WPS8428 Policy Research Working Paper 8428 Political Connections and Firms Network Dimensions Maurizio Bussolo Simon Commander Stavros Poupakis Europe and Central Asia Region Office of the Chief Economist May 2018 Policy Research Working Paper 8428 Abstract Business and politicians’ interaction is pervasive but has short path length. Matching the data to firm-level infor- mostly been analyzed with a binary approach, i.e. either mation, the paper examines the association between being a firm is connected to a politician or not. Yet the network connected and firm-level attributes. The originality of the dimensions of such connections are ubiquitous. This paper analysis is to identify how location in a network, including uses use a unique data set for seven economies that documents the extent of ties and centrality, is correlated with firm scale politically exposed persons and their links to companies, and performance. In a binary approach, such network char- political parties, and other individuals. The data set is used acteristics are omitted and the scale and economic impact to identify networks of connections, including their scale of politically connected business may be significantly mis/ and composition. The analysis finds that all country net- under-estimated. By comparing the results of the binary works are integrated having a Big Island. They also tend to approach with the network approach, the paper also assesses be marked by small-world properties of high clustering and the biases that result from ignoring network attributes. This paper is a product of the Office of the Chief Economist, Europe and Central Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at mbussolo@worldbank.org, scommander@alturapartners.org, s.poupakis@ucl.ac.uk. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Political Connections and Firms: Network Dimensions ∗ Maurizio Bussolo Simon Commander Stavros Poupakis May, 2018 JEL codes: L14; L53; P26 ∗ Author contact details: Maurizio Bussolo (World Bank) mbussolo@worldbank.org, Simon Comman- der (Altura Partners, IE Business School and IZA) scommander@alturapartners.org, Stavros Poupakis (University College London) s.poupakis@ucl.ac.uk. The authors thank Wenyu Zhu and Esther Bartl for research assistance and Boyko Amarov, Beata Berkovics, Adrian Hernandez, Marta Kahancova, Leo Rollinger, Alina Scherbakova and Roxana Voicu-Dorobantu for help with the country-level data matching, as also Katya Ushakova. Thanks to Luis-Felipe Lopez-Calva, Ishac Diwan, Bob Rijkers and Hans Timmer for very helpful comments, as well as seminar participants at the World Bank and Nuffield College, Oxford. 1 Introduction The interaction of firms and politicians and the outcomes – notably in terms of rent seeking – that result has been documented across a wide range of political and economic systems.1 State-owned enterprises have been found to be particularly responsive to politi- cians, although family and other network ties to politicians can deliver advantage to private companies. Regulatory privilege has been found to be an important way in which profits and market share can be boosted. The channels include preferential access to credit, assets and infrastructure (Khwaja and Mian, 2005; Diwan et al., 2015). The flow of transactions may not be one-way, as firms may also favour politicians, including by creating employment at opportune moments in the electoral cycle (Bertrand et al., 2004). These multiple modes through which connections are realised is further qualified by differences across countries in political and other institutions. High corruption levels and weak institutions tend to be asso- ciated with more politically connected firms (Faccio, 2006), while, not surprisingly, political connections tend to be endemic in autocratic regimes. The focus in this paper is on how businesses and politicians interact in a set of transition economies, specifically Bulgaria, Hungary, Romania, the Russian Federation, Serbia and the Slovak Republic. Existing studies already indicate that political influence over business and commercial decisions has been – and often continues to be – significant, not least because of the legacy of large public sectors and their subsequent unwinding, including through discre- tionary asset sales and privatization (Commander, 2016). In addition, public contracting in the region has tended to be associated with the application of preferential treatment or fa- voritism. Lack of integrity in public procurement practices not only appears widespread but has also been explicitly linked to the political cycle (Palguta, 2014; Doroftei and Dimulescu, 2015; Koren et al., 2015). As is the case more generally, autocratic regimes in the region ap- pear particularly prone to cronyism with widespread resort to connections in the allocation of assets and contracts. We also include a Western European comparator, Spain, in the analysis. While the latter sits higher on most rankings of competitiveness and/or governance than the transition economies and has a longer history of democratic institutions, scores for diversion of public funds, favoritism by government officials or irregular payments by firms, show Spain to be comparable to some transition economies (World Economic Forum, 2014). Existing research on political connections has mostly concentrated on establishing the in- cidence and consequences of connections, generally at the level of the connected firm. Some papers have also tried to establish the welfare consequences (Cingano and Pinotti, 2013). 1 See, inter alia, from a large literature, Faccio (2006, 2010); Fisman (2001); Boubakri et al. (2008); Johnson and Mitton (2003). 2 Yet, it is evident that connections are rarely, if ever, simply binary in nature. Their network properties tend to be important. Indeed, the ubiquity of networks in the social and economic life of humans is now well understood. Networks summarize sets of relationships between agents, often with particular structures. As such, they tend to shape or reflect specific opportunities that link individuals, companies and other entities, such as political parties. Available evidence indicates that network structures vary significantly - some have a large number of ties or edges while others have higher centrality secured by bridges through which those ties or edges pass. Some networks may be more hierarchical than others, maintaining more edges and higher centrality than so-called distributed networks. Dense networks, de- fined by high clustering and a low number of edges, tend to generate bonding capital that can be useful in leveraging valuable assets or connections and hence in ensuring coopera- tion.2 By contrast, more diffuse networks with low clustering and many edges (high degree) tend to be more suited to providing information or access to information; bridging capital, in other words (Granovetter, 1973).3 A network may also have small world properties where locally dense clusters of nodes are connected to many others through a small number of bridging connections (Humphries and Gurney, 2008). For political connections with firms, dense networks aimed at maximizing cooperation with peer pressure might be expected to predominate. Despite their evident relevance, curiously little attention has been paid to mapping the format and nature of connections between firms, politicians and political parties and to understanding how different types of networks may affect outcomes, including the benefits. Our paper uses a unique dataset that documents politically exposed persons (PEPs) and their links to firms – private and state-owned – as well as to political parties and other individuals. We also are able to match these individuals through ownership and shareholder status to firm level balance sheet and performance data. Not only does this particularly rich dataset document political connections across countries, but it also allows identification of the types of connections that link individuals, parties and companies at a country level. As such, our main objectives are to understand how connections are actually configured, while also analyzing how different types of connections may have an impact on firms. The existing literature has concentrated on identifying political connections in a binary manner and then exploring that association – sometimes in a causal way – with outcomes. What is particularly original in our paper is that connections are analyzed for their network properties, including the type and location. The paper is organized as follows. Section 2 reviews the main findings of the large 2 Dense networks can be composed of few agents closely linked or many agents with small average distances between them. 3 Bonding and bridging capital are terms employed by Putnam (2000). 3 literature on the political connections of companies. Section 3 describes the datasets that are used. Section 4 examines the network features of connections, picking out differences across countries. Section 5 then examines the association between firm characteristics and performance, using a number of indicators, and being politically connected. Although the base estimates – as is common in the literature – relate variables to whether or not a firm is connected, we augment the analysis by taking account of the nature of those connections and their network properties, such as the extent to which a connection is strategically important as measured by the extent to which a node lies between other nodes. Section 6 concludes. 2 Consequences of connections Being politically connected has mostly been viewed through the lens of the benefits that such connections confer. The scale of benefits is generally calibrated relative to not having connections. In the broad literature, the focus has mostly been on indicators such as growth in revenues, leverage, taxation, subsidy receipt, productivity and accounting performance, as well as stock market valuation. When researchers have matched information linking politically connected persons with firm or establishment level data, the evidence across a variety of sectors and economies suggests that political connections can – but not always – support superior financial perfor- mance. This may not necessarily be associated with productivity gains. Rather, higher mar- ket shares and attenuation of competition can be one outcome (Cingano and Pinotti, 2013). Connections may primarily facilitate survival and employment but not growth in productiv- ity and innovation (Akcigit et al., 2017). Faccio (2010) has argued that although connected firms may have larger market power, they tend to have poorer accounting performance than non-connected firms and this difference is accentuated by the nature of connection. In the case of newly privatized companies, there is some evidence that political connections impede performance relative to non-connected companies (Boubakri et al., 2008). As regards dynamics, Ben-Nasr et al. (2012) use an event study approach with multi- country observations to argue that firms’ performance changes after political connections are secured. Such connections support firm performance and are also associated with an increase in financial leverage, long term debt and liquidity ratios. In general, it appears that connected firms