Does Medieval Trade Still Matter? Historical Trade Centers, Agglomeration and Contemporary Economic Development

Fabian Wahl∗ University of Hohenheim

August 2, 2013

Abstract This study empirically establishes a link between medieval trade, agglomeration and contemporary regional development in ten European countries. It documents a statistically and economically significant positive relationship between prominent involvement in medieval trade and regional economic development today. This finding is robust to inclusion of various historical, economical and geographical control variables and to controlling for endogeneity via IV estimations. A mediation analysis shows that, as theoretically postulated, the majority of this long-lasting effect trans- mits via the impact of medieval trade on contemporary agglomeration and industry concentration. Thus, this research highlights the long-run importance of medieval trade in shaping contemporary spatial patterns of economic activity throughout Europe. The path-dependent regional development processes caused by medieval trading activity can also provide an explanation for the observed persistence of regional differences in development across the considered European countries.

Keywords: Medieval Trade, Agglomeration, Regional Economic Development, Path- Dependency, New Economic Geography JEL Classiﬁcation: F14, N73, N93, O18, R12

∗Department of Economics, University of Hohenheim. Chair of Economic and Social History, Speise- meisterﬂ¨ugel,Stuttgart, Germany. [email protected]. The author especially would like to thank Bas van Bavel, Sibylle Lehmann, Alexander Opitz, Alfonso Sousa-Poza, Oliver Volckart and Nicole Waidlein for the helpful comments and discussions. Additionally he is indebted to T. Matthew Ciolek for his helpful suggestions and for discussing his medieval European trade route maps.

1 1 Introduction

There is ample evidence that trade is an important determinant of both long- and short- run economic development. However, most of the existing literature focuses on the impact of 19th century trade on market integration or the “Great Divergence” (e.g. Galor and Mountford 2008 or O’Rourke and Williamson 2002) or on the impact of contemporary, Post-World War II trade activities on recent economic growth and development performance across countries (Dollar and Kraay 2003, Frankel and Romer 1999). There is only one study (Acemoglu et al. 2005) considering the effect of cross country trade in earlier periods. They investigate the impact of long-distance overseas trade on institutional developments and the pre-industrial development process across European countries. Hence, until now there is no study exploring the possible long-lasting effects of trade in European cities throughout the High and Late Middle Ages. The importance of medieval trade for the development of cities and regions in the Middle Ages and the following centuries is well-known and widely acknowledged. Apart from this, no research acknowledged the fact that medieval trade could have also long-term influences on regional development persisting until today. This despite the fact that medieval trade through its influence on agglomeration and spatial concentration of industry could have lead to path-dependent regional development processes resulting in development differences outlasting the centuries in between. The aim of this study is to investigate whether medieval trade had caused differences in regional development which are still visible today due to its its impact on agglomeration. If this is the case it could provide a new explanation for the uneven distribution of economic activity and significant spatial concentration of industries throughout Europe (e.g. Chasco et al. 2012, Koh and Riedel 2012, Roos 2005). Furthermore, it can contribute to the understanding of persistent differences in regional economic development (Becker et al. 2010, Maseland 2012, Tabellini 2010 or Waidlein 2011). To establish a link between medieval trade, agglomeration and contemporary performance we link typical characteristics of medieval trade and cities to the determinants of agglomeration suggested by New Economic Geography (NEG) and agglomeration economics (e.g. Krugman 1991, Glaeser et al. 1992). In a second step, based on studies combining NEG and endogenous growth models and the theory of path-dependence (David 2007) we propose a positive connection between agglomeration, industrial concentration and contemporary development. Afterwards, we test the causal chain from medieval trade through agglomeration to

2 contemporary regional economic development using a rich data set (where we choose a NUTS-3 region as unit of observation) and a wide range of empirical methods. In general, the detailed empirical analysis shows indeed medieval trade is robustly associated with contemporary regional economic performance. Moreover, we also find that the majority of the effect of medieval trade on contemporary regional development can be explained by its influence on agglomeration patterns. Most importantly, we show that our hypotheses are robust to the inclusion of many geographical, political, economical and historical covariates of development and agglomeration and are not biased by endogeneity. Finally, a mediation analysis shows that medieval trade activities are strong predictors of today’s spatial distribution of economic activity and population and that around two third of the influence of medieval trade on contemporary regional GDP per capita can be attributed to this influence of medieval trade on agglomeration. The remainder of the article proceeds as follows. First, we theoretically establish the link between medieval trade, agglomeration and present-day’s economic development. Afterwards, we introduce and discuss the most important variables and data and explain the empirical setting. Next, we conduct our empirical analysis and interpret and discuss the results in detail. At last, we conclude and summarize the main findings.

2 Theory and Hypotheses

It is a well established idea that trade was a decisive factor in the development of medieval cities and the revival of city growth during the period of the so called “Commercial Rev- olution” (e.g. B¨ornerand Severgnini 2012, Epstein 2000, Habermann 1978, Holtfrerich 1999, King 1985, Postan 1952, Pounds 2005 and van Werveke 1952). History provides many examples of cities owing their importance primarily to their function as centers of trade, like the German cities of Nuremburg (Nicholas 1997), Frankfurt (Holtfrerich 1999) or Cologne (King 1985) or the Polish city of Gdansk.1 Using concepts developed by NEG (Krugman 1991) and agglomeration economics, one

1 Obviously, there are exceptions from this story, i.e. cities and regions becoming large and important agglomerations without being important in medieval trade. This is true for example for Stuttgart (the sixth largest German city today) and Munich two of the richest and economically prosperous cities and agglomeration areas in present day’s Germany. Stuttgart was not important until after the Napoleonic Wars it became the capital of the newly founded kingdom of W¨urttemberg. The rise of Munich (today the third largest city of Germany) followed a similar pattern, albeit the capital of a kingdom and residence of a bishop (and later archbishop) Munich began to become a large city not before late 18th century. Again, it experienced large population growth in the nineteenth century after the Napoleonic Wars until World War I. Even more, Bavaria and Munich as it’s center stayed relatively poor till the 1950ies (when e.g. the Siemens corporation moved its headquarter from Berlin to Munich). Additionally, the largest agglomeration in Germany the Ruhr Area largely results from

3 can explain why medieval trade was important for the rise of cities in medieval Europe. This can be done by linking the characteristics of medieval trade and trade cities to second nature causes of agglomeration (for an overview over these see e.g. Christ 2009, Glaeser et al. 1992, Henderson et al. 2001). In medieval times, the economy, especially the urban economy was characterized by a high degree of regional specialization (Am- mann 1955, King 1985, Lopez 1952, Nicholas 1997,Postan 1952, Pounds 2005 and van Werveke 1963).2 The Southern German cities that became important trade centers in the later medieval for instance were specialized in textile (Barchent etc.) and paper production, while other areas had specialized in mining (like e.g. the Saxon town of Freiberg or Liègein today’s Belgium that had the this times most productive coal field), or in food and salt (where the cities at the French Atlantic coast were the main exporters). The different regions exported in what they were specialized in – or had an comparative advantage in e.g. due to natural endowments– and imported what they did not have themselves.3 This specialization of trade cities on a particular industry or sector gave rise to the existence of technological (non-pecuniary) externalities like Marshall-Arrow- Romer (MAR) externalities (Marshall 1890, Romer 1986) or Porter externalities.4 Those type of externalities arise as knowledge spillovers between firms in the same industry and contribute therefore to the growth of both industry and city (Glaeser et al. 1992).5. In- deed Epstein (1998) and more broadly Epstein and Prak(2008) in an anthology about the Guilds and Innovation they edited show that the guild as the dominant economic institution of the later medieval city indeed could have fostered innovation and enable knowledge spillovers and diffusion within the urban economy (and through migration also between cities).6 A second important characteristic of medieval trade cities was the comparatively high

the rich endowments with coal and iron making it to one of the most important nucleus of German industrialization. 2A comprehensive illustration of medieval trade activities is provided in Postan (1952) and Lopez (1952). 3A review of the general geographical patterns of trade and industry specialization in the middle ages is provided among others by King (1985). 4Nicholas (1997) additionally points to the fact that over the course of the Middle Ages the industry dominating in a city e.g. the textile industry did more and more diversify. This intra-industry diversification could be an additional channel through which technological externalities could had been arisen. 5Such knowledge spillovers between firms might appear because of imitations, movements of skilled workers between the different firms in the industry etc. 6For evidence about the high mobility of skilled craftsmen in this period see Reith (2008) in this anthology. Of course, among historians there is no consensus about the role of guild and whether their negative or positive effects for economic development are more dominant. However, at least the more recent contributions clearly brought forward evidence that guilds indeed could had large positive impacts through their positive influence on innovativeness.

4 variety of available goods. Those varieties of goods were available first at the local markets, then at the big trade fairs in the Champagne and other important trade cities (like Frankfurt, Cologne, Ulm etc.) and then, in the late medieval age in the branches and kontors of the Hanseatic League and the trading companies (“super-companies”) like the Fugger in Augsburg78 Especially the latter two also provide supply with luxury goods and exotic commodities from far east, as long-distance trade was reestablished at the beginning of Late Middle Ages. This high variety can be considered as an important demand-side driven agglomeration force, because it makes it more attractive to settle in a city.9 Additionally, the large variety of goods and prospering industry gave rise to the self- reinforcing circular causation caused by backward and forward linkages and leading to agglomeration and core-periphery patterns in NEG models (Krugman 1991, Ottaviano and Thisse 2004). Because trade cities provided a higher variety of goods, employment for high-skilled specialized workers and –as consequence of the higher labor demand– also higher wages, they attracted additional workers. When more and more workers made use of the opportunity to work in the city as e.g. textile workers or craftsmen, employment and the number of firms increased. This decreased the price index, raised real wages and therefore resulted in the immigration of even more workers to the city. Consequently, this pecuniary externality (forward linkage) caused increased agglomeration and industry concentration in the city. Supplementary, more workers lead to a higher demand for goods produced and/or traded in the city. The higher demand once more lead to the expansion of markets and industries, raising labor demand and real wages resulting again in additional immigration. This is the so called “home market effect” or the backward linkage. In short, this amounts to the logic that industry will tend to concentrate where there is a large market, whereas the market is large at the area where industry is already located. Thus, forward and backward linkages constitute the virtuous circle that generates agglomeration and uneven spatial distribution of population and economic activity.10

7For a detailed description of the business activities of the Hanseatic League the reader is reffered to Dollinger (1966). An illustration of the medieval early medieval markets and fairs is found in van Werveke (1963). 8A comprehensive description of the medieval super-companies can be found in Hunt and Murray (1999). An transaction economic analysis of the super companies using the example of the Fugger is provided by Börner(2002). 9This follows clearly from love of variety preferences commonly assumed in NEG models. Additionally, one can make a transaction cost argument, because e.g. when living in a city there are no costs of transporting the sold commodities back to the village. 10Of course, the medieval city was a highly cartelized and regulated economy with dominant guilds and significant rent-seeking activities (e.g. Braudel 1986). However as Braudel (1986) concludes since the

5 Furthermore, after the process of agglomeration lasted for some time other types of technological externalities occurred. Conditional on certain factors (i.e. geographical position or natural endowments) other industries located in the previously specialized cities, e.g. in the Southern German city of Ravensburg (an important trade center in the 15th century) the traditional textiles industry was supplemented by paper production at the beginning of the 15th century (Schelle 2000). In addition, there were also incentives to locate in a trade city for firms using special commodities as inputs or that produced inputs used in the industry the city was specialized in.11 Therefore also Jacobs (1969) externalities occurred in the late medieval cities.12 A first test whether the story fits to stylized empirical facts about city population and city growth in the Middle Ages delivers the regressions in Table1. There we regress However, the main argument of this paper is that medieval trade has significant conse- quences on economic development today. Reassuringly, the self-reinforcing nature of the described agglomeration and concentration processes implies a path-dependent process of city development. This path-dependent development process results in differences in concentration of economic activity and population that are persistent until today. Cities involved in medieval trade activities over a sufficient period of time got locked in on a superior development path as compared to other cities which were not involved. This is a typical characteristic of processes caused by increasing returns or positive feedback and that are characterized by multiple equilibria (David 2007). There are many examples of historical events and phenomenons having long-run impacts on economic development, e.g. Colonization (e.g. Acemoglu et al. 2001, 2002), Slave Trade (Nunn 2008, 2011), the Neolithic revolution (e.g. Ashraf and Galor 2011, Olsson and Hibbs 2005 or Putter- man 2008), the capacity to adopt and develop new technologies (Comin et al. 2010) or the timing of human settlement (Ahlerup and Olsson 2012).13 Additionally, Maseland (2012) shows, that regional development disparities in Germany are persistent and can

13th century something like market integration (to some extent) existed with prices varying in the markets of cities every week according to supply and demand. Furthermore, the increasing spread of the “Verlagssystem” sometimes might had limited the power of the guilds. Concerning the urban rural wage differential evidence in general is limited for this period in time Braudel (1986) notes that in general, also due to the power of guilds the wages in the city can be considered to be usually higher than in rural areas. In line with this, Munro (2002) comparing the real wages in England and Flanders between 1300 and 1500 found that the real wages in the cities were higher than in rural areas and showed a higher downward rigidity. van Bavel and van Zanden (2004) in addition notice that in pre industrial societies the relationship between city size and nominal wages usually was positive. 11 The idea that vertical linkages along the supply chain can lead to agglomeration is developed in Krugman and Venables (1995). 12Jacobs externalities are knowledge spillovers arising between firms of different industries. 13A comprehensive review of such events caused path-dependent developments is Nunn (2009).

6 largely be explained by strong and increasing diﬀerences between core areas and the periphery. We argue that medieval trade can be added to the list of such events. Finally, the positive connection between agglomeration, industry concentration and regional economic growth is reported by several theoretical studies (e.g. Baldwin and Martin 2004, Martin and Ottaviano 2001, Yamamoto 2003 or Bertinelli and Black 2004) linking growth e.g. through innovations and agglomeration by combining standard NEG and endogenous growth models. In addition, studies like Hohenberg and Lees (1995) or Fujita and Thisse (2002) also establish empirically the positive relationship between agglomeration and regional growth. In conclusion, we postulate the following two hypotheses about the relationship between medieval trade and contemporary regional development:

Hypothesis 1. There is a positive and signiﬁcant relationship between involvement in medieval trade activities and regional economic performance today, i.e. cities that were centers of medieval trade show a higher GDP per capita today than cities not involved in medieval trade.

Hypothesis 2. Medieval trade activities influence contemporary regional economic development through their positive effect on agglomeration and industry concentration, i.e. there is a positive and significant relationship between medieval trade centers, agglomeration and industry concentration measures and current regional economic development.

3 Data and Setting

3.1 Setting and Level of Analysis

Because medieval trade took place in cities and agglomeration is a regional phenomenon, our empirical analysis is based on regional level data. We stick to the NUTS (“Nomen- clature of Units for Territorial Statistic”) regional classification, the official regional reference unit systematic used in the European Union (EU).14 Furthermore the official regional statistics of Eurostat are available for those territorial units. Additionally, different regions on the same NUTS level have the advantage of being relatively comparable to each other since they are defined according to a particular range of inhabitants.15 We

14A detailed description and overview over the NUTS classiﬁcation scheme and the regions can be found in the data appendix and the references mentioned there. 15Although the population thresholds are deﬁned very widely, e.g. a NUTS-3 region can have 150.000 and 800.000 inhabitants. Again, sometimes there are exceptions so that some NUTS-3 regions show a larger population. From this it follows also, that more densely populated regions cover on average a smaller area. To overcome potential biases resulting from the this, we will control for the area of

7 choose to conduct our analysis on the most disaggregated level for which our essential data (e.g. GDP per capita) is available. Therefore we conduct our analysis with a NUTS-3 region as observational unit. NUTS-3 regions are identical to existing administrative units in most of the countries in our sample, which is an additional advantage of using them. In Germany for example they are mostly identical to districts or district-free cities, in France to Departments and in Italy to Provinces. A potential bias resulting from considering regions instead of actual cities that were subject to medieval trade is limited as heterogeneity within NUTS- 3 regions should not be of significant size. However, some control variables are available only at NUTS-2 or NUTS-1 level. In these cases we include the respective variables at the level where they are provided. Another advantage of sticking to the NUTS classification is that it enables to use fixed effects for the different NUTS-levels (countries, federal states etc.). This allows to appropriately handle all kinds of heterogeneity on country and regional levels. Besides this, one can also account for cross-sectional and spatial dependence among the regions in the dataset. The latter being a important advantage of regional empirical analyses especially when compared to country level investigations.16

3.2 Dependent Variables and Agglomeration Measures

As dependent variable we use the natural logarithm (ln) of GDP per capita in a NUTS-3 region, originating from the Eurostat regional statistics database. We take the latest available values from the year 2009. All other time-variant variables are also taken from the year 2009 to enable comparability. As measure of spatial industry agglomeration we follow Roos (2005), Chasco et al. (2012) and others in using the ln of the relative GDP density as measure for the spatial distribution of economic activity. The measure is calculated by dividing a region’s share of GDP per capita through its share of the country’s total area. This means it shows whether the concentration of economic activity in a region is below or above the country’s average.17 As such this is a more direct measure of economic agglomeration than population density. Additionally, we present results using the ln of a regions population density in 2009 as a more general measure of agglomeration, i.e. as a variable identifying more densely populated places. These results are reported in Appendix C.

a region and introduce dummy variables for city districts, city states and districit-free cities (regions with a high population density, i.e. a large population but a small area). 16Chasco et al. (2012) discuss further advantages of using NUTS-3 regions as observational units in the context of spatial economic analyses. 17The exact formula according to which the relative GDP Density is calculated is shown in the data appendix.

8 We think that the relative GDP Density is a more direct measure of industry agglomeration and concentration and is therefore should more suitable for our empirical analysis. However, since population density might capture additional aspects of agglomeration that might be important for economic activities indirectly and therefore can provide additional insights. Table A.1 in the data appendix gives a descriptive overview over all variables used in the following empirical analysis. The exact sources and further explanations of all variables are provided in the data appendix.

