The Growth Effect of Democracy and Technology: an industry disaggregated approach

The Growth Effect of Democracy and Technology: an industry disaggregated approach Izaskun Zuazu Abstract The theoretical and empirical sides of democracy-growth literature fail to offer a consensus on the impact of democracy on growth. This paper offers a disaggregated manufacturing approach that reveals different effects of democracy across industries within countries. I surmise that the interplay between democracy and technological development is crucial to the economic performance of industries. A panel dataset of 61 manufacturing industries from 72 countries between 1990 and 2010 is employed, along with a wide variety of democracy measures. The results point to a technologically-conditioned effect of democracy. Political regime changes towards democracy are growth-enhancing for industries close to the World Technology Frontier. By contrast, the same institutional changes exert a negative effect on backward industries. This evidence is robust to specification changes and alternative estimation techniques, and prevails once the possible dynamics of manufacturing growth are tackled. Keywords: frontier, panel data JEL: H7, O14, O43 Democracy, manufacturing industries, economic growth, world technology 1. Introduction Scholars have long been addressing the question of whether political regimes foster economic development and, if so, whether democracy is more conducive to economic development than autocracy. Flourishing scholarship on that precise debate has produced a Foundations of the Economic Analysis I Department, University of the Basque Country (UPV/EHU), Lehendakary Aguirre 83, 48015 Bilbao (Spain) Email address: izaskun.zuazu@ehu.eus (Izaskun Zuazu ) I am grateful to Adam Przeworski, Joan María Esteban, Francesco Trebbi, Jan-Egbert Sturm and two anonymous referees for their comments. I thank Manuel Arellano, Juan Mur, Carlos Velasco and my thesis advisor Josu Arteche for guidance in econometrics. I also thank the audiences at the 3rd Ph.D. Workshop on Industrial and Public Economics (WIPE) in Reus (Spain) 2015, Dynamics, Economic Growth and International Trade (DEGIT XX) in Geneva (Switzerland) 2015 and the Association of Southern European Economic Theorists (ASSET) in Granada (Spain) 2015. All errors are mine alone. This research is supported by the Spanish Ministry of Science and Innovation and ERDF grant ECO2016-76884-P. Preprint submitted to SAEe - job market 2017 August 25, 2017

5 10 15 20 25 democracy-growth literature that, as of today, is still inconclusive 2. In the words of Przeworski (2016), the functions and limits of democracy are a never-ending quest. This article challenges this lack of consensus by addressing a disaggregated approach to the effects of democracy on economic growth. In this sense, I study whether political changes towards more democratic regimes impact industries differently within a country. I conjecture that the interplay of democracy with technological development determines the sign and magnitude of the effect of democracy on growth. I find that technologically advanced industries benefit from more democratic political institutions, whereas backward industries are worse off in more democratic environments. In a nutshell, this paper offers a mechanism through which democracy exerts influence in favour of growth of advanced industries by lowering market-entry barriers and bringing economic freedom. By contrast with backward industries, advanced industries are able to adapt and thrive in such an economic and political regime. The literature features three main opposing arguments that have fired up the research agenda on the effects of democracy on economic development 3 : the conflict view, the compatibility school and a sceptical approach. The conflict view suggests a negative effect of democracy on growth. Its proponents thus argue that autocratic regimes are better able to implement unpopular but profitable investments needed for growth, whereas democracies lend themselves to inefficient popular demands (Huntington, 1968). This approach is reinforced by the experiences of the Tiger economies -South Korea, Taiwan, Hong Kong and Singapore- which introduced democracy after economic reforms, and the case of China 4 (Huntington et al., 1976). By contrast, the compatibility school states that democracy promotes economic growth (Gerring et al. (2005); Acemoglu et al. (2014); Madsen et al. (2015)) and encourages more stable, 2 See inter alia Przeworski & Limongi (1993), Alesina & Perotti (1994) and Sirowy & Inkeles (1990), De Haan & Siermann (1996), Baum & Lake (2003) for an understanding of the background of democracygrowth literature. 3 These three approaches intermingle with those of the debate on democracy and economic freedom: see De Haan & Sturm (2003) and Lundström (2005) 4 As noted in Bell (2016), meritocracies could select better political leaders. See other studies on the Chinese case in Guo (2007), and Xu (2011) 2