tend to have higher leverage and more exposure to debt financing, measured in terms of long-term debt. Connected firms commonly pay lower taxes but the difference is not always significant. The ability to evade taxes or, for example, to avoid tariffs and non-tariff barriers can be ways in which connected companies achieve preferential status (Rijkers et al., 2015). In most contexts, however, data limitations restrict the scope 4 of analysis, notably in quantifying the benefits flowing to unlisted, privately held entities, while benefits may also be industry-wide rather than firm specific. The network dimension of connections can be seen in some of the existing research. For example, Bertrand et al. (2004) argue that networks in France get formed around attendance at a common university, prior employment as a civil servant, as well as the political party of the government under which a CEO has served. Companies managed by connected CEOs have lower rates of return on assets than those managed by non-connected CEOs and have particularly weak financial performance and high labor to capital ratios in cases where the companies are located in politically contested areas. Connected firms also tend to make lower tax payments and receive more subsidies. Such firms also tend to reciprocate benefits by taking actions supportive to politicians at opportune moments. The body of existing research also shows that political connections to firms exist across a wide range of economies, including at different levels of income and institutional devel- opment, as well as at differing degrees of de-centralization. They also cut across political systems, although autocracies seem to select and generate rents for connected parties in rather different ways than more competitive political systems. Finally, severing, or at least limiting, political connections has motivated the adoption of policies for privatizing assets. Boycko et al. (1995), for example, argued that large-scale (and rapid) privatization in Russia was the principal way to ensure that politicians could no longer influence company decisions. Yet two decades later, evidence from Russia and other transition economies indicates that those objectives have, at best, been partially achieved. In fact, a striking feature has not just been the persistence of state owned enterprises (and even the reversal of policy, as in Russia) but also the persistence of networks tying former SOE managers – subsequently owners – to politicians, as well as to others of their own type. Part of this can be attributed to the way in which privatization has been engineered (Megginson and Netter, 2001), but also to the deep roots of the networks linking key players. 3 Data description We use two main data sources in the paper. The first is a database compiled using publicly available information that contains an exhaustive list of Politically Exposed Persons (PEP) in each country. These data will, henceforth, be referred to as PEPData. In addition, we use Bureau van Dijk’s Orbis dataset that provides ownership and shareholder information, along with balance sheet and financial information, for companies in each of the countries. PEPData classifies PEPs consistent with widely used definitions, such as that provided by the Financial Action Task Force (FATF), but also extends the definition to incorpo- 5 rate the business relationships of PEPs. The dataset has eight categories of PEPs. These comprise individuals in International and Regional Organizations, National, Sub-national and Local Government, State-Owned Enterprises and State-Invested Enterprises, as well as Non-Governmental Organizations. They include all senior political and government leaders and officials but exclude lower level officials. In addition, based on their connection to the primary PEP, links with direct family members and close business associates or advisers are included. Detailed information for each PEP has been compiled from several sources, including from sanctions, regulatory and legal lists. In addition, a variety of media sources have been used to identify individuals and entities not found on official lists. Consequently, the dataset contains a list of names, aliases and other individual details and evidence on the positions that an individual has held. Each identified individual’s links to either a political party, other PEPs or to specific companies or financial institutions is noted, as are links to sources. For the purposes of our analysis, the data are organized around five main categories. The first two include PEPs, i.e., persons, while the other three comprise entities: (a) Political Individuals – persons currently holding or having held a political position, including in a political party or having been elected, (b) Other Individuals – any person appointed (as opposed to elected) to a PEP position or appointed to a government position, as well as immediate relatives or close associates of primary PEPs, (c) Political Party – any registered and active political party, (d) State-Owned or Invested Enterprises and, (e) Private companies or financial institutions. PEPData also includes links between all of these five categories. This allows us to go beyond the existing literature. The latter has focussed on having, or not having a (0-1 binary-type) link between the first two ((a) and (b)) categories and the latter three ((c), (d) and (e)). However, as our analysis shows, there is a web of relationships, direct and indirect, that matter and which are not captured by this approach. We additionally link individual PEPs in PEPData to firm level information. This is done in the following manner. The first step involves taking the name of the PEP and the explicit link to a company that is directly specified in PEPData. We then associate through the name of that company with firm-specific financials and other information (e.g., on employment) contained in the Bureau van Dijk’s Orbis dataset. Since firm names are 6 used to match the two datasets, this is referred to throughout the paper as the Firms’ Name Method (henceforth, FN). In addition, we take the names of the PEPs and then search to see whether they are shareholders or owners of firms, as listed in the Orbis dataset for each country. Because of common names and other possible sources of error, each possible match was subsequently reviewed manually to ensure the integrity of the match.4 A substantial number of matches were discarded as false positives and a revised set of matches adopted.5 As names of people are used to match the two datasets, this is henceforth referred to as the PEPs’ Name Method (PN). Matches from both methods are then used to build the final dataset that contains the total known connections between PEPs and entities. Table 1 provides some basic descriptive statistics including the number of PEPs identified under each method for each country. Using the FN Method gives a significant amount of variation in the number of companies associated with PEPs. For example, in Bulgaria, Hungary and Serbia only 46-88 firms are identified by this method, as against between 631- 800 in Russia and Romania respectively. The number increases significantly when including the PEP’s Name Method (PN). For the consolidated measure (FN+PN), the total number ranges between 384 in Serbia and 4,568 in Russia.6 Using the assets of PEP-connected companies to get an approximation of the aggregate scale of connected companies, Table 1 shows that they account for around 0.8 percent of GDP in Bulgaria and Hungary, between 2.5-5 percent in the Slovak Republic, Serbia and Romania and >15 percent in Russia in 2013. The massive share for Spain relative to the other countries is due to the fact that in the latter SOEs play a more significant role. In Spain, by contrast, some massive private firms are connected, swelling the share.7 Concerning the properties of connected firms depending on the method of identifica- tion, applying the size criteria (incorporating revenues, assets and employment) indicated in Appendix Table B1, it appears that the share of small firms among connected firms is far higher when applying the PN Method; 77 percent as against 40 percent for the FN Method. Correspondingly, the share of large and extra-large firms comprises 30 percent for FN iden- tification but only 6 percent when using the PN approach (see Appendix Table B2). This disparity is true for all countries, bar Bulgaria. In addition, regarding the size distribution, Spain looks different as between 27-52 percent of connected firms are large or very large – a 4 Researchers based in each country and with extensive local knowledge checked each match using a range of complementary sources and documentation. Only verified matches were maintained and false positives and ambiguous matches were discarded. 5 We explored alternative approaches, such as stochastic matching through location used by Koren et al. (2015), but found that data gaps were too large. 6 Note that for Russia, when using the FN Method a very large number of firms appeared, attributable to duplication of common firm names (e.g., Sputnik). To avoid this, we use only those firms with a unique name. For the PN Method, each entry was carefully cross-checked to ensure no duplication. 7 For example, the total assets of just three connected companies amount to over 20% of GDP. 7 far higher share than for the others where the average is between 4-11 percent.8 Concerning sector affiliation, there are differences depending on method of identification, but they are not that significant. The share of firms in wholesale and retail trade, as well as professional and scientific, is clearly higher for the PN Method, while the reverse is true for financial and insurance services. Comparing connected and non-connected firms, the former are mostly under-represented in manufacturing, construction and trade while being over-represented in all countries in professional and scientific activity. Table 2 provides mean and median values for a range of firm characteristics and perfor- mance indicators distinguishing between connected – as measured by FN+PN – and non- connected firms. These variables are taken from Orbis and comprise levels of assets, capital, sales, employment, wage bill and wage per worker, output per worker, as well as the return on assets (ROA) – defined as the ratio of net income to total assets – leverage (a proxy for access to debt financing) and the tax rate. For a number of variables, connected firms are consistently larger than non-connected ones. This discrepancy is mostly the case for sales but also for employment and average wages. In the case of ROA, connected firms mostly have lower values than non-connected ones. For leverage, mostly there is little difference although for both Hungary and Romania, connected firms are on average less leveraged. There is no clear pattern for the tax rate. The table also indicates that for both groups of firms, there is a large difference between mean and median values that indicates skewness to the right. Finally, given that widespread evidence shows that politicians are particularly prone to trying to influence the actions of state owned enterprises (SOEs), such as the amount of hiring, Orbis data show that for employment, the SOE share is particularly large in Serbia and Bulgaria (18-21%) and, to a lesser extent in Russia. The employment share for Spain (<6%) is roughly comparable to Hungary and the Slovak Republic. For sales, the SOE share is huge in Russia (>58%) and also substantial (>18%) in Serbia. Elsewhere, the share falls in the range of 4-6%. 