3.3 Independent Variables

This study aims to investigate the impact of trade between cities during the medieval age.18 To be able to identify the theoretically assumed effect of medieval trade on agglomeration we focus on the most important trade cities, i.e. cities where trade probably had the most powerful and long-lasting impact. Since agglomeration is a long-lasting process unfolding its effects only after some time, it is important to ensure that trade took place long enough in a city to influence agglomeration there in a sufficient way. Stated differently, trade had to take place long enough in a city to lock it in on a superior development path. To account for this fact, we focus on important trade cities at the end of the medieval period (i.e. around 1500 AD). This is due to the fact that cities important at the end of the medieval period are most likely also having experienced noticeable trade activities in the years before (i.e. over a longer time period). Our main source of information about important medieval trade activities are maps printed in historical atlases or monographs. We focus on maps because they provide a much more comprehensive source of trade cities and activities then the information available historical monographs. In addition, their information usually can assigned to a certain period much clearer than that contained in books. In consequence, we collect information about cities prominently involved in trade from four historical maps provid- ing evidence about cities located on “major” or “important” trade routes around 1500

18It is important to note that between the breakdown of the Roman Empire and the early medieval (the foundation of Francia) there were comparatively small trade activities. Trade began to increase not before the tenth century (Postan 1952, Braudel 1986). Furthermore, after the end of the medieval in the course of the 16th century, overseas trade (e.g. with the colonies) and long-distance trade became increasingly important. Due to this, the character of trade (e.g. rising importance of slaves trade) as well as the leading trade centers changed (Spain and Portugal came to rise). Compatibly with that, the leading actors of medieval trade like e.g. the Hanseatic League lost their importance in the period following the medieval. In consequence, it is possible to isolate the medieval trade activities in cities from trade activities before and after the medieval. This ensures that the eﬀects we measure empirically can actually be attributed to medieval trade activities and not trade in general or trade in other periods.

9 AD (i.e. the late medieval). Due to the fact that there is no consensus or quantitative evidence about the exact importance of trade cities and trade routes during the medieval period we have to consult several different sources to become sufficiently reliable data. The first is a map printed in Davies and Moorehouse (2002), the second is a map printed in King (1985). The third source is a map on Central European trade published in Magocsi’s (2002) Historical Atlas of Central Europe.19 At last, we consult several maps included in “Westermanns Atlas zur Weltgeschichte” (Stier et al. (1956). More information about the kind of information and the geographical and temporal scope of those maps is provided in the Data Appendix. There, we also list the primary sources on the basis of which the maps are drawn – if we were able to identify them. We include a city if it is mentioned in one of these maps. We include only cities located in EU countries, since only for those the Eurostat regional statistics database provides data. Despite this, in some cases we included cities in the sample not mentioned by the maps but by other sources of information. For example, we include the eastern German city of “Zwickau” because it is prominently recognized in Spufford (2002) standard account on medieval commerce and is known for its importance in salt trade. In other cases, we included cities not mentioned in the maps but in other sources for robustness checks. Furthermore, we stick to other qualitative information in our judgment of the importance of the included trade cities. For example, we look whether a city was an important member of the Hanseatic League or a capital of a quarter or a third (like e.g. Dortmund or Cologne). Information about this is provided by Dollinger (1966). Additionally, we also look whether, especially for not so prominent trade cities (Paderborn, Soest, Harfleur, Tarent etc.) they lied on well-known trade routes like the “Hellweg” in German (as it is the case e.g. for Soest). Moreover, we consult several historical standard sources about medieval trade activities in different Central European regions (e.g. Dietze 1923, Hunt and Murray 1999, Schulte 1966, Spufford 2002 etc.) and look whether they mention a city as being prominently involved in trade or having an over-regional importance as market, trading place or fair city. Finally we also draw on other historical atlas like that by Kinder and Hilgemann (1970) and other e.g. regional trade route maps (e.g. Schulte 1966) as sources for validating the information in the primary maps.In the Data Appendix (Table A.3) we report and discuss all these source and provide information about which city is mentioned by which source. Overall these sources have left us with 119 trade cities located in 10 European coun-

19As we are not interested in information about only regionally important trade cities an additional reason for choosing this particular maps is that they provide cross-national information about trade activities.

10 tries. Our dataset encompasses all 839 NUTS-3 regions in these countries.20 The Data Appendix offers a detailed description of how we construct our database of important late medieval trade cities. Even with the information in these sources, the relative importance of cities is not always clear. Additionally, there is also a different degree of uncertainty about the extent and location of trade activities and the course of main routes, i.e. the actual importance of a certain trade route at a particular point in time is not always clear. However, there are cities that undoubtedly were important centers of trade like the Northern Italian city states (Milan or Genoa etc.), some Southern German imperial cities (like Augsburg, Nuremburg or Ulm) and the leading centers of the Hanseatic League (Hamburg, Bremen, Lübeck, Cologne etc.) . On the other hand, there are cases were only some sources mention the city as important trade center or lying on a main trade route, like in the case of Paderborn, Minden or some port cities in France e.g. Harfleur or some smaller cities in Italy (Brindisi,Mantoa or Udine). This uncertainty is a natural result of the qualitative — and therefore to some extent always subjective — nature of the collected information and the scares amount of overall information about the medieval period and the trade activities back then. To account for this uncertainties, we will re-estimate the most important of our empirical results with alternative samples of trade cities where we first remove cities mentioned only by one of our sources. Second, we exclude cities reported in some of the maps or sources but actually do not lay on a well-known important trade route, where no important member of the Hanseatic League (according to Dollinger and Stier et al. 1956) or are not mentioned by any of our other historical sources as being of notable importance in later medieval trade (albeit there was probably some extent of trade activity). Those cities include e.g. Amberg, Einbeck, Como, Paderborn, Parma or St. Melo. 21 What is more, we also conduct our empirical analysis with a sample of trade cities including additional cities (Dijon, Piacenza or Aigues Mortes) that are mentioned by some of the sources, but for which we — after consulting several different information about the history of the respecitve places — are in doubt of their actual importance in medieval trade, at least over a longer period. At last, we try to ensure that we do not include trade cities that only experienced significant trade activities for a a short period and therefore not long enough for resulting in a lock-in to a superior development path. To overcome this problem, that would downward bias our results, we construct a fourth alternative sample of trade centers

20We exclude the islands of Elba, Corse and Sicily from our sample because they are not comparable with regions on the continent with respect to trade ﬂows. (This follows Chasco et al. 2012 who also exclude island regions). 21A full list of excluded cities is reported in the Data Appendix.

11 only considering cities for which we found records of recognizable trade activities in earlier periods than the late 15th and early 16th century. The sources consulted here are e.g a volume about medieval trade in the Levant by Heyd (1879a,b) and the already mentioned volume about the history of German trade by Dietze (1923). Furthermore, also Dollinger (1966) presents some evidence about trade activities in the periods preceding the late medieval in a map, where he e.g. depicts cities lying on the Hellweg and “other important trade routes” in the period between 1286 and 1336. Additional, this map also reports the signers of the treaty of Smolensk in 1229 (i.e. the most important trade cities in this times Western Dvina trade) and additionally some information of maps digitized by the Old World Trade Routes Project (OWTRAD) website, primarily containing information about trade in Eastern Europe, especially Poland.22 Exact information about the construction of this alternative sample is provided in the Data Appendix. Such information about earlier trade activities could be collected for 70 of the originally 119 trade cities. As such, this last sample represents the most selective one and probably contains only cities for which important medieval trade activities are most sure. Overall, we consulted fifteen different sources to construct our different samples of trade cities. However, even with this amount of sources one cannot be sure that the coding of the trade city dummy variables is perfect. Regardless of this fact, there seems to be no reason why the inclusion of cities that were probably not that important than other cities or experienced trade activities for only a short period of time should more than downward bias our estimates. The estimates obtained using this kind of dummy variable should therefore considered to be a lower bound of the actual long-term effect of medieval trade. We will use two different variables as measures of late medieval trade and its impact on contemporary regional development. First, we will use a dummy variable “Trade Center” that is equal to one if a region includes at least one medieval trade city. The lack of quantitative information and the limited availability of qualitative judgments leads us to use a simple dummy variable coding important trade cities. Of course, this implies that we treat all trade cities being the same with respect to the scale of trade activities and agglomeration forces working there. However, since we try to focus on cities located on “major” or “important” trade cross-national trade routes and also rely on qualitative judgments of importance —when available– we should be able to reduce the heterogeneity among the trade cities. Additionally, the construction of a dummy variable allows also for the construction of a second variable “Distance to Trade Center” representing the distance (in degrees) between a region and the closest medieval trade

22http://www.ciolek.com/owtrad.html

12 city.footnoteThe variable is zero in regions that are coded as trade centers. This variable oﬀers a very useful direct test of our hypothesis that medieval trade contributed to the emergence of time persistent core-periphery patterns and therefore can act as a notable explanation for contemporary regional income diﬀerences. Table 1 provides a summary of our trade city data. For each country, the total number of NUTS-3 regions, the number of regions with trade cities, the share of trade center regions and the average distance of a region to the closest trade city is listed.

[Table 1 about here]

As reported in the table, the average distance to a medieval trade center is about 1.5 degrees (e0.432) that is approximately 170 km. Overall around 14% of all regions are considered as containing medieval trade centers. A list with the name, NUTS-3 region and country of all trade cities is provided in Table A.2 in the data appendix. Furthermore Figure 1 shows a map that depicts all included NUTS-3 regions and the regions with medieval trade centers (reddish colored).23

[Figure 1 about here]

4 Empirical Analysis

4.1 Medieval Trade and Contemporary Development

4.1.1 Descriptive Evidence

Some ﬁrst insights about the relationship between medieval trade centers, agglomeration and contemporary economic performance can be obtained from a descriptive look on the relevant variables. At ﬁrst, we consider simple bivariate correlations between the ln of GDP per capita, the trade center dummy, the ln of the distance to the next trade center and our two measures of agglomoration, ln population density and ln relative GDP density. These correlations are shown in Table 2.

[Table 2 about here]

In general, we see that there is a high and signiﬁcant correlation between all the variables. Additionally, the sign of the correlation coeﬃcients are as expected (e.g. there is a

23The geographical distribution of medieval trade cities in the map is largely consistent with what King (1985) wrote about the location of leading trade and economic centers in medieval Europe (King 1985, p. 220)

13 strong positive relationship between agglomeration measures and GDP per capita. Vice versa we found a negative association between distance to a trade center and both agglomeration and GDP). The correlation between GDP per capita and the trade center dummy is significant and positive, but comparatively low. On the one hand, this low correlation could be the result of considerable heterogeneity of GDP per capita across regions and countries in the sample that is not accounted for in these simple pairwise correlations. On the other hand, the high correlation between the trade center dummy and the agglomeration measures on the on side and agglomeration measures and GDP per capita on the other side indicates that the effect of trade centers is largely running through agglomeration. Therefore the observed correlations provide preliminary support for our theoretical reasoning. Another way to illustrate the stylized relationship between medieval trade, agglomeration and present day’s regional economic development is to compare averages values of GDP per capita and agglomeration measures for late medieval trade centers and non- trade centers. This is done in Table 3 both separately for each country as well as for the whole sample of regions. From the last line of Table 3 we can infer that in total, i.e. pooled over all regions and countries in the sample, regions with late medieval trade cities have a significant “GDP Advantage”, that is, their average GDP per capita is around 5000 Euro higher than that of regions without trade cities. Furthermore, they also exhibit significantly higher population and relative GDP densities.24 This result does also hold within all countries apart from Lithuania where trade center regions show a higher GDP per capita but the differences is insignificant. For relative GDP Density the within country results are not that clear. In Belgium and the Netherlands the relative GDP Density is lower, although the difference is not significant.25 However, in Austria, Germany, France and Poland the countries account for three quarters of the sample, there is a statistically and economically significant advantage of trade centers with respect to both regional economic development and relative GDP Density.

[Table 3 about here]

Finally, we estimate the kernel densities of ln relative GDP for all regions, for regions with medieval trade cities and for regions without them. The kernel density of ln relative GDP density is shown in Figure 2. The density for regions with and without medieval

24The significance of the Difference between trade regions and non trade regions is tested by a two-sample t test. 25In the smaller countries (like Lithuania, the Czech Republic or Belgium) the insignificance of the differences is probably attributable to the insufficient total number of regions/ trade centers. Here, the numbers should be treated with caution.

14 trade centers is depicted in Figure 3. A comparison of those kernel densities reveals that the variable’s kernel density over all regions is clearly leftly skewed and shows an additional notable local peak on the right.26 The latter indicates that there is a cluster of regions showing a relatively high spatial concentration of economic activity. However, what is more interesting for our argumentation is the comparison of the density across groups of regions with and without medieval trade cities. One can infer from Figure 3, that as expected the kernel density across both groups diﬀers considerably.27 Most importantly, the density function for regions with medieval trade centers clearly show a larger mass in the right tail supporting the idea that agglomeration and concentration of economic activity are higher in regions with medieval trade centers. We also run similar estimations using population density as agglomeration measure. The result of this task are shown in Appendix C (Figure C.1). In sum, the descriptive analysis of the data delivers strong preliminary evidence for our hypotheses.

[Figure 2 and 3 about here]

4.1.2 OLS Regressions

To test our main hypothesis, i.e. that regions with cities involved in medieval trade exhibit higher levels of economic development today we estimate the following regression using Ordinary Least Squares (OLS):

ln(GDP )cijk = α + βT Ccijk + γ10 Xcijk + γ20 Xcij + δc + θi + λj + cijk (1)

Where ln(GDP )cijk is the natural logarithm of GDP per capita in NUTS-3 region k

NUTS-2 Region j in NUTS-1 region i of country c. TCcijk is a dummy variable “Trade Center” that is equal to one if a NUTS-3 region includes a medieval trade city and zero otherwise. Xcijk and Xcij are vectors of NUTS-3 or NUTS-2 level covariates, respectively.

δc, θi and λj are country, NUTS-1 and NUTS-2 region fixed effects. At last, cijk is the error term capturing all unobserved factors.28 Equation (1) is a straightforward way to establish a significant direct link between late medieval trade activities and contemporary economic performance. Our expectation is that β > 0 and significantly different from zero.

26Accordingly, a Shapiro-Wilk test clearly rejects the null hypothesis of normality for the kernel density 27Conversely, a Kolomogorov-Smirnov rejects the equality of both group’s densities. 28 As mentioned before, all time-variant variables are measured in the year 2009 so we do not report an index for the period of measurement.

15 But, even when medieval trade still matters today, does its impact transmit via agglomeration and concentration of economic activities in places where it took place? A simple way to test this additional hypothesis is to look whether GDP per capita becomes lower when the distance to medieval trade centers increases. Expressed diﬀerently, if the eﬀect of trade works through agglomeration, then, a “classical” core-periphery pattern should emerge, with the medieval trade cities as core and the regions far away as periphery. One can therefore modify equation 1 by substituting the trade center dummy through a variable representing the distance between a region’s centroid and the closest trade city. Equation 1 can be rewritten as:

ln(GDP )cijk = α + ρln(Dist TC)cijk + γ10 Xcijk + γ20 Xcij + δc + θi + λj + cijk (2)

Where Dist TCcijk is the natural logarithm of the distance from a region’s centroid to the closest trade city measured in degrees. We expect ρ to be negative and signiﬁcant.

4.1.3 Baseline Results

First, we estimate equations one and two using NUTS-1, NUTS-2 and country fixed effects. They are included to account for shocks common to all observations at the respective geographical unit. Additionally, they are included to exploit the pure variation between NUTS-3 regions.29 We also add a set of basic geographical controls, including latitude, longitude and altitude of a NUTS-3 region. The latter set of variables should capture the general geographical pattern of development in Central Europe. This means, that economic development roughly increases from South to North (i.e. with increasing latitude) and decreases - in our sample- from West to East (i.e. with increasing longitude). Furthermore, it is a well known fact that regions with a higher latitude are more difficult to reach - what seems relevant for trade- and show a less favorable climate so that we expect a negative influence of altitude. The results of these regressions are shown in Table 4. There, we report three different standard errors above each coefficient. At first, in parentheses there are reported heteroskedasdicity robust standard errors. Below those, in brackets we present standard errors obtained by multiway clustering on NUTS-1 and NUTS-2 region level according to the methodology of Cameron et al. (2011). The use of multiway clustering is justified because it is likely that the development in NUTS-3 regions is not independent of that

29Overall, there are 49 NUTS-1 regions and 143 NUTS-2 regions in our dataset. In the regression some of them are omitted, because of multi-collinearity. The multi-collinearity is most often caused by the fact, that sometimes, like in the case of the German city states Berlin, Hamburg or Bremen NUTS-1, NUTS-2 and NUTS-3 regions are identical.

16 in NUTS-1 or NUTS-2 regions.30 Supplementary, because multiway clustering allows for arbitrary residual correlation across both included dimensions, it also accounts for possible spatial correlation. Finally, the third standard errors (in curley brackets) are adjusted for two-dimensional spatial correlation using the method proposed by Conley (1999).31

[Table 4 about here]

A look at the estimation results confirms our expectations and the descriptive evidence brought forward before. Regions with medieval trade centers show a significantly higher GDP per capita than regions without such cities. The coefficient of the trade center dummy remains relatively stable and significant at 1 % level, regardless which combination of control variables and fixed effects is used. According to column (3) of Table 4, where we include the full set of country and region dummy as well as the basic geographic controls, regions with medieval trade centers on average have around 30 % higher GDP per capita than regions without. This means that the effect of medieval trade is not only statistically but also economically of considerable significance. This holds also true for the coefficients of the distance to trade center. They are always highly significant and are quantitatively in the same range as that of the trade center dummy. Furthermore, they show the anticipated negative sign. The clear positive relationship between contemporary GDP per capita and medieval trade centers is also illustrated graphically in Figure 3, a partial regression plot of the Trade Center Dummy based on the full baseline specification in column (3). And in Figure 4 the same is done for the negative relationship between the distance to a medieval trade center and present days GDP per capita. Regarding the geographical controls latitude and longitude turn out to be insignificant throughout all estimations. Altitude, to the contrast, is always significant and its coefficient shows the expected negative sign. Furthermore, the NUTS-2 dummies are often not significant and do – according to the adjusted R2 – add nothing to the explanatory power of the model. For this reason, they would only introduce additional noise in the estimation and are therefore excluded from the remaining regressions.

30It might even be the case that the development of included variables regional variables is correlated within countries. However, due to the fact that we only have ten countries in our sample and clustered standard errors are only consistent asymptotically, clustering at country level is no option. 31Conley’s (1999) standard errors are obtained using a cutoff point of 3 degrees (approx. 330 km) after which the spatial correlation is assumed to be zero. We experimented with several different cutoff points and this cutoff produced the most conservative standard errors.

