilian states as a whole, there were added incentives for plants to be opened in the Northern, North-eastern and Central-Western states (Bonelli & Veiga, 2003; Laplane & Sarti, 1997; Santos & Burity, 2002). According to Santos & Burity (2002), the plan allowed a reduction of 50% on import tax for vehicles for companies that were already producing or involved in producing in the country. Furthermore, there were drastic tax reductions on the import of industrial goods, tools and molds for raw materials. The results of the tax cut were seen in the spare parts industry, affecting its prices. Besides the import tariffs, the IPI (tax on industrialized products) for industrial goods, raw materials, spare parts, pneumatics and packaging material were also reduced. In the case of previously established companies, average nationalization indices of 60% were required. For new companies, this index was 50%, as well as a compensation system for imports and exports. In 1995, at which time the import tax rate was 70%, with the Regime Automotivo program, the same taxes were set at 35%. A special trade agreement between Brazil and Argentina established the Brazil-Argentina Automotive Agreement of 1995. The difficulties of commercial relations between the two countries, however, were made clear with the devaluation of the real against the peso in January, 1999. From 2000, a common policy was established between the two countries for the industry to be in force from 2000 to 2005 (Bonelli & Veiga, 2003; Bonelli, 2001). The concession of incentives, however, was not restricted to the federal government. Among the policies of the states to attract new investments, Santos & Burity (2002) and Bonelli (2001) have witnessed the use of measures that vary from the use of direct expenditure with financing and participation of capital, to support in supplying infrastructure and the process of simplifying bureaucracy. The most widely criticized measures, however, are those which 5 compromise the revenue of the ICMS (Tax on the Circulation of Goods), the main source of income for state governments. It is these last measures that justify the term fiscal war used to characterize the behavior of the governments of several states during this period. The widespread use of tax reductions and exemptions5 from the ICMS tax was accompanied by a variety of other measures. In the case of Rio de Janeiro, the new units of Volkswagen were given a five-year tax deferral for 75% of their ICMS tax, benefits of infrastructure and had natural gas, digital telephones, water and electric energy at their disposal. These were practically the same benefits offered for the installation of the bus and truck factory in Resende. The donation of industrial plots of land was the measure adopted by the government of the state of Paraná to entice Renault to open their factory in the state. The effects of the fiscal war, however, have been a matter of debate in literature. According to Perobelli & Piancastelli (1996), the fiscal war is simply a fiscal renunciation and is harmless. The authors argue that the adoption of a similar set of instruments by states reinforces the question of location as a decisive factor in the flow of investments. Bonelli (2001) suggests that the real beneficiaries of the reduction of the ICMS tax are the multinational companies who assemble cars. According to this author, companies were given an opportunity to set up their plants at an extremely low cost and that they would have set up in Brazil anyway even without all these benefits. The author argues that the states may be the losers here since, at the time of his study, there were no papers to estimate the cost-benefit relationship for regions that had given these tax incentives, simultaneously assessing the impact on the creation of jobs and income6; and the cost resulting from the loss of revenue and the expenditure with infrastructure, electricity, water and sanitation, donation of land etc. Arbix & Rodrigues-Pose (2001) agree with the former viewpoint by defending that territorial competition was nothing more than high expenditure. In the view of these authors, any well-being that the industry stimulates is neutralized by the direct and indirect costs of attracting investments. From a global stance, this territorial competition implies closing other plants and, therefore, leads to the reduction of economic activity and increased unemployment nationwide. The set of policies in favor of the automobile industry in the nineties led to the effective widening of production capacity and modernization of industry. The efforts of this restructuring were seen, especially in production levels, jobs and foreign trade, which we will examine more closely in the next section. 5 Deferral of ICMS tax means exemption for the first agent in the productive chain, allowing for the sale of cheaper products. The next agent, even having to pay it in full, has a financial gain, as the ICMS is a tax payable on the value added to the product (Bonelli, 2001). 6 For a discussion on this topic, see Cavalcanti & Udenrman in this volume. 6 2.2.2 The panorama of the nineties in the Brazilian automotive industry From 1990 to 1993, the production of autovehicles – cars, light commercials, trucks and buses – doubled in the Brazilian economy, rising from 914,000 to 1,800,000 units. This growth was sharper in the production of passenger vehicles, the volume of which rose from 663,000 units in 1990 to 1,500,000 in 2003. This growth mostly took place from 1990 to 1997, with a sharp retraction in 1998/1999, picking up again in 2000. This retraction is linked to the slowing of the internal market, and the larger part of production is destined for domestic consumption7. Increased production reflects in part the modernization and increased capacity for production in this sector of Brazilian industry. Besides the reflections in production, the outcome of modernization in the industry has been less use of labor. In 1990, the production of autovehicles provided 117,396 jobs in Brazil. In 2002, that number had fallen to 81,737, in other words, only 70% of those registered, according to the data presented in Figure 1. 120 115 110 105 100 95 90 85 80 75 70 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Figure 1 – Employment in the production of autovehicles. Brazil 1990-2002 (in thousands of people) Source: Anfavea (2004) The reduction in the number of people employed is largely a result of automation and robotization of some plants during this period. The industry has also undergone intense internal restructuring for production8. 