4 How are connections configured? Connections are rarely atomistic tending, rather, to have strong network dimensions. In this section, we try to understand more about the properties of such networks and, in particular, the links running from individuals and/or entities to firms, both private and state-owned. We are interested in identifying the scale of network connections, the factors that appear to sort individuals into networks – such as kinship or allegiance to a political 8 Country level size and sector breakdowns are available on request. 8 party – as well as the scope, density and likely robustness of those networks. In a subsequent section, we go on to look at how network attributes correlate with the firm level variables summarized in Tables 1 and 2. 4.1 Some network descriptives Our objective is to study how firms and political exposed persons (PEPs) are linked and the economic impact of those links. The main innovation in this paper is to look beyond the binary character of the connection, i.e., whether a firm is connected or not, and identify whether connections are direct or, as in most cases, indirect through a network. So for example, in a binary case, some firms may look as if they are not connected, yet in reality they may be connected to PEPs indirectly as through connections to third parties that are, in turn, directly connected to PEPs. These third parties may be other firms, or political parties, or some other entity. In addition, some firms may be connected to multiple PEPs as well as many other firms, and thus be ‘super’ connected. Likewise, some PEPs may act as hubs with many spokes emanating from them. In a binary approach, all these network characteristics will be lost and the scale and impact of politically connected business may be significantly mis/under-estimated. By definition, a network is composed of nodes and edges. In our case,9 the nodes are firms, individuals, and political parties and the edges are the links between these entities. Firms are split into two groups, private and SOEs while individuals can also be separated between political individuals, and others.10 A network consequently represents relationships between agents, while also providing some form of structure for those relationships.11 Before deploying all the network analytics – betweenness, clustering, density and so on – it is useful to begin with some simple statistics and to look at the network from the point of view of the firms. In other words, this is to describe the network by looking at how many connections firms have, and with what types of entity. Table 3 gives the percentage of firms (i.e. nodes) grouped according to the number of connections and whether a firm has connections exclusively with other firms or with firms and/or other entities. It is useful to single out the group of firms that have connections only with other firms, because this is a group where firms are linked with PEPs only by an indirect link (and this indirect link may or may not be intentional). How large is the group 9 Firms can also be connected through inter and intra-industry trading, but such transactions (as for intermediate goods and services) are not measured in PEPdata nor is this dimension being considered in the paper. 10 PEPdata also contain other entities, such as international organizations. 11 A good review of the wider literature on networks is Ward et al. (2011), see also, inter alia, Do et al. (2015); Goyal et al. (2006). 9 of firms with only connections to other firms? In Spain, for example, firms with only one link (in network terms with degree 1) are split almost 40-60 between the group of firms that have connections with other entities and those that have connections only with other firms. However, for the other - transition - countries, the share of firms with degree 1 and with connections to other entities is much larger, indicating that indirect connections (of a firm to a PEP via another firm) are less common, except perhaps in Russia. Interestingly, this split is not constant when considering firms with degree 2, 3, 4 or ≥5. It could be expected that as the number of connections increases, and with it the size of the firm, the proportion of firms with connections with other entities would increase. For example, in Russia, firms with connections with only other firms account for 40 percent of the group of firms with one connection, but only 2 percent for the group of firms with ≥5 connections. A similar pattern is observed for Spain and the other countries. The rightmost column displays the split at the economy-wide aggregate level. Shares are not uniform across countries. In Russia and Spain, about three-quarters of firms belong to the group that have connections with diverse types of entities. In the other countries, this group has a larger share that reaches 97 and 99 percent in Hungary and the Slovak Republic respectively.12 What is the distribution of firms in terms of number of connections (or degree)? Figure 1 shows that the distribution is quite left-skewed, with the most common group of firms being that with the smallest number of connections. This skewness parallels that of the size of firms, as large firms to the right of the distribution are both those that have many connections and are much fewer than smaller firms with limited connections. For firms not exclusively linked with other firms, Table 4 indicates that connections with ‘individuals’ are the most common type. This table focuses only on the rows of Table 3 that contain firms that have diverse connections, i.e. connections that are not exclusively with other firms. An example may be useful to understand how the percentages in Table 4 are calculated. Consider the 50 percent that represents the share of Russian firms with 2 connections (not exclusively with other firms) that are linked to individuals. This percentage is calculated as the ratio of the number of firms with two connections that are linked to individuals (515 firms) over the number of firms with two connections (with other firms, individuals, political parties and so on; a total of 1,034 firms).13 Note that the percentages in Table 4 cannot be summed across rows, as they are calculated as the ratio of the number 12 Note that the bottom panel, which includes all countries, does not provide much insight since it basically mirrors Russia, as this country dominates the others in terms of number of firms in the dataset. 13 Note also that these 1,034 firms are the firms which do not have connections exclusively with other firms, and represent 57 percent of the total number of Russian firms with two connections (as shown in Table 1), which includes both the group of firms with exclusive connections with other firms and the group with diverse connections. 10 of firms with a fixed number of connections with a specific entity to the total number of firms with that specific number of connections. Apart from the first column, the rows are not mutually exclusive, as one firm with, say, 3 connections can appear in multiple rows as it may have one connection with an individual, one connection with a political individual, and one with a party. Table 4 highlights three important features of these networks. First, the largest share of firms with ‘diverse’ connections have links to individuals, followed by links to other firms, and then to political individuals. Second, the larger the number of connections (remembering that this is a group with fewer firms), the higher is the percentage of connections with individuals (at ≥90 percent). This indicates that when firms have multiple connections, it is almost certain that they will have a connection with an individual (a PEP). Third, in no country save Spain, do firms have direct connections with political parties. The latter, most likely, connect people rather than firms. This descriptive section has highlighted the fact that individuals (PEPs) are indeed the hubs of the network and, further, that they become increasingly important for larger, and more connected, firms. However, the percentages reported in Tables 3 and 4, while infor- mative, do not capture the whole story. To do that, a set of network analytics needs to be deployed. 4.2 Network analytics Networks are commonly represented in terms of degree (the number of links sent to a node) and density (as indicated by the ratio of ties in a network to the total possible number of ties). In addition, measures of betweenness (the extent to which a node falls between other nodes); closeness (how close nodes are to each other); clustering (the extent of locally dense clusters of nodes); path length or distance (the number of steps to connect a pair of nodes), as well as neighbourhood or proximity measures can be employed. For the purposes of our analysis, the idea of centrality, or how important particular nodes are in a network, is of particular interest. Centrality is best captured by the betweenness measure. However, there may be significant variation in how centrality is configured. For example, in a small-world network, nodes are located in locally dense clusters but can reach other nodes through a small number of bridging connections (Goyal et al., 2006). In what follows, we apply these measures in order to describe the main features of networks across the selected countries. But, first, some definitions: For a node i we have Degreei = N umber of edges connected to i 11 Geodesic Distanceij = M inimum number of edges required to reach node j N umber of distances f rom j to k, through i Betweennessi = i=j =k N umber of distances f rom j to k We can take the average for all nodes to get a network-level measure, noting that this will grow with log(N ) where N equals the number of nodes. Moreover, for a network we have: N umber of edges Density = N umber of N odes ∗ (N umber of N odes − 1) N umber of closed triplets Clustering Coef f icient = N umber of triplets A triplet is a set of three connected nodes and a closed triplet a set of three connected nodes where each one is connected with both others. Geodesic distance is also called shortest path length and the way the clustering coefficient is calculated is called transitivity. In Figure 2 we give a simple illustration for two, stylized examples. On the left hand side, node A has degree 3; the distance to B is 1 (and is the same for C and D). The betweenness of A is 3 (viz., A is between BC, BD and CD). All the others have degree 1, distance to A, 1 and distance to others, 2 with betweenness 0. The network density is 3/12=25% and the clustering coefficient 0. For the right hand example, the betweenness of A becomes 0 (in fact, everyone’s betweenness becomes 0), the density is 6/12=50%, and the clustering coefficient becomes 1. 