17 The three different types of standard errors in general do not differ substantially. If any, the standard errors in brackets, adjusted fro multiway clustering are a little bit larger than the other two. Because of that, we will use standard errors clustered on NUTS-1 and NUTS-2 level, for all remaining specifications if possible.

[Figures 3 and 4 about here]

4.1.4 Controlling for Determinants of Agglomeration and Development

To ensure that the significant positive relation between medieval trade and contemporary economic development is not driven by omitted variables bias we have to control for relevant determinants of both agglomeration and economic development. As a next step, we therefore add several sets of control variables to the baseline specification. In agglomeration economics, the causes of agglomeration are categorized in first nature (physical and political geography, climate etc.) and second nature causes of agglomeration (man-made factors, i.e. agglomeration resulting from spatial spillovers or scale effects) (e.g. Chasco et al. 2012, Christ 2009, Ellison and Glaeser 1999, Krugman 1993, Roos 2005). This literature assumes that there are direct effects of both types of causes, as well as an additional indirect effect of second nature through its interaction with first nature. Because medieval trade is supposed to be a first nature cause of agglomeration, this indirect effect geography and other natural factors exert on first nature causes is what we especially have to control for. Additionally to standard economic agglomeration and growth literature we also have to account for potentially important historical causes of agglomeration and development. This clearly follows from our argument that medieval trade influenced regional development processes through its impact on agglomeration and industry concentration. In conclusion, we decide to group the control variables in four set of variables we add separately to the baseline specification (without NUTS-2 dummies). The first set of variables controls for the “geographic centrality” of regions. It includes variables measuring the distance of a region to the closest important infrastructure facilities (airports, roads and railroads) and to important political and physical geographic features (coasts and borders).32 Especially, the last two are found to be important first nature determinants of agglomeration (e.g. Roos 2005, Ellison and Glaeser 1999). Ad- ditionally, the ln of the distance of each region to the geographically nearest major river

32Holl (2004) and Martin and Rogers (1995) establish empirical and theoretical evidence on the importance infrastructure facilities for industry location. This justiﬁes the inclusion of distance to road, airports and railroads as control variables.

18 is included as control.33 Rivers are geographical features important for both medieval trade, industry and city location (Börnerand Severgnini 2012, Bosker and Buringh 2010, Ellison and Glaeser 1999, Roos 2005 and Wolf 2009). The idea behind this set of controls is to ensure that we do not simply capture the impact of many medieval trade cities being located at geographically favorable places today or in the past. A second set of variables controls for relevant contemporary characteristics of the included regions. It comprises out of dummy variables for district-free cities in Germany (which are by definition larger or more densely populated places than others), for the regions that include a country’s capital or the capital of an autonomous region.34 Addi- tionally, a categorical variable identifying the degree to which a region can be considered as a“mountain regions” is included. Furthermore the set includes dummies for regions with coal or ore mines (or mining firms), for regions located in the former GDR and for regions located in Easter European post-communistic transition countries. At last, it includes the ln of a region’s area. In consequence, this set of controls accounts for many important first nature causes of agglomeration (political geography and resource endowments) as well as for relevant historical facts that could have influence the contemporary economic performance of a region (like communism). The next set of controls captures historical characteristics of the regions that could matter for both present day’s agglomeration and economic performance. Here we consider dummy variables indicating regions with a university founded before 1500 AD and regions that adopted printing technology before 1500 AD. As unearthed by Cantoni and Yuchtman (2012), Dittmar (2011) and Rubin (2011) both universities and printing technology are important factors in explaining the late medieval commercial revolution and city growth. To account for the positive influence Protestantism probably had on economic development (Woesmann and Becker 2009, Rubin 2011) we also include ln distance to Wittenberg as variable in this set of controls. Furthermore, we also include dummies for regions containing at least one imperial city or at least one city that was member of the Hanseatic League. Finally, we also control for the possible long-lasting influence of roman heritage and low transport costs for trade and agglomeration in including a dummy for cities located at an important imperial road (Postan 1952).35 The fourth set controls for the most important covariates of economic growth and development. Here we use the share of people aged between 25 and 64 with tertiary

33In Germany for example we consider Elbe, Danube, Rhine and the Oder as major rivers. 34An autonomous region is considered to be a Belgian or Italian Region or a German or Austrian federal state (“Bundesland”). 35This variable considers the Via Regia, the Via Regia Lusatiae Superioris and the Via Imperii as the probably important imperial roads more or less following the route of former Roman roads.

19 education (on NUTS-2 level) as measure for regional human capital.36 As variable to measure the quality of regional economic and political institutions we use the quality of government index developed by the Quality of Government Institute at the university of Gothenburg which provides a measure for regional institutional quality design similar to the World Governance Indicators (WGI) of the World Bank.37 As measure for regional inequality we construct the ratio of average workers compensation to GDP per capita. As measure of innovative activity in a region we use the number of patents registered by a region’s firms again at NUTS-2 level. Furthermore, we include a region’s unemployment rate, ln of the average workers compensation and the ln of the average fixed capital of a region’s firm. Finally, the last set of controls include all robust covariates from the regressions before. The robust controls are obtained by including all variables in one regression that were significant both times when added with one of the other four sets of controls to the baseline specification. In the next step, we did remove the variables becoming insignificant in that regression. We repeat this procedure until only significant controls remain in the specification.38 This procedure results in a set of 12 variables robustly associated with GDP per capita. These are altitude, the ln distances to airports, railroads and rivers, dummies for district free cities, capital cities, capital cities of autonomous regions, post- communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. Once more, this highlights the importance of human capital and political geography. Furthermore the robust influence of printing confirms Dittmar’s (2011) claim that printing technology fostered - similar to medieval trade- localized spillovers and forward- and backward linkages. The results of the regressions are shown in Table 5. There we first add the first four set of controls separately to the baseline specification and then we include as fifth set all robust covariates to the country and NUTS-1 region fixed effects. We see that the coefficient of the trade center dummy and the distance variable remain significant in every of the specifications although the sizes of the coefficients is considerably reduced compared to the baseline estimates.

[Table 5 about here]

The coeﬃcient is smallest (e.g. 0.07 in the case of the trade center dummy) in the

36Again, we take the values for the year 2009. 37This variable is for some countries available at NUTS-1 level and for others it is available at NUTS-2 level. For details consult the data appendix. 38These regressions are not shown but are available from the author upon request.

20 specification with all robust covariates added to the baseline model. This is not surpris- ing since in this specification we added only the variables with the highest explanatory power to the regression. It suggests, that medieval trade center regions have today a GDP per capita around 7 % higher than other regions. Based on the average regional GDP per capita in our sample this corresponds to a GDP per capita that is approximately 1200 Euros higher. When looking at the different set of controls it is evident from the adjusted R2, that region characteristics and growth covariates add most additional explanatory power to the model. Apart from mountain and mining region dummies, each variable in the regional characteristics set is significant and especially the effects of political geography (capital regions or regions with a capital of a autonomous region) seem to be important. And regarding the growth covariates especially inequality (with an remarkable negative sign) and human capital exert a strong effect on GDP per capita.39 In general, the historical region characteristics are least important in explaining contemporary regional economic development. But regions with universities and cities adpoted printing technology before 1500 AD seem to have a significantly higher GDP per capita even today, once again highlighting the importance of human capital.40 How- ever, the university before 1500 AD dummy becomes insignificant when added jointly with the measure of current regional human capital. This suggests universities lead to advantages of regions concerning their human capital persisted until today. Finally, the robustly negative impact of the distance to river variable again shows the already widely acknowledge role of first nature geography for regional economic development. Overall, we see that the relationship between medieval trade and contemporary regional development is robust to the inclusion of a wide range of control variables and other important determinants of agglomeration and economic performance. The one exception is the estimation in column (10) where distance to trade center becomes insignificant.

4.1.5 Accounting for Endogeneity

Even after controlling for many factors endogeneity of the medieval trade variables remains a serious issue. Endogeneity primarily could arise through unobserved factors, inﬂuencing both contemporary regional development and medieval trade. Geography

39This finding is for example in line with Simon (1998) and Gennaioli et al. (2013) who highlight the importance of human capital for regional development and city growth. 40In the specification with the distance to trade center variable and historical region characteristics (column (7)) also the other historical region characteristics seem to be significant (at least at 10% level). This indicates that some of the effects captured in distance to trade centers are in fact e.g. are attributed to the course of important imperial roads like the Via Regia.

21 might be a prominent factor for which this holds true. However, we can control for geography in our regressions. But there are many other unobservable factors that might affect both our right- and left-hand side variables. A prominent example is institutional quality in medieval cities an important factor in medieval trade and the commercial revolution (e.g. Greif, 1992, 1993 and 1994).Other cases are cultural differences between regions and countries –apart from being protestant or not– or historical differences in politics between regions. To solve the endogeneity issue, we therefore run IV Regressions using the Limited Information Maximum Likelihood (LIML) method.41 In order to be able to test the validity of the exclusion restriction we choose two instrument variables. The first considered instrument variable is a categorial variable (taking the values zero, one, two and three) indicating whether a region is classified as a mountain region by the official EU regional statistics. The variable is zero if a region is not classified as a mountain region. It is equal to two or three if the region is a mountain region according to two different set of criteria (for details about the exact definition consult the Data Appendix).42 The idea behind this variable is intuitively plausible. In mountain regions, characterized by higher trade costs, less favorable climate and many other adverse features trade activities were lower than in region located at large rivers, the coast or in low altitude areas with fertile soils and less rugged terrain. Especially in the medieval age, where no advanced transport technologies are available — especially for over-land transport — mountains constituted a severe hindrance of trade (Spufford 2002).43 Furthermore, as highlighted by Bosker and Buringh (2010) high elevation (as well as differences in elevation between places) has a considerable negative effect on city

41This estimation method has better small sample properties and is most often more efficient than the standard 2SLS method,especially in the presence of weak instruments. Its confidence intervals are more reliable and it is unbiased in the median when the instruments are weak (Stock and Yogo 2005). 42Albeit this variable is of categorial nature we choose to include it as a single variable and not by using three different dummies as instruments. This is primarily motivated by guaranteeing a parsimonious set of instruments since the IV estimates are biased towards the OLS estimates when the number of instruments increases. Furthermore, the test of overidentifying restrictions wouldn’t be valid if one include several instruments following the same reasoning or originating from the same measure phenomenon as excluded instruments in the first stage. However, the results are fully robust to using the three different categories of the mountain region variable as separate instruments. They are also robust to recoding the three categories to one and include the variable as binary dummy variable. Results not shown but available from the author. 43Evidently, the large amount of trade activities between the northern Italian city states and the southern German cities (Ulm, Ravensburg etc.) require that the traded goods are transported over the alps, e.g. through the SplügenPass (Schulte 1966). However, the transport probably took place over only a few important passes and none of the small villages and populated places along those mountain routes could develop to an remarkable center of trade.

22 growth and urban potential of a place. The exogeneity of this geographical characteristic of a region should not be a concern. The second instrument variable we will use is a dummy variable for cities that were residential cities of bishops before 1000 AD. The church as political, spiritual and economical power had a significant impact on the development of cities in the medieval age (e.g. Baker and Holt 2004, Isenmann 1988, King 1985 and).44 Because of this it is probable that ecclesiastical centers, like residential cities of bishops did grow larger and had a higher probability of becoming a trade center. In line with this reasoning, Börner and Svergnini (2012) could show that trading activity (in- and outflows of commodities) were higher in bishop residence cities. Additionally, Bosker and Buringh (2010) found that the presence of a bishop was a important factor in the foundation and development of cities during the Middle Ages. The exogeneity of this measure is not as sure as in the case of distance to river. But nevertheless, since we can control for geography it is hard to find a variable that could potentially influence both the location of bishop residences in 1000 AD and contemporary regional development. First, in 1000 AD most of the political and economical institutions emerged in the late medieval did not exist. Even the central political power of our sample countries during the middle age, the Holy Roman Empire, was found in the second half of the 10th century and couldn’t therefore have larger influences on bishops residences founded before 1000 AD. This is especially true because many of the considered dioceses or archbishoprics are already established when the Empire was found in 962 AD. Second, we control for many other historical factors like being located on an important imperial road or early adoption of printing that might had influenced both the location of trade cities, bishop residences and economic development today. Third as explained e.g. in Pounds (2005) the dioceses built in the early medieval period were virtually identical to the territory of predated Roman cities. In consequence, their location was determined centuries before the early medieval period which makes it even more unlikely that they are endogenous to contemporary economic development. In other words, there are many reason to conclude that bishop residences before 1000 AD are exogenous and can be used as instrument. Additional to those instruments, we make use of Lewbel’s (2012) approach that ex-

44King (1985) describes the importance of the church for commercial activities and trade, i.e. they mentioned that in many cases the local fairs and markets are managed and organized by the church. Pounds (2005) and Nicholas (1997) additionally emphasize the importance of bishops for the development of cities in the early middle age, when traditional trade declined during the economic depression in the eighth and ninth century. Finally, Hunt and Murray (1999) notice the signiﬁcance of the church for city development and commerce arising from fostering ecclesiastical tourism and pilgrim activities.

23 ploits heteroskedastic first stage errors terms to generate artificial instruments not correlated with the product (covariance) of the first stage’s heteroskedasdic errors.45 This method can provide more reliable estimates if it is doubtful, that the instruments meet the exclusion restriction or are weak. Since at least the exogeneity of the bishop seats can be disputed in principle this method ensures that we do not produce invalid IV estimates. The strength of these generated instruments depend on the amount of scale heteroskedasdicity in the error. The presence of heteroskedasdicity in our first stage regression is tested with a Pagan-Hall test. The test clearly rejects the presence of a homoskedasdic disturbance (p-value<0.000). Therefore, the method can yield reliable estimates although first stages statistics are not available. 46 We run LIML IV regressions using the instruments outlined above and using Lewbel’s (2012) approach with generated instruments for the trade center dummy and the distance variable. We include the set of robust covariates as well as NUTS-1 region and country fixed effects as controls, i.e. we reestimate columns (5) and (10) of Table 5. The results of these estimations are shown in Table 6. The first important result is that throughout all specifications the trade center dummy and the distance to trade center variable are significant and retain there signs. Even more, the size of the coefficients increased remarkably, at least in the case of the con- ventional IV regressions in columns (1) and (3). Moreover, the distance to trade center variable that was insignificant before in column (10) of Table 5 regains significance at 1 % level. This can be interpreted as endogeneity downward biased the OLS results, probably due to measurement error or a negative correlation between an unobserved factor and our medieval trade measures. Concerning the validity of the instruments the overidentification tests (Hansen J-statistic) informs us that the validity of the exclusion restriction cannot be rejected in almost all case at the common levels of significance. The exception is the last specification where we cannot reject the null at all levels of significance. Due to this, one should be cautious in interpreting the results from the last columns here. Nevertheless, in line with our arguments above it seems the case that the being a mountain region and bishop residences before 1000 AD affect contemporary levels of development solely through their impact on which cities became medieval trade centers.47 Furthermore, at least in the case of the trade center dummy, Lewbel’s (2012)

45 The vector of instruments Zj is constructed by multiplying the first stage error terms with each of the included exogenous, mean-centered regressors (all or a subset of the first stage regressors), i.e. Zj = (Xj X) (Lewebel 2012). − 46Lewbel 2012 mention several papers that already applied this method resulting in plausible estimates e.g Sabia (2007) or Kelly and Markovitz (2009).Thus, the method has proven to provide reliable estimates in different empirical settings. 47 In fact, it is very likely that geographic characteristics like being a region in the mountains also

24 approach show, that our results hold even when we do not use external instruments but instruments that are exogenous by construction. However, the coefficients obtained with LIML IV are much larger as that resulting from Lewbel’s (2012) approach that are in much closer to the original OLS estimates. Since Lewbel’s (2012) approach relies on second moment conditions and additionaly produces a comparatively large number of instruments it is likely that this results reflect the lower bound of the true estimates. Turning to the first stage results, it emerges that both instruments are indeed significant and strong predictors of medieval trade. The bishop dummy is highly significant in both specifications. This is also true for mountain region dummy, although it is only marginally significant when the trade center dummy is instrumented . The underidentification test and the Angrist-Pischke F statistic of excluded instruments always indicate that the instruments are strong and relevant. Altogether, the IV estimations show that endogeneity does not affect the detected significant relation between medieval trade and contemporary economic development. If anything, endogeneity downward biases the OLS estimations and therefore lead us to underestimate the true effect.48

[Table 6 about here]

4.2 Further Results - Index of Medieval Commercial Importance

Although the evidence brought forward in the previous section provide robust empirical support for a signiﬁcant relationship between medieval trade and contemporary regional development, the data on which the results are based has its limitations. First and foremost, the evidence so far is solely based on a dummy variable constructed according to whether a city was located at an important trade route and few other qualitative judgments about their importance. In treating all trade cities as equal this approach is probably not able to capture all the dimensions and factors that made a city an important center of commercial activity throughout the medieval. In consequence, we possibly do not catch the true eﬀect of medieval trade or commercial activities on contemporary development levels. However, based on the data set at hand and historical evidence about important determinants of trade, economic and commercial activities in the middle ages one can construct an “Index of Commercial Importance” for each region in our sample. Among the many potential determinants of medieval commercial activity, we choose eight

inﬂuenced which cities became residence cities of medieval bishops but since we include both variables jointly in the ﬁrst stage we take into account this correlation. 48A test of endogeneity of the instrumented variables rejects the null of actual exogeneity in at 1 % level in every LIML IV regression.