30 35 40 45 50 higher quality development (Rodrik (2000);Bhagwati (1995)). This view notes that democracy enforces growth-enhancing features (e.g. property rights) precisely through the provision of political rights and civil liberties 5 (North, 1990). A third way in democracy-growth literature is the sceptical approach. Some papers find that democracies make little or no difference, e.g. Przeworski (2000) and Murtin & Wacziarg (2014). Bhagwati (1995) states that the market can deliver growth in both regimes. This view is also shared by Bardhan (1993), who suggests that collective action problems and quick responses to changes in technology and market conditions are more essential to growth than political regimes 6. Positioned between the three approaches above are the proponents of a non-linear effect of democracy on economic development. The hypothesis of a curvilinear relationship dates back to Lipset s hypothesis (1959), under which some level of development is a pre-requisite for democracy to function properly. Barro (1996) also finds a nonlinear effect: growth is increasing in democracy at low levels of democracy, but the relation turns negative once a moderate amount of political freedom is attained (p. 58). Other publications suggest that a curvilinear relationship is plausible by conjecturing different trade-offs between property rights, entry barriers and redistribution taxes (Plümper & Martin (2003);Rodrik & Wacziarg (2005);Acemoglu et al. (2008)). The opposing approaches taken in democracy-growth literature can be seen as a direct consequence of the complexity surrounding the conceptualisation of political institutions in general (Fukuyama, 2007) and of democracy in particular. Different conceptualisations and measurements of democracy constitute an issue that is exacerbated by alternative modelling and research designs (Sirowy & Inkeles, 1990). Notwithstanding that complexity, some novel conceptualisations and measurements have been made in the literature by using optimization rather than aggregation techniques to convey the political attributes that democracies 5 The influential paper by North & Weingast (1989))also emphasises the importance of commitment and the elimination of confiscatory government. It shows how 17th-century English democracy secured property rights, but only the propertied enjoyed political rights. 6 Murrell & Olson (1991) also point to th resolution of collective action issues and raise the issue of institutional sclerosis. 3

55 60 65 70 75 should have 7. One such stride is the democracy index developed by Gründler & Krieger (2016), who present evidence that higher levels of democracy are always growth-enhancing, using an algorithm-based index of democracy. The different results concerning the economic effect of democracy have called into question the direct effect of the democracy-growth relationship, which has opened up the stage to the conjecture of more accurate questions. In their meta-analysis of 84 studies of the effect of democracy on GDP growth, Doucouliagos & Ulubaşoğlu (2008) state that the overall effect of democracy on growth does not seem to be detrimental. Indeed, a large body of research suggests that the indirect effect of democracy through socioeconomic, political and cultural features is conducive to growth, e.g. mechanisms for the accumulation of physical or human capital (Helliwell (1994);Baum & Lake (2003); Dawson (1998)), political stability conjecture 8 ( Alesina et al. (1996);Rodrik (1998)) and economic freedom ((Sturm & De Haan, 2001). The contribution of this paper to the empirical strand of democracy-growth literature is threefold. First, I focus on the effect of democracy on manufacturing. This approach 9 allows certain patterns to be revealed that remain hidden when aggregate variables are used, and accounts for a within country-industry variation of the effects of democracy. Second, I combine that disaggregated approach with the old suspicion that democracy needs a prerequisite to exert a positive effect on economic performance (Lipset (1959); Przeworski & Limongi (1993)). I explore a technological channel through which a beneficial effect of democracy in technologically advanced industries is expected. By so doing, I reconcile two competing results in the relevant literature regarding a positive aggregate effect of democracy (Acemoglu et al., 2014) and a technologically-conditioned disaggregated effect (Aghion et al., 2009). Finally, I also contribute to the field by offering an extensive comparative analysis of the conceptualisations and consequent measures of democracy available in the 7 This debate is extended in the Data section of this paper 8 However, Haber et al. (2003) deny the political stability argument by considering that the empirical evidence is not robust and thus there is no clear causal link. 9 This disaggregated, manufacturing approach is inspired by the works of Rajan & Zingales (1998) and more directly by Aghion et al. (2009), which I supplement by using 61 ISIC industries rather than their 28 ISIC industries. 4

80 85 90 95 literature. I not only employ well-known indices of democracy, namely the Boix index (Boix et al., 2013), the extended Democracy-Dictatorship (DD) 10 index (Cheibub et al., 2010), the Polity2 index (Marshall et al., 2014) and the Vanhanen index (Vanhanen & Lundell, 2014), but also use more recent, more technically refined measures such as the Polyarchy index from the Varieties of Democracy (V-Dem) index as defined in Teorell et al. (2016) and the Support Vector Machine Democracy Indicator (SVMDI) constructed by Gründler & Krieger (2016). This paper offers reasons to believe that the effect of democracy on the economic performance of manufacturing industries is influenced by the level of technological development at which industries operate. Advanced industries benefit from political regime changes towards more democratic institutions. By contrast, the same political changes harm the economic performance of backward industries. Additional results presented in this research do not support the contention that more economic freedom and better regulation are conductive to higher manufacturing growth rates. The paper is organised as follows: Section 2 offers theoretical reasons and extant empirical findings for a technologically-conditioned effect of democracy on growth. Section 3 explains the data employed and introduces the six measures of democracy that are employed in the econometric analysis. Section 4 specifies a baseline model and shows the main results. Section 5 presents various robustness checks. Section 6 concludes. 2. The interplay between democracy and technology 100 This paper provides evidence of a technologically conditioned effect of democracy on the economic performance of manufacturing industries. Drawing on previous scholarship on democratic institutions, regulation and economic freedom, I offer some arguments here to explain that striking result. Democracy has long been associated with lower barriers to market-entry and more economic freedom. Djankov et al. (2002) show that more democratic, limited governments have lighter regulations by addressing two competing theories on economic regulation; the public interest 10 Originally developed in Alvarez et al. (1996). 5