4.3 Network properties We turn to explicit consideration of the characteristics of networks. Specifically, we de- scribe the extent and detail of connections running between individuals, as well as between individuals and institutions in the seven countries. As mentioned above, PEPData docu- ments the links that PEPs may have both with other individuals (including other PEPs), as well as with specific firms and political parties. We document the size and composition of the networks, paying particular attention to their components and extent of integration. We consider the extent to which these networks have dense or diffuse features along with evidence of small world properties, where clusters have the ability to link to other parts of the network through a limited number of bridges. The focus also falls on location in the net- work specifically in the context of centrality. These attributes are subsequently incorporated explicitly in the empirical analysis of Section 5. 12 4.3.1 Network size Table 5 provides information about the size of the network, as measured by the number of nodes, in each country. The information is broken down by type of actor, namely; private firm; state-owned enterprise; political party; political individual; other individuals (including relatives). We report an adjusted measure of network size that excludes those family relatives (the overwhelming majority) that have no betweenness – in other words, do not lie between any other nodes and hence are largely irrelevant from a network perspective. There are significant differences in the size of the network across countries. In abso- lute terms, Russia has the largest network followed by Spain. Both Hungary and Serbia’s networks are far larger than neighbouring Slovak Republic and Bulgaria. Adjusting for pop- ulation, the network ranking has Russia very clearly at the top followed by Serbia, Hungary, Romania and the Slovak Republic and, lastly, Spain. There are also clear differences in com- position: Spain has significantly more political parties and political individuals. In contrast, for Russia not only is there a relatively small number of political parties, but the level and share of SOEs is particularly high. The latter comprise around 9% of the total network, as against an average <3% for the other countries. Romania has a relatively large share of private firms in its network. In all countries, individuals and political individuals comprise between 85-94% of the total network size. Given our interest in the links to SOEs and private firms, Table 6 reports shares for different types of connections by country. SOE connections are mostly with individuals and, then, politicians14 (the mean share for both is >5%). In Spain, connections to politicians for both SOEs and private firms comprise particularly high shares. In Bulgaria and Russia there is also a relatively strong connection of SOEs to other SOEs. For private firms, the picture is rather more diverse. Links to politicians are again important in Spain but, elsewhere, the shares of individuals are mostly the largest although that for other firms is also a substantial component; in Romania it is the largest share. 4.3.2 Network components: Big Islands Aside from the absolute size of the network, it is necessary to look at the constituent parts of the network. A component is composed by taking together all the nodes that are connected to each other (irrespective of their distance) so that the whole network can be divided into components (islands). The issue is whether there exists a component in which the greater part of the network falls, rather than, say, being composed of small fractional 14 Note that the terms ‘political individual’ (as per definition (a) on page 6) and ‘politician’ are used interchangeably. 13 parts. We call the largest component a ‘big island’ as its size is much greater than all the others. Table 7 indicates that for all the countries there is indeed a giant component or big island that comprises up to 76 percent (in Romania) of the adjusted network size. Spain’s big island holds two-thirds of the network; a situation roughly comparable to that in Hungary and Romania. Russia’s relatively small big island may reflect a difference originating in political systems as in democratic settings, the big island tends to be more prominent. In addition, there is less integration through the network and hence more fragmentation. In Russia’s case, this may partly be a function of geographical size. Finally, it should be noted that none of the countries has any second level component of significant magnitude, indicating that network activity is concentrated in the big island. Table 7 also breaks down by category the respective shares contained in the big island. For political parties, there is some – but not massive – cross-country variation. The main outlier is Spain, where only 18 percent of its (many) parties are in the big island (as against an average of 66 percent for the other countries). This may reflect the decentralized nature of Spanish politics. With the exception of political parties, Spain’s respective shares are, however, broadly comparable to the others. For political individuals, their inclusion in the big island is everywhere substantial, with the exception of Russia with only 29 percent. For both private firms and SOEs, with the exception of Bulgaria in the latter instance, >60->90 percent fall in the big island. Appendix Figures C1-C7 also permit visualization of the big island of each country with scaling by degree (or number of edges). It can be seen that in Russia political parties are not only less numerous but also less connected to other entities. The network in Russia is heavily influenced by the SOEs, as also private companies, and the links between the two types of firm. By contrast, in Spain the larger number of political parties stands out, as does the relative absence of SOEs and private firms. Although the scale and location of political parties varies significantly across the other economies, the SOEs’ place in the network is clearly significant and, perhaps, the dominant feature. Private firms’ place in the respective networks also varies but their presence is more notable when compared with Spain. 4.3.3 Properties of the Big Island Turning to the properties of the big island, Table 8 shows limited difference in the average degree (or number of links sent to a node) across countries, although Spain, Russia and Romania are somewhat higher. As regards density, both Bulgaria and the Slovak Republic have a higher ratio of ties in the network relative to the total possible number of ties, while Russia and Spain have the lowest ratios. Although there is some clear variation, most of 14 the networks we observe are not that tightly connected, suggesting that diffuse networks are present. Pursuing this point, the clustering coefficient further indicates what share of a person/entity’s neighbours are neighbours of each other and hence whether dense clusters of nodes are present in the network. Although there is some variation across countries, with higher clustering in Serbia, in general the clustering coefficient is quite low and especially so in Russia. The complementary indicator of average distance or path-length exhibits less variation across countries. An obvious question concerns whether these networks have small world properties charac- terized by high clustering and low path lengths. This is where nodes tend to be concentrated in clusters but with the ability to reach other nodes in the population through a small number of bridging connections. Most generally, such properties exist if the average distance between nodes is very small relative to the size of the population. To assess the ‘small-world-ness’ of the networks, one option is to use the method of Humphries and Gurney (2008) by generating an Erdos-Renyi (E-R) random graph with the same number of nodes and edges, and after calculating their Clustering Coefficient and Average Distance, use them to get the following; Cluster Coef freal Cluster Coef frandom S= Average Distancereal Average Distancerandom A value of S > 1 is evidence of the existence of a small-world network.15 The intuition lies in the properties that Average Distance will be a bit greater (almost equal) to the random one, whereas the Cluster Coefficient will be much greater than the random one if the network under examination is a small world. Table 8 accordingly reports the clustering coefficient and average distance generated by the random process, as well as the values for S for each country. They show a significantly higher clustering coefficient than those randomly generated alongside a lower average distance. The S values for each country are all >1 implying that small world properties are indeed present. Several findings emerge from the analysis so far. The first concerns the importance of the big island in all countries. These networks are not a collection of un-integrated small islands. Rather, there is significant integration in the big island. Higher clustering and lower path length relative to the random also provide indications of a small world. This suggests that relatively dense clusters of nodes are able to tie to other nodes in the big island through a small number of bridging connections. Such structure has been noted to be present in many human networks, possibly because the combination of the identity advantages of dense clustering alongside the information advantages of short average distance 15 Note a key finding in Humphries and Gurney (2008) is that the small-worldness indicator tends to scale linearly with network size and reflects behavioural characteristics. 15 (Watts, 2003). With this in mind, it is clear that the ways in which nodes are connected, and the role of specific nodes in integrating the network, need more understanding. This brings us to the issue of centrality. 4.3.4 Betweenness and centrality The extent to which a node lies between other nodes and has, or has not, high betweenness is linked to the broader issue of centrality or the importance of specific nodes in a network. Table 9 reports the betweenness shares - the extent to which nodes lie between other nodes - for each of the components (the columns sum to 100).16 The betweenness shares are highest for politicians and political parties, as might be expected. Strikingly, betweenness is also high for SOEs; in Hungary and the Slovak Republic the share is higher than for either political parties or politicians. Private firms have low betweenness shares across the board.17 Russia also looks different. The share for both political parties and political individuals is significantly lower than elsewhere, while the share for SOEs is significantly higher. If we extend this analysis to the neighbours of both SOEs and private firms, the betweenness of both SOE and private firm neighbours is particularly high in Russia (77% and 47% respectively). 4.3.5 Summary All countries are marked by the presence of a big island. The second largest component is very small in all cases. But the properties of the big island differ significantly across countries. The most obvious difference – reflecting differences in political systems – con- cerns political parties. These differences relate not only to the number and scale, but also the nature of connections flowing to and from those parties. Mostly, the links are from political individuals to parties. SOEs are in almost all instances a significant component of the big island. In autocratic contexts, such as Russia (but also its neighbours, Belarus and Kazakhstan)18 , SOEs have relatively high betweenness shares. In the case of private firms, betweenness shares are lower than for SOEs but there is also less difference across 16 The betweenness share is defined as the ratio of the sum of the betweenness of all nodes in a particular group (parties, SOEs, etc.) over the total network betweenness, i.e. over the sum of the betweenness of all the nodes in the network. 