25 to construct the index. At first, we include out trade center dummy, representing cities located on important trade routes. Second, we consider the variable indicating cities that were residence of a bishop or archbishop before 1000 AD. As already outlined, the church was found to be one of the most important factors in the development of medieval cities and trade. Hence, the presence of a bishop should be a valid proxy variable for cities of notable commercial importance. Third, we include the ln distance to coast of each region’s centroid, representing the distance of each city to a potential sea harbor and the significant trade cities located at the coast (like e.g. many of the Hanseatic cities). Fourth, we include the dummy variable identifying important members of the Hanseatic League. Since the Hanseatic League was one of the leading actors in medieval commerce, its important members cities very likely were subject to significant commercial activity. Fifth, we adopt the dummy variables representing cities that had the status of an imperial city or that were located at an important imperial road. As transport cost were a crucial factor in medieval trade, the presence of a paved and protected road should be an important economic advantage for the cities located at it (e.g. Spufford 2002). On the other hand, most of the important trade cities in the Holy Roman Empire that were not member of the Hanseatic League were free or imperial cities. Due to this, imperial cities, with their political and institutional microcosm can be seen as germ cells of commercial activity in the medieval period(Cantoni and Yuchtman 2012). Sixth, we include a variable depicting regions in which medieval mining activties (copper and salt mining) took place. This accounts for the fact that salt and copper —as raw materials in general— were some of the major commodities trade in medieval Europe (e.g. Postan (1952), King 1985, Spufford 2002). Finally, we follow the reasoning of a recent study by Cantoni and Yuchtman (2012) showing that universities decisively fostered commercial activities and market establishment in the area around them. Consequently, we include the dummy variable reporting cities with universities founded before 1500 AD as last variable. The index is constructed by simply adding up this variables combining them in one index ranging from zero to eight.Thereafter, we substract the mean of the index from all its values so that the average region would have a value of zero. Regions with an below average value therefore have a negative and regions with an above average value have a positive value. We also construct an alternative version where we include the ln distance to trade center variable instead of the trade center dummy.49 Clearly, there are other determinants of commercial activity in the middle age. Never- theless, we choose this set of variables because these variables are significant predictors

49We recode this variable so that it is positively associated with economic development and agglomeration as the other seven variables in the index.

26 of the original trade center dummy when jointly included in probit model. Together, they produce a pseudo R2 of around 0.2.50 This result serves as a initial hint confirming the relevance of our variables for explaining commercial activity in the medieval age. We now perform OLS and instrumental variable regressions (as before with the LIML and Lewbel’s (2012) method) using both versions of the index of medieval commercial importance as independent and the ln of GDP per capita in 2009 as dependent variable. We include the complete baseline specification (NUTS-1, NUTS-2 and country fixed effects as well as the basic geographic controls) and the set of robust covariates employed in Tables 5 and 6 supplemented by NUTS-1 region and country fixed effects. This ensures that the results are comparable to that obtained before using the simple trade center dummy and the distance variable. The results are shown in Table 7.

[Table 7 about here]

All in all, the index of commercial importance, in both the original and the alternative version, shows up significant with a positive sign in every regression. Reassuringly, the LIML IV regressions using the same instrumental variables as before and a version of the index without the bishop before 1000 AD dummy, yield a more significant and remarkably higher coefficient. This is similar to the IV regressions using the dummy variable. The coefficient obtained with Lewbel’s (2012) generated instruments is much closer to the original OLS estimate but keeps its significance. Furthermore, the Lewbel estimate has to be treated with some care since the overidentification test does reject the null of a valid exclusion restriction at the marginal significance level. To sum up, the index of commercial importance confirm the results of the regressions using a simple dummy variable. Therefore, it is fair to conclude that there is a statistically robust relationship between medieval trade and commerce and today’s regional economic development.

4.3 Medieval Trade, Agglomeration and Contemporary Economic Development - Establishing Causality

Until now, we only indirectly show that medieval trade inﬂuences present-day’s regional economic development through its impact on agglomeration. We did so by showing that the distance of a region to the next trade city is robustly negatively associated with regional GDP per capita. In this section we will conduct a more direct test of the

50Regression not shown but available from the author.

27 proposed causal chain going from medieval trade activities to medieval city growth to contemporary agglomeration patterns and from there to regional economic performance.

4.3.1 Trade and City Growth in the Medieval Age

The first building block of our argument is that there should be a positive association between involvement in medieval trade activities and city growth during that period. To illustrate that the theoretically proposed relationship between medieval trade and city growth does actually exist, we run a set of regressions were we explain ln city growth in the medieval period by the trade center dummy and other covariates of medieval city growth identified in the literature.The population data on which the city growth variable is based originates from Bairoch’s (1988) compilation of European city population data from 800 to 1850. We include every city for which there is population data in Bairoch (1988) in 1500 AD and that is located in one of our ten sample countries. This leaves us with 361 cities from which 90 are coded as trade cities based on the same information than in the NUTS-3 region sample. A list of all included trade cities is provided in the Data Appendix. The estimated results are depicted in Table 8. There, in columns (1) to (3) we run cross-sectional OLS regressions with the natural logarithm (ln) of city growth between 1500 AD (the end of the medieval period) and 1200 AD, 1300 AD and 1400 AD. We choose these three variables to demonstrate that the results are not dependent on the chosen period and furthermore are stronger when we consider a longer period of city growth. The latter would be an indication that the impact of trade on city growth works trough agglomeration processes unfolding there effect only after a longer period of time. In every of the regressions we include country fixed effects as well as a set of other set of historical determinants of city development as controls. We control for first nature agglomeration forces by including the distance of a city to the next river or coast and also a city’s latitude and longitude and whether it is classified as a mountain region and was therefore difficult to reach(e.g. Bosker and Buringh 2010, Spufford 2002). Furthermore, we consider several dummy variables indicating whether a city was residence of a bishop before 1000 AD, had the status of imperial city, was located at an important imperial road or was a member of the Hanseatic League.51 At last, we always include the ln of the initial city population at the beginning of the considered growth period. This

51This variables were already used in the preceding empirical analysis on regional level data. However, the NUTS-3 level variables do not always fully coincide with the city level variables. This is due to the fact that a NUTS-3 region could harbor an archbishop in 1000 AD but none of the cities we consider in this sample and are located in this region.

28 accounts for the fact that city growth is concave in city size and in consequence the growth rate of a city strongly depend on there initial size.52 This is, we estimate the following regression speciﬁcation:

ln(CityGrowth) 1500 = α + βT Ci + γP OPt + δ0Xi + θc + i (3) i, t

Where ln(CityGrowth) 1500 is the ln the growth in population in a city between 1500 AD i, t and period t with t=1200, 1300 or 1400 AD. TCi is the trade center dummy POPt is the ln city population begin of the period and Xi is a set of time-invariant covariates and θc are country fixed effects.We also estimated this equation using the Index of Commercial Importance insteas of the trade center dummies. These results, that do not generally not differ from that reported here are available in Appendix C (Table C.2). Turning to the interpretation of the results, we clearly find that the trade cities show significantly higher growth throughout the medieval than non trade cities. This is clear evidence in favor of our theoretical reasoning that medieval trade contributed to city growth and agglomeration. Furthermore, we also see a highly significant and negative effect of the initial population level showing that indeed already large cities did grow slower. What is more, in columns (4) and (5) we also run random effect (RE) estimations using a panel data set comprising out of the same sample and variables as the cross section. In these estimations we first regress the ln of the city population in every of our considered years (1200, 1300, 1400 and 1500 AD) on the trade center dummy and the same set of controls as previously in the cross sectional estimates and additionally we add year fixed effects. Again, pooled over all years, the population of a city is significantly higher if the city is a important medieval trade city. At last, we regress the change in ln population between every of our base years on the trade center dummy and additionally include the lagged population in the regression (what is similar to the cross sectional estimations). Once more, we found a significantly positive association between being a trade center and changes in population throughout the period from 1200 AD through 1500 AD. In sum, this results suggest that medieval trade can indeed be regarded as an important determinant of city growth and agglomeration during the middle ages. Having established this, in the following we will focus on a detailed investigation of the relationship between medieval trade activities, contemporary agglomeration patterns and regional economic growth.

52A descriptive overview over all variables used in the city level estimations is available in Table A.2 in the Data Appendix

29 [Table 8 about here]

4.3.2 The Medieval Legacy of Contemporary Economic Agglomeration Patterns

The next step in our causal chain is to link medieval city growth and contemporary economic agglomeration patterns, i.e. we have to establish that there is a significant amount of path-dependency in city development throughout the regions in our sample. To do so, we regress the ln of the relative GDP density of a region on the three medieval city growth variables used in the previous subsection, the initial city population at the beginning of the considered growth period and again NUTS-1 region and country fixed effects and the robust covariates used already in the preceding estimations. More formally spoken following regression equation is estimated using OLS:

ln(RGDP D)cijk = α+βln(CityGrowth) 1500 +γP OPcijk,t+δ0Xcijk +θc+i+cijk (4) cijk, t

Where ln(RGDP D)cijk is the ln of the relative GDP Density in a NUTS-3 region, ln(CityGrowth) 1500 is the ln of a city’s population in 1500 AD divided by its pop- cijk, t ulation in t with t being either 1200, 1300 or 1400 AD. γP OPcijk,t represents the ln of the city’s population at the t, i.e. the beginning of the considered growth period. Xcijk is the set of robust covariates used several times before. θc and i are NUTS-1 or country

fixed effects, respectively. cijk finally is the error term. The final step, is then to establish the relationship between medieval trade, contemporary economic agglomeration (via path dependent agglomeration processes as shown above) and regional economic development. We will achieve this by conducting a causal mediation analysis (estimation of mediation effects) following the method developed by Imai et al. (2010, 2011).53 Me- diation analysis enables to disentangle direct and indirect effects –via determining agglomeration– of medieval trade on contemporary development. Since we cannot rule out that there are direct effects or –what amounts to the same– indirect effects of medieval trade working via other channels this methodology seems to be appropriate for our setting. The estimation of mediation effects is based on a set of three different linear estimation equations (Imai et al. 2010):

53The method suggested by Imai et al. is a generalization of the traditional mediation analysis (MacK- innon 2008) that implement it as a variant of linear structural equation modeling (LSEM).

30 Ycijk = α1 + β1Tcijk + γ110 Xcijk + γ120 Xcij + δc + θi + λj + cijk1 (5)

Mcijk = α2 + β2Tcijk + γ210 Xcijk + γ220 Xcij + δc + θi + λj + cijk2 (6)

Ycijk = α3 + β3Tcijk + πMcijk + γ310 Xcijk + γ320 Xcij + δc + θi + λj + cijk3 (7)

Where Ycijk represents ln GDP per capita in a NUTS-3 region, Tcijk represents our variables of interest (treatment variable), i.e. the trade center dummy, the ln distance to trade center measure and the index of medieval commercial importance. Mcijk represents the mediating variable, that is ln relative GDP density as measure of the spatial distribution of economic activity. Xcijk is defined as before and stands for a set of NUTS-3 level covariates. Accordingly, Xcij is a set of NUTS-2 level covariates. δc, θi and λj are again country, NUTS-1 and NUTS-2 region fixed effects. The epsilons represent the error terms. This means that equation (4) is identical to equation (2) or (3) respectively, while in equation (5) we regress the medieval trade variables on the agglomeration measures and in equation (6) finally we include both the medieval trade variables and the agglomeration measures in one regression on ln GDP per capita. The “average causal mediation effect” (ACME) is estimated by the product of the coefficients β2 and π (β2π) and is obtained through a two-step procedure described in detail in Imai et al. (2011).54 The ACME represents the indirect effect of medieval trade on contemporary GDP per capita, i.e. that part of the overall effect of medieval trade running through agglomeration. Correspondingly, β1 measures the total (average) effect of medieval trade on regional GDP per capita and β3 represents the direct effect of medieval trade, i.e. that part of the effect not mediated by agglomeration (but maybe other factors). In consequence, this methodology of separating direct and indirect effects enables to calculate which amount of the total effect of medieval trade works via increased agglomeration. We expect β2 > 0 in the case of the trade center dummy and β2 < 0 in the case of the distance to trade center variable. Even more, we also hypothesize that on average, the majority of the effect of medieval trade should run through agglomeration. This leads us expecting the ACME being significantly different from zero and greater than the direct effect ( β π > β ). Moreover, since it holds that β = β π + β | 2 | | 3| 1 2 3 equation (4) is redundant given equations (5) and (6) and therefore we only estimate

54In the classical case, where the mediation analysis is conducted using LSEM the coeﬃcients are obtained by separately estimating equations (5) and (6) using OLS.

31 those two equations.55 Last, we assume π > 0, i.e. a signiﬁcant positive direct eﬀect of agglomeration on regional GDP per capita. The results of both the regressions of medieval city growth on ln GDP density and the mediation analysis are presented in Table 9. Supplementary to those result, we estimated Table 9 with ln population density as mediating agglomeration measure. The results are similar and available in Appendix C (Table C.1).

[Table 9 about here]

Columns(1) to (3) show the results for the estimation of equation (4). We clearly see that there is a robust and positive relationship between medieval city growth in different time periods and the contemporary relative GDP density of the NUTS-3 regions in which the cities are located. The smallest estimate, resulting from the estimation with city growth between 1400 and 1500 AD as regressor, implies that on average, one percentage of city growth in this period leads to a around 0.17 percent higher relative GDP density. This shows that there is indeed a considerable amount of path-dependency in the development of European cities, i.e. cities that grew larger in the medieval age due to trade are the economic centers and agglomeration areas still today. Turning to the results of the mediation analysis (columns (4) to (6)) we again find strong empirical support for our theory. As expected based on the previous empirical results, all three measures of medieval trade (the dummy, the distance variable and the index of commercial importance) are strong predictors of contemporary relative GDP density. The coefficients are both significant from a statistical and economical point of view. The coefficient of the trade center dummy for instance implies that regions with an important medieval trade center shows on average a around 40 % higher relative GDP density than non trade center regions. What is more, the results clearly show that a higher distance to a trade center largely corresponds to a higher distance to areas where the economic activity is concentrated.Thus, according to those estimates, there is a significant and robust positive relation between present day’s spatial distribution of economic activity and medieval trade. Moreover, from the estimations of equation (7) we see that the significant effect of the medieval trade measures on the ln GDP per capita does completely disappear when we include the relative GDP density in the regression estimation. The relative GDP density in contrast, enters with a positive and significant sign in each of the three regressions. Thus, areas with a high concentration of economic activity are also the regions with a higher GDP per capita. Most importantly,

55Finally, this also implies that the share of the total eﬀect of medieval trade running through agglomeration is (β2π) . β1

32 this also implies that the vast majority of the observed strong effect of medieval trade on regional development levels works through its impact on the patterns of spatial industry agglomeration. In line with this, the ACME is always significant and on average above 100 % indicating that the insignificant remaining effect of medieval trade is even negative in some cases. Thus, it is fair to conclude that the effect of medieval trade indeed runs through agglomeration as proposed in this paper.

4.4 Robustness of the Results

Our results have proven to be robust to the inclusion of many important covariates and to endogeneity issues. However, there remain some additional concerns about the robustness of the obtained estimates. To account for these objections, we conduct various robustness checks. The results of these tasks are reported in appendix B (Tables B.1 to B.8). At first, we account for the effect some additional variables might have on both the current level of regional development and/or medieval trading activities. In order to do so, we add four different variables to the set of control variables used in Tables 5 and 6.56 We add a dummy variable indicating regions with copper mining sites in the medieval age to look whether such type of economic activities at least partly causes the significant effects we attribute to medieval trade activities. This could be possible if e.g. mining activities actually led to higher trade activities in the regions they took place. We add this variable to the specifications three and eight in Table 6, i.e. we add the variable to the set of control variables capturing historical region characteristics. Additionally, we include an interaction term of latitude and longitude of a region’s centroid to the set of basic geographic controls and re-estimate specifications three and six of Table 5 including this interaction effect. The justification for this is to look whether development levels systematically differ when changing latitude and changing longitude and vice versa. In this way we can for example identify effects of different climatic conditions varying along different latitudes for countries located at the same longitudes. Furthermore, we add the share of Roman Catholic people in a country’s population in 2009 to the set of growth covariates and the re-run the regression in Table 6 columns (4) and (9). This takes account of the fact that the impact of Protestantism (or religion in general) on economic outcomes might not be captured adequately by the Distance to Wittenberg variable, at least not today 500 years after the Reformation.

56A descriptive overview over these variables is provided in Table B.9. A detailed description of the variables and their sources is available in appendix B.

33 At last, we add a dummy variable equal to one if a region includes an important residence city of a clerical or secular ruler. Residence cities of important rulers were the centers of political and economic power in the territory of the ruler. Therefore, it is quite likely that they showed high growth rates of population and economic activity and maybe explain a significant part of medieval trade and its long-lasting effects on agglomeration and development (e.g. Ringrose 1998). The results obtained when adding these supplementary variables to the mentioned regression specifications are shown in Table B.1. The dummy for medieval copper mining regions and the latitude longitude interactions are not significant (Columns one to four in Table B.1). Apart from the fact, that some of the included covariates seem to be significant (e.g. the catholic variable) the trade center dummy and the distance to trade center variable retain there significance and the size of the coefficients is comparable to that obtained in the original estimates or larger. A second robustness check is to look whether our results are sensitive to removing influential observations. To test this we re-estimate Table 6 but remove regions that show a high leverage, i.e. have a large impact on the coefficient estimate. This can be done by computing the DFITS statistics, developed by Belsely et al. (1980). They suggest to consider an observation as influential if DFITS > 2 k N (with k indicating | j| \ the number of regressors and N denoting the number of observationsp in the sample). Following their suggestion in each regression the regions having a DFITS statistic above this threshold are removed from the sample and then the estimations are based on this reduced sample. The results of this task are shown in Table B.2. Once again, the exclusion of influential observations does only lead to minor quantitative changes in the coefficient values (in both directions). Qualitatively, the results seem to be completely unaffected by influential observations. As already discussed in the data section, there is a considerable amount of uncertainty in the historical sources and information on which our identification of important medieval trade centers is based. In consequence, it is adequate to test, whether our empirical results hold, when alternative sample of trade cities are used in the regressions. We therefore re-estimate the all important results that depend on the trade center dummy using the four different alternative samples of trade regions introduced in section 3.3 and further elaborated in Appendix B. For each of this four alternative trade center dummies we re-run the regression specification in Table 5 column (5) where we employed all robust covariates from the previous regressions as controls. This specification is used —as in most parts of the analysis above– because it yields the most conservative estimates. We further repeat the LIML and Lewbel (2012) instrumental

34 variables regressions from Table 6 columns (1) and (2) as well as the estimation in Table 8 column (1) where we regress the ln city growth between 1200 and 1500 AD with the trade center dummy, the inital population level and appropriate historical controls. At last, we re-do the mediation analysis with ln relative GDP density as mediator variables (originally reported in Table 9 column (4)). The results of this re-estimations are shown in Appendix B, Tables B.3–B.7. As one can infer from the results in these Tables the results most often do only marginally change with the alternative trade center variables. They coefficients even tend to be a little bit larger than with the original sample of trade cities. However, this does not hold for the estimations from Table 8. At least, with the last sample of trade cities containing cities with reported trade activities in earlier periods. The coefficient of the trade center dummy becomes insignificant when using this alternative sample. However, in sum, none of our conclusions and general results is invalidated by the alternative samples of trade cities. As such, the results are robust to considerable changes in the sample due to uncertainty of historical information and underlying data selection criteria.