7 8 In 2003, around 71% of autovehicle production was destined for the domestic market (Anfavea, 2004) for further details on the internal restructuring of the companies with a view to establishing more suitable standards of efficiency with lean production methods, see Salerno (2002) 7 According to the estimates given by Anfavea (2004), in 2003, the automobile industry comprised a total of 48 plants, in seven states and 27 cities and towns. Of these 48 plants, around 22 were inaugurated from 1996 to 2002. With capacity to produce 3.2 million vehicles per year, this industry accounted for around R$10 billion in direct taxes and R$1.5 billion in indirect taxes. According to Anfavea, the automobile industry, including agricultural machinery and automotives, is responsible for creating 92,000 direct jobs and in that year established inter-industrial relations with over 200,000 companies and 3,700 showrooms. Its GDP, including the spare parts industry, accounts for around 13.5% of industrial GDP in Brazil and 4.5% of the GDP on the whole. According to what has been seen in the production of autovehicles, the spare parts industry grew until 1997. In the following years, the volume of sales in the industry cooled. Exports grew by 82% from 1990 to 1992. on the other hand, the installation of new foreign companies in Brazil and the global strategies adopted by these companies led to a significant increase in the imported volume of the sector. Thus, the volume of imports that was US$837,000,000 in 1990, had increased around five times by 2002, reaching US$3.9 billion. When it comes to the investments made in the sector during this period, the volume grew until the late nineties, especially from 1995 to 1998. in the next section, some aspects of the investments in this industry in the nineties are highlighted, as well as their reflections on regional configurations. 2.2.3 Investments and recent regional configuration In the nineties, worldwide companies announced investments in the Brazilian economy with a view to establishing new plants or restructure the old ones. Companies such as Fiat, Ford, General Motors and Volkswagen increased their presence on the domestic market, while others such as Mercedes and Renault set up new plants. The huge investments in this industry in the nineties are similar to those seen during the initial period of their installation (Arbix & Rodrigues-Pose, 2001). Indeed, the investments made in the automobile industry were expressive. The amount exclusively put into the production of autovehicles from 1990 to 1992 was over US$18 billion, with 80% of this total made after 1993 (Anfavea, 2004). The regional guidance of these investments, however, varied a great deal. According to data from Anfavea (2004) and from Santos & Pinhão (1999), it is possible to systematize them by region, as shown in Table 2. Several reasons are mentioned to explain the regional deconcentration of this industry and the degree of importance attributed to each of them differs throughout the literature. Arbix & Rodrigues-Pose (2001) highlight that in the past, there was qualified labor and superior infrastructure in the South-east and that led to greater development in this region. More recently, the relatively lower cost of labor in the other regions of the country and improvement in the quality of skilled labor have led to deconcentration. When it comes to salaries, there is certainly better organization of trade unions in the South-east with greater bargaining power in the 8 industrial ABC region of São Paulo. In the city of São Paulo, there are also the effects of pollution and traffic congestion. Therefore, the Brazilian automobile industry, which has always historically been concentrated in São Paulo state and the South-eastern region of the country, has recently shown signs of deconcentration to other parts of the country. Table 3 below shows the current regional configuration of the main companies in this industry 9 Table 2. Regional distribution of direct foreign investment in automobile assembly plants in Brazil 1996-2001 Location São Paulo São Bernardo do Campo São Carlos Indaiatuba Sumaré Paraná São José dos Pinhais São José dos Pinhais Campo Largo Minas Gerais Juiz de Fora Betim Sete Lagoas Belo Horizonte Rio Grande do Sul Gravataí Caxias do Sul Rio de Janeiro Porto Real Resende Bahia Camaçari Company Country of Origin Date of Minimum Planned investment planned annual investment capacity (in millions of (in thousands) US$) BMW/ L.Rover UK 1998 150 150 Volkswagen Toyota Honda Germany Japan Japan In operation 1999 In operation 250 150 100 300* 15 30 Renault France 1999 750 100/110 Audi Germany 1999 600 120 Chrysler/ BMW USA/Ger many 1999 600 120 Mercedes Fiat Iveco Fiat Germany Italy Italy Italy 1999 1998 1998 1999 820 500 250 200 70 5.000* 20 100 GM Navistar USA USA 1999 1998 600 50 120 5 PSA-Peugeot Volkswagen France Germany 2000 In operation 600 250 1.000 50 Ford USA 2001 1300 250 Source: Anfavea (2004), Santos & Pinhão (1999) 9 9 The Ford project, initially set up in Guaíba, Rio Grande de Sul, expected investments of half a million dollars and a capacity for 100,000 units of vehicles (Cavalcante & Uderman, 2003). Other investments such as those announced by Mitsubish, Ásia and Hyundai were not made owing to the Asian crisis. 10 Table 3. Regional Distribution of Brazilian Automobile Industry Region State City/Town Main Companies Products North Amazonas Manaus Motorcycles Northeast Ceará Bahia Minas Gerais Catole Horizonte Camaçari Betim Sete Lagoas Cross Lander Pólos de duas Rodas Agrale Troller Ford Fiat Automóveis Iveco Iveco-Fiat Southeast Rio de Janeiro Juiz de For a Duque de Caxias Porto Real Resende São Paulo Indaiatuba São Bernardo do Campo DaimlerChrysler Óbvio Citroën Peugeot VW Caminhões Toytota DaimlerChrysler Assembly of motorcycles and scooters Cars Cars, light commercials, engines Light commercials, trucks and buses (Iveco) Light commercials and trucks (Fiat) Cars Cars, light commercials, trucks, Engines. Trucks and buss chassis. Cars Trucks, buses and assemblages Ford Cars, light commercials, trucks Karmann-Ghia Tool room, devices, stamping sets, bodies, aria, sets and subsets, prototypes, assembly of vehicles Light commercials Land Rover Trucks, buses, industrial and maritime engines. Scania Car parts São Jose dos Campos South Paraná São Caetano do Sul Sumaré Taubaté São Jose dos Pinhais Curitiba Caxias do Sul CentralWest Goiás Gravataí Catalão Toyota Volkswagen General Motors Cars, light commercials General Motors Honda Volkwagen Audi/Volkswagen Nissan Renault Volvo Agrale International Cars, light commercials, foundry, CKD preparation for export, engines and transmissions. Cars Cars Cars Cars, light commercials, Light commercials Cars, engines, light commercials Trucks, truck cabins, buses and engines Trucks, buses Truck assembly General Motors Mitsubishi Cars Light commercials Source: Automotivebusiness (2004); Anfavea (2004) 11 Even with the recent regional deconcentration moves, the volume of production in the state of São Paulo is still higher than the other states. According to Anfavea (2004), around 53.4% of Brazilian car production is done in this state. Minas Gerais, Paraná and Bahia are next in line with 20.1%, 7.6% and 7.5% respectively in volume of production. 