105 110 115 120 125 130 (Pigou (2013)) and public choice theories (Tullock (1967); Stigler (1971);Peltzman (1976)). The public interest theory argues for the ability of government regulation to avoid market failures such as fly-by-night operators and externalities and consequently, to ensure minimum quality standards of products and services. The public choice theory offers a less benevolent view of governments and is further divided into two standpoints: the first is the capture view (Stigler, 1971), which suggests that industries close to political elites will control marketentry and regulatory policy. Thus, incumbent industries use regulation to perpetuate their privileged position to the detriment of new, probably more innovative, industries. In this line of thought, incumbent industries have less incentives to innovate because the favourable policies derived from their political power, and a substitution effect arises between political power and the need for innovation. The second standpoint of the public choice theory of regulation is the tollbooth view (Shleifer & Vishny, 2002), which sees regulation as a clear sign of the rent-seeking attitudes of politicians and bureaucrats. Regarding the link between democracy and economic freedom, the literature offers a direct channel through which political liberalisation is conducive to economic liberalisation. De Haan & Sturm (2003) use both cross-section and panel data techniques along with different measures of democracy to show that the increase in economic freedom between 1975 and 1990 in developing countries was driven to certain extent by the level of political freedom. A straight-forward association of better regulation and economic freedom with a pro-growth effect is also found in the literature. De Haan & Sturm (2000) show that although economic freedom does not seem to affect the steady state level it does increase the speed at which countries attain their steady states. An overall positive direct association between economic freedom and economic growth is suggested by Doucouliagos & Ulubasoglu (2006). This evidence supports the idea of a channel through which democracy -in its impact on economic freedom and regulation- fosters economic growth. I examine these landmark findings from the literature regarding the interplays between democracy, regulation and economic freedom with a disaggregated approach to study how different levels of technological development may display different effects within countries. I surmise that the aftermath of rising levels of democracy is specifically beneficial for technologically advanced industries for two reasons. First because advanced industries are able to 6

135 140 145 150 155 adapt to more competitive and innovative markets and second because of the promotion of new technologies that democracy entails 11. In this paper I note that the degree of democratisation shapes the intervention of the government in the market. This is of crucial importance for all economic activities in general and for manufacturing industries in particular because policies on market-entry, adoption of new technologies or trade might be determinant for their growth (McGillivray, 2004). The argument offered in this paper is close to that of Aghion et al. (2009), who also surmise a technologically-conditioned effect of democracy on manufacturing industries. Freedom of entry is determinant for sectors close to the technological frontier since, as suggested in Aghion et al. (2008), entry of new firms and competition spurs innovation at high levels of technological development but discourages innovation in backward sectors. An important feature of political regimes concerning investments and new technologies is worth noting at this point in the argumentation. New technologies that are not fully implemented depend on investments the potential of which is difficult to evaluate 12. Consequently, such investments are made based on hunches about the future (Boschini (2006); Hodgson (2015)). Through their positive effects on political stability (Alesina et al. (1996);Rodrik (2000)), democracies promote a better environment for technological innovation. The argument in Acemoglu (2008) 13 that democracy is better able to promote new technologies than oligarchies applies here. Due to its positive effects on economic freedom 14, democracy is less likely than other, less free regimes to block the entry of industries with comparative advantages in new technology. New investment opportunities (e.g. new technologies) are 11 All the proxies of democracy used here are negatively correlated with the measure of technological development. See Sclove (1995) or Milner (2006) for an extension of this idea. 12 An interesting case is made in Boschini (2006) regarding the ability and incentives of governments to promote industrialisation. In that regard, she brings to the debate the distribution of wealth, skills and political power, and compares the cases of autocracies and democracies, where the decisive agents may vary in education and where running elites may strategically extend the franchise. 13 He posits the case of a trade-off between redistribution taxes and market-entry barriers in alternative political regimes. 14 Dawson (1998) also finds that economic freedom fosters investment as well as features commonly associated with democracy such as political and civil rights. 7

160 165 170 175 180 reduced when high market-entry barriers are faced and property rights are not secure and properly enforced 15. Autocratic regimes may be more likely to keep political power in the hands of producers which ex-ante have comparative advantages in political terms. Democracies prosper because they reap the benefit of adapting to new technologies, which allows more efficient production systems and reduces inequalities (which are potential sources of conflict) and economic instability. I implicitly assume that before changes towards more democratic institutions take place backward industries enjoy some measure of favouritism on the part of the government maybe in the form of market entry barriers that limit competition and the entry of more innovative but politically weaker industries. However, as pointed out in Acemoglu (2008), protection becomes costly as the economy approaches the world technology frontier, and entrepreneur selection based on efficiency becomes more important. My main result supports that view so long as I show that political regime changes towards democracy have a pro-growth effect on advanced industries in technological terms. By contrast, the same institutional changes harm backward industries. Finally, an illustrative case of how political regimes may affect the entry of new firms and consequent industrialisation and the adoption of technology is the Porfirio Díaz dictatorship in Mexico from 1876 to 1911 16. As cited in Acemoglu (2008): Manufacturers who were part of the political coalition that supported the dictator Porfirio Diaz were granted protection, everyone else was out in the cold. (p. 18 in Haber et al. (2003)) Restrictions on imports and regulatory policies without legislative approval and custody over banking services and administering of federal taxes to the creditors of the government were common practices in early 20th century Mexico (Maurer & Gomberg (2004); Haber et al. (2003)). The Porfiariato was able to specify and enforce property rights, which were private 15 As in Haber et al. (2003), one might think in terms of selective property rights rather than low property rights, since properties are secure for some bit not for those not favoured by the government. An interesting point in their theory is that property rights are regarded as private rather than public goods. 16 Haber et al. (2003) and Maurer & Gomberg (2004) are two comprehensive studies of the Mexican economy and industrialisation process during the dictatorship of Porfirio Diaz. 8