5 Conclusion

This paper argues that medieval trade led to agglomeration and concentration of economic activities in the region it took place. It further postulates that the observed spatial distribution of population and economic activity across Europe today is still shaped by the self-reinforcing and long-lasting agglomeration processes originating from medieval trade activities. An empirical tests of these hypotheses brought forward that, as expected, there is a statistically and economically significant positive relationship between medieval trade activities and contemporary regional economic development. The analysis further unearthed that this relationship is indeed caused by the influence medieval trade exerted on the emerging patterns of agglomeration and spatial concentration of industrial activities throughout European regions. Based on the result of this paper we are able to confirm a causal chain running from medieval trade activities through medieval city growth to contemporary industry concentration and regional economic development. Medieval trade therefore can considered to be an important determinant of modern economic development. Further quantitative analyses of medieval trade activities maybe based on more detailed historical data can therefore help to significantly improve our understanding of the sources of long-lasting economic and social prosperity.

35 Tables and Figures

Figure 1: NUTS-3 Regions with Medieval Trade Cities

36 iue3: Figure iue2: Figure enlDniyEtmtsfrTaeCnesadNnTaeCenters Trade Non and Centers Trade for Estimates Density Kernel Density Density 0 .05 .1 .15 .2 .25 0 .1 .2 .3 .4 -5 2 enlDniyEtmt o nRltv D Density) GDP ln(Relative for Estimate Density Kernel 4 0 ln(Relative GDPDensity) Kernel DensityTradeCenter=0 Kernel DensityTradeCenter=1 ln(Population Density) 37 6 5 8 10 10 iue5: Figure iue4: Figure e( ln(GDP per capita) | X ) e( ln(GDP per capita) | X ) -.5 0 .5 1 -.5 0 .5 1 D ..adDsac oTaeCnes-PrilRgeso Plot Regression Partial - Centers Trade to Distance and p.c. GDP -1 -1 D . n rd etr ata ersinPlot Regression Partial - Centers Trade and p.c GDP -.5 e( ln(DistancetoTradeCenter)|X) -.5 e(Trade Center|X) 38 0 0 .5 .5 1 Table 1: The Data on Medieval Trade Centers

No. of No. of Trade Share Trade Mean ln(Distance Country Regions Centers Centers to Trade Center) Austria 35 7 20 0.36 Belgium 44 3 6.8 0.41 Czech Republic 14 4 28.6 0.43 France 94 20 21.3 0.53 Germany 429 37 8.6 0.39 Hungary 20 2 10.0 0.69 Italy 90 25 27.8 0.41 Lithuania 7 2 28.6 0.56 Netherlands 40 7 17.5 0.29 Poland 66 12 18.18 0.55 Total 839 119 14.8 0.425

39 Table 2: Bivariate Correlations of the Main Variables

Trade Center ln(Distance to ln(Population ln(GDP per ln(Relative Trade Center) Density) capita) GDP Density) Trade Center 1 ln(Distance to -0.529*** 1 Trade Center) (0.000) ln(Population 0.228*** -0.36*** (0.000) 1 Density) (0.000) ln(GDP per 0.12*** -0.356*** 0.461*** 1 capita) (0.108) (0.000) (0.000) ln(Relative 0.218*** -0.303*** 0.921*** 0.434*** 1 GDP Density) (0.000) (0.000) (0.000) (0.000) Notes. Correlation coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. Reported are pairwise correlation coefficients using the whole sample of NUTS-3 regions.

40 Table 3: Medieval Trade, Agglomeration and Regional Development - Descriptive Overview

country Av. GDP p.c. GDP p.c. non “GDP Advantage” Rel. GDP Dens. Rel. GDP Dens. “Rel. GDP Dens. trade centers trade centers trade centers trade centers non trade centers Advantage” trade centers Austria 37428.71 26885.71 10542.28*** 19.21 0.453 18.76** (2569.8) (8.5) Belgium 35566.66 25014.6 10552.03** 1.02 3.00 -1.98 (4669.6) (8.43) Czech Republic 15950 11100 4850* 31.94 0.247 31.7 (2574.7) (18.79) France 29680 24513.5 5166.48** 137.07 13.71 123.36* (2267.2) (72.72) Germany 34381.08 26342.86 8038.22*** 14.02 5.91 8.1*** (1692.8) (2.5) 41 Hungary 13500 6677.78 6822.23*** 75.51 .174 75.34*** (2049) (18.73) Italy 27576 24095.38 3480.62*** 3.04 2.23 0.818 (1220.9) (1.73) Lithuania 8200 6439.99 1760 1.64 0.71 0.924 (2397.35) (0.471) Netherlands 36142.86 30430.3 5712.56* 1.81 2.97 -1.15 (2883.3) (2.0) Poland 10475 6822.22 3652.78*** 42.9 4.16 38.74*** (921.2) (9.00) Total 28652.9 23779.2 4873.77*** 35.99 5.48 30.51*** (1050.28) (9.7) Notes. The statistical significance of differences in GDP per capita, population density and relative GDP density between trade centers and non trade centers is tested by a two-sample t test (assuming equal variances). Differences between trade centers and non trade centers are statistically different from zero at the ***1 %, **5 % and *10 % level. Standard errors of the t tests are reported in parentheses. Table 4: Medieval Trade and Contemporary Economic Development - Baseline Esti- mates

Dep. Var. ln(GDP per capita) (1) (2) (3) (4) (5) (6)

Trade Center 0.244***0.272***0.264*** (0.026) (0.028) (0.028) [0.03] [0.033] [0.031] 0.03 0.029 0.27 { }{ }{ } ln(Distance to -0.232*** -0.31*** -0.29*** Trade Center) (0.039) (0.046) (0.046) [0.047] [0.053] [0.055] 0.038 0.045 0.043 { }{ }{ } Country Dummies Yes Yes Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes Yes Yes NUTS-2 Dummies No Yes Yes No Yes Yes Basic Geographic No No Yes No No Yes Controls

Obs. 839 839 839 839 839 839 Adj. R2 0.78 0.778 0.778 0.765 0.762 0.763 Notes. Below each coefficient three standard errors are reported. First, heteroskedasdicty robust standard errors are reported in parentheses. Second, standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in square brackets. Third, standard errors adjusted for two-dimensional spatialcorrelation according to Con- ley’s (1999) method are reported in curley brackets. The standard errors are constructed assuming a window with weights equal to one for observations less than 3 degrees apart and zero for observations further apart. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The basic geographic controls include a NUTS-3 region’s latitude, longitude and altitude. Each regression contains a constant not reported.

42 Table 5: Medieval Trade and Contemporary Economic Development - Adding Further Controls

Dep. Var. ln(GDP per capita) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Trade Center 0.175***0.105***0.181***0.0701***0.045** (0.025) (0.024) (0.024) (0.027) (0.021)

ln(Distance to -0.105** -0.0857* -0.135** -0.138*** -0.0529 Trade Center) (0.044) (0.044) (0.053) (0.041) (0.041)

Country Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Basic Geographic Controls Yes Yes Yes Yes No Yes Yes Yes Yes No Geographic Centrality Controls Yes No No No No Yes No No No No Region Characteristics No Yes No No No No Yes No No No Historical Region Characteristics No No Yes No No No No Yes No No

43 Growth Covariates No No No Yes No No No No Yes No All Robust Controls No No No No Yes No No No No Yes

Obs. 839 839 839 518 818 839 839 839 518 818 Adj. R2 0.809 0.873 0.784 0.878 0.878 0.798 0.859 0.776 0.872 0.877 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. The basic geographic controls include a region’s latitude, longitude and altitude. The geographic centrality controls include the ln distances of a region’s centroid to the nearest airport, railroad, road, border and coast point. Region characteristic controls include a dummies for regions in Germany that are district-free cities, for regions including a country’s capital, are classified as mountain regions, with ore or coal mines, located in the former GDR and located in an Eastern European post-communistic transition country. Furthermore it encompasses the ln of a regions area. The historical region characteristics consist of a dummy variables indicating regions with a university founded before 1500 AD, that adopted printing technology before 1500 AD, contain cities that were members of the Hanseatic League, with former imperial cities and were located on an imperial road. Moreover it includes the ln of the distance of a region’s centroid to Wittenberg. The growth covariates encompass a region’s unemployment rate, number of registered patents, average firm ln fixed capital stock, average worker compensation. Furthermore, it includes the share of people aged between 25-64 with tertiary education on NUTS-2 level, the quality of government index on NUTS-1/ NUTS-2 level and the ratio of an average workers compensation to a region’s GDP per capita as inequality measure. The set of all robust covariates encompasses altitude, the ln distances to airports, railroads and rivers, dummies for district free cities, capital cities, capital cities of autonomous regions, post-communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. Each regression includes a constant not reported. Table 6: Medieval Trade and Contemporary Economic Development - IV Regressions

(1) (2) (3) (4) Method LIML Lewbel (2012) LIML Lewbel (2012)

2. Stage Results Dep. Var. ln(GDP per capita)

Trade Center 0.306*** 0.0787*** (0.105) (0.0247) ln(Distance to Trade Center) -0.519*** -0.155*** (0.173) (0.0503) R2 (centered) 0.563 0.632 0.508 0.880 F-value 55.02 86.43 51.52 131.85 Overidentiﬁcation Test 0.307 66.64 0.0981 78.26 (Hansen J statistic) p-value 0.580 0.116 0.754 0.008

1. Stage Results Dep. Var. Trade Center ln(Distance to Trade Center)

Mountain Region -0.0232* 0.0259*** (0.013) (0.01) Bishop before 1000 AD 0.2553*** -0.1342*** (0.071) (0.039)

Country Dummies Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes All Robust Controls Yes Yes Yes Yes

Obs. 818 818 818 818 Angrist-Pischke F statistic of 8.39 44.51 9.32 13.47 excluded IV’s (p-value) R2(centered) 0.273 0.837 0.206 0.699 Underidentification Test 14.06 194.6 16.25 158.2 p-value 0.000 0.000 0.000 0.000 Notes. Robust standard errors are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. The set of all robust covariates encompasses altitude, the ln distances to airports, railroads and rivers, dummies for district free cities, capital cities, capital cities of autonomous regions, post-communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. Each regression includes a constant not reported. The Overidentification test reporst the Hansen J-statistic and the Underidentification test reports the Kleibergen-Paap rk LM statistic (null hypothesis: equation is underidentified). Lewbel’s (2012) approach uses a vector of generated instruments that are uncorrelated with the product of the heteroskedasdic first stage’s errors as instruments. These instruments are not included in the table due to space restrictions, but their coefficients and standard errors are available from the author upon request.

44 Table 7: Medieval Commercial Importance and Contemporary Regional Development

Dep. Var ln(GDP per capita) (1) (2) (3) (4) (5) (6) OLS LIML IVLewbel (2012)

Commercial Importance 0.0964***0.0211** 0.153*** 0.0232** (0.014) (0.009) (0.055) (0.01) Commercial Importance 0.0972***0.0181* Alternative (0.016) (0.011)

Country Dummies Yes Yes Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes Yes Yes NUTS-2 Dummies Yes No Yes No No No All Robust Controls No Yes No Yes Yes Yes

Obs. 839 818 839 818 818 818 Adj.R2 R2 0.776 0.877 0.77 0.877 0.502 0.621 \ Underidentification Test 16.45 224.5 p-value 0.000 0.000 Overidentificaton Test 0.129 69.41 p-value 0.719 0.0772 AP F-statistic of excluded 9.15 32.72 IV’s p-value 0.000 0.000 Notes.Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. In column (5) and (6) heteroskedasdicity robust standard errors are reported. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. The index of commercial importance of a medieval city is constructed by adding up the coast region dummy, the trade center, bishop in 1000 AD, imperial city and road, hanseatic league, medieval mining region and university before 1500 AD dummy variables. The alternative index of commercial importance includes the distance to trade center variable instead of the dummy (recoded to be positively related to GDP). In the case of the LIML IV regression a version of the index is used that does not include the bishop before 1000 AD dummy since this variable is used as excluded instrument in that estimation. The set of covariates encompasses altitude, the ln distances to airports, railroads and rivers, dummies for district free cities, capital cities, capital cities of autonomous regions, post-communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. Each regression includes a constant not reported. The Overidentification test reporst the Hansen J-statistic and the Underidentification test reports the Kleibergen-Paap rk LM statistic (null hypothesis: equation is underidentified). Lewbel’s (2012) approach uses a vector of generated instruments that are uncorrelated with the product of the heteroskedasdic first stage’s errors as instruments. These instruments are not included in the table due to space restrictions, but their coefficients and standard errors are available from the author upon request. The first stage regressions are also not reported but are available from the author.

45 Table 8: Medieval Trade Activity and City Growth

Dep. Var. ln( P opulation1500 )ln( P opulation1500 )ln( P opulation1500 )ln(Population)ln(∆ Population) P opulation1200 P opulation1300 P opulation1400 (1) (2) (3) (4) (5) Method OLS RE

Trade City 0.65*** 0.49*** 0.448*** 0.777*** 0.393*** (0.215) (0.121) (0.151) (0.094) (0.072) ln(Population 1200 AD) -0.66*** (0.148) ln(Population 1300 AD) -0.62*** (0.068) ln(Population 1400 AD) -0.427*** 46 (0.08) ln(Populationt 1) -0.433*** − (0.049)

Obs. 86 199 180 826 390 Adj. R2 overall R2 0.39 0.398 0.222 0.288 0.369 \ Number of Clusters 361 194 Notes. Robust standard errors are reported in parentheses in columns (1) - (3). Standard errors clustered at city level are reported in parentheses in columns (4) and (5). Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a city. The set of covariates encompasses the ln distances of a city to the next river or coast, dummies indicating cities that were residence of a bishop before 1000 AD, had the status of an imperial city, were located at a main imperial road, were member of the Hanseatic League or are classified as a mountain region by the EU regional statistics. Furthermore, we control for a city’s latitude and longitude and include country fixed effects. In columns (4) and (4) we additionally include year fixed effects. Each regression includes a constant not reported. Table 9: Medieval Trade, Relative GDP Density and Regional Economic Development

(1) (2) (3) (4) (5) (6)

Method OLS Mediation Analysis City Growth from to 1200–15001300–15001400–1500 Equation (7) Dep. Var. ln(Relative GDP Density) ln(GDP per capita)

P opulation1500 0.337*** 0.178*** 0.172*** P opulationt (0.105) (0.067) (0.062) ln(Relative GDP Density) 0.202*** 0.203*** 0.205*** (0.011) (0.011) (0.011) Trade Center 0.0048 (0.017) ln(Distance to Trade Center) 0.0103 (0.023) Commercial Importance -0.0074 (0.007)

R2 0.964 0.955 0.947 0.919 0.919 0.919 ACME 0.0661***-0.0786***0.0317*** Direct Eﬀect 0.0054 0.0111 -0.0072 Total Eﬀect 0.0715*** -0.0675** 0.0246*** % of total mediated 92.1*** 115.1** 128.1***

Equation (6) ln(Relative GDP Density)

Trade Center 0.3316*** (0.063) ln(Relative GDP Density) -0.3799*** (0.103) Commercial Importance 0.1565*** (0.023)

Country Dummies Yes Yes Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes Yes Yes All Robust Controls Yes Yes Yes Yes Yes Yes

Obs. 85 179 197 818 818 818 R2 0.939 0.938 0.94 Notes. Robust standard errors are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. The set of all robust covariates encompasses altitude, the ln distances to airports and railroads, dummies for district free cities, capital cities, capital cities of autonomous regions, post-communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. Each regression includes a constant not reported. ACME is the “Average Causal Mediation Effect” and means how much of the effect of medieval trade is mediate, i.e. works indirectly through the relative GDP density. 47 References

Acemoglu, D., Johnson, S., Robinson, J.A. (2001): The Colonial Origins of Compar- ative Development: An Empirical Investigation. American Economic Review 91(5), pp. 1369–1401.

Acemoglu, D., Johnson, S., Robinson, J. A. (2002): Reversal of Fortune: Geogra- phy and Institutions in the Making of the Modern World Income Distribution. The Quarterly Journal of Economics 117(4), pp. 1231–1294.

Acemoglu, D., Johnson, S., Robinson, J. (2005): The Rise of Europe: Atlantic Trade, Institutional Change, and Economic Growth. American Economic Review 95(3), pp. 546–579.

Ahlerup, P., Olsson, O. (2012): The Roots of Ethnic Diversity. Journal of Economic Growth 17(2).

Ammann, H. (1955): Vom geographischen Wissen einer deutschen Handesstadt des Sp¨atmittelalters. Ulm und Oberschwaben: Zeitschrift f¨urGeschichte, Kunst und Kul- tur 34, pp. 39 –65.

Ashraf, Q., Galor, O. (2011): Dynamics and Stagnation in the Malthusian Epoch. The American Economic Review 101(5), pp. 2003–2041.

Bairoch, P., Batou, J., Che, P. (1988): The Population in of European Cities from 800 to 1850. Librairie Droz, Geneva.

Baker, N., Holt, R. (2004): Urban Growth and the Medieval Church: Gloucester and Worcester. Ashgate Publishing Limited.

Baldwin, R. E., Martin, P. (2004): Agglomeration and Regional Growth. Handbook of Regional and Urban Economics 4, pp. 2671–2711.

Bavel, B. J. P. van, Zanden, J. L. van (2004): The Jump-Start of the Holland Economy During the Late-Medieval Crisis, c. 1350–c. 1500. The Economic History Review 57(3), pp. 503–532.

Becker, S. O., Egger, P. H., Von Ehrlich, M. (2010): Going NUTS: The Eﬀect of EU Structural Funds on Regional Performance. Journal of Public Economics 94(9), pp. 578–590.

48 Becker, S.O., Woessmann, L. (2009): Was Weber Wrong? A Human Capital Theory of Protestant Economic History. The Quarterly Journal of Economics 124(2), pp. 531– 596.

Belsey, D. A., Kuhn, E., Welsch, R. E. (1980): Regression Diagnostics. New York, John Wiley and Sons.

Bertinelli, L., Black, D. (2004): Urbanization and Growth. Journal of Urban Eco- nomics 56(1), pp. 80–96.