3 METHODOLOGY10 3.1 Input-Output Model This paper is based on the input-output model developed by Wassily Leonteif. The model relates the information of flows of input and output for each of the sectors of an economic region during a determined period of time (Feijó et al., 2001; Leontief, 1986). In mathematical terms, the relationship between sectors may be represented by a system of linear equations in the form of a matrix so that each of the equations describes within the economy the distribution of a product from a certain industry or sector. The economic interpretations to be had from the use of the model come from the resolution of this system of equations by way of an inverse matrix and other algebraic derivations of the model (Miller & Blair, 1985). According to Miller & Blair (1985), the fundamental aspect of the model can be expressed by the equation: X I  A Y 1 in which, (I – A)-1 represents the Leontief inverse matrix. X represents the production vector; Y represents the final demand vector; 10 For a detailed discussion about the data base, inter-regional matrix see Cecchini (2005). (1) 12 A represents the matrix of technical coefficients of the sectors; Equation (1), by multiplying the final demand vector Y by the Leontief inverse matrix, allows us to measure the total impacts caused by exogenous shocks resulting from the variations of the final demand components (consumption by families, exports and private investment). The inter-regional input-output models are more suitable when the intention is to analyze the interactions between economic regions. According to Miller & Blair (1985), an inter-regional model for two regions L and M, can have its coefficients matrix represented in matricial terms as:  A LL A   ML A A LM   A MM  (2) The vectors XL and XM will constitute the total production vector, X XL  X  M X  (3) The final demand vector, Y will be composed of the vectors YL and YM Y L  Y  M Y  (4) The complete inter-regional model can be represented as: ( I  A) X  Y I  A X LL LL  (5)  A LM X M  Y L   A ML X L  I  A MM X M  Y M (6) (7) 13 Considering in equation (7) the variation of final demand of region M to be null, i.e., YM =0, we have: I  A X LL L   A LM I  A  MM 1  X M  I  A MM A ML X L  Y L  1 A ML X L (8) (9) Comparing equation (9) with the equation I  A  LL 1 X L  Y L , which would be equivalent to a model from a single region, we can see in equation (9) the following additional term:  A LM I  A MM  1 A ML X L (10) The above equation represents the regional feedback effect. 3.2 Fuzzy clusters11 One of the more relevant contributions in literature concerning the methods for identifying clusters based on the input-output theory can be found in Czamanski & Ablas (1979). These authors discuss and analyze fourteen studies that seek to identify clusters and industrial complexes. Other works also deserve to be mentioned, such as Streit (1969), Bergsman et al (1972, 1975), Roepke et al (1974), Blin & Cohen (1977), Bergman (1997) and Hewings et al.(1998), Oosterhaven et al. (1999) and Hoen (2002). 11 For a detailed discussion on this topic see Cecchini (2005) 14 One aspect in which the method used in this paper differs from most of the methods used in literature is the type of cluster identified. Most methods allow for the identification of clusters formed by a limited number of sectors, identifying crisp clusters. In these analyses, the sectors belong to only one cluster and it is not possible to allow participation in more than one cluster at a time, which leads to analyses that are far from economic reality. To this end, Czamanski (1974) utilizes four different coefficients in one of the first attempts to evaluate clusters according to the most important sectors. More recently, Dridi & Hewings (2002) utilize the analysis of fuzzy clusters, seeking to take into consideration the complexity of the productive relationships in the establishment of sector groupings. The method does not require the researcher to choose arbitrarily the values of restrictions required by other methods. It is fundamental to add that the groupings are identified in such a way that all sectors of the economy belong to a certain cluster in varying degrees. These characteristics were preponderant in the choice of method used in this paper, which will be dealt with in further detail later. The sector groupings, or clusters as they are called in this paper, do not restrict their relationships for the purchase and sale of inputs to a small number of activities in the productive structure, even when there is great similarity between them. On the contrary, the sectors establish input purchase and sale relationships with a high number of sectors. In other words, it is possible that some activities, when grouped together according to the input purchase and sale relationships, belong to more than one cluster at a time. Utilizing the fuzzy set theory, it is possible to carry out an analysis that takes this fluidity of the productive structure into consideration. Dridi & Hewings (2002a) present the utilization of the analysis of fuzzy clusters resulting from this theory as a methodological proposal for assessing sector clusters and do an exercise with the matrix of the American economy. By recognizing the limitations of the traditional crisp approach, the fuzzy approach aims to analyze the complexity of the productive structure more coherently. An application of this method in a study of the Brazilian economy can be found in Simões (2003). The author utilizes the fuzzy cluster approach to identify spatial industrial complexes in the state of Minas Gerais, applying the method in a way that is similar to its use in 15 this study. This paper, by utilizing the use matrix data, requires a decomposition of that matrix in order to obtain distances between sectors. This method, called dual scaling, will be outlined in item 3.2.2 of the following section, which seeks to specify the methodology utilized. 3.2.1 Identification of fuzzy clusters The method for identifying fuzzy clusters is described below, as demonstrated by Dridi & Hewings (2002a). Considering X as a finite set of points, and a generic point, x, a fuzzy subset of X, denominated A, will be characterized by a function of membership, A(x) that will associate, at each x point, a coefficient within a real interval of [0.1]. Thus, the subset, fuzzy A is a set of ordered pairs ( x  A x ; x  X  , in which A(x) is the membership coefficient of the x element in A. If we let Ak,, k = 1, ...K, all the subsets of a universal set X, have the following properties:  A x   0,1; x  X , k  1,..K k   x   1 K k 1 Ak (11) The authors mention controversies concerning the format of the membership function, especially because this type of function is determined ad hoc, and adopt the proposal of Kaufman & Rosseeuw (1990). According to this alternative, the clusters are obtained minimizing the following objective function: min  K  ik v 1  16 r i , i´ 2 iv  2 i ´v d i , i ' 2   2 i ´v (12) restricted to:  iv  0; i  1,..., r; v  1,..., k  k v 1 iv  1; i  1,..., r In which:  iv represents the values of the membership coefficients of each of the i sectors in relation to the v clusters, which will minimize the function; d i ,i ' represents the Euclidian distance calculated between the sectors of the matrix submitted to the cluster analysis, defined in the following section. To the authors, this method has advantages over the other models of fuzzy cluster analysis. This is because this method minimizes errors by utilizing a unitary distance exponent whereas in other methods, the distance is squared. The fanny algorithm classifies r objects (in this case sectors) into k clusters based on the observation of s characteristics, or observations of a variable. Obtaining the s characteristics requires a decomposition of the use matrix by using a dual scaling technique which will be demonstrated below. 