185 rather than public goods. Certain manufacturing industries were favoured by less democratic governments, whereas others did not enjoy such luck. Linking the case of the Profiriato to current debate is not far-fetched, as Robinson (2003) finds similarities between the contemporary Russian economy and that of Mexico in 1900. Indeed, some of the countries in the sample employed in my research score lower on democratic values than early 20th-century Mexico 17. 3. Data 190 195 200 I employ an unbalanced panel dataset covering 61 International Standard Industrial Classification (ISIC) manufacturing industries from 72 countries over the period 1990-2010. The dependent variable is the annual growth rate computed by log-differencing the output 18 of 61 International Statistical Industrial Classification (ISIC) manufacturing industries 19. Appendix A shows the 61 ISIC industries and the countries in the sample. Disaggregated data regarding manufacturing industries are collected from the United Nations Industrial Development Organisation (UNIDO) Industrial Statistics Database at the 3-digit level of ISIC (INDSTAT4) 20. Previous studies using a disaggregated manufacturing data offer comprehensive arguments in favour of this approach and the employment of the UNIDO Industrial Statistics Database (Rajan & Zingales (1998); Imbs & Wacziarg (2003); Aghion et al. (2009); Vaz & Baer (2014)). The three main explanatory variables are technological development, democracy and the interaction between them. Table 1 shows the descriptive statistics of the variables included in the empirical analysis, and the appendix lists the 72 sample countries and 61 ISIC industries. 17 Eritrea and Qatar obtain the lowest scores in the Polyarchy V-Dem index of democracy for the whole period 1990-2010: 0.085 in the case of Mexico 1990 and an average value of around 0.07 and 0.09 respectively for Eritrea and Qatar. 18 The same empirical analysis using value added growth rate as the dependent variable yields similar results and is available from the author. 19 Prior to log-differencing, output is converted into real terms. 20 UNIDO is one of the most comprehensive data sources in the context of manufacturing industries, although as stated in Aghion et al. (2009) it relies on national statistics that tend to be especially noisy. 9

3.1. Distance to the World Technology Frontier I use the concept of distance to the world technology frontier (henceforth WTF) defined in Acemoglu et al. (2006) to measure the degree of technological development of industries. The measure is expressed in the following formula: Distance ict = 1 log(v A ict /EMP ict ) log(max c (V A ic t/emp ic t)) i = industry; c = country; t = year 205 210 215 220 Where V A ict and EMP ict stand respectively for value added and the number of employees, max c (V A ic t/emp ic t) refers to the world maximum V A/EMP ratio of industry i. Distance to the world technology frontier is a yearly assessment of how far each industry is to the industry that marks the technology frontier, i.e. the country that operates with the highest ratio of value added per worker for each industry. Note that each industry is compared with its counterparts in the rest of the world and it is not compared to the value added per worker ratio of other industries either within a country or across countries. The variable ranges between zero (meaning that the industry operates with the most advanced technology) and one (meaning that the industry operates with the lowest value added per worker ratio). Appendix B contains a histogram of the distribution of the distances to the WTF for the whole sample (Figure B1) and 61 histograms, i.e. one for to each industry (Figures B2-B4). The sample industries that show the highest average levels of technological development (i.e. are closest to the WTF) are the processing of nuclear fuel, electricity distribution and control apparatus, insulated wire and cable and aircraft and spacecraft. At the other end of the technological spectrum are tobacco products, coke oven products, basic chemicals and motor vehicles. The countries that are located on average closest to the WTF are the USA, Japan and New Zealand, whereas Madagascar, Vietnam and Georgia are the furthest away. In terms of the technological development of each industry across countries, coke oven products, tobacco products and textiles show the greatest variation. Processing of nuclear fuel and refined petroleum products show the least distance variation, partly due to the fact that these industries operate in few countries. 10