Borner¨ , L. (2002): “Die Fugger. Ein Grossunternehmen des 16. Jahrhunderts transak- tionskostenkonomische erkl?” In: Staffelbach, B., Senn, M. (eds.). onomik im Mittelalter. Eine Zeitreise mit modernen mikrokonomischen Theorien. Chronos Ver- lag, Zurich, pp. 119 –139.

Borner¨ , L., Severgnini, B. (2012): Epidemic Trade. Working Paper.

Bosker, E., Buringh, E. (2010): City Seeds. Geography and the Origins of the Euro- pean City System. CEPR Discussion Paper No. DP8066.

Braudel, F. (1986): Sozialgeschichte des 15.–18. Jahrhunderts. Vol. Band 2: Der Han- del. Munich, Kindler Verlag.

Cameron, A.C., Gelbach, J., Miller, D.L. (2006): Robust Inference With Multi-Way Clustering. NBER Technical Working Paper No. 327.

Cantoni, D., Yuchtman, N. (2012): Medieval Universities, Legal Institutions, and the Commercial Revolution. NBER Working Paper No. 17979.

Chasco, C., Lopez, A., Guillain, R. (2012): The Inﬂuence of Geography on the Spa- tial Agglomeration of Production in the European Union. Spatial Economic Analysis 7(2), pp. 247–263.

Christ, J. (2009): New Economic Geography Reloaded: Localized Knowledge Spillovers and the Geography of Innovation. FZID Discussion Paper 1-2009.

Comin, D., Easterly, W., Gong, E. (2010): Was the Wealth of Nations Determined in 1000 BC? American Economic Journal. Macroeconomics 2(3), pp. 65–97.

Conley, T.G. (1999): GMM Estimation With Cross Sectional Dependence. Journal of Econometrics 92(1), pp. 1–45.

49 David, P.A. (2007): Path Dependence: A Foundational Concept for Historical Social Science. Cliometrica 1(2), pp. 91–114.

Davies, N., Moorhouse, R. (2002): Microcosm: Portrait of a Central European City. Jonathan Cape, London.

Dietze, Hugo (1923): Geschichte des deutschen Handels. Vol. 97. Gloeckners Handels- B¨ucherei. G. A. Gloeckner Verlagsbuchhandlung.

Dittmar, J.E. (2011): Information Technology and Economic Change: The Impact of the Printing Press. The Quarterly Journal of Economics 126(3), pp. 1133–1172.

Dollar, D., Kraay, A. (2003): Institutions, Trade, and Growth. Journal of Monetary Economics 50(1), pp. 133–162.

Ellison, G., Glaeser, E.L. (1999): The Geographic Concentration of Industry: Does Natural Advantage Explain Agglomeration? The American Economic Review 89(2), pp. 311–316.

Epstein, S. R., Prak, M. (2008): “Introduction: Guilds, Innovation and the European Economy, 1400 - 1800”. In: Epstein, S. R., Prak, M. (eds.). Guilds, Innovation and the European Economy, 1400 - 1800. Cambridge UK, Cambridge University Press, pp. 1–24.

Epstein, Stephan R (1998): Craft Guilds, Apprenticeship, and Technological Change in Preindustrial Europe. Journal of Economic History 58, pp. 684–713.

Frankel, J.A., Romer, D. (1999): Does Trade Cause Growth? The American Eco- nomic Review 89(3), pp. 379–399.

Fujita, M., Thisse, J. F. (2002): Economics of Agglomeration: Cities, Industrial Loca- tion, and Regional Growth. Cambridge University Press, Cambridge, UK.

Galor, O., Mountford, A. (2008): Trading Population for Productivity: Theory and Evidence. The Review of Economic Studies 75(4), pp. 1143–1179.

Gennaioli, N., La Porta, R., Silanes, F. Lopez-de, Shleifer, A. (2013): Human Capital and Regional Development. The Quarterly Journal of Economics 128(1), pp. 105–164.

Glaeser, E.L., Kallal, H.D., Scheinkman, J.A., Shleifer, A. (1992): Growth in Cities. Journal of Political Economy 100(6), pp. 1126–1152.

50 Greif, A. (1992): Institutions and International Trade: Lessons from the Commercial Revolution. The American Economic Review 82(2), pp. 128–133.

Greif, A. (1993): Contract Enforceability and Economic Institutions in Early Trade: The Maghribi Traders’ Coalition. American Economic Review 83(3), pp. 525–548.

Greif, A. (1994): On the Political Foundations of the Late Medieval Commercial Rev- olution: Genoa During the Twelfth and Thirteenth Centuries. Journal of Economic History 54(2), pp. 271–87.

Habermann, K., Nieße, B., Oertel, D., Preis, R., Rich, H. (1978): Historische, politische und ¨okonomischeBedingungen der Stadtentwicklung. Von den Anf¨angen der Stadtentwicklung in Mitteleuropa bis zum Ende des 2. Weltkrieges. SOAK Verlag, Hannover.

Henderson, J.V., Shalizi, Z., Venables, A.J. (2001): Geography and Development. Journal of Economic Geography 1(1), pp. 81–105.

Heyd, W. (1879a): Geschichte des Levantehandels im Mittelalter. Vol. 1. Stuttgart, Verlag der J. G. Cotta’schen Buchhandlung.

Heyd, W. (1879b): Geschichte des Levantehandels im Mittelalter. Vol. 2. Stuttgart, Verlag der J. G. Cotta’schen Buchhandlung.

Heyd, W. (1890): Die Große Ravensburger Gesellschaft. Verlag der J. G. Cotta’schen Buchhandlung Nachfolger, Stuttgart.

Hohenberg, P. M., Lees, L. H. (1995): The Making of Urban Europe, 1000-1994. Harvard University Press, Cambrigde, MA.

Holl, A. (2004): Manufacturing Location and Impacts of Road Transport Infrastruc- ture: Empirical Evidence from Spain. Regional Science and Urban Economics 34(3), pp. 341–363.

Holtfrerich, Carl-Ludwig (1999): Frankfurt as a Financial Centre: From Medieval Trade Fair to European Banking Centre. C.H. Beck’sche Verlagsbuchhandlung, Mu- nich.

Hunt, E. S., Murray, J. M. (1999a): A History of Business in Medieval Europe 1200 - 1500. Cambridge University Press, Cambridge UK.

Hunt, E. S., Murray, J. M. (1999b): A History of Business in Medieval Europe 1200 - 1500. Cambridge University Press, Cambridge UK.

51 Imai, K., Keele, L., Tingley, D. (2010): A General Approach to Causal Mediation Analysis. Psychological Methods 15(4), p. 309.

Imai, K., Keele, L., Tingley, D., Yamamoto, T. (2011): Unpacking the Black Box of Causality: Learning About Causal Mechanisms from Experimental and Observa- tional Studies. American Political Science Review 105(4), pp. 765–789.

Isenmann, E. (1988): Die deutsche Stadt im Spittelalter. Eugen Ulmer Verlag, Stuttgart.

Kelly, I.R., Markowitz, S. (2009): Incentives in Obesity and Health Insurance. In- quiry 46(4), pp. 418–432.

Kinder, H., Hilgemann, W. (1970): Atlas zur Weltgeschichte. Von den Anf¨angenbis zur Gegenwart. R. Piper & Co. Verlag, Mnchen.

King, D. (1985): “Currents of Trade. Industries, Merchants and Money”. In: Evans, Joan (ed.). The Flowering Middle Ages. Thames and Hudson, London.

Koh, H., Riedel, N. (2012): Assessing the Localization Pattern of German Manufactur- ing and Service Industries: A Distance-based Approach. Regional Studies pulished online ﬁrst, pp. 1–21.

Krugman, P. (1991): Increasing Returns and Economic Geography. Journal of Political Economy 99(3), pp. 483–499.

Krugman, P. (1993): First Nature, Second Nature, and Metropolitan Location. Journal of Regional Science 33(2), pp. 129–144.

Krugman, P., Venables, T. (1995): Globalization and the Inequality of Nations. Quarterly Journal of Economics 110(4), pp. 857–880.

Lopez, R. S. (1952): “The Trade of Medieval Europe: The South”. In: Postan, M., Rich (eds.). The Cambrigde Economic History of Europe. Vol. 2. Cambridge University Press, Cambridge UK, pp. 257–354.

MacKinnon, D. (2008): Introduction to Statistical Mediation Analysis. Routledge, New York, NY.

Magocsi, P. R. (2002): Historical Atlas of Central Europe. University of Washington Press, Seattle, Washington.

Marshall, A. (1890): Principles of Economics. Macmillian, London.

52 Martin, P., Ottaviano, G.I.P. (2001): Growth and Agglomeration. International Eco- nomic Review 42(4), pp. 947–968.

Martin, P., Rogers, C.A. (1995): Industrial Location and Public Infrastructure. Jour- nal of International Economics 39(3), pp. 335–351.

Maseland, Robbert (2012): Does Germany Have an East–West Problem? Regional Growth Patterns in Germany since Reuniﬁcation. Regional Studies iFirst article.

Munro, J. H. (2002): Wage-Stickiness, Monetary Changes, and Real Incomes in Late- Medieval England and the Low Countries, 1300 - 1500: Did Money Matter? MPRA Paper No. 10846.

Nicholas, D. (1997): The Later Medieval City. 1300 – 1500. Addison Wesley Longman, New York, NY.

Nunn, N. (2008): The Long-Term Eﬀects of Africa’s Slave Trades. Quarterly Journal of Economics 123(1), pp. 139–176.

Nunn, N. (2009): The Importance of History for Economic Development. Annual Review of Economics 1(1), pp. 65–92.

Nunn, N., Wantchekon, L. (2011): The Slave Trade and the Origins of Mistrust in Africa. American Economic Review 101(7), pp. 3221–3252.

Olsson, O., Hibbs, D.A. (2005): Biogeography and Long-Run Economic Development. European Economic Review 49(4), pp. 909–938.

O’Rourke, K.H., Williamson, J.G. (2002): When Did Globalisation Begin? European Review of Economic History 6(1), pp. 23–50.

Ottaviano, G., Thisse, J.F. (2004): Agglomeration and Economic Geography. Hand- book of Regional and Urban Economics 4, pp. 2563–2608.

Porter, M. E. (1990): The Competitive Advantage of Nations. Free Press, New York, NY.

Postan, M. (1952): “The Trade in Medieval Europe: The North”. In: Postan, M., Rich, E. E. (eds.). The Cambrigde Economic History of Europe. Vol. 2. Cambridge University Press, Cambridge UK, pp. 119–256.

Pounds, Norman (2005): The Medieval City. Greenwood Press Westport, CT.

53 Putterman, L. (2008): Agriculture, Diﬀusion and Development: Ripple Eﬀects of the Neolithic Revolution. Economica 75(300), pp. 729–748.

Reith, R. (2008): “Circulation of Skilled Labour in Late Medieval and Early Modern Central Europe”. In: Epstein, S. R., Prak, M. (eds.). Guilds, Innovation and the European Economy, 1400 - 1800. Cambrigde, UK, Cam, pp. 114–142.

Ringrose, David R (1998): Capital Cities, Urbanization, and Modernization in Early Modern Europe. Journal of Urban History 24(2), pp. 155–183.

Romer, P. M. (1986): Increasing Returns and Long-Run Growth. Journal of Political Economy 94, pp. 1002 –1037.

Roos, M.W.M. (2005): How Important is Geography for Agglomeration? Journal of Economic Geography 5(5), pp. 605–620.

Rubin, J. (2011): Printing and Protestants: An Empirical Test of the Role of Printing in the Reformation. MPRA Paper No. 34294.

Sabia, J. J. (2007): The Eﬀect of Body Weight on Adolescent Academic Performance. Southern Economic Journal, pp. 871–900.

Schelle, K. (2000): Die Große Oberschw¨abische Handelsgesellschaft. Biberacher Verlags-Druckerei, Biberach a.d. Riß.

Schnyder, W. (1973): Handel und Verkehr ber die Bndner Pe im Mittelalter. Zurich, Schulthess Polygraphischer Verlag.

Schulte, A. (1923): Geschichte der Großen Ravensburger Handelsgesellschaft: 1380 - 1530. Vol. 1 – 3. Deutsche Verlagsanstalt, Stuttgart.

Simon, C.J. (1998): Human Capital and Metropolitan Employment Growth. Journal of Urban Economics 43(2), pp. 223–243.

Spufford, P. (2002): Power and Proﬁt. The Merchant in Medieval Europe. London, UK, Thames and Hudson.

Stier, H.-E., Kirsten, E., Quirin, H., Trillmich, W., Czybulka, G. (1956): West- ermanns Atlas zur Weltgeschichte. Ed. by Kirsten E. Whr W. Quirin H. Trillmich W. Czybulka G. PInnow H. Stier H.-E., H. Ebeling. Vol. Vorzeit/ Altertum, Mittelalter, Neuzeit. Braunschweig, Georg Westermann Verlag.

54 Stock, J, Yogo, M. (2005): “Test for Weak Instruments in Linear IV Regression”. In: Andrews, D. W. K. (ed.). Identiﬁcation and Inference for Econometric Models. Cambridge University Press, New York, pp. 80 –108.

Tabellini, G. (2010): Culture and Institutions: Economic Development in the Regions of Europe. Journal of the European Economic Association 8(4), pp. 677–716.

Waidlein, N. (2011): Causes of Persistent Productivity Diﬀerences in the West German States in The Period From 1950 to 1990.

Werveke, H. van (1963): “The Rise of Towns”. In: Postan M., Rich E. E., Miller, E. (eds.). The Cambrigde Economic History of Europe. Vol. 3. Cambridge University Press, Cambridge UK, pp. 3–41.

Wolf, N. (2009): Was Germany Ever United? Evidence From Intra-and International Trade, 1885-1933. Journal of Economic History 69(3), p. 846.

Yamamoto, K. (2003): Agglomeration and Growth With Innovation in the Intermediate Goods Sector. Regional Science and Urban Economics 33(3), pp. 335–360.

55 A. Data Appendix

The level of an observation is a NUTS-3 region ( For example, in Germany this corresponds to the “Landkreise”, in France to the “Departments” and in Italy to the “Provinicas”). If the variables are defined on an other NUTS level, this is indicated in the description of the respective variable. City level information is matched to the NUTS-3 regions by the use of Eurostat (2007). We use the NUTS-2006 classification, since the most data is available only for this version of the NUTS classification. An descriptive overview over all variables used in the empirical analysis is given in Table A.1 below.

Main Variables

Trade Centers. Primarily, the data on historical trade cities is based on four different maps. The first is a map printed in Davies and Moorhouse (2002) and includes ”Main trade routes in the Holy Roman Empire and nearby countries” for the period around 1500 AD. It contains the trade routes and the cities located on them. Davies and Moorhouse (2002) is a book about the history of the Polish city of Wrcolaw written by a renowned expert for Polish and Eastern European history Norman Davies and his student Roger Moorhouse. According to google scholar it is cited around 60 times (at 24th June 2013) e.g. in articles in the Journal of the Royal Statistical Association. Therefore it considered to be a reliable source for information about medieval trade activities. Because this map only covers the area of Austria, Belgium ,Czech Republic, Eastern France, Germany, Hungary Lithuania, the Netherlands, Poland and North Italy we make use of a second map published in King (1985) including ”Chief trade routes in Europe, Levant and North Africa 1300-1500 CE”. The map covers a wide area including parts of North Africa and the Near East. From this map, we primarily take the information about French trade cities, but we also include cities from other countries that are not mentioned in the first map. The original map is printed in a chapter about the “Currents of Trade. Industry, Merchants and Money” in the medieval age as part of a volume about the “Flowering of the Middle Ages” edited by the Oxford-based medieval art historian Joan Evans. In this chapter Donald King illustrates the most important goods of the medieval economy, discusses how they were produced and traded. He lays special emphasis on the patterns of commerce and trade. He describes the most important centers of commerce and trade activity (Fair and market cities etc.) and also discusses the importance of institutions (like contract security) etc. played for trade activities. Again, this volume seems to be an often cited source with around 50 citations in google

48 scholar (24th June 2013). According to the bibliography of the volume King (1985) heavily draws on standard sources about medieval trade like Heyd (1879a,b), Lopez and Raymond (1955) or Postan and Rich (eds.)(1952). As third source we employ an overview map of late medieval trade printed in Magocsi (2002) a historical atlas of central Europe and an often cited source for historical information about economic and cultural and political features. He is cited 222 (at 24th June 2013) at google scholar. Among the papers using information provided by the atlas are the historical economic papers by B¨ornerand Severgnini (2012) and Dittmar (2011) as well as Becker et al. (2011). It contains information on ”economic patterns” in Central Europe around the year 1450. From this map, we primarily took the information about Southern Italian trade cities not included in the other maps. Again, we also include cities mentioned there but not in the other two sources. From this map, a city is considered if it is located on a ”major” or ”important” trade route. The map also contains also information about members of the Hanseatic League (and their importance) as well as commercial oﬃces and foreign depots of the Hanseatic League. Further, it also depicts the goods traded over the particular routes and the areas where they are the commodities are typically produced. The map drawn in Magocsi’s atlas relies on other regional and general historical atlases like the that of Darby and Fuller (eds.)(1978) or Lendl and Wagner (1963) for Austria. However, Magocsi also consulted books about the history of certain cities like Dubrovnik (Carter 1972) or Wroclaw (Ochmanski 1982). At last, we consult several maps included in “Westermanns Atlas zur Weltgeschichte” (Stier et al. 1956). To be precise, we consider the information of a map depicting the “Hanseatic League and its Opponents in the 15th century after the piece of Utrecht”. The map reports the location of Hanseatic cities, contours of the Hanseatic League in other countries and the main trade routes of the time as well as the traded goods. The geographical scope of the map is limited to the part of Germany northern of Prague, the Netherlands, the most part of today’s Belgium and Poland. We include a city, if it is located at one of the trade routes but regardless of whether it was a member of the Hanseatic League or not. Second, we draw on a map in this atlas that limns “Western European Trade” in the late medieval and reports the course of “important trade routes” and the cities located on them. The scope of the map is south-west Europe (Spain and France) but it also includes West Germany and the north-western Italy. Here again, we include a city if it is located on a major trade route. At last, we use the information contained in a map about “Levant Trade in the Late Medieval and the Ottoman Invasion”. This map among other information, limns the course of “important” trade routes (both on land and sea) and the cities located at them. We recognize cities

49 on trade routes in the southern part of Germany, Hungary, Italy and the most parts of France as well as parts of Poland. Although not the only sources of information about medieval trade activities, these four maps seem to contain the most complete cross-national information about important trade activities in the later medieval period. To validate the information of these maps and obtaining additional evidence about medieval trade we consult other sources like a list depicting members of the Hanseatic league from Dollinger (1966) a standard source for the history of the Hanseatic League. We only recognize cities that according to Dollinger ”played an important role in the Hanseatic League” or that were capitals of thirds and quarters. Furthermore we consulted a map containing information about “North-South Trade Routes in the Alps Area in the Medieval Period” from Schulte (1966), two very general maps printed in Kinder and Hilgemann (1970) focusing on Baltic Sea and Levant trading activities in 1400 AD, a map published in Ammann (1955) focusing on trade routes for Southern Germany textile products (Barchent) and the map “Business Centers and Maritime Trade Routes High Middle Ages” printed in Hunt and Murray (1999).1 Furthermore, we draw on qualitative information about the importance of a trade cities from Spuﬀord’s (2002) standard work about medieval trade and commerce and the monograph about the history of German trade written by Dietze (1923). In Table A.2, all trade cities and the corresponding regions for which the dummy variable is equal to one and the source(s) mention the respective city as trade center are shown. However, due to space restrictions we do not report any of the sources we consulted for becoming information about the validity of our sample of important trade centers. For example, there is a three volume anthology by Escher and Hirschmann (eds.) (2005) where a group of researches developed an index of urban centrality for cities in the “Rhine-Meuse area” in the period from 1000 to 1350 AD (i.e. south-west Germany, and western Switzerland, east France , large parts of Belgium and the South of the Netherlands). As part of the index of urban centrality they collected data about the existence and number of markets, fairs, trade hall and the presence and importance of long-distance trade activities. They also have data about the presence of certain manufacturing activities also being a good indicator for the presence of trade. They develop a categorical index of centrality from the qualitative information the collect. From the trade cities in our sample Aachen, Antwerp, Cologne, Dordrecht, Dortmund, Frankfurt, Maastricht, Metz, M¨unster, Paderborn, Rotterdam, Soest and Straßburg are included in

1Geographical scope, time period and level of generality sometimes diﬀer between these maps, so a cross-validation is always possible only with limitations.