17 3.2.2 Dual scaling The dual scaling method is a descriptive, multivaried analysis technique introduced by Nishisato (1994). Its application in the input-output matrix allows for the decomposition of the complexity of associations that are established between the sectors of the productive structure. According to Dridi & Hewings (2002a)12, the technique applied to a contingency table, in this case an input-output table, allows us to determine a vector of rows weight and a vector of columns weight that maximize the relationship: 2  SS b SS i (13) Being: F = [fi,j] (r+1) x (c+1) ; flow matrix in the input-output table. fr : vector of total output of the input-output table. fc: vector of total input of the input-output table. Dr : diagonal matrix with row total in the main diagonal; Dc : diagonal matrix with columns total in the main diagonal; y: a vector of weights for the supplying sectors; x: a vector of weights for the demanding sectors; ft : the total value or intensity of the input-output table. Where, SSb expresses the variation between the F rows, and SSt expresses the total variation in the whole input-output table. 12 The description of the technique is based on Dridi & Hewings (2002a) 18 Thus, SS b  x' F' Dr1Fx , SS t  x' D c x The problem of maximization can be solved by setting SSt and maximizing SSb, in which the Lagrangian will be solved by: Lx,    x' F' Dr1 Fx   x' Dc x  f t  (14) With the first order conditions: L  F' D r1 Fx  D c x  0 x L  x' D c x  f t  0  By pre-multiplying by x’ and organizing the terms, equation (14) can be rewritten thus: F' D 1 r  F   2 Dc x  0 , (15) Which, pre-multiplied by D c1 , gives us the following eigenequation: D F' D c 1 r  F  2I x  0 , (16) Once a solution of  2 has been established, a eigenvector x, associated to the highest value of  2 is found in equation (16). The eigenvector y is found utilizing the dual relation: 1 y   D r1 Fx   19 (17) Thus, the first solution can be obtained, in other words, the first vectors of weights x and y of the new resulted matrix. If the first solution is insufficient to explain the correlation between rows and columns, new solutions are found, generating other vectors of weights x and y. In this case, all the possible s solutions are found. The application of this technique in the input-output table allows us to obtain two matrixes: one which establishes s weights for the columns of dimensions c versus s, in which c is the number of columns in the original matrix. The second matrix establishes s weights for the rows and their dimension will be r versus s and r is the number of rows in the original matrix. Therefore, the resulting matrixes are utilized to calculate the Euclidean distance between the rows and columns. The next step consists of the realization of the cluster analysis based on the distances calculated from the pondered matrixes. The number of clusters analyzed must be the highest possible (Simões, 2003). For this work, S-PLUS software was used, whose fuzzy cluster analysis is obtained by the fanny algorithm and allows for a maximum number of clusters, so that k  s  1 , where s is the number of solutions found, as described in the dual 2 scaling procedure. Thus, the result of the cluster analysis consists of a matrix formed by vectors expressing the membership coefficients of each of the sectors that form the clusters identified in the economy. These matrixes are called membership matrixes. The membership information subsidizes a preliminary analysis of the productive clusters. Other measures, like those given below, can be employed to characterize better the importance of the sectors in each of the clusters identified. 20 3.2.3 Involvement of the Sectors According to Dridi & Hewings (2002b), the relative importance of a cluster can be obtained through the coefficient called involvement, defined as: Inv A xi    A  xi  card  A ; i  1,..., n ; (18) Where,  A xi  represents the membership coefficient of the sector to the cluster A card  Ak     AA xi ; k  C , represents the cardinality of the sectors, X is a set of points, in this case, sectors, xi finite and countable, i  I  1,..., N , Ak are the fuzzy X sets. k  C  1,..., K . Thus, the highest values of this involvement coefficient indicate the most important sectors of the cluster, with leadership of the cluster in question. In their turn, the lower values indicate the sectors with secondary importance in the cluster, in other words, the sectors that supply support to the main activity of the cluster. 21 3.2.4 Subsethood This is an indicator for assessing how much a fuzzy set is present in another fuzzy set. It is defined as: card B  A D  B, A    card B   min  x ,  x    x  A i B i B i (19) i i Therefore, D (B,A) expresses the subsethood of cluster A in relation to cluster B. The subsethood matrix expresses the mutual dependence of the clusters in the productive structure. 4 RESULTS & DISCUSSION The clusters identified in this paper are called fuzzy clusters because the participation of all economic activities in a productive grouping are considered. The most important activities in a cluster are called leader activities, and the others are called support activities to the production of the final goods of the cluster. The first step in the identification of clusters consists of decomposing the use matrix into a similarity matrix using the dual scaling method. This procedure results in two similarity matrixes: one referring to the relations of purchase and the other to the relations of sales in the economy. The two resulting matrixes have m x n dimensions, where m is the number of sectors in the economy and n the number of variables of solutions found in the decomposition of the original use matrix. The vectors that constitute these matrixes, therefore, match each of the n similarity variables found in the decomposition of the matrix to the different economic activities. 22 The number of variables found will be utilized to define the maximum number of clusters in the economy13. The resulting matrixes are submitted to a fuzzy cluster analysis and as a result produce membership matrixes, m x c dimensions, in which m represents the number of sectors in the economy and c is the number of clusters identified. These matrixes present the degree of relationship of each of the sectors of the economy to the clusters. The sum along the row that represents each sector of this matrix is one. Based on the information of the membership matrix, we obtain the involvement matrix. This matrix has m x c dimensions, where m represents the number of sectors in the economy and c is the number of clusters, as in the original matrix. Contrary to the previous matrix, the values along each column add up to one, so that the degree to which an economic activity belongs to or is involved with a cluster can be measured in relation to the other economic activities belonging to the same cluster. According to Dridi & Hewings (2002a), this information is a more precise measurement to assess the relationship of economic activities in established clusters. Once the clusters have been identified, the subsethood between clusters is measured. In this paper, the clusters were identified from the viewpoint of the inter-regional and isolated region system for 1999, and the national viewpoint, based on the national input-output matrixes of 1990 and 2002. The results are discussed in more detail below. 