17 Regarding closeness – which measures how close nodes are to one another and hence of the ability to connect to many, even when not between – shares are higher for both political individuals and other individuals across all countries. For both private firms, and particularly SOEs, the closeness shares are very much higher in Russia than elsewhere. 18 The betweenness share for Belarus and Kazakhstan is 60% and 50% respectively, while the closeness shares are 17% and 10% - all significantly higher than in either Spain or the other economies covered in this paper. 16 countries. Not surprisingly, betweenness shares for political individuals and political parties are relatively high, although lower in the case of Russia. Relative to Spain where networks are more located around political parties and individuals, SOEs, in particular, have a more salient role in all the transition economies. In the following section, in looking at the asso- ciation between scale and performance and firm level attributes, we explicitly incorporate these network features. 5 Correlates of being connected We now look at whether being connected and the manner and location of any connection is associated with specific firm level indicators, including sales, output and return on assets. Our measures of being connected may not be exogenous and unobserved factors affecting firm outcomes could also be explanatory factors for having political connections. With the data that we have available, as well as the lack of a significant temporal dimension, instru- mentation is problematic. As such, the estimates we report should be viewed as correlates, rather than indicating causality. We estimate the following: Outcomeict =β0 + β1 Connectedic + β2 SOEic + β3 F irm Size dummiesict (1) + Sector F E + Y ear F E + Country F E + uict for each firm i, in country c, at year y , where several outcome measures are related to a dummy variable for whether a firm is connected, as indicated by the FN+PN method. We also account for whether the firm is an SOE, using information reported directly in Orbis. The reason for including SOEs is that due to their ownership and governance they will, almost by definition, be connected to politicians; features that have, of course, been well documented. The outcome measures include Return on Assets, Leverage, as well as the logarithm of Sales, Output and Wages. We also include fixed effects for firm size, sector, country and year. The definition of firm size (Small/Medium/Large/Very Large) is taken from Orbis (see Appendix Table B1). The results from estimating Equation (1) are reported in Table 10 using a pooled country specification. Note that Russia is excluded due to the severe limitations of coverage in Orbis. The same equation has also been estimated separately for each country (see Appendix Table A1). For Log Sales, Log Output and Log Wages, it can be seen that being connected is positively associated across the board and, in most instances, highly significant. For SOEs, the coefficients on sales and wages are positive, large and highly significant but negative and 17 significant for output. With respect to the performance variable – ROA – the coefficient on the connected variable is actually negative, large and significant as is the case for SOEs. For Leverage, being connected is negatively signed but insignificant while for SOEs the sign is also negative but in this instance, significant. The picture that emerges from the pooled regression reported in Table 10 is that connec- tions tend to be associated with higher levels of sales, output and wages for both connected firms and SOEs. In numerical terms, some of these differences are large. Sales, output and wages are respectively 53, 16, and 24 percent higher for a connected firm than a non- connected firm. In contrast, Leverage and ROA are mostly lower for connected firms. Taking SOEs as a benchmark for politically connected business, apart from Sales, all the other coef- ficients for SOEs are not only very significant but also larger than those of connected private firms. This is particularly true in the case of ROA where SOEs have a far larger, negative sign. 5.1 Network regressions The results reported up to this point are part of the standard binary approach of existing analyses of politically connected business. The focus is on the magnitude, sign and signifi- cance of the dummy that discriminates connected as against non-connected firms. However, and this is the originality of our study, simply ‘being connected’ is quite imprecise. A network approach can refine that identification. In a network, a firm can be connected to political power in different ways and with different intensity. We therefore extend the analysis to incorporate the network measures discussed in detail in Section 4 above. We are interested in explicitly testing conjectures relating to the nature of the connection, as well as the impact of variation in network properties. More specifically, we focus on the following measures: (a) Big Island: whether the firm is in the big island or giant component; (b) Betweenness – whether betweenness >0, compared to having no betweenness (condi- tional on being in the Big Island); (c) Log Degree: the logarithm of the number of connections; (d) Politician: whether the firm has a shareholder who is a politician. A common prior for all the above variables is that they should influence the intensity of the connection, and thus the impact on the outcomes from being connected, in a positive way. For example, we would expect that being in the largest component of the network confers a stronger advantage than being connected to some other more peripheral parts of 18 the network. Similarly, having a larger number of connections will be better than having just a few, and so on. We augment Equation 1 by now including these network measures. Implementation is for the same indicators as reported in the baseline estimates reported in Table 10 and includes country, sector, year and firm size dummies. Since the network variables are defined only for connected firms, all non-connected firms would be excluded from this analysis. As such, we replace all network values with zero for all non-connected firms as well as including a dummy for being connected in all specifications. The table below shows the different specifications we estimate. Network estimations Variable: (1) (2) (3) (4) (5) (6) SOE Connect Big Island (BI) Has Betweenness Log Degree BI × Log Degree Politician Politician × Has Between. Constant For example, the estimation of column (1) in equation form is as follows: Outcomeict =β0 + β1 SOEic + β2 Connectic + β3 BigIslandic + β4 F irm Size dummiesict + Sector F E + Y ear F E (2) + Country F E + uict So, it is the same as Equation (1) with the Big Island network variable added to it. Similarly, for the other five columns of the table. Note that these correspond exactly to columns 1 to 7 in Tables 11-15.19 The idea behind this set of specifications is to consider the network properties as different layers of the connection. The first layer is simply being connected (examined in equation 1). Then, it is being connected but also being in the Big Island (column (1) in the network specifications). Further the ‘location’ or centrality matters, so that the betweenness or the 19 Country-specific estimates are reported in Appendix Tables A2-A6. 19 number of connections is examined. Finally, we would like to distinguish the impact of the type of the connection, such as whether being connected with a political individual or with another PEP produces different results. Interactions are also considered and these allow us to assess whether the effects of two network properties, for example, being in the Big Island and having multiple connections (degree), reinforce or weaken each other. Tables 11-15 contain the main results using these different specifications. We provide detailed comments for the results for the outcome ‘Log Sales’ in Table 11, and then summarize the main differences for the results for the other outcome variables. For Log Sales, the coefficient for being connected is always large, positive and highly significant. That is also true for the SOE variable. Being in the Big Island yields a positive and significant coefficient (column 1 in Table 11). Indeed, being connected and being in the Big Island (across all countries and years), is associated with a level of sales that is 58 percent higher (viz., exp(0.310+0.148) - 1). Note that this value lies above that identified in Eq. 1 (53 percent) which did not distinguish between being connected and having a presence in the Big Island. In other words, when we control for the network and the locus of the firm in that network, we can distinguish the impact of a connection when a firm is outside of the Big Island, from the impact of being inside it, and we indeed find that there is a difference. The next specification, column (2) in the table, shows that what really matters is not just being in the Big Island, but having centrality. A firm could be inside the Big Island but this does not produce much of an effect if that firm is located at the periphery, i.e. if the firm has no betweenness. Indeed, in column (2) of Table 11, the coefficient for the Big Island becomes statistically not different from zero, and all of its previous effect (in column (1)) is taken over by the betweenness variable. A different result is found when comparing columns (3) and (4). The new variable is the log of Degree and this is defined for firms within and outside of the Big Island.20 Specification (3) illustrates that, without controlling for belonging to the Big Island, the number of connections matters. Specification (4) allows to assess whether the value of connections is the same inside or outside the Big Island. The data tell that with just one connection, there is no difference between having it inside or outside the Big Island. However, starting from two connections onwards, the value of the connections (i.e. the degree) is much higher inside the Big Island than outside. A numerical example can clarify this. Using the coefficients from specification (4), a firm outside of the Big Island with one connection would 20 Note that log degree for one connection is equal to 0. To avoid this problem, the degree variable has been transformed to into a new variable called degree* which is equal to degree+1. This means that firms with one connection have now degree* = 2 and log degree* = 0.693. Note also that degree is not defined for not connected firms, but with this transformation, we assign degree*=1 to firms with no connection, so that for these firms log degree*=0. 