50 the volume. For every of those cities, one or more markets, a fair or differently important long-distance trade are mentioned. But here, the range goes from Cologne (having 4 markets, and ”very important” fairs and long-distance trade activities) to e.g. Pader- born where it is stated that it have a fair and long-distance trade. Due to this, it is not an easy task to say, that the information provided by this source can be used to validate whether a city was important enough to be included in the sample. Furthermore, the period for which the index is constructed ends in the middle of the 14th century and therefore earlier than our period of observation. Nevertheless, the information provided in the anthology of Escher and Hirschmann (eds.) (2005) can be useful to select cities that were probably not that important because e.g. the markets, fairs or trade there was comparably limited in scope (i.e. according to the number of markets, halls, fairs or there importance) or time. Additionally, it provides clear evidence for the outstand- ing importance of Cologne and e.g. the over-regional importance (“very important” long-distance trade or fair) of Dortmund, Frankfurt, Münster and Soest. As already mentioned, the information in those sources primarily is used to validate that the information printed in the maps. However, as indicated in the main text we sometimes also include cities mentioned in these sources but not in the maps when we are in doubt about the actual importance of a city in medieval trade. Furthermore, we construct several trade center dummies using alternative samples of trade cities (as discussed in the main text). At first, we exclude cities mentioned by only one of our sources. These cities are Amberg, Bruck, Fulda, Maastricht, Malbork, Mantoa, Minden, Orleans, Parma, Pecs, Piotrkow Trybunalski, Plock, ,Rotterdam, St. Melo, Udine, Utrecht and Zwickau. Second, we exclude cities for which we are not sure about there importance, altough they are reported in more than one of our sources. Those cities are Paderborn, Einbeck, Greifswald, Braniewo, Görlitz, Metz, Palanga, Como and Stargard. For example, we exclude Paderborn because despite the fact that it was a member of the Hanseatic League and layed on the Hellweg, no other source mentioned it and Dollinger (1966) did not consider it as being a Hanseatic city of special importance. Furthermore, the data collected by Escher and Hirschmann (eds.) (2005) group implies that the existing trade activity in Paderborn was of relatively lower importance compared to e.g. Cologne, Münster,Dortmund or other leading trade cities. Third, we add some cities to the original sample of trade cities. These cities are cases were a first look at the available information lead to the decision not to include the trade city. Even though, the city is mentioned somewhere in one of the sources as a place of certain relevance for trade. This is for example the case for Anklam, a member city of the Hanseatic League lying on an important trade route according to a map in Stier et al. (1956). However, none of

51 the other sources mention Anklam as important trade center and Dollinger (1966) did not intend a special role for Anklam within the Hanseatic League. Finally, we build a last alternative sample of trade cities that only includes cities for which historical sources indicate long-run trade activities (i.e. cities that are important trade cities around 1500 AD and that were important also in the period before). An overview over these cities the earliest period in which trade activities are reported and the source mentioned the respective city are depicted in Table A.4. This re-coding is based on information primarily derived from the 2 Wilhelm Heyds two volumes about medieval Levant trade (Heyd 1879a and 1879b). He provides information about medieval trade activities in the Levant and the most important involved parties in a chronological order beginning with the end of migration period (“Barbarian Invasions”). We take the period mentioned in the chapter headings of the chapter where the trade activities of a city are firstly mentioned as the period with the earliest authenticated trade activities. If Heyd explicitly reports a date or a period we take this date. Heyd (1879a,b) provides information about trade activities of Austrian, Belgian, French, German and Italian cities. Additionally, the monograph about the Hanseatic League written by Dollinger (1966) includes a couple of maps depicting e.g. the main Hanseatic trade routes and trade cities before 1250, between 1250 and 1350 and 1350 and 1500 (always AD). Another map report important trade routes (e.g. the salt way) and the cities that signed the treaty of Smolensk in 1229 AD a trade agreement between German trade cities and the Duke of Smolensk. According to Dollinger (1966), this map covers the period from 1286 to approximately 1336. We stick to the dates given in these maps when assigning the respective cities the dates when they are mentioned first. All in all, this and the other maps in Dollinger (1966) contain information about trade activities in France, Germany, Lithuania and Poland. Finally, for Germany, Italy and France the book of Dietze (1923) about the history of German trade reports significant trade activities and places since the ”pre-historical” period. We include a city in the sample if Dietze (1923) reports a city to be an important player in early and high medieval trade. For Austria, the Czech Republic and Poland information is provided by three digitized maps from T. Matthew Ciolek’s OWTRAD website. The first is based on a map printed in Humnicki and Borawska (1969) and shows “Central European Trade Routes 800 – 900 CE”.2 The second map originates from Wojtowicz (1956) and according to the OWTRAD website reports “Major trade roads in Poland and adjacent border regions

2The map can be found under the following URL: http://www.ciolek.com/OWTRAD/DATA/tmcCZm0800. html; accessed at June 11th, 2013.

52 1340 – 1400 CE”.3 Form this map we include information about Polish trade cities. The last map from the OWTRAD project is based on Rutkowski (1980) and is about ‘Major trade roads in Poland and adjacent border regions in 1370 CE”.4 From this map we solely include the German city of Görlitzsince all the other relevant cities in the map were mentioned by another source depicting trade in an earlier period. Overall are able to found information about 68 of our 115 medieval trade cities. ln(Distance to Trade Center). This variable is calculated using the ArcGIS Near Tool. It represents the natural logarithm (ln) of the distance between a region’s centroid and the closest trade city in degrees. The variable takes the value 0 for regions that contain medieval trade cities (i.e. for which the trade center dummy is equal to one). Trade City. Variable used for the city-level regressions in Table 3. The collection of cities coded as trade cities stem from Bairoch’s (1988) data, as explained in the main text. The cities are coded according to the procedure described in detail below in the explanation of the trade center dummy on regional level. The cities coded as trade cities are: Amsterdam, Antwerp, Augsburg, Avignon, Bari, Berlin, Bordeaux, Braniewo, Brunswick, Bremen, Brno, Bruges, Budapest, Chalon-Sur-Saone, Como, Deventer, Dordrecht, Dortmund, Einbeck, Elblag, Erfurt, Florence, Frankfurt (Main), Frankfurt (Oder), Gdansk, Genoa, Ghent, Görlitz, Graz, Hamburg, Hannover, Hildesheim, Imola, Innsbruck, Kampen, Cologne, Cracow, Leipzig, Linz, Lübeck, Lucca, Lyon, Maastricht, Magdeburg, Mantoa, Marseille, Metz, Milan, Minden, Montpellier, Münster, Naples, Narbonne, Nuremberg, Orleans, Osnabrück, Padoa, Paris, Parma, Perpignan, Plock, Poznan, Prague, Prato, Ravensburg, Regensburg, Reims, Rome, Rostock, Rotterdam, Salzburg, Soest, St. Malo, Stralsund, Straßbourg, Torun, Toulouse, Tours, Treviso, Troyes, Udine, Ulm, Utrecht, Venice, Verona, Warsaw, Vienna, Wismar and Wroclaw. ln (GDP per capita). The natural logarithm of GDP per capita on NUTS-3 level is from the Eurostat regional statistics database (http://appsso.eurostat.ec.europa. eu/nui/show.do?dataset=nama_r_e3gdp&lang=en; accessed at October 10th 2012). It is in measured in current market prices. We took values from 2009 the latest year for which data is available. Commercial Importance. Variable that should measure the commercial importance of a city according to different, historically relevant characteristics. The exact construction is explained in the main text. It is the sum of following five dummy variables: trade

3The original title of the map is (according to the OWTRAD website) “Trade roads at the times of Casimir the Great”). The map is available at the OWTRAD website under this link http: //www.ciolek.com/OWTRAD/DATA/tmcPLm1370a.html; accessed at June 11th, 2013. 4The map can be accessed under the URL http://www.ciolek.com/OWTRAD/DATA/tmcPLm1370.html; accessed at June 11th, 2013.

53 center, imperial city, hanseatic league, imperial road, medieval mining, coast region and university before 1500 AD. This variable is constructed by the author. Commercial Importance Alternative. Identical to the variable commercial importance but instead of the trade center dummy, it constains the distance to trade center variable, recoded in a way that it is positively associated with the GDP per capita (as the other variables). ln(Population Density). A region’s Population Density comes from the Eurostat regional statistics database (http://appsso.eurostat.ec.europa.eu/nui/ show.do?dataset=demo_r_d3dens&lang=en; accessed at October 10th 2012). The values are from 2009. ln(Relative GDP Density). This variable is calculated using the following formula (Roos 2005): Yi/ Yi rdi = Ai/ P Ai

Where rdi is the relative GDP Density of a region.P Yi is a region’s GDP (calculated by multiplying the GDP per capita with the population density) and Ai is a region’s area. Therefore, the relative GDP Density is the GDP density of a region (GDP per km2) relative to the average density of all other regions. Alternatively, it is the ratio of a regions share of GDP relative to its share of a country’s overall area. In consequence, if the relative GDP Density is larger than one this means that a region shows concentration of economic activity higher than the average region in a country (Roos 2005). For the empirical estimations, we take the natural logarithm of the variable, so that it is greater than zero for above average levels of spatial economic concentration. GDP per capita, the population density and the area of a region are all from the sources listed in this appendix.

Control Variables and Instruments

Altitude. The Altitude of a region is from the website gpsvisualizer.com (accessed at November 8th 2012) and based on the coordinates of its centroid. Bishop before 1000 AD. Dummy variable equal to one if a region includes a city that was seat of a bishop (or in France and Italy of an archbishop) before the year 1000 AD. The variable is coded according to information from the website http://www.catholic-hierarchy.org (accessed at November 27th, 2012). For bishoprics in the Holy Roman Empire additionally Oestreich and Holzer (1970b) is consulted. When there were doubts on whether the diocese or archbishopric was founded before 1000 AD wikipedia and the catholic encyclopedia (http://www.newadvent.org/cathen/;

54 accessed at November 27th, 2012) are consulted. Capital. A dummy variable equal to one if a region includes the capital of a sovereign state. Coded by the author. Capital Autonomous Region. A Dummy Variable equal to one if a region includes the capital of a partly autonomous administrative unit, i.e. a German or Austrian State (“Bundesland”) or an Italian or Belgian Region. Coded by the author. District-Free City. A dummy variable equal to one for German NUTS-3 regions being district-free cities (“Kreisfreie St¨adte”or “Stadtkreis”). Coded by the author. Eastern German Region. Binary variable equal to one if a region in Germany is located in the former GDR. Coded by the author. Education. We measure human capital of a NUTS-2 region with the share (in percent) of persons aged 25-64 with tertiary education attainment. The variable is obtained from the Eurostat regional statistics database (http://appsso.eurostat.ec.europa.eu/ nui/show.do?dataset=edat_lfse_11&lang=en; accessed at October 10th, 2012). We took the values from 2009. Hanseatic League. Binary variable equal to one if a region contains at least one city that was a member of the Hanseatic League. Coded according to Dollinger (1966). Imperial City. A Dummy Variable equal to one if a region includes at least one city that was an imperial city in the Holy Roman Empire. The variable is coded following Oestreich and Holzer (1970a). Imperial Road. Dummy variable equal to one if a region contains at least one city that was located on an important imperial city, i.e. the Via Imperii, the Via Re- gia or the Via Regia Lusatiae Superioris. The variable is coded according to information provided by K¨uhn (2005), the entry “Hohe Landstraße” in the online version of “Meyers Großes Konversations-Lexikon” a general german encyclopedia (http://www.zeno.org/Meyers-1905/A/Hohe%20Landstra%DFe; accessed at Decem- ber 18th 2012), a map from a website of the federal government of the German State Saxony on regional development (http://www.landesentwicklung.sachsen. de/download/Landesentwicklung/ED-C_III_Via_Regia_Verlauf.jpg; accessed at December 18th, 2012) and wikipedia entries. Inequality. We measure inequality as ratio of average workers compensation to the GDP per capita. The Sources of GDP per capita and average workers compensation are as listed in this appendix. Latitude. The values of this variable represent the latitude in decimal degrees of a region’s centroid and are obtained from a GIS map of NUTS territories provided by the Eurostat GISCO Database.

55 (http://epp.eurostat.ec.europa.eu/cache/GISCO/geodatafiles/NUTS_2010_03M_ SH.zip; accessed at November 8th, 2012). ln(Area). The natural logarithm of a region’s area is taken from the Eurostat regional statistics database http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset= demo_r_d3area&lang=en; accessed at January 10th, 2013. As always, we use the values from 2009. ln(Distance to Airport). The variable represents the natural logarithm of the distance between a region’s centroid and the closest international airport in degrees. It is calculated using the ArcGIS Near Tool. The coordinates of airports are from the GIS map “Airports and Ports” from ArcGIS Online Database (accessed at November 9th, 2012). ln(Distance to Border). The variable represents the natural logarithm of the distance between a region’s centroid and the closest point of the country’s border. It is calculated using the ArcGIS Near Tool. The coordinates of borderlines are taken from a GIS map of EU countries provided by the Eurostat GISCO Database (http://epp.eurostat.ec. europa.eu/cache/GISCO/geodatafiles/CNTR_2010_03M_SH.zip; accessed at January 10th, 2013). ln(Distance to Coast). The variable represents the natural logarithm of the distance between a region’s centroid and the closest point of a country’s coastline. It is calculated using the ArcGIS Near Tool. The coordinates of a country’s coastlines are taken from the GIS map “Corine land cover 2000 coastline” provided by European Environment Agency (EEA) (http://www.eea.europa.eu/data-and-maps/data/ corine-land-cover-2000-coastline; accessed at November 8th, 2012). ln(Distance to Railroad). The variable represents the natural logarithm of the distance between a region’s centroid and the closest point of a country’s major railroad. It is calculated using the ArcGIS Near Tool. The coordinates of the railroads are obtained from the map “World Railroads” from ArcGIS Online Database (accessed at November 9th 2013). ln(Distance to River). The variable represents the natural logarithm of the distance between a region’s centroid and the closest point of a country’s major waterway (e.g. in Germany these are Elbe, Danube, Rhine and Oder). It is calculated using the ArcGIS Near Tool. The coordinates of the rivers are taken from the GIS map “WISE Large rivers and large lakes” provided by European Environment Agency (EEA) (http:// www.eea.europa.eu/data-and-maps/data/wise-large-rivers-and-large-lakes; accessed at November 8th, 2012). ln(Distance to Road). The variable represents the natural logarithm of the distance between a region’s centroid and the closest point of a country’s roads. It is calculated

56 using the ArcGIS Near Tool. The coordinates of the roads are obtained from the GIS Map “World Roads” from ArcGIS Online Database (accessed at November 9th, 2012). ln(Distance to Wittenberg). Variable containing the geodesic distances between each region’s centroid and the city of Wittenberg in the German State of Saxony-Anhalt. The coordinates of Wittenberg are taken from the website geonames.com (accessed at November 8th, 2012). ln(Employees Compensation). Natural logarithm of average of employees compensation (wages, salaries and employer’s social contributions) at NUTS-2 level measured at current prices and from the year 2009. Data was obtained from the Eurostat regional statistics database (http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset= nama_r_e2rem&lang=en; accessed at October 10th, 2012). ln(Fixed Capital). Gross fixed capital formation by NUTS-2 regions measured for 2009. Data is obtained from the Eurostat regional statistics database (http:// appsso.eurostat.ec.europa.eu/nui/show.do?dataset=nama_r_e2gfcfr2&lang=en; accessed at October 10th, 2012). Longitude. The values of this variable represent the longitude in decimal degrees of a region’s centroid and are obtained from a GIS map of NUTS territories provided by the Eurostat GISCO Database (http://epp.eurostat.ec.europa.eu/cache/GISCO/ geodatafiles/NUTS_2010_03M_SH.zip; accessed at November 8th, 2012). Medieval Mining. Binary Variable depicting regions with medieval copper or salt mining sites. The variable is coded according to a map in Elbl (2007) as well as information in Spufford (2002). Mining Region. Dummy variable equal to one if in a region at least one ore or coal mine (or mining firm) is located. The information on which the coding is based origi- nate from the structural business statistics included in the Eurostat regional statistics database (http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=sbs_r_ nuts06_r2&lang=en accessed at January 28th, 2012). Mountain Region. Categorial variable equal to one if in a region more than 50% of their population living in mountain areas according to the ESPON (European Observation Network for Territorial Development and Cohesion) regional typologies project. The variable is equal to one if more than 50% of a region’s population live in a mountain area. It is two if more than 50% of a region’s surface is covered by mountain areas. At last, it is three for regions with more than 50% of their surface covered by mountain areas and with more than 50% of their population living in mountain areas. It is zero when a region fulfills none of this criteria. The data and an explanation of the classifications can be downloaded from http://www.espon.eu/export/sites/default/Documents/