4.1 Inter-regional system The decomposition of the inter-regional system utilized, composed of 186 economic activities, allows us to identify the maximum number of 91 clusters. Table 4 shows each of the clusters identified according to the purchase profiles, according to the activities that show the larger membership coefficient. These activities are called leader activities of the cluster. The 13 The procedure is carried out using SPLUS software. 23 clusters are shown in the order in which they are identified by the method. Thus, the first cluster shown is led by the activity of Food products in the North. In Table 5, we see the economic activities that lead the clusters, considering the sales profiles of the inter-regional system. This table can be read in the same way as Table 3. in this way, the first cluster identified according to the purchase profile of the inter-regional system is the cluster led by the Crops & Livestock activity in São Paulo. The analyses according to the two viewpoints show the same leader activities for the clusters of the automotive industry. They highlight the region of São Paulo, where automotive activity (cars, spare parts and other vehicles) lead the different clusters. It is also worth mentioning that the commerce of spare parts and vehicles in São Paulo, seen in the automotive industry, also leads a cluster in the system. Other leader activities in the clusters are: trucks and buses in the rest of the South-east, cars, parts and other vehicles in the South, trucks and buses in the Central West, cars and parts and other vehicles in the North-east and trucks in the North. 24 Table 4. Clusters according to the purchase profiles in the inter-regional system, 1999 (ISPU = Industrial Services & Public Utilities) Leader Activities Leader Activities Leader Activities N - Food Products 31 NE - Plastics 61 SP - Public administration N - Mineral extraction 32 NE - Clothing & footwear 62 SP - Clothing & Footwear N - Commerce 33 NE - Various industries 63 S - Crops & Livestock N - Non-metallic minerals 34 NE - Sale Vehicles/spare parts 64 SP - Various industries SP - Man. Machinery & tractors 35 NE - Communications 65 SP - Sales of Vehicles & parts N - Steelworks 36 SP - Crops & Livestock 66 SP - Communications N - Electrical/Electronic Equipment 38 CO - Mineral extraction 67 SER - Food products N - Trucks & buses 38 CO - Private services 68 SER - Private services N - Wood & furnishing 39 CO - Non-metallic minerals 69 SER - Non-metallic minerals N - Private services 40 CO - Steelworks 70 SER - Steelworks N - Rubber industry 41 CO - Electrical/Electronic equip. 71 SER - Electrical/electronic equip. N - Oil refinery 42 CO - Trucks & buses 72 SER - Trucks & buses SP - Various chemicals 43 CO - Wood & furnishing 73 SER - Wood & furnishing N - Pharmacy & veterinary 44 CO - Rubber industry 74 SER - Rubber industry SP - Plastics 45 CO - Food products 75 SER - Oil refinery N - Textile industry 46 CO - Oil refinery 76 SER - Pharmacy & veterinary N - ISPU 47 CO - Pharmacy & veterinary 77 RSE - Textile industry N - Transport 48 CO - Textile industry 78 SER - ISPU NE - Crops & Livestock 49 S - Clothing & Footwear 79 SER - Transport NE - Public administration 50 CO - ISPU 80 S - Public administration NE - Ext. oil, gas, coal 51 CO - Commerce 81 S - Ext. oil, gas, coal NE - Building 52 CO - Transport 82 S - Building NE - Steelworks 53 RSE - Ext. minerals 83 S - Steelworks NE - Man. Machinery/tractors 54 SP - Ext. oil, gas, coal 84 S - Man. Machinery/tractors NE - Cars 55 SP - Building 85 S - Cars NE - Parts & other vehicles 56 SP - Steelworks 86 S - Parts & other vehicles NE - Cellulose, paper & printing 57 SP - Cars 87 S - Cellulose, paper & printing NE - Chemical elements 58 SP - Parts & other vehicles 88 S - Chemical elements RSE – Commerce 59 SP - Cellulose, paper & printing 89 S - Plastics NE - Diverse chemicals 60 SP - Chemical elements 90 S - Diverse chemicals 91 S - Various industries 25 Table 5. Clusters according to the sales profiles in the inter-regional system, 1999 Leader Activities Leader Activities 31 NE - Plastics 61 SP – Sale vehicles/parts 2 N - Ext. mineral 32 NE - Clothing & Footwear 62 SP - Public administration 3 NE – Ext. oil, gas, coal 33 NE - Various industries 63 No activity 4 N - Non-metallic minerals 34 NE - Public administration 64 RSE – Mineral ext. 6 SP - Mach & tractors 35 NE – Sale vehicles/spare parts 65 RSE – Non-metallic minerals 5 N - Steelworks 36 No activity 66 RSE - Steelworks 7 N - Commerce 37 CO – Mineral ext. 67 RSE - Commerce 8 N - Electrical/electronic equip. 38 CO – Non-metallic minerals 68 RSE - Electrical/electronic equip. 9 N - Trucks & buses 39 CO – Steelworks 69 RSE - Trucks & buses 10 N - Wood & Furnishing 40 CO - Private services 70 RSE - Wood & Furnishing 11 N - Rubber industry 41 CO - Electrical/electronic equip. 71 RSE - Rubber industry 12 SP - Diverse chemicals 42 CO – Trucks & buses 72 RSE - Oil refinery 13 N - Oil refinery 43 CO - Wood & Furnishing 73 RSE - Pharmacy & Veterinary 14 N - Pharmacy & Veterinary 44 CO - Rubber industry 74 RSE - Textile industry 15 N - Textile industry 45 CO - Oil refinery 75 RSE - ISPU 16 N - Food products 46 CO - Pharmacy & Veterinary 76 RSE - Transport 17 N - ISPU 47 CO - Textile industry 77 RSE - Private services 18 N - Private Services 48 CO – ISPU 78 S - Crops & Livestock 19 N - Transport 49 CO – Commerce 79 S - Mach & tractors 20 NE – Crops & Livestock 50 CO – Transport 80 S - Ext. oil, gas, coal 21 SP - Plastics 51 SP - Ext. oil, gas, coal 81 S - Building 22 NE – Steelworks 52 SP - Building 82 S - Steelworks 23 NE - Mach & tractors 53 SP – Steelworks 83 S - Plastics 24 NE – Cars 54 SP – Cars 84 S - Cars 25 NE - Parts & other vehicles 55 SP - Parts & other vehicles 85 S - Parts & other vehicles 26 NE – Communications 56 SP - Cellulose, paper & printing 86 S - Cellulose, paper & printing 27 NE - Cellulose, paper & printing 57 SP - Chemical elements 87 S - Chemical elements 28 NE – Chemical elements 58 SP - Communications 88 S - Diverse chemicals 29 NE - Diverse chemicals 59 SP - Clothing & Footwear 89 S - Clothing & Footwear 30 NE - Building 60 SP - Various industries 90 S - Various industries 91 S - Public administration 1 SP – Crops & Livestock Leader Activities It is worth highlighting that the clusters identified according to the purchase profiles of the North, Central West and North-east regions (N – Trucks and buses, NE – Cars, and NE – Parts and other vehicles) are characterized as support activities of these very regions. In general, the clusters of São Paulo (SP – cars, SP – Parts and other vehicles and SP Commerce of vehicles and parts) in their turn are those belonging to the region of São Paulo, the South and the rest of the South-east and shown to be main support activities. The cluster of cars in the South shows 26 the importance of the activities in that region and the rest of the South-east. The clusters with the highest degree of regional diversity in support activities are parts, in the South and trucks and buses in the rest of the South-east. When it comes to sales, the presence of support activities in the North, North-east and Central West in the clusters of the automotive industry is much more expressive, especially in the cluster led by the Commerce of vehicles and parts. It is worth mentioning that the identification of a cluster in the. North, North-east and Central West must be assessed with