3.2. Democracy: conceptualisation and measurement 225 230 Although this is a fruitful area of research, the literature available fails to show a consensus on the conceptualisation and measurement of democracy 21. There is however a predominant consensus that it is preferable to quantify the degree rather than the stock of democracy (Grundler and Krieger, 2016). There are three major debates surrounding the measurement of democracy, regarding its constituent components (Boix et al., 2013), the numerical form of the measure and how different components are combined into a single measure. These debates and the wide variety of conceptualisations of democracy have produced two broad types of definition, known as minimal and extensive 22 definitions. Different definitions mean alternative combinations of regime components of democracy and different forms of aggregation, so data in terms of countries and periods of time vary from one index to another 23. 235 240 This paper employs six different measures of democracy which are conceptually different, use alternative means of aggregation and focus on different political attributes of regimes 24. I thus supplement dichotomous categorisations such as that of Boix et al. (2013) and Cheibub et al. (2010) with polychotomous measures, and minimal definitions with maximal ones. I benefit from the availability of more data in terms of countries and periods of time. Ultimately, this approach serves as a robustness check on the results reported in this paper. The Boix index applies a minimal conceptualisation of democracy that focuses on two main attributes: contestation and participation. It is a dichotomous measure in which zero means no democracy and one is democracy 25. 21 The case that conceptual disagreement underlies the lack of conclusive results on the growth effect of democracy can be made straightforwardly (Sirowy & Inkeles (1990); Gründler & Krieger (2016)). 22 This is also called the maximalist conceptualisation (Teorell et al., 2016) 23 For a deeper discussion see Gründler & Krieger (2016) 24 Tables C1 and C2 in the appendix include correlations across the six measures considered and information on how they are constructed. 25 The Boix index considers a political regime as democratic if 1) the executive is directly or indirectly elected in popular elections and is responsible either directly to voters or to a legislature; 2) the legislature (or the executive if elected directly) is chosen in free and fair elections; and 3) a majority of adult men have the right to vote. 11

245 250 255 260 Another dichotomous measure (with zero meaning autocracy and one democracy) included in this analysis is the Cheibub index (Cheibub et al., 2010), which is an extension of the Democracy-Dictatorship (DD) index constructed by Alvarez et al. (1996). The Cheibub index is a minimal conceptualisation of democracy and is based on office and contestation as determinants of political attributes of democracies 26. The popular Polity2 index of the Polity IV Project, provided by Marshall et al. (2014) 27, is also included here. Polity2 is a polychotomous, categorical variable ranging from -10 to 10 which modifies the combined annual Polity score by applying a special treatment to fix instances of political interregnum, transition or interruption. This is an extensive conceptualisation of democracy that differs from those above in terms of the underlying definition of democracy, the nature of the data and the type of measurement and aggregation. Although widely used, the Polity 2 index is surrounded by a great deal of controversy: for instance Treier & Jackman (2008) cast doubts on the precision of the Polity IV measures. Vanhanen & Lundell (2014) offer a minimal conceptualisation of democracy. The Vanhanen index is polychotomous and is the outcome of multiplying two electoral attributes: competition (the percentage of votes going to the largest party) and participation (voter turnout) 28. Nevertheless, as stated in Gründler & Krieger (2016) and in Munck & Verkuilen (2002), the Vanhanen (2000) measure suffers from a lack of theoretical justification of the two attributes that make up the index. The Vanhanen index ranges from 0 to 45.6 in the dataset. 265 Because of the criticism levelled at the measures above, I include two additional indices 26 In order for a country to be a democracy as envisaged by Cheibub 1) the chief executive must be chosen by popular election or by a body that was itself popularly elected; 2) the legislature must be popularly elected; 3) there must be more than one party competing in the elections; 4) an alternation in power under electoral rules identical to the ones that brought the incumbent to office must have taken place. 27 Due the similarities and high correlation between the Polity2 index and the Freedom House (FH) index I only include the former. However, the same results are reached using the FH political rights and civil liberties indices, which are available from the author. 28 The Vanhanen index results from multiplying participation ( the percentage of the total population who actually voted in the election) by competition (the percentage of votes gained by the smaller parties in parliamentary and/or presidential elections) and dividing by 100. 12

270 275 280 285 in my analysis here that correspond to maximalist conceptualisations of democracy, and strive to include novel techniques for computing and scoring political regime features. The Polyarchy measure of Teorell et al. (2016), based on Varieties of Democracy (V-Dem) data, focuses on five categories: elected officials, free and fair elections, freedom of expression, associational autonomy and inclusive citizenship. This index is inspired by the works of Schumpeter (2013) and Downs (1957). More specifically, as these authors argue, the index anchors core institutional guarantees in Dahl s (1956, 1989) concept of polyarchy 29. Consequently, the conceptualisation of the V-Dem data views elections and the institutions that uphold the democratic qualities of elections as the core of the concept of democracy. Finally, I include the Support Vector Machines Democracy Index (SVMDI) provided in Gründler & Krieger (2016). This measure is a response to the low level of sophistication in the aggregation of components shown by the above-mentioned democracy indices. Gründler & Krieger (2016) translate the aggregation of political attributes issue into an optimisation problem by using Support Vector Machines and machine learning algorithms. The calculation of the SVMDI index involves an unambiguous characterisation of highly democratic and highly autocratic political regimes. Based on that characterisation, the authors employ the Support Vector regressions (a mathematical algorithm for pattern recognition), along with eleven observable political variables 30 to provide a continuous scale of democracy in the [0,1] interval. This novel approach to measuring democracy enables the authors to achieve highly accurate classifications. Nevertheless, as the authors cautiously argue, the accuracy of the index relies on the input variables. 4. Econometric specification and results This section sets out a panel data model and presents within-group estimates of the effect of democracy and technology on manufacturing growth rates. Equation (1) shows the 29 As stated in Teorell et al. (2016), minimalist conceptions of democracy correspond to empirical reasons rather than being settled by a definitional fiat, so maximalist conceptualisations might be better. 30 The eleven variables all together account for four broader aspects: political participation, independence of the judiciary, civil liberties and freedom of the press. 13