57 ToolsandMaps/ESPONTypologies/Typologies_metadata_data_final.xls (accessed at November 8th, 2012). Patents. Total number (over all IPO section and classes) of patent applications to the European Patent Oﬃce (EPO) in each region in 2009. Data available from the Eurostat regional statistics database (http://appsso.eurostat.ec.europa.eu/nui/show.do? dataset=pat_ep_ripc&lang=en; accessed at October 10th, 2012). Post Communistic Country. A binary variable equal to one if a region lies in an Eastern European post communistic transition country, i.e. the Czech Republic, Hungary, Lithuania or Poland. Coded by the author. Printing Press before 1500 AD. Dummy variable equal to one if at least one city in a region had adopted printing technology before 1500 AD. The coding is based on information in Benzing (1982), Clair (1976) and the Incunabula Short Title Catalogue (ISTC) of the British library (http://www.bl.uk/catalogues/istc/index.html; accessed at November 18th, 2012). A region is included if any of these sources mentioned a city in this region. Quality of Government. The European Regional Quality of Government Index (EQI) as developed by the Quality of Government Institute at the university of Gothenburg in Denmark. The index is constructed in a similar way than the World Governance (WGI) Indicators of the World Bank (further information on the index design and the data can be found here: http://www.qog.pol.gu.se/digitalAssets/1362/1362471_ eqi---correlates-codebook.pdf; accessed at January 28th 2013). The data on which the indix is based are collected in 2009. In Belgium, Germany, Netherlands and Hungary the index report values at NUTS-1 level in the other countries in our dataset it reports values at NUTS-2 level. The data can be downloaded from http://www.qog.pol.gu. se/digitalAssets/1362/1362473_eqi-and-correlates--qog-website-.xlsx (accessed at January 28th, 2013). Unemployment. The average annual unemployment rate (in percent) in a region in 2009 (including people above the age of 15). Data is from the Eurostat regional statistics database (http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=lfst_r_ lfu3rt&lang=en; accessed at October 10th, 2012). University before 1500. Dummy variable equal to one if at least one city in a region has a university founded before 1500 AD. Coding according to Eulenburg (1994), Kinder and Hilgemann (1970) and R¨uegg(1993). The a city is recognized if it is mentioned by any of these sources. If there were doubts on the founding date of a university (or contradicting dates) Cantoni and Yuchtman (2012) or wikipedia are used as validation.

58 59 Table A.1: Descriptive Data Overview – Regional Level Variables

Variable Obs Mean Std. Dev. Min Max

Altitude 839 279.230 320.194 -6.200 2472.600 Bishop before 1000 AD 839 .064 .246 0 1 Capital 839 0.011 0.103 0 1 Capital Autnomous Region 839 0.051 0.221 0 1 Commercial Importance 839 0.67 0.955 0 5 Commercial Importance Alt. 839 1.46 0.866 0 5.357 District-Free City 839 0.147 0.354 0 1 Eastern German Region 839 0.122 0.327 0 1 Education 832 24.211 6.319 8.4 48.6 Hanseatic League 839 0.108 0.311 0 1 Imperial City 839 0.069 0.254 0 1 Imperial Road 839 0.045 0.208 0 1 Inequality 825 1.134 0.921 0.037 8.425 Latitude 839 49.460 3.088 38.245 55.939 ln(Area) 839 7.032 1.297 3.575 9.400 ln(Distance to Airport) 839 -0.645 0.727 -4.142 0.792 ln(Distance to Border) 839 -0.825 1.083 -5.532 1.16 ln(Distance to Coast) 839 0.308 1.204 -5.566 1.882 ln(Distance to Railroad) 839 -2.111 1.390 -7.365 0.429 ln(Distance to River) 839 -.675 1.322 -7.185 1.944 ln(Distance to Road) 839 -4.001 1.376 -10.868 -1.194 ln(Distance to Trade Center) 839 0.432 0.272 0 1.665 ln(Distance to Wittenberg) 839 6.027 0.804 -7.447 7.335 ln(Employees Compensation) 825 9.867 0.924 7.086 12.331 ln(Fixed Capital) 803 9.141 0.818 6.802 11.494 ln(Population Density) 839 5.351 1.137 2.709 9.964 ln(Relative GDP Density) 839 -.077 1.262 -2.461 6.194 Longitude 839 10.228 5.012 -4.091 25.573 Medieval Mining 839 0.027 0.16 0 1 Mining Region 839 0.228 0.420 0 1 Mountain Region 839 0.479 1.022 0 3 Patents 803 83.094 89.654 0.286 764.717 Post Communistic Country 839 0.111 0.314 0 1 Printing Press before 1500 839 0.199 0.4 0 1 Quality of Government 839 72.130 17.163 10.18 97.61 Trade City 361 .249 .433 0 1 Trade Center 839 0.137 0.344 0 1 Unemployment 582 8.237 3.435 1.9 19.1 University before 1500 839 0.052 0.223 0 1

60 Table A.2: Descriptive Data Overview – City Level Variables

Variable Obs Mean Std. Dev. Min Max

Bishop 10000 AD 361 0.127 0.334 0 1 Imperial Road 361 0.078 0.268 0 1 Imperial City 361 0.122 0.328 0 1 Hanseatic League 361 0.155 0.363 0 1 Latitude 361 48.453 3.633 40.11 54.473 Longitude 361 8.727 5.048 -4.29 22 Mountain Region 361 0.385 0.887 0 3 ln(Distance to Coast) 361 -0.24 1.326 -5.566 1.762 ln(Distance to River) 361 -0.541 1.504 -7.185 1.944 ln(Population 1200 AD) 86 9.533 0.812 6.908 11.608 ln(Population 1300 AD) 199 9.114 1.104 6.908 11.918 ln(Population 1400 AD) 180 9.053 1.063 6.908 12.524 ln(Population 1500 AD) 361 8.817 0.983 6.908 12.324 Trade City 361 .249 .433 0 1

61 Table A.3: Overview over the included Trade Cities and Regions

Trade City NUTS-3 Region country Map Sources (Primary) Other Historical Records

Bruck Ostliche¨ Austria Magocsi (2002) Obersteiermark Innsbruck Innsbruck Austria Davies and Moorhouse Schulte (1966), Spufford (2002), King (1985), Magocsi (2002) (2002) and Stier et al. (1956) Graz Graz Austria Magocsi (2002), Stier et al. (1956) Linz Linz-Wels Austria Davies and Moorhouse (2002), Magocsi (2002), Stier et al. (1956) 62 Vienna Wien Austria Davies and Moorhouse Kinder and Hilgemann (2002), Magocsi (2002), (1982), Spufford (2002) Stier et al. (1956) Villach Klagenfurt-Villach Austria Magocsi (2002) Schulte (1966) Salzburg Salzburg und Austria Davies and Moorhouse Schulte (1966), Spufford (2002) Umgebung (2002), Magocsi (2002), Stier et al. (1956) Antwerp Arr. Antwerpen Belgium Davies and Moorhouse Ammann (1955), Hunt (2002), Stier et al. (1956) and Murray (1999), Kinder and Hilgemann (1982), Spufford (2002) Bruges Arr. Brugge Belgium Davies and Moorhouse Hunt and Murray (1999), (2002), King (1985), Stier et Kinder and Hilgemann al. (1985) (1982), Spufford (2002) Table A.3 – Continued Ghent Arr. Gent Belgium Stier et al. (1956) Hunt and Murray (1999), Kinder and Hilgemann (1982), Brno Jihomoravskýkraj Czech Republic Davies and Moorhouse Spufford (2002) (2002), Magocsi (2002) Kutna Hora Stredoceskýkraj Czech Republic Magocsi (2002) Spufford (2002) Olmouc Olomouckýkraj Czech Republic Davies and Moorhouse (2002), Magocsi (2002) Prague Hlavn´ımesto Praha Czech Republic Davies and Moorhouse Kinder and Hilgemann (2002), Magocsi (2002), (1982), Spufford (2002) Stier et al. (1956) Avignon Vaucluse France King (1985), Stier Hunt and Murray (1999), et al. (1956) Spufford (2002) Bayonne Pyrénées-Atlantique France Stier et al. (1956) Spufford (2002) 63 Bordeaux Gironde France Stier et al. (1956) Spufford (2002) Chalon-sur-Saône Saône-et-Loire France Stier et al. (1956) Schulte (1966), Spufford (2002) Harfleur Seine-Maritime France King (1985), Stier et al. (1956) Limoges Haute-Vienne France King (1985), Stier et al. (1956) Lyon Rhône France Stier et al. (1956) Ammann (1955), Hunt and Murray (1999), Kinder and Hilgemann (1982), Schulte (1966), Spufford (2002) Marseille Bouches-du-Rhône France King (1985), Stier Kinder and Hilgemann et al. (1956) (1982), Spufford (2002) Metz Moselle France Davies and Moorhouse (2002) Schulte (1966) Montpellier Hérault France King (1985) Spufford (2002) Table A.3 – Continued Narbonne Aude France King (1985), Stier et al. (1956) Orleans Loiret France Stier et al. (1956) Paris Paris France Davies and Moorhouse Kinder and Hilgemann (1982), (2002), King (1985), Stier Hunt and Murray (1999), Schulte et al. (1956) (1966), Spufford (2002) Perpignan Pyrénées-Orientales France King (1985) Spufford (2002) Reims Marne France Stier et al. (1956) Schulte (1966), Spufford (2002) St. Melo Ille-et-Vilaine France Stier et al. (1956) Strasbourg Bas-Rhin France Davies and Moorhouse (2002), Kinder and Hilgemann (1982), Stier et al. (1956) Schulte (1966), Spufford (2002) Toulouse Haute-Garonne France Stier et al. (1956) Spufford (2002) Tours Indre-et-Loire France Stier et al. (1956) Spufford (2002) Troyes Aube France Stier et al. (1956) Schulte (1966), Spufford (2002) 64 Amberg Amberg, Germany Magocsi (2002) District-Free City Augsburg Augsburg, Germany Davies and Moorhouse Dietze (1923), Kinder and District-Free City (2002), King (1985), Magocsi Hilgemann (1982), Schulte (2002), Stier et al. (1956) (1966), Spufford (2002) Berlin Berlin Germany Davies and Moorhouse (2002), Magocsi (2002), Stier et al. (1956) Brunswick Braunschweig, Germany Davies and Moorhouse (2002), Dollinger (1966), Kinder and District-Free City King (1985), Magocsi (2002), Hilgemann (1982) Stier et al. (1956) Bremen Bremen, Germany Davies and Moorhouse Dollinger (1966), Kinder and District-Free City (2002), Stier et al. (1956) Hilgemann (1982), Spufford (2002) Bremerhaven Bremerhaven, Germany Davies and Moorhouse Dollinger (1966), Kinder and District-Free City (2002), Stier et al. (1956) Hilgemann (1982), Spufford (2002) Table A.3 – Continued Cologne Cologne, Germany Davies and Moorhouse Ammann (1955), Dollinger (1966), District-Free City (2002), King (1985), Stier Hunt and Murray (1999), Kinder and et al. (1956) Hilgemann (1982), Spufford (2002) Constance Konstanz Germany Davies and Moorhouse Dietze (1923), Schulte (1966), (2002), Stier et al. (1956) Spufford (2002) Dortmund Dortmund, Germany Stier et al. (1956) Dollinger (1966) District-Free City Einbeck Northeim Germany Stier et al. (1956) Erfurt Erfurt, Germany Davies and Moorhouse (2002), Dietze (1923), Kinder and District-Free City Magocsi (2002), Stier et al. (1956) Hilgemann (1982) Frankfurt (Oder) Frankfurt (Oder), Germany Davies and Moorhouse District-Free City (2002), Magocsi (2002), Stier et al. (1956) 65 Frankfurt (Main) Frankfurt am Main, Germany Davies and Moorhouse (2002), Kinder and Hilgemann (1982), District-Free City Magocsi (2002), Stier et al. (1956) Schulte (1966), Spufford (2002) Fulda Fulda Germany Stier et al. (1956) Görlitz Görlitz, Germany Magocsi (2002) Spufford (2002) District-Free City Greifswald Greifswald, Germany Stier et al. (1956) Dollinger (1966) District-Free City Hamburg Hamburg Germany Davies and Moorhouse Ammann (1955), Dollinger (2002), King (1985), Magocsi (1966), Kinder and Hilgemann (2002), Stier et al. (1956) (1982), Spufford (2002) Hannover Region Hannover Germany Magocsi (2002), Stier et al. (1956) Hildesheim Hildesheim Germany Stier et al. (1956) Dollinger (1966) Leipzig Leipzig, District-Free Germany Davies and Moorhouse (2002), Ammann (1955), Kinder and City Magocsi (2002), Stier et al. Hilgemann (1982), Spufford (2002) (1956) Table A.3 – Continued

Lübeck Lübeck, Germany Ammann (1955), Dietze (1923), Davies and Moorhouse District-Free City Dollinger (1966), Kinder and (2002), King (1985), Magocsi Hilgemann (1982) and Hunt and (2002), Stier et al. (1956) Murray (1999) Lüneburg Lüneburg Germany Davies and Moorhouse Dollinger (1966), Kinder and (2002), Magocsi (2002), Stier Hilgemann (1982), Spufford et al. (1956) (2002) Magdeburg Magdeburg, Germany Davies and Moorhouse Dollinger (1966), Kinder and District-Free City (2002), Magocsi (2002), Hilgemann (1982) Stier et al. (1956) Minden Minden-Lübbecke Germany Stier et al. (1956) Münster Münster, Germany Stier et al. (1956) 66 District-Free City Nuremberg Nuremberg, Germany Davies and Moorhouse Ammann (1955), Dietze (1923), District-Free City (2002), Magocsi (2002), Kinder and Hilgemann (1982), Stier et al. (1956) Spufford (2002) Osnabrück Osnabrück, Germany Stier et al. (1956) Dollinger (1966) District-Free City Paderborn Paderborn Germany Stier et al. (1956) Dollinger (1966) Ravensburg Ravensburg Germany Stier et al. (1956) Dietze (1923), Spufford (2002) Regensburg Regensburg, Germany Davies and Moorhouse, Magocsi Schulte (1966), District-Free City (2002), Stier et al. (1956) Spufford (2002) Rostock Rostock, Germany Davies and Moorhouse (2002), Dollinger (1966),Kinder and District-Free City King (1985), Magocsi (2002), Hilgemann (1982) Stier et al. (1956) Soest Soest Germany Stier et al. (1956) Dollinger (1966) Table A.3 – Continued Stralsund Stralsund, Germany Magocsi (2002), Stier et Dollinger (1966) District-Free City al. (1956) Ulm Ulm, Urban District Germany Davies and Moorhouse (2002), Dietze (1923), Kinder and Stier et al. (1956) Hilgemann (1982), Schulte (1966), Spufford (2002) Wismar Wismar, Germany Stier et al. (1956) Dollinger (1966) District-Free City Budapest Budapest Hungary Davies and Moorhouse (2002), Spufford (2002) Magocsi (2002), Stier et al. (1956) Pecs Baranya Hungary Magocsi (2002) Ancona Ancona Italy Magocsi (2002), Stier et al. (1956) Spufford (2002) Bari Bari Italy Magocsi (2002), Stier et al. (1956) Spufford (2002) Bologna Bologna Italy King (1985), Magocsi (2002), Schulte (1966) 67 Stier et al. (1956) Bozen Bolzano-Bozen Italy Magocsi (2002), Stier et Dietze (1923), Kinder and al. (1956) Hilgemann (1982), Schulte (1966) Como Como Italy Stier et al. (1956) Schulte (1966) Florence Firenze Italy Magocsi (2002), King (1985), Dietze (1923), Kinder and Stier et al. (1956) Hilgemann (1982), Hunt and Murray (1999), Spufford (2002) Genoa Genova Italy Davies and Moorhouse Ammann (1955), Dietze (1923), (2002), King (1985), Stier Hunt and Murray (1999), at al. (1956) Kinder and Hilgemann (1982), Schulte (1966), Spufford (2002) Lucca Lucca Italy Stier et al. (1956) Dietze (1923), Spufford (2002) Mantoa Mantova Italy Magocsi (2002) Table A.3 – Continued Milan Milano Italy Davies and Moorhouse Dietze (1923), Hunt and (2002), King (1985), Murray (1999),Kinder and Stier et al. (1956) Hilgemann (1982), Schulte (1966), Spufford (2002) Naples Napoli Italy King (1985), Magocsi (2002), Hunt and Murray (1999), Stier et al. (1956) Kinder and Hilgemann (1982), Schulte (1966), Spufford (2002) Padoa Padova Italy Magocsi (2002) Schulte (1966) Parma Parma Italy Magocsi (2002) Prato Prato Italy King (1985) Spufford (2002) Rome Roma Italy Hunt and Murray King (1985), Magocsi (1999),Kinder and (2002), Stier et al. (1956) Hilgemann (1982), 68 Spufford (2002) Siena Siena Italy King (1985),Magocsi Spufford (2002) (2002), Stier et al. (1956) Trento Trento Italy Magocsi (2002) Schulte (1966) Treviso Treviso Italy Magocsi (2002) Schulte (1966) Udine Udine Italy Magocsi (2002) Venice Venezia Italy Davies and Moorhouse Dietze (1923),Hunt and (2002), King (1985), Murray (1999), Kinder and Magocsi (2002), Stier Hilgemann (1982), Schulte et al. (1956) (1966), Spufford (2002) Verona Verona Italy Magocsi (2002), Stier et al. Schulte (1966) (1956) Klaipeda Klaipedos apskritis Lithuania Davies and Moorhouse (2002), Magocsi (2002) Table A.3 – Continued Kovno Kauno apskritis Lithuania King (1985), Magocsi (2002) Kinder and Hilgemann (1982) Palanga Klaipedos