caution since the method identifies clusters according to the similarity in inter-sector relations, not taking into account the value of the total production of the sector. In these regions, the results may indicate the potential for development of a cluster. A fundamental aspect is that the productive relevance of the automotive industry in the rest of the South-east and São Paulo is not captured by the method when using the inter-regional system. In other words, the automotive industry of the rest of the South-east is not among the activities that present the highest values for the involvement of sectors in the automotive clusters in the North, North-east and Central West. This behavior is expected since involvement is a measure of the importance of sectors in the clusters. However, the method does not consider the value of the total production of the sector, but rather the input values. In order to observe how clusters relate to one another, it is possible to assess their subsethood. The subsethood results in a matrix of c x c dimensions, where c represents the number of clusters identified in the economy. The main diagonal shows values equal to 1, expressing the maximum subsethood of the sector in relation to itself. The other cells in the matrix show values that vary from zero to one, equaling the mutual subsethood of clusters. As suggested by Dridi & Hewings (2002b), the option was to consider the subsethood of those values that were above the average, i.e. over 0.5. Thus, seeking to observe the dependence between clusters in the whole inter-regional system, Figure 3 shows a matrix with these values estimated for the clusters according to the purchase profiles. In this graphic matrix, values above 0.5 appear highlighted in red, whereas those below average are in blue. 27 Note that in the main diagonal of this matrix, all the values are above average. In truth, the diagonal expresses the subsethood of a certain cluster with itself. Thus, the values found in the main diagonal will always be maximum, therefore equal to one. The clusters appear according to the order in which they were found in the cluster analysis. According to the Figure, for instance, it is possible to say that cluster 88 (led by activity S – Chemical elements) shows dependence on clusters 63 (S – Crops and livestock), 64 (SP – Various industries), 65 (SP – Commerce of vehicles and parts), 66 (SP – Communications), 67 (RSE – Food products), 68 (RSE - Private Services ), 69 (RSE - Non-metallic minerals), 70 (RSE - Steelworks), and so forth. When it comes to the automotive industry, the clusters led by the respective economic activities stand out: Cars (57), Parts and other vehicles (58) and Commerce of vehicles and parts (65) in São Paulo; and Cars (85) in the South, suggesting greater importance of these activities in the functioning of the economic system than those that do not show relevant subsethood. We can also highlight the clusters in the economy led by transport activity in the North (18); Crops & Livestock (19), Public administration (20), Extraction of oil (21), Chemical elements (28), Diverse chemicals (30) and Communications (35) in the North-east (37), Steelworks (40), Commerce (51), Transport (52) in the Central West, Mineral extraction in the rest of the South-east (53); Cellulose & paper (59), Chemical elements (60), Public administration (61), Various industries (64), Communications (66) in São Paulo; Food products (67) Private services (68), Non-metallic minerals (69), Steelworks (70) & Transport (79) in the rest of the South-east; Crops & Livestock (63), Public administration (80), Cellulose & paper (87), Plastics (89) and Various industries (91) in the South. 28 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 6 52 55 58 61 64 67 70 73 76 79 82 85 88 91 1 4 7 10 13 16 19 22 25 28 0,0 -0,5 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 0,5 -1,0 Figure 3 – Subsethood of clusters according to the inter-regional system purchase profiles - 1999 29 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 6 52 55 58 61 64 67 70 73 76 79 82 85 88 91 1 4 7 10 13 16 19 22 25 28 31 34 0,0 -0,5 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 0,5 -1,0 Figure 4 – Subsethood of clusters according to the inter-regional system sales profiles - 1999. 30 In Figure 4 we can see the dependence subsethood of clusters according to Sales profiles. The interpretation of this figure is similar to Figure 3. In general, this subsethood is much lower than the clusters according to purchase profiles. This behavior may be explained by the fact that many sectors that produce final goods, supply their products directly to final demand. In the case of the automotive industry, only the clusters led by the activities of parts and other vehicles in São Paulo (55) and commerce of vehicles and parts (61) from São Paulo deserve to be highlighted. Other clusters also stand out in terms of a dependence or relationship with the other above-average productive clusters. The following deserve to be highlighted: Public administration (34) in the North-east; Steelworks (39), Private services (40), Commerce (49) and transport (50) in the Central West; Building (52), Parts and other vehicles (55), Cellulose, paper & printing (56), Various industries (60) and Public administration (62) in São Paulo; Mineral extraction (64), Steelworks (66), Commerce (67), Transport (76) and private services in the rest of the South-east (77), Crops and livestock (78), Machinery & tractors (79), Plastics (83), Cellulose, paper & printing (86), Chemical elements (87), Diverse chemicals (88), Various industries (90) and Public administration (91) in the South. 4.2 Clusters in isolated regions Evaluating regions in an isolated way and, therefore, not considering the inter-regional flows, the results are seen to be very similar from region to region. The clusters identified in relation to the sales profiles and purchase profiles were mostly the same in all the regions analyzed, as can be seen in Tables 6 and 7 respectively. These tables indicate the economic activities that led clusters in six regions. Thus, we see, for example, that the mineral extraction leads cluster in each of the regions. 31 Table 6. Clusters identified according to the purchase profiles of isolated regions Economic Activities Mineral Ext. Non-metallic minerals Steelworks Electrical/electronic equip. Trucks & buses Wood & furnishing Rubber industry Oil refinery Pharmacy & veterinary Textile industry Food products ISPU Commerce Public administration North North-east Central West São Paulo Rest of the South-east South X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X x x Table 7. Clusters identified according to sales profiles in isolated regions Economic Activities Ext. mineral Non-metallic minerals Steelworks Electrical/electronic equip. Trucks & buses Wood & furnishing Rubber industry Oil refinery Pharmacy & veterinary Textile industry Food products ISPU Commerce Public administration North X X X X X X X X x x x x North-east Central West São Paulo X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Rest of the South South-east x x x x x x x x x x x X X X X X X X X x x x x x Considering the 31 sectors, 14 clusters were identified in each region; nevertheless, in some regions, there were clusters that were hardly