290 baseline model. log(y ict ) = β 0 + β 1 Distance ict + β 2 Democracy ct + β 3 Distance ict Democracy ct + X ictα + Z ctω + u ict (1) log(y ict ) = log y ict+1 log y ict u ict = γ ic + δ t + ε ict i = industry; c = country; t = year where Y ict is the output of industry i in country c in year t. The dependent variable is defined as the log-different of t + 1, so all the explanatory variables are one year lagged with respect to the dependent variable in order to solve possible endogeneity issues. Distance ict refers to distance to the most highly developed industry in the world, so higher values indicate less technological 295 development. Democracy ct can be any of the six measures mentioned above. Along with their interaction, these two covariates are the main regressors of the model. Table 2 shows within-group estimates of the model in equation (1), using in each column respectively the indices of Boix, Cheibub, Polity2, Vanhanen, Polyarchy and SVMDI. The interaction between democracy and distance to the WTF is statistically significant over the six models, thus 300 providing clear evidence of a technologically-conditioned effect of democracy on manufacturing growth rates. The distance to the WTF is associated with a statistically significant, positive effect (Columns 1-6), which might be showing a convergence of those industries far from the frontier. This positive effect of distance is also found in Aghion et al. (2009), and may be a sign that a catch-up effect is at work 305 in technologically backward industries (Barro & Sala-i Martin (1995); Bloom et al. (2002)). The democracy variable is associated with a statistically significant and positive coefficient, except in the Cheibub and Polyarchy 31 indices. The marginal effects of democracy vary for different levels of technological development (measured here by the distance to the WTF). The interaction between democracy and distance is statistically significant and negative in all six models estimated 310 in Table 2. Thus, the marginal effect of democracy is contingent on the level of technology (at the value added per worker ratio) at which industries operate. Considering the model in Column (1), 31 The Polyarchy index is negative but not statistically significant. 14

a change in the Boix index from zero to one is associated with a 25 percentage points (henceforth, pp) increase in manufacturing growth rates in industries located at the WTF. As industries get further from the frontier the positive effect of the Boix index vanishes. As a matter of fact, for 315 industries with distance scores higher than 0.15 32 a turn towards democratic in the political regime harms manufacturing growth. Similar results are found in the rest of the Columns in Table 2. To illustrate this contingent effect, Figures 1 and 2 show the marginal effect of democracy depending on the values of distance to the WTF considering those indices that are continuous and statistically significant at 1%, namely the Vanhanen and SVMDI indices 33. 320 Equation (1) includes the term γ ic that captures time-invariant effects that may arise because of countries idiosyncrasies or industrial peculiarities, whereas δ t shows time effects, which that also can be an unobserved source of heterogeneity. To consistently estimate their effects of, the withingroup estimator yields consistent estimations by getting rid of the individual and time fixed effects 325 330 335 by demeaning the equation in (1). Strong exogeneity of regressors isare assumed, butalthough this assumption is relaxed in the following section, in which a dynamic model is estimated using the system-gmm technique. A clustered-robust variance matrix estimator at country-industry level (White (1980);Newey & West (1987)), which is consistent in the presence of any kind of heterosckedasticity or serial correlation, is calculated. Within-group correlation is allowed to take any form but not across groups. The term X ict includes three covariates at country-industry level. I attempt to control for the lobbying ability of the industries and unequal political power that might be conductive to favourable policies (such as market-entry barriers, subsidizes or import taxes), in three different ways 34 and the results are displayed in Table 2. First, I include industry output as a share of the total manufacturing sector output at the 61 ISIC industry-level (Output Share) to capture how economically inuential and nancially powerful each industry is relative to other industries. The weight of indus- 32 The average distance to the WTF in the sample is 0.16. 33 Industries with distance scores higher than 0.25 and 0.37 respectively for changes in the Vanhanen and SVMDI indices are harmed by increasing levels of democracy. 34 The logic of including these three variables at country-industry level is based largely on the idea that certain characteristics of industries might effect the economic performance of industries and simultaneously have political and electoral relevance. In this sense, I consider workers and industries as potential electoral targets of certain policies. I acknowledge in a sense the risks associated with electoral democracy regarding the passing of popular policies at the expense of growth-enhancing but less popular policies. 15