20 have sales 34% higher (viz. exp(0.223+0.103×0.693)-1) than a non-connected firm. This is similar to the 31% (viz. exp(0.223 - 0.081 + 0.102×0.693 + 0.088×0.693) - 1) increase for a firm with one connection inside the Big Island. But for firms with two connections the difference, with respect to a non-connected firm, is 39.8% when outside of the Big Island, versus 42.0% when inside the Big Island. In sum, the value – in terms of increased volume of sales – of having multiple connections is higher for firms which are already in the main component of the network. Together, specifications (1) to (4) tell a compelling story. Connections matter, but their importance comes from being in specific locations of the network. In particular, the largest impact of the connections is correlated with firms that not only are in the Big Island, but are there in a central position (with betweenness) and have a high degree. Specifications (5) and (6) add another layer. Connections with a politician would, a priori, seem important, but a closer look at the data shows a more complex narrative. Without discriminating between groups of firms that have betweenness and those without, specification (5) does not detect any effect of controlling for a connection with a politician. But when the two groups are split, connections with a politician matter for firms with no betweenness, while politicians’ relevance is diminished for firms with betweenness, as the interaction term is negative. Politicians can act as ‘substitutes’ for not having betweenness. Although there is some variation across the other left-hand side variables that indicate scale of operation – namely Log Average Output and Wage – the results obtained for Log Sales are largely replicated. Turning to Table 14 that reports estimations for a performance variable – namely Return on Assets – some similar patterns emerge. For ROA, being connected is mostly negatively associated but not always significant, unlike for SOEs where the coefficient is always large, negative and highly significant. In column (3) of Table 14, although being in the Big Island is positive and significant, having betweenness enters negatively and significantly. As in the case of Log Sales of Table 11, this means that for firms that are at the periphery of the Big Island the connection is not so important. Actually, in this case, for firms at the periphery the impact of the connection seems to be diminished, as the coefficient is of the opposite sign compared with firms that are central. This is also true in the estimation that incorporates Politicians. The Politician variable is itself positive and significant, both when entered by itself or along with the betweenness measures. A clear conclusion can be drawn from these results. Not distinguishing these network attributes may not bias the coefficient of the connected dummy of equation (1), but certainly obscures the large heterogeneity captured in the specifications of Tables 11-15. Taking into account the network dimensions of connections is essential. 21 6 Conclusion Politically connected businesses are not a rarity, nor are they limited to SOEs. Their incidence cuts across political systems, regions and levels of development. Our paper has focused primarily on economies in East and Central Europe, Russia and, as a comparator, Spain. Using a new and unique dataset that identifies Politically Exposed Persons (PEPs) and their links to companies, politicians and political parties, we are able to identify the broad scale of the phenomenon in each of the countries. Yet, the originality of the paper lies principally in our ability to identify the links between individuals, politicians, political parties and different types of firms, as well as the complex configurations of the resulting networks in each of the countries. These networks in turn could be expected to shape behaviour. Drawing on the tools of network analysis, we show that each of these countries is charac- terized by a giant component or Big Island. The second largest component in all instances is small. Nevertheless, we find evidence that their networks have small-world properties, with high clustering and short path length due to a relatively small number of bridging connec- tions. In the network space, however, there are also very different configurations that reflect, inter alia, differences in political and other institutions. Given the region’s recent past, it is perhaps not surprising that State-Owned Enterprises (SOEs) are prominent and tend to have positions in the network with relatively high betweenness or centrality. There is a clear difference in this regard when compared with Spain. Matching the information on connections to firm level data from Orbis, we demonstrate that there is a positive and robust correlation between the levels of sales, output, wages as well as a performance variable – the return on assets – and being connected. We cannot pin down causality, given the limitations in our data, but it seems reasonable to assume that the likely direction of association is from connections to outcomes. However, the principal interest from our data analysis concerns how network features can influence these associations. We find clear evidence that the location in a network, the extent of ties and betweenness or centrality is often positively, and significantly, associated with firm level indicators for the scale of activity. Network variables are also seen to be significant in an estimation that has a performance indicator – Return on Assets – on the left-hand side. In other words, not only are networks likely to be important in shaping how connections arise and propagate (something that we cannot directly measure) but they are also important in shaping the returns to connections. Of course, the configuration of networks, as well as their respective paybacks, is likely to be materially affected by other factors, including political institutions, resource endowments and neighbourhood. We are exploring these features in continuing research for a far larger group of countries and regions using the same dataset. 22 References Akcigit, U., Baslandze, S., and Lotti, F. (2017). Connecting to power: Political connections, innovation, and firm dynamics. Technical report, University of Chicago, mimeo. Ben-Nasr, H., Boubakri, N., and Cosset, J.-C. (2012). The political determinants of the cost of equity: Evidence from newly privatized firms. Journal of Accounting Research, 50(3):605–646. Bertrand, M., Kramarz, F., Schoar, A., and Thesmar, D. (2004). Politically connected CEOs and corporate outcomes: Evidence from France. University of Chicago, mimeo. Boubakri, N., Cosset, J.-C., and Saffar, W. (2008). Political connections of newly privatized firms. Journal of Corporate Finance, 14(5):654–673. Boycko, M., Shleifer, A., and Vishny, R. (1995). Privatizing Russia. MIT Press, MA, Cambridge. Cingano, F. and Pinotti, P. (2013). Politicians at work: The private returns and social costs of political connections. Journal of the European Economic Association, 11(2):433–465. Cohen, L., Frazzini, A., and Malloy, C. (2008). The small world of investing: Board connec- tions and mutual fund returns. Journal of Political Economy, 116(5):951–979. Commander, S. (2016). Politically connected business in transition economies: Scale, con- sequences and political economy. Paper prepared for World Development Report, London. Diwan, I., Keefer, P., and Schiffbauer, M. (2015). Pyramid capitalism: political connections, regulation, and firm productivity in Egypt. World Bank Policy Research Working Paper 7354. Do, Q.-A., Lee, Y. T., and Nguyen, B. D. (2015). Political connections and firm value: Evidence from the regression discontinuity design of close gubernatorial elections. CEPR Discussion Paper No. DP10526. Doroftei, M. and Dimulescu, V. (2015). Corruption risks in the Romanian infrastructure sector. In Mungiu-Pippidi, A., editor, Government Favouritism in Europe. The Anticor- ruption Report Volume 3, volume 3, chapter 2, pages 19–34. Barbara Budrich, Opladen. Earle, J. S. and Gehlbach, S. (2015). The productivity consequences of political turnover: Firm-level evidence from Ukraine’s Orange Revolution. American Journal of Political Science, 59(3):708–723. 23 Faccio, M. (2006). Politically connected firms. The American Economic Review, 96(1):369– 386. Faccio, M. (2010). Differences between politically connected and nonconnected firms: A cross-country analysis. Financial Management, 39(3):905–928. Faccio, M., Masulis, R. W., and McConnell, J. (2006). Political connections and corporate bailouts. The Journal of Finance, 61(6):2597–2635. Fisman, R. (2001). Estimating the value of political connections. The American Economic Review, 91(4):1095–1102. alez, J. L. (2006). Economics: An emerging Goyal, S., Van Der Leij, M. J., and Moraga-Gonz´ small world. Journal of Political Economy, 114(2):403–412. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78:1360–1380. Humphries, M. D. and Gurney, K. (2008). Network ‘small-world-ness’: A quantitative method for determining canonical network equivalence. PloS One, 3(4):e0002051. Johnson, S. and Mitton, T. (2003). Cronyism and capital controls: Evidence from Malaysia. Journal of Financial Economics, 67(2):351–382. Khwaja, A. I. and Mian, A. (2005). Do lenders favor politically connected firms? Rent provi- sion in an emerging financial market. The Quarterly Journal of Economics, 120(4):1371– 1411. Koren, M., Szeidl, A., Szucs, F., and Vedres, B. (2014). Political connections and favoritism in Hungary. Central European University, Budapest, mimeo. Koren, M., Szeidl, A., Szucs, F., and Vedres, B. (2015). Political favor exchange in a democracy. Central European University, Budapest, mimeo. Krueger, A. O. (1974). The political economy of the rent-seeking society. The American Economic Review, 64(3):291–303. Megginson, W. L. and Netter, J. M. (2001). From state to market: A survey of empirical studies on privatization. Journal of Economic Literature, 39(2):321–389. Palguta, J. (2014). Concealed ownership of contractors, manipulation of tenders and the allocation of public procurement contracts. CERGE-EI Working Paper Series 501. 24 Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. Simon & Schuster Paperbacks, NY: New York. Rijkers, B., Baghdadi, L., and Raballand, G. (2015). Political connections and tariff evasion evidence from Tunisia. The World Bank Economic Review. Ward, M. D., Stovel, K., and Sacks, A. (2011). Network analysis and political science. Annual Review of Political Science, 14:245–264. Watts, D. J. (2003). Six degrees: The science of a connected age. W.W. Norton & Company, NY: New York. Watts, D. J. and Strogatz, S. H. (1998). Collective dynamics of ‘small-world’ networks. Nature, 393:440–442. World Economic Forum (2014). Global competitiveness report. Technical report, Geneva. 