defined. These clusters showed no economic 32 activity in leadership; on the contrary, there were several economic activities with extremely low membership coefficients. Even though the same economic activities led most of the clusters between regions, the dependence between clusters, not considering the flow between regions, is quite different. Following this logic for the analysis of clusters in the inter-regional system, Figures 5 and 6 show in red the subsethood between above-average clusters in the region. In general lines, we see greater relationship between the clusters when it comes to purchase profiles. As mentioned beforehand, in the case of the inter-regional system, this accounts for why sectors that produce final goods supply their products directly to final demand. We also see that the clusters of the North-east, North and Central West regions are less dependent on each other in comparison to the other regions analyzed. When it comes to the automotive industry, the only economic activity that appeared as a leader in a cluster is Trucks and Buses in all regions. This activity is represented in the following figures by clusters (7) in the North, North-east and Central West regions and São Paulo, (8) in the rest of the South-east and (6) in the South, according to the purchase profiles. As for the sales clusters, it leads clusters (7) in the North, (9) in the North-east, (8) in the Central West and the rest of the South-east and (6) in São Paulo and the South. In general, this activity shows a subsethood in relation to the other above-average clusters. 33 66 1 N 1 2 3 4 5 6 7 8 9 1 2 3 4 5 1 6 7 8 9 6 10 11 12 13 14 10 2 3 4 5 7 8 9 11 12 13 1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 1 6 10 6 1 6 10 11 11 12 12 13 CO 13 14 1 2 3 4 5 6 7 8 9 14 10 11 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 9 6 1 10 10 11 11 12 12 13 13 14 SP 1 2 3 4 5 6 7 8 9 10 11 12 13 14 RSE 8 9 16 NE 14 10 11 12 13 14 14 1 2 3 4 5 6 0 -0 ,5 7 8 9 10 11 12 13 14 S 0 ,5-1 Figure 5 – Subsethood of clusters according to the sales profile of the regions 34 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 66 1 N NE 14 1 2 3 4 5 6 7 8 9 14 10 11 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 6 1 6 CO 9 16 10 10 11 11 12 12 13 13 14 14 1 2 3 4 5 6 7 8 9 10 1 11 12 13 14 2 3 4 5 6 7 8 9 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 16 9 16 SP 9 10 10 11 11 12 12 13 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 RSE 10 11 12 13 14 14 1 2 3 4 5 6 0 -0 ,5 7 8 9 10 S 11 12 13 14 0 ,5-1 Figure 6 - Subsethood of clusters according to the purchase profiles of the regions 35 4.3 National Clusters The identification method was applied to the matrixes of the Brazilian economy for 1990 and 2002. In Table 8, we can see the clusters identified according to the purchase profiles and according to the sales profiles for the two years. As for the automotive industry, the activity Other vehicles and parts stands out as it leads a cluster during both periods. Table 8. Clusters identified in the Brazilian economy in 1990 and 2002. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 1990 Purchases Public administration Mineral ext. Non-metallic minerals Non-ferrous metallurgy Mach. & Tractors Electronic equip. Other vehicles & parts Commerce Paper & Printing Chemical elements Diverse chemicals Plastics Clothing Coffee industry Slaughter of animals Sugar industry Other food products ISPU Communications 2002 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Sales Commerce Mineral ext. Public administration Non-metallic minerals Non-ferrous metallurgy Mach. & Tractors Electronic equip. Other vehicles & parts Paper & Printing Diverse chemicals Chemical elements Plastics Clothing Coffee industry Slaughter of animals Sugar industry Other food products ISPU Communications 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Purchases Commerce Mineral ext. Paper & Printing Non-metallic minerals Non-ferrous metallurgy Mach. & Tractors Public administration Electronic equip. Other vehicles & parts Chemical elements Diverse chemicals Plastics Clothing Coffee industry Slaughter of animals Sugar industry Other food products ISPU Communications 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Sales Commerce Mineral ext. Non-metallic minerals Non-ferrous metallurgy Mach. & Tractors Electronic equip. Other vehicles & parts Paper & Printing Chemical elements Diverse chemicals Plastics Clothing Coffee industry Slaughter of animals Sugar industry Other food products ISPU Communications Public administration The subsethood of clusters identified in both years can be seen in Figure 7. Comparing the two periods, we see that the productive structure in 2002 shows clusters with greater subsethood. This pattern is observed in both the purchase and sales profiles of the economy. A possible explanation for this behavior may be the opening of the economy in the nineties. With the increase of imports, domestic economic relationships became less dense since imported input utilized in some industries had prices and/or quality that was superior to their domestic equivalents. 36 purchase 1 1 3 3 5 5 7 7 9 9 11 11 13 13 15 15 17 17 19 1 3 5 7 9 11 13 15 17 sales 19 1 19 3 5 7 9 11 13 15 17 19 1 1 3 3 5 5 7 7 9 9 11 11 13 13 15 15 1990 2002 17 17 19 19 1 3 5 7 9 11 13 15 17 19 1 3 5 7 9 - - 0,500 Figure 7 – Subsethood of clusters in 1990 & 2002 11 13 15 0,500 - 1,000 17 19 37 It is worth remembering that the methodology applied is still preliminary, proposed by Dridi & Hewings (2002a & 2002b) and applied by the authors in the analysis of the use matrixes of the economy of the United States. There have been no applications by other authors aiming to analyze the flow of input in the Brazilian economy. Simões (2003) applied the method and obtained interesting results. 5 CONCLUSIONS The aim of this paper has been to carry out an analysis on the importance of the Brazilian automotive industry through the analysis of fuzzy clusters. Studies of this nature, especially those dealing with the Brazilian economy, face difficulties and limitations owing to the utilization of input-output matrixes and estimated inter-regional systems. In the case of national matrixes, the latest official data made public by the IBGE (Brazilian Institute of Geography and Statistics), used in this paper, were from 1996. Therefore, it was necessary to estimate the matrixes for 1996 to 2002 using information from the National System of Accounting as a basis for these years. The information about the inter-regional system was obtained from set of data that comprises several research projects. Even with all the criteria utilized to obtain the information on the flow of Brazilian inputoutput matrixes, such as methods of consistency between regional and national systems, the data could still be refined. However, it is necessary to emphasize that possible improvements depend above all on the availability and precision of information at regional and sector levels from complementary data bases. At the moment, even with the advances in the research of the IBGE and some sector institutions, the difficulty to obtain more detailed information on the North, North-east and Central West regions of Brazil and some sectors, especially those informally involved in the economy, still remains. The