tries in the total manufacturing sector is statistically significant and is associated with a negative effect on growth rates, which may reinforce the convergence hypothesis. Second, I isolate the effect of the size of each industry (in terms of employment) on growth by controlling for the proportion of the total population employed in each industry (Employment). The effectiveness of lobbying and 340 the electoral power of industries, and thus the consequent effects on growth, might be driven by the number of workers (voters) that an industry employs. This idea is partially based on the collective action and lobbying literatures 35. Thirdly, I include the ratio of the number of establishments in each industry to the total population 36 (Establishments). These two covariates are associated with a negative coefficient but are not statistically significant. 345 The six models explained above include both country-industry level control variables and country level variables to reduce any problems of bias due to omitted variables. All together they account for economic, socio-demographic and politico-institutional features that have been shown by the literature to alter the effect of democracy on growth. Doucouliagos & Ulubaşoğlu (2008) show how the literature on the effects of democracy on growth 350 has developed augmented neoclassical growth models that account for indirect effects through hu- man capital, trade openness and economic freedom, among other channels. The term Z ct in equation (1) accounts for such claims because they can be essential in isolating the effect of democracy and its interaction with technology on manufacturing industries. Regarding economic country-level covariates, I first control for the initial total output of the sector as a whole in logarithms (Total 355 output (log)) to capture potential shocks that could impact the manufacturing sector as a whole. This variable is statistically significant and is associated with a negative coefficient in all the specifications shown in the robustness check section. This negative sign could be explained by the findings in Imbs & Wacziarg (2003) regarding the U-shaped relationship between sectoral concentration and economic development. 360 The growth rate of gross domestic product per capita (GDP pc Growth Rate), obtained from Penn World Tables (mark 7.1), is included to control for the effect of the economic development of countries on manufacturing growth. All the models in Table 2 associate the GDP per capita growth rate with a positive effect, but it is only statistically significant for the Polity2 and Polyarchy models. 35 For further insight on this thought see Olson (1965); Becker (1983) and Esteban (2001). 36 This argument is supported by the idea that occupation matters to voting political participation (Carnes (2012)) and that workplaces actually work as political agoras. 16

The exchange rate (Exchange Rate) is included to rule out any possible effect of exchange rate 365 370 375 380 385 movements on the main results. The exchange rate seems to exert no effect on manufacturing industry growth rates, and it is only statistically significant in the case of the Vanhanen index (Column 4). The importance of trade in GDP (Openness) as shown by the World Bank is included in all the estimates. The results in Table 2 show that the recurrent finding regarding the effect of trade on aggregate growth also applies to the context of manufacturing growth. In line with previous results in the literature (Frankel & Romer (1999); Dawson (1998)), I find that trade openness exerts an only moderately statistically significant positive effect on income. These results support those of Rodrik et al. (2002), who suggest that trade has little or no impact on growth once institutions are controlled for. As pointed out above, regulation is crucial to understanding how the growth effect of democracy interplays with technology (by lowering market-entry markets and providing a better economic environment regarding property rights and secure contracts which - as North (1990) suggests - go hand to hand with political rights). Additionally, this research is also interrelated with economic freedom insofar as regulatory areas are embedded in the concept of economic freedom. Indeed, features that were formerly attributed to political freedom are nowadays defined as economic freedom 37 (Gwartney & Lawson (2003); Doucouliagos & Ulubaşoğlu (2008)). Previous attempts to test the hypothesis of a technology-contingent effect of democracy on manufacturing growth were based on cross sectional data and account for market entry barriers as the number of procedures, as in the World Bank Doing Business project 38. I depart from that literature by employing the regulation indicator provided by the Fraser Institution as the fifth component of their economic freedom index 39 published in the annual report Economic Freedom of the World (EFW) Gwartney et al. (2016)). The Fraser Institute proxy of regulation outperforms the other alternatives available in two main ways. First, the Fraser Institute index of regulation benefits from a multifaceted concept of regulation by considering regulatory standards of credit market, labour market and busi- 37 As defined by Gwartney & Lawson (2003), economic freedom expresses a variety of policies consistent with i) smaller governments; ii) secure property rights; iii)access to sound money; iv) freedom of exchange; and v) freer credit and labour markets. 38 Aghion et al. (2009) replicate the cross-section estimates of Djankov et al. (2002). 39 The index measures the degree to which the policies and institutions of countries are supportive of economic freedom. 17