25 Figure 1: Firms with only one connection are the largest group Figure 2: Two stylised networks 26 Table 1: Number of Connected Firms and Size/GDP Bulgaria Serbia Slovak Rep. Hungary Romania Russian Fed. Spain No. of Connected Firms 46 88 136 63 851 631 638 (FN) No. of Connected Firms 407 296 1,110 612 1,236 3,937 542 (PN) No. of Connected Firms 453 384 1,246 675 2,087 4,568 1,180 27 (FN+PN) No of All Firms 669,642 110,432 439,497 551,846 837,779 6,194,392 1,181,296 Tot Assets GDP% FN 0.29% 1.92% 1.06% 0.17% 2.12% 4.00% 27.18% Tot Assets GDP% FN+PN 0.76% 4.61% 1.92% 0.78% 2.59% 15.34% 74.94% Tot Assets GDP% (All) 397.25% 300.85% 247.67% 461.10% 201.21% 304.90% 494.39% Tot Assets Ratio FN/All 0.07% 0.64% 0.43% 0.04% 1.05% 1.31% 5.50% Tot Assets Ratio FNPN/All 0.19% 1.53% 0.78% 0.17% 1.29% 5.03% 15.16% Calculations are based on 2013. Table 2: Descriptive Statistics for Connected and Non-Connected: FN+PN Bulgaria Serbia Slovak Rep. Non-Connected Connected Non-Connected Connected Non-Connected Connected Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Employees 7.9 2.0 18.9 3.0 20.7 4.0 116.8 7.0 13.2 2.0 15.8 3.0 Assets 826k 41k 1,510k 165k 3,366k 223k 12,000k 793k 1,685k 63k 1,977k 158k Sales 464k 20k 762k 62k 2,243k 234k 5,321k 450k 1,500k 57k 1,223k 61k Capital 143k 4k 123k 4k 1,030k 3k 4,553k 12k 352k 9k 428k 9k Wage 78k 8k 183k 27k 261k 31k 672k 80k 258k 23k 287k 32k ROA 10.29 4.35 6.53 2.45 2.49 1.76 3.13 1.33 0.21 0.54 -0.69 0.03 Leverage 0.98 0.42 1.43 0.53 0.63 0.67 0.62 0.69 35.96 0.70 11.05 0.75 Ave Output 48,927 9,866 51,773 18,224 133,527 57,884 167,032 53,033 180,214 51,676 169,088 53,314 Ave Wage 3,938 2,537 6,231 4,201 9,022 6,435 12,292 9,216 15,897 11,228 17,789 14,191 Tax Rate 0.04 0.00 -0.02 0.00 -0.02 0.01 0.03 0.01 0.24 0.00 0.10 0.00 28 Hungary Romania Spain Non-Connected Connected Non-Connected Connected Non-Connected Connected Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Employees 10.1 2.0 27.8 2.0 6.3 1.0 26.3 3.0 16.9 3.0 752.5 17.0 Assets 1,494k 28k 1,864k 56k 624k 23k 4,397k 154k 8,728k 463k 1,110,000k 11,100k Sales 1,351k 47k 1,329k 52k 527k 16k 5,874k 64k 4,134k 298k 539,000k 4,404k Capital 171k 2k 367k 14k 104k 0k 689k 0k 1,043k 17k 69,800k 827k Wage 126k 12k 307k 11k 87k 8k 446k 33k 704k 104k 53,100k 870k ROA 4.13 2.53 6.29 1.86 1.44 0.00 0.68 0.12 -1.60 0.33 -1.37 0.51 Leverage 16.72 0.53 1.16 0.47 46.64 0.98 5.62 0.76 8.89 0.67 0.79 0.59 Ave Output 151,585 37,678 124,922 47,946 54,977 18,084 124,031 26,353 273,516 102,774 1,513,735 226,204 Ave Wage 12,213 8,009 11,503 8,543 4,082 3,323 6,424 4,731 39,125 33,221 72,459 55,951 Tax Rate 0.09 0.10 0.10 0.10 0.06 0.00 0.18 0.00 0.23 0.25 0.13 0.27 Calculations are based on 2013. Table 3: Connection types (Share of firms by country, number of connections, and whether they have connections only with other firms, or with firms and other entities) Each firm’s number of connections: Connection with: 1 2 3 4 >=5 tot Russian Federation Diverse entities 60 57 83 90 98 75 Exclusively other firms 40 43 17 10 2 25 Spain Diverse entities 43 85 100 75 98 71 Exclusively other firms 57 15 0 25 2 29 Slovak Republic Diverse entities 96 100 100 100 100 99 Exclusively other firms 4 0 0 0 0 1 Serbia Diverse entities 94 100 77 100 100 97 Exclusively other firms 6 0 23 0 0 3 Bulgaria Diverse entities 63 100 100 100 100 91 Exclusively other firms 37 0 0 0 0 9 Romania Diverse entities 89 97 92 100 99 95 Exclusively other firms 11 3 8 0 1 5 Hungary Diverse entities 88 98 100 100 99 97 Exclusively other firms 12 2 0 0 1 3 All countries Diverse entities 65 60 84 91 98 78 All countries Exclusively other firms 35 40 16 9 2 22 All countries All firms 100 100 100 100 100 100 Source: Firm name connected firms’ database. Note: In the dataset, firms can be connected with diverse entities, which include other firms, individuals, political party, etc., or exclusively with other firms, the second line for each country in the table above. 29 Table 4: Share of firms with ‘diverse’ connections by country, entity and number of connec- tions Connections with: Each firm’s number of connections: 1 2 3 4 >=5 tot Russian Federation Corporate 0 74 93 98 88 77 Individual 39 50 70 73 95 70 Pol. Individual 23 10 5 7 21 13 Pol. Party 0 0 0 0 0 0 Spain Corporate 0 39 33 33 50 34 Individual 39 74 89 67 97 77 Pol. Individual 9 22 44 33 43 31 Pol. Party 0 0 0 33 0 1 Slovak Republic Corporate 0 8 17 20 42 22 Individual 70 92 83 100 100 90 Pol. Individual 22 24 33 20 28 25 Pol. Party 0 0 0 0 0 0 Serbia Corporate 0 42 40 71 28 26 Individual 69 100 100 100 100 93 Pol. Individual 25 0 10 0 24 20 Pol. Party 0 0 0 0 0 0 Bulgaria Corporate 0 74 40 67 76 59 Individual 33 63 80 100 76 68 Pol. Individual 50 42 0 0 52 41 Pol. Party 0 0 0 0 0 0 Romania Corporate 0 21 39 47 40 24 Individual 23 100 96 87 98 72 Pol. Individual 66 45 48 47 47 53 Pol. Party 0 0 0 0 0 0 Hungary Corporate 0 38 81 61 69 56 Individual 65 98 100 100 100 95 Pol. Individual 21 11 12 4 34 23 Pol. Party 0 0 0 0 0 0 Source: PEPData with Firm name (FN) listing 30 Table 5: Network size and components (by country) Bulgaria Serbia Slovak Rep. Hungary Romania Russian Fed. Spain Party 31 96 25 38 60 38 319 SOE 218 195 110 349 205 5,660 476 31 Firm 504 434 1,344 817 2,731 5,802 892 Political Individuals 1,745 2,264 1,092 2,322 5,244 18,066 9,631 Individuals 1,901 5,144 2,130 4,265 7,068 31,445 9,403 Network size 4,399 8,133 4,701 7,791 15,308 61,011 20,721 Table 6: SOE and Firm Connection Shares Bulgaria Serbia Slovak Rep. Hungary Romania Russian Fed. Spain SOE Firm SOE Firm SOE Firm SOE Firm SOE Firm SOE Firm SOE Firm Firm 3% 30% 1% 15% 4% 37% 3% 25% 6% 50% 7% 18% 3% 7% 32 Individual 56% 39% 71% 59% 68% 47% 73% 48% 56% 33% 49% 45% 45% 43% Party 1% 1% 1% 1% 0% 0% 0% 1% 0% 0% 0% 0% 1% 1% Politician 25% 26% 20% 24% 20% 14% 14% 21% 37% 17% 24% 25% 45% 47% SOE 15% 5% 7% 1% 8% 2% 10% 5% 1% 1% 19% 12% 6% 2% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% Table 7: Size of Big Island and components (by country) Bulgaria Serbia Slovak Rep. Hungary Romania Russian Fed. Spain Network Size 4,399 8,133 4,701 7,791 15,308 61,011 20,721 Big Island (BI) 1,934 2,499 2,620 5,106 11,593 22,016 13,888 BI (%) 44% 31% 56% 66% 76% 36% 67% Rest 2,465 5,634 2,081 2,685 3,715 38,995 6,833 Second BI (in rest) 51 1004 27 18 39 51 98 Parties in BI 26 35 19 19 43 24 56 % 84% 36% 76% 50% 72% 63% 18% SOEs in BI 58 98 69 296 184 4770 303 % 27% 50% 63% 85% 90% 84% 64% Firms in BI 352 250 982 630 2,480 3,693 666 % 70% 58% 73% 77% 91% 64% 75% Politicians in BI 997 1,073 748 1,898 4,368 5,270 8,557 % 57% 47% 68% 82% 83% 29% 89% Individuals in BI 501 1,043 802 2,263 4,518 8,259 4,306 % 26% 20% 38% 53% 64% 26% 46% Table 8: Network features (by country) Bulgaria Serbia Slovak Rep. Hungary Romania Russian Fed. Spain Nodes 1,934 2,499 2,620 5,106 11,593 22,016 13,888 Log(Nodes) 7.6 7.8 7.9 8.5 9.4 10.0 9.5 Edges 4,153 6,811 5,077 13,604 32,923 66,246 46,369 Average Degree 2.1 2.7 1.9 2.7 2.8 3.0 3.3 Density 0.1% 0.1% 0.1% 0.1% 0.0% 0.0% 0.0% Clustering Coefficient 1.3% 8.6% 1.7% 0.8% 1.7% 0.4% 1.4% Average Distance 6.3 7.6 6.6 5.9 5.3 5.2 5.1 Random Clust Coeff 0.1% 0.1% 0.2% 0.1% 0.0% 0.0% 0.0% Random Ave Dist 8 7.8 8.2 7.9 8.2 8.6 7.8 S for small-world 21.65 87.31 13.1 11.3 115.77 47.65 86.26 Table 9: Betweenness by type and country Bulgaria Serbia Slovak Rep. Hungary Romania Russian Fed. Spain Betweenness parties 33% 26% 24% 21% 32% 13% 38% Betweenness SOEs 23% 18% 32% 40% 15% 52% 15% Betweenness firms 2% 8% 2% 1% 1% 4% 5% Betweenness politicians 36% 39% 30% 25% 38% 20% 35% Betweenness Individuals 7% 9% 11% 13% 13% 11% 7% 100% 100% 100% 100% 100% 100% 100% 33 Table 10: Baseline Estimates Pooled (1) (2) (3) (4) (5) VARIABLES ROA logSales logAvOutput logAvWage logLev Connect -0.616*** 0.427*** 0.152*** 0.221*** -0.031 (0.230) (0.028) (0.023) (0.013) (0.020) SOE -3.749*** 0.175*** -0.335*** 0.263*** -0.045*** (0.135) (0.019) (0.016) (0.007) (0.014) Constant 12.331*** 10.236*** 9.402*** 7.700*** -0.932*** (0.062) (0.007) (0.006) (0.003) (0.005) Wald 138.017 55.677 297.127 8.142 .322 p-value <0.001 <0.001 <0.001 .004 .570 Observations 10,541,077 11,178,831 8,846,434 7,465,385 10,945,643 R-squared 0.039 0.499 0.444 0.660 0.065 Sector Dummies Size Dummies Year FE Country FE Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 34 Table 11: Network regressions: log Sales Pooled (1) (2) (3) (4) (5) (6) logSales logSales logSales logSales logSales logSales SOE 0.175*** 0.175*** 0.174*** 0.174*** 0.175*** 0.175*** (0.019) (0.019) (0.019) (0.019) (0.019) (0.019) Connect 0.310*** 0.310*** 0.180*** 0.223*** 0.392*** 0.279*** (0.052) (0.052) (0.054) (0.070) (0.040) (0.050) BigI 0.148** 0.053 -0.081 (0.062) (0.068) (0.107) HasBetween 0.189*** 0.254*** (0.065) (0.082) logDegree 0.167*** 0.102* (0.032) (0.060) BigI×logDegree 0.088 (0.075) Politician 0.075 0.128* (0.056) (0.067) Politician×HasBetween -0.082 (0.119) Constant 10.236*** 10.236*** 10.236*** 10.236*** 10.236*** 10.236*** (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) Observations 11,178,831 11,178,831 11,176,694 11,176,694 11,178,831 11,178,831 R-squared 0.499 0.499 0.499 0.499 0.499 0.499 Sector Dummies Size Dummies Year FE Country FE Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 35 Table 12: Network regressions: log Average Output Pooled (1) (2) (3) (4) (5) (6) logAvOut logAvOut logAvOut logAvOut logAvOut logAvOut SOE -0.335*** -0.335*** -0.335*** -0.335*** -0.335*** -0.335*** (0.016) (0.016) (0.016) (0.016) (0.016) (0.016) Connect 0.079* 0.079* 0.042 -0.008 0.208*** 0.110*** (0.047) (0.047) (0.042) (0.060) (0.034) (0.040) BigI 0.093* 0.047 0.096 (0.054) (0.057) (0.084) HasBetween 0.093* 0.229*** (0.052) (0.070) logDegree 0.075*** .100*** (0.025) (0.049) BigI×logDegree -0.046 (0.059) Politician -0.113** -0.002 (0.046) (0.053) Politician×HasBetween -0.267*** (0.097) Constant 9.402*** 9.402*** 9.402*** 9.402*** 9.402*** 9.402*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Observations 8,846,434 8,846,434 8,844,757 8,844,757 8,846,434 8,846,434 R-squared 0.444 0.444 0.444 0.444 0.444 0.444 Sector Dummies Size Dummies Year FE Country FE Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 36 Table 13: Network regressions: log Average Wage Pooled (1) (2) (3) (4) (5) (6) logAvWag logAvWag logAvWag logAvWag logAvWag logAvWag SOE 0.263*** 0.263*** 0.263*** 0.263*** 0.263*** 0.263*** (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) Connect 0.154*** 0.154*** 0.073*** 0.077* 0.261*** 0.176*** (0.028) (0.028) (0.027) (0.041) (0.017) (0.024) BigI 0.084*** 0.008 -0.008 (0.031) (0.035) (0.054) HasBetween 0.146*** 0.192*** (0.028) (0.034) logDegree 0.097*** 0.098*** (0.015) (0.030) BigI×logDegree 0.001 (0.035) Politician -0.083*** -0.031 (0.025) (0.033) Politician×HasBetween -0.104** (0.051) Constant 7.700*** 7.700*** 7.700*** 7.700*** 7.700*** 7.700*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Observations 7,465,385 7,465,385 7,463,907 7,463,907 7,465,385 7,465,385 R-squared 0.66 0.66 0.66 0.66 0.66 0.66 Sector Dummies Size Dummies Year FE Country FE Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 37 Table 14: Network regressions: ROA Pooled (1) (2) (3) (4) (5) (6) ROA ROA ROA ROA ROA ROA SOE -3.749*** -3.748*** -3.748*** -3.748*** -3.749*** -3.747*** (0.135) (0.135) (0.135) (0.135) (0.135) (0.135) Connect -0.825* -0.826* 0.574 0.005 -1.334*** -0.602 (0.481) (0.481) (0.460) (0.665) (0.307) (0.411) BigI 0.263 1.415** 1.197 (0.547) (0.612) (0.921) HasBetween -2.161*** -1.562** (0.521) (0.615) logDegree -0.705*** -1.115** (0.251) (0.476) BigI×logDegree 0.189 (0.578) Politician 1.610*** 1.406** (0.462) (0.597) Politician×HasBetween 0.126 (0.945) Constant 12.331*** 12.329*** 12.332*** 12.332*** 12.329*** 12.329*** (0.062) (0.062) (0.062) (0.062) (0.062) (0.062) Observations 10,541,077 10,541,077 10,538,873 10,538,873 10,541,077 