methodology of analysis used here has some advantages over traditional cluster analysis methods that identify crisp clusters, which generally require arbitrariness on the part of the researcher and do not consider the fact that a sector may be present in more than one cluster 38 of the economy. In the fuzzy approach, the groupings do not exclude all economic activities belonging to the groupings. The need for methodological sophistication is more evident in the analysis of clusters in the inter-regional system. This is because the data does not clearly indicate the relevance of the activities of São Paulo and the rest of the Southeast in the clusters of the other regions. An interesting alternative to perfect the analysis, may be to study the possibility of considering the value of production in the pondering of the data submitted to cluster analysis. The method, in its turn, provided interesting results in the analysis of the national productive structure. The analyses show that with the opening of the economy and the consequent entry of imported products into the Brazilian economy in the nineties, the national clusters showed less dependence on each other, reflecting the lower dynamism in the productive relationships of the national economy. In other words, the results suggest that with the opening of the economy to international commerce, the productive relationships between domestic clusters became more fragile. This paper has contributed to applying an analysis methodology that has been little used and explored in studies that make use of the input-output theory. To this end, the difficulties and limitations found in the application of the method can be seen as stimuli do developing future studies. 39 REFERENCES ASSOCIAÇÃO NACIONAL DE VEÍCULOS AUTOMOTORES (2004). Anuário da indústria automobilística brasileira 2003. http://www.anfavea.com.br (10 abr. 2004) ASSOCIAÇÃO NACIONAL DE VEÍCULOS AUTOMOTORES (2004). Anuário da indústria automobilística brasileira 2004. http://www.anfavea.com.br (07 ago. 2004) AUTOMOTIVEBUSINESS (2004). O mapa da produção. http://www.automotivebusiness.com (12 jul. 2004) ARBIX, G.; RODRIGUEZ-POSE, A. (2001). Strategies of waste: bidding wars in the brazilian automomotive sector, International Journal of Urban and Regional Research, v.25, n.1, pp.134-154. BAER, W. (1995) Economia brasileira. São Paulo: Nobel. BEDÊ, M.A. (1997) A política automotiva nos anos 90, in: ARBIX, G.; ZILBOVICIUS, M. (Org.). De JK a FHC: a reinvenção dos carros. São Paulo: Scritta. BERGSMAN, J.; GREENSTON, P.; HEALY, R. (1972) The agglomeration process in urban growth. Urban Studies, v.9, n.3, pp. 263-288. BERGSMAN, J.; GREENSTON, P.; HEALY, R. (1975) A classification of economic activities based on location patterns, Journal of Urban Economics, v.2, n.1, pp.1-28. BLIN, J.M.; COHEN, C. (1977) Technological similarity and aggregation in input-output systems: a cluster-analytic approach. Review of Economic Statistics, v.59, n.1, pp.82-91. BONELLI, R. (2001) Políticas de competitividade industrial no Brasil: 1995/2000. 44pp. Brasília: IPEA (Texto para discussão, 810) BONELLI, R.; VEIGA, P.M. (2003) Dinâmica das políticas setoriais no Brasil na década de 1990: continuidade e mudança. Revista Brasileira de Comércio Exterior, n.75, pp.1-24. CAVALCANTE, L.C.; UDERMAN, S. (2002) The cost of a structural change: a large automobile plant in a brazilian less developed region. Urbana: University of Illinois, Regional Economics Applications Laboratory. (Discussion paper, 04-T-5). CECCHINI, K. Setor automotivo brasileiro: evolução da estrutura produtiva e sua importância regional nos anos 90. Piracicaba, 2005. Dissertação (Mestrado) – Escola Superior de agricultura “ Luiz de Queiroz”, Universidade de São Paulo. CZAMANSKI, S.; ABLAS, L.A.Q. (1979) Identification of industrial clusters and complexes: a comparison of methods and findings, Urban Studies, v.16, pp.61-80. CZAMANSKI, S. (1974) Study of clustering of industries. Halifax: Dalhouse University, Institute of Public Affairs. 40 DRIDI, C.; HEWINGS, G.J.D. (2002a) Toward a quantitative analysis of industrial clusters I: fuzzy clusters vs crisp cluster. Urbana: University of Illinois, Regional Economics Applications Laboratory. (Discussion paper, 02-T-1) DRIDI, C.; HEWINGS, G.J.D. (2002b) Toward a quantitative analysis of industrial clusters II: shapley value, entropy, and other fuzzy measures. Urbana: University of Illinois, Regional Economics Applications Laboratory. (Discussion paper, 02-T-6) FEIJÓ, C.A.; RAMOS, R.L.O.; YOUNG, C.E.F.; LIMA, F.C.G.C.; GALVÃO, O.J.A. (2001) Contabilidade social: o novo sistema de contas nacionais do Brasil. Rio de Janeiro: Editora Campus. FERRO, J.R. (1992) A produção enxuta no Brasil, in: WOMACK, J.P.; JONES, D.T.; ROSS, D. et al. A máquina que mudou o mundo. pp.311-337. Rio de Janeiro: Campus. HEWINGS, G.J.; SCHINDLER, G.R; ISRAILEVICH, P.R; SONIS, M. (1998) Agglomeration, clustering, and structural change: interpreting changes in the Chicago regional economy, in: STEINER, M. (Ed.). Clusters and regional specialization. Heidelberg: Springer Verlag. HOEN, A.R. (2002) Identyfing linkages with cluster based methodology. Economic Systems Research, v.14, n.2, pp.131-146. HUMPHREY, J.; LECLER, Y.; SALERNO, M. (Ed) (2000) Global strategies and local realities: the auto industry in emerging markets. New York: St. Martins Press. KAUFMAN, L.; ROSSEEUW, P.J. (1990) Finding groups in data: an introduction to cluster analysis. New York: Wiley LAPLANE, M.F.; SARTI, F. (1997) The restructuring of brazilian automobile industry in the nineties. Actes du GERPISA, n.20, pp.31-48. LEONTIEF, W. (1986) Input-output economics. 2.ed. New York: Oxford University Press. MILLER, R.E.; BLAIR, P.D. (1985) Input-output analysis: foundations and extensions. New Jersey: Englewood Cliffs. NISHISATO, S. (1994) Elements of dual scaling: an introduction to practical data analysis. New Jersey: Lawrence Erlbaum Associates. OOSTERHAVEN, J.; EDING, G.J.; STELDER, D. (1999) Clusters, forward and backward linkages, and bi-regional spillovers: policy implications for the two dutch mainport regions and the rural north (compact disc). In: EUROPEAN CONGRESS OF THE REGIONAL SCIENCE ASSOCIATION, 39., Dublin Proceedings. Dublin: Carfax. ORENSTEIN, L.; SOCHACZEWSKI, A.C. (1999) Democracia com desenvolvimento, in: ABREU, M. (Org.). A ordem do progresso: cem anos de política econômica 1889-1989. 10.ed. Rio de Janeiro: Campus. PIANCASTELLI, M.; PEROBELLI, F. (1996) ICMS: evolução recente e guerra fiscal. Brasília: IPEA. (Texto para Discussão, 402) ROEPKE, H.; ADAMS, D.; WISEMAN, R. (1974) A new approach to the identification of industrial complexes using input-output data. Journal of Regional Science, v.14, pp.15-29. 41 SANTOS, A.M.M.M.; BURITY, P. (2002) Complexo automotivo, in: BANCO NACIONAL DE DESENVOLVIMENTO ECONÔMICO E SOCIAL. BNDES 50 anos: histórias setoriais. Rio de Janeiro. SANTOS, A.M.M.M.; PINHÃO, C.M.A. (1990) Pólos automotivos brasileiros. BNDES Setorial, n.10, set. 1999, pp.173-200 SIMÕES, R.F. (2003) Complexos industriais no espaço: uma análise de fuzzy cluster. Belo Horizonte: Universidade Federal de Minas Gerais (UFMG)/Cedeplar. (Texto para discussão, 209) STREIT, M.E. (1969) Spatial association and economics linkages between industries. Journal of Regional Science, v.9, p.177-188.