390 ness environment 40. Second, other regulation measures used as components of economic freedom indices (such as the Heritage Foundation/Wall Street Journal 41 ) use rankings rather than scores of business regulation, and thus their use would be inaccurate in a panel data regression context. All estimates in Table 2 include the regulation indicator of the Fraser Institute (Regulation) as a score out of ten where higher values mean freer and more business-friendly environments. The 395 regulation covariate is strikingly associated with a negative effect on manufacturing growth and is generally statistically significant across different democracy indices. Once the effect of better quality regulation is accounted for, this may show that certain manufacturing industries are better off in lower quality regulatory environments, although the overall effect of greater freedom is positive for the economy as a whole. This finding reinforces the disaggregated approach taken here since 400 it uncovers conflicting results that are not apparent in studies that use aggregate data (Djankov et al. (2006); Jalilian et al. (2007)). The baseline model in (1) also controls for population 42 (expressed in logarithms) (Population (log)) and human capital 43 (Human Capital), which are thought to play a role in growth literature and electoral studies (Lipset (1963); Lucas (1988); Glaeser et al. (2004)). Table 2 shows that these two 405 covariates are generally associated with negative coefficients, but they are not statistically signif- 410 icant in some of the models (Column 1 in the case of population and Columns 2-6 in the case of human capital). The results here are consistent with those of Benhabib et al. (2013), though the dependent variable in their case is expressed in aggregate terms. Finally, the number of consecutive years of the current regime type (Age of Democracy), as provided by Boix et al. (2013), is also controlled for. As shown by Persson & Tabellini (2009), the age of democracy is crucial to the dynamics of economic and political regime changes. Democratic capital, in Persson and Tabellinis parlance, is an essential determinant for growth through its impact on stability. The age of democracy is also included as a response to the correlation found in Clague et al. (1996) and Gerring et al. (2012) between the age of democracy with secure property 40 See Gwartney et al. (2016) for the methodology and see Dawson et al. (2007) for a comprehensive explanation and a survey on the use of the EFW index. 41 I chose the regulation index of the Fraser Institute because it is more precise and more transparent in its methodology than the HF/WSF alternative (Gwartney & Lawson, 2003)) 42 Collected from World Data Bank, December 2014 43 Human capital is measured by the average years of education of citizens aged over 15, obtained from the V-Dem dataset. 18

415 rights and contracts, which are the ultimate determinants of economic development. The age of democracy is always associated with a negative effect, but is statistically significant in only three of the six models shown in Table 2. The main results provide empirical support for the idea that democracy is conditional for the economic performance of industries. Additionally, these results offer two conflicting findings regarding 420 the up-to-date understanding of the growth effects of regulation and democratic experience. Both regulation quality and the age of democracy seem to exert a negative effect on manufacturing industries. These findings led me to take the disaggregated approach adopted in this paper, which enables patterns to be uncovered that differ from those revealed by analysis conducted at an aggregate level. Although it would be interesting for its own sake, going further into the effect of 425 regulation and duration of democracy lies beyond the scope of this paper, since for the purpose of the hypothesis explored these covariates are treated as control variables. 5. Robustness analyses This section explores whether the results above are sensitive to the model specied and to the 430 estimation strategy. Two main alternatives to the baseline model are offered. First, using withingroup estimates and focusing on the SVMDI measure of democracy 44, I change the specifications of the model in (1). Second, I employ the Generalised Method of Moment (GMM) as developed in Arellano-Bover (1995) and Blundell-Bond (1998) to estimate a dynamic panel approach. Overall, the results provided by these two different robustness checks confirm that higher levels of democracy are conducive to higher growth rates in technologically advanced industries. However, the same 435 democratic changes harm the economic performance of backward industries. 5.1. Alternative static panel data models Table 3 displays within-group estimates for five alternative models. I first estimate the interaction democracy and distance to the WTF along with its constituent terms (Column 1). The exclusion of control variables from the analysis results in a reduction in the marginal effect of 440 SVMDI from 1.03 (baseline model, Table 2 Column (6)) to 0.6. The interaction between democ- racy and technology drops in Column (2), with the coefficients of both terms showing the same 44 I focus on this proxy of democracy due to the novelty and computational sophistication of the index. Robustness checks using the alternative measures yield similar results and are available from the author. 19

sign and being statistically significant, although these results should be taken with caution due to potential problems caused by the omission of relevant variables. Columns (3) and (4) show first that the main results of the paper are not driven by regulation. 445 450 455 Second, they show that other measures of good regulation are also associated with a negative effect on the manufacturing performance of industries, though it is not statistically significant. In column (4) I use the regulatory quality of the World Bank Worldwide Governance Indicators (WGI) 45 project, as defined in Kaufmann et al. (2011), as an alternative to the regulatory measure of the Fraser Institute. The last column (5) accounts for regional effects on both democratisation and economic growth and estimates the same model as in Table 2 Column 6 but using data on OECD economies only. The results are consistent with the above finding of a technologically-conditioned growth-effect of democracy on manufacturing industries. However, with the sample of OECD countries the distancedemocracy interaction is associated with a greater effect (from -2.7 in Table 2, Column 6 to -9.8 in Table 3, Column 5). Regulation is associated with a positive but not statistically significant effect. 5.2. Growth dynamics I now specify a dynamic panel data model in equation (2), where for the sake of comparison 460 with Table 2 the same six measures of democracy are employed. I thus explore whether the dynamics of growth might be at work in the effects of democracy and technology on manufacturing performance. The AR(2) p-value suggests that when the Boix index is used there is in fact secondorder correlation, so the second lag of the dependent variable is included as a regressor (Table 4, Column (1)). log(y ict ) = φ log(y ic,t 1 ) + β 1 Distance ict + β 2 Democracy ct + β 3 Distance ict Democracy ct + X ictα + Z ctω + u ict (2) 45 The WBGI index on regulation captures perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development (ranging from -2.5 to 2.5). 20