10,541,077 R-squared 0.039 0.039 0.039 0.039 0.039 0.039 Sector Dummies Size Dummies Year FE Country FE Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 38 Table 15: Network regressions: log Leverage Pooled (1) (2) (3) (4) (5) (6) logLev logLev logLev logLev logLev logLev SOE -0.045*** -0.045*** -0.045*** -0.045*** -0.045*** -0.045*** (0.014) (0.014) (0.014) (0.014) (0.014) (0.014) Connect -0.079* -0.079* -0.017 0.007 0.006 -0.036 (0.043) (0.043) (0.038) (0.054) (0.030) (0.037) BigI 0.06 0.036 -0.036 (0.048) (0.052) (0.077) HasBetween 0.046 0.087 (0.045) (0.060) logDegree -0.012 -0.097** (0.022) (0.044) BigI×logDegree 0.099* (0.053) Politician -0.083** -0.042 (0.039) (0.049) Politician×HasBetween -0.086 (0.080) Constant -0.932*** -0.932*** -0.932*** -0.932*** -0.931*** -0.931*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Observations 10,945,643 10,945,643 10,943,381 10,943,381 10,945,643 10,945,643 R-squared 0.065 0.065 0.065 0.065 0.065 0.065 Sector Dummies Size Dummies Year FE Country FE Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 39 Appendix A Country-specific regressions Table A1: Baseline Estimates using Connected Dummy ROA lnSales lnAvOutput lnAvWage lnLev BG Connect -2.804*** 0.635*** 0.303*** 0.229*** 0.162** (0.996) (0.08) (0.063) (0.045) (0.067) SOE -14.319*** 0.282*** -0.717*** 0.458*** 0.145*** (0.522) (0.047) (0.042) (0.019) (0.04) SE Connect -3.000*** 0.024 -0.153* 0.233*** -0.094* (0.829) (0.089) (0.078) (0.035) (0.050) SOE -7.463*** 0.099 -0.949*** 0.288*** -0.235*** (0.430) (0.073) (0.060) (0.024) (0.039) SL Connect -0.904* -0.039 -0.078 0.061 0.075* (0.496) (0.055) (0.055) (0.038) (0.041) SOE -2.046*** 0.336*** 0.032 0.402*** -0.043 (0.771) (0.089) (0.09) (0.05) (0.067) HU Connect 0.708 0.200*** 0.084 0.053 -0.267*** (0.766) (0.076) (0.066) (0.043) (0.057) SOE -2.829*** 0.668*** -0.169 0.529*** -0.095 (0.717) (0.103) (0.111) (0.050) (0.067) RO Connect 0.259 0.644*** 0.220*** 0.262*** -0.203*** (0.376) (0.037) (0.027) (0.017) (0.027) SOE -5.342*** 0.338*** -0.448*** 0.496*** -0.284*** (0.363) (0.049) (0.04) (0.022) (0.036) SP Connect -0.937*** 0.624*** 0.237*** 0.219*** 0.210*** (0.340) (0.056) (0.048) (0.018) (0.045) SOE -0.949*** 0.111*** -0.101*** 0.162*** -0.031* (0.140) (0.021) (0.020) (0.008) (0.019) Year FE Sector Dummies Size Dummies Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 40 Table A2: Network estimates log Sales Model BigI Big+Big×Betw Politician Between×Politician LogDegree Big×LogDegree Given Bet=0 Bet=1 Pol=0 Pol=1 B=0 B=1 Effect BigI Big Big Politician Bet Bet LogDegree LogDegree LogDegree BG 0.525 *** 0.393 ** 0.935 *** -0.091 0.530 0.603 ** 0.425*** 0.277 0.264 (0.176) (0.180) (0.255) (0.177) (0.437) (0.274) (0.109) (0.234) (0.283) SE -0.144 -0.297 0.017 -0.259 0.291 0.282 -0.022 0.024 0.074 (0.197) (0.228) (0.241) (0.203) (0.368) (0.349) (0.090) (0.166) (0.219) SL 0.334 *** 0.295 ** 0.418 *** 0.019 0.327 -0.071 0.074 -0.005 0.017 (0.114) (0.123) (0.151) (0.108) (0.201) (0.202) (0.066) (0.117) (0.153) 41 HU -0.207 -0.287 -0.135 -0.158 0.098 0.212 -0.124 0.056 -0.255 (0.187) (0.207) (0.204) (0.148) (0.238) (0.261) (0.110) (0.203) (0.260) RO -0.099 -0.241 0.030 -0.136 0.384 * 0.221 * 0.105* -0.046 0.206 (0.177) (0.185) (0.185) (0.108) (0.212) (0.123) (0.056) (0.149) (0.161) SP -0.095 -0.205 -0.033 -0.066 0.130 0.496 0.171** 0.061 0.197 (0.124) (0.153) (0.138) (0.151) (0.215) (0.342) (0.070) (0.101) (0.143) Year FE Sector Dummies Size Dummies Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 Table A3: Network estimates Log Ave Output Model BigI Big+Big×Betw Politician Between×Politician LogDegree Big×LogDegree Given Bet=0 Bet=1 Pol=0 Pol=1 B=0 B=1 Effect BigI Big Big Politician Bet Bet LogDegree LogDegree LogDegree BG 0.400 *** 0.313 ** 0.667 *** -0.184 0.081 0.494 *** 0.337*** 0.478** -0.129 (0.135) (0.141) (0.179) (0.142) (0.302) (0.181) (0.085) (0.200) (0.235) SE -0.078 -0.154 -0.006 -0.100 0.415 0.156 -0.015 -0.032 0.149 (0.174) (0.185) (0.213) (0.173) (0.279) (0.254) (0.073) (0.183) (0.209) SL 0.249 *** 0.240 * 0.266 * -0.161 0.315 * -0.251 0.064 0.032 -0.069 (0.122) (0.127) (0.159) (0.110) (0.19) (0.204) (0.072) (0.102) (0.153) 42 HU -0.170 -0.246 -0.105 -0.212 0.100 0.289 -0.022 -0.018 0.042 (0.169) (0.181) (0.185) (0.132) (0.189) (0.226) (0.098) (0.169) (0.210) RO -0.043 -0.051 -0.035 -0.100 -0.002 0.019 -0.01 0.051 -0.054 (0.117) (0.122) (0.123) (0.073) (0.133) (0.087) (0.045) (0.091) (0.104) SP -0.054 -0.233 ** 0.070 -0.452 *** 0.269 * -0.057 0.092** 0.099 0.000 (0.106) (0.114) (0.125) (0.095) (0.145) (0.216) (0.046) (0.082) (0.101) Year FE Sector Dummies Size Dummies Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 Table A4: Network estimates Log Ave Wage Model BigI Big+Big×Betw Politician Between×Politician LogDegree Big×LogDegree Given Bet=0 Bet=1 Pol=0 Pol=1 B=0 B=1 Effect BigI Big Big Politician Bet Bet LogDegree LogDegree LogDegree BG 0.140 0.127 0.173 -0.230 ** -0.117 0.022 0.154*** 0.283** -0.111 (0.118) (0.124) (0.141) (0.113) (0.236) (0.126) (0.059) (0.115) (0.137) SE -0.182** -0.301 *** -0.069 -0.175 ** 0.221 0.299 *** 0.056* 0.093 0.098 (0.082) (0.102) (0.087) (0.078) (0.177) (0.119) (0.033) (0.066) (0.081) SL 0.140 0.103 0.211 ** -0.067 0.338 *** -0.087 0.095** 0.101 -0.053 (0.093) (0.100) (0.103) (0.081) (0.121) (0.131) (0.047) (0.074) (0.100) 43 HU 0.290 ** 0.272 ** 0.305 ** -0.170 * 0.140 -0.135 0.119** 0.167 -0.135 (0.122) (0.127) (0.133) (0.09) (0.123) (0.129) (0.056) (0.131) (0.144) RO -0.082 -0.150 * -0.021 -0.162 *** 0.121 0.111 ** 0.055** 0.012 0.065 (0.080) (0.084) (0.082) (0.048) (0.095) (0.054) (0.026) (0.071) (0.076) SP 0.084 ** -0.014 0.150 *** -0.258 *** 0.064 0.081 0.074*** 0.123*** -0.092** (0.040) (0.047) (0.046) (0.046) (0.055) (0.112) (0.021) (0.034) (0.044) Year FE Sector Dummies Size Dummies Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 Table A5: Network estimates ROA Model BigI Big+Big×Betw Politician Between×Politician LogDegree Big×LogDegree Given Bet=0 Bet=1 Pol=0 Pol=1 B=0 B=1 Effect BigI Big Big Politician Bet Bet LogDegree LogDegree LogDegree BG 4.148 * 4.804 * 2.347 5.798 ** -5.486 -0.671 0.87 2.968 -5.534 (2.497) (2.591) (3.017) (2.865) (5.656) (2.905) (1.072) (3.987) (4.336) SE 1.490 1.979 1.003 -0.538 -0.062 -0.952 -0.959 -2.931* 1.46 (1.902) (2.269) (2.181) (1.863) (3.575) (3.100) (0.826) (1.543) (1.855) SL -0.979 -0.109 -2.613 * 3.106 *** -2.572 -3.017 * -1.938*** -3.012*** 1.354 (1.184) (1.251) (1.446) (0.978) (1.749) (1.747) (0.552) (0.880) (1.209) 44 HU -1.050 1.407 -3.270 * 1.521 -6.498 ** -3.097 -1.508 -1.832 0.338 (1.769) (2.064) (1.953) (1.600) (2.599) (2.719) (0.978) (1.507) (2.086) RO -0.386 0.979 -1.576 1.575 -4.283 * -1.630 -0.979* 1.767 -3.089** (1.816) (1.910) (1.884) (1.170) (2.260) (1.247) (0.534) (1.462) (1.575) SP -0.667 0.385 -1.198 -0.047 -1.497 * -1.911 -0.635* -1.184* 0.735 (0.767) (0.847) (0.864) (0.842) (0.906) (2.144) (0.347) (0.698) (0.819) Year FE Sector Dummies Size Dummies Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 Table A6: Network estimates Log Leverage Model BigI Big+Big×Betw Politician Between×Politician LogDegree Big×LogDegree Given Bet=0 Bet=1 Pol=0 Pol=1 B=0 B=1 Effect BigI Big Big Politician Bet Bet LogDegree LogDegree LogDegree BG -0.168 -0.248 0.291 -0.530*** 0.657* -0.757* 0.166* 0.202 0.187 (0.151) (0.155) (0.184) (0.179) (0.375) (0.400) (0.094) (0.168) (0.224) SE -0.139 -0.02 -0.226* -0.159 -0.105 -0.21 -0.099* -0.093 0.005 (0.111) (0.115) (0.127) (0.105) (0.159) (0.235) (0.057) (0.117) (0.142) SL 0.238** 0.198* 0.113 -0.093 -0.101 0.021 0.035 -0.005 -0.039 (0.096) (0.102) (0.098) (0.081) (0.095) (0.181) (0.052) (0.071) (0.108) 45 HU 0.029 -0.09 0.222* 0.144 0.302** -0.327 -0.021 -0.352*** 0.526*** (0.145) (0.157) (0.125) (0.115) (0.152) (0.234) (0.077) (0.120) (0.173) RO 0.185 0.196 -0.021 0.014 -0.144 0.241 0.038 -0.101 0.141 (0.131) (0.136) (0.074) (0.078) (0.132) (0.155) (0.038) (0.134) (0.140) SP 0.219** 0.208* 0.016 -0.007 0.146 -0.14 0.032 -0.108 0.135 (0.094) (0.110) (0.104) (0.095) (0.101) (0.284) (0.049) (0.075) (0.103) Year FE Sector Dummies Size Dummies Robust standard errors in parentheses clustered at the level of firm *** p<0.01, ** p<0.05, * p<0.1 Appendix B Other tables Table B1: Size firm definitions Companies meeting at least one of the criteria are included: Very large companies* - Operating revenue (turnover) >= 100 mln EUR; - Total assets >= 200 mln EUR; - Number of employees >= 1 000; - Listed company. Large companies** - Operating revenue (turnover) >= 10 mln EUR; - Total assets >= 20 mln EUR; - Number of employees >= 150; - Not related to the category of very large companies. Medium sized companies*** - Operating revenue (turnover) >= 1 mln EUR; - Total assets >= 2 mln EUR; - Number of employees >= 15; - Not related to the category of large companies. Small companies - Companies not classified by any other of the above categories are included. *Companies with Operating revenue per employee or Total assets of less than 100 EUR, are ex- cluded from this category. Companies with unknown operating revenue, total assets or number of employees, but with capital exceeding 5 mln EUR, are included in this category. **Companies with Operating revenue per employee or Total assets of less than 100 EUR, are excluded from this category. Companies with unknown operating revenue, total assets or number of employees, but with capital from 500 th. EUR to 5 mln EUR, are included in this category. ***Companies with Operating revenue per employee or Total assets of less than 100 EUR, are excluded from this category. Companies with unknown operating revenue, total assets or number of employees, but with capital from 50 th. EUR to 500 th. EUR, are included in this category. 46 Table B2: Size by Matching Method (N and %) FN PN Total SOE FN PN SOE (FN+PN) Small 535 2,595 3,130 4,259 40% 77% 31% Medium 402 574 976 6,749 30% 17% 48% Large 245 103 348 2,267 18% 3% 16% Very large 165 94 259 651 12% 3% 5% Total 1,347 3,366 4,713 13,926 100% 100% 100% Table B3: Sector Distribution and by Matching Method (N and %) FN PN Total SOE FN PN SOE (FN+PN) A - Agriculture, fore 22 120 142 227 1.6% 3.6% 1.6% B - Mining and quarry 23 9 32 47 1.7% 0.3% 0.3% C - Manufacturing 125 301 426 747 9.3% 8.9% 5.4% D - Electricity, gas 93 21 114 541 6.9% 0.6% 3.9% E - Water supply, sew 23 19 42 632 1.7% 0.6% 4.5% F - Construction 133 212 345 984 9.9% 6.3% 7.1% G - Wholesale 183 632 815 932 13.6% 18.8% 6.7% H - Transportation 30 53 83 511 2.2% 1.6% 3.7% I - Accommodation 46 161 207 238 3.4% 4.8% 1.7% J - Information 99 217 316 372 7.3% 6.4% 2.7% K - Financial 145 103 248 890 10.8% 3.1% 6.4% L - Real estate 124 260 384 846 9.2% 7.7% 6.1% M - Professional 148 784 932 960 11.0% 23.3% 6.9% N - Administrative 84 209 293 537 6.2% 6.2% 3.9% O - Public admin 3 2 5 359 0.2% 0.1% 2.6% P - Education 7 42 49 3,969 0.5% 1.2% 28.5% Q - Human health 15 121 136 613 1.1% 3.6% 4.4% R - Arts, entertain 15 64 79 410 1.1% 1.9% 2.9% S - Other service act 29 36 65 108 2.2% 1.1% 0.8% Total 1,347 3,366 4,713 13,923 100.0% 100.0% 100.0% 47 Appendix C Network Mapping by country Figure C1: Bulgaria Big Island Figure C2: Serbia Big Island 48 Figure C3: Slovak Republic Big Island Figure C4: Hungary Big Island 49 Figure C5: Romania Big Island Figure C6: Russian Federation Big Island 50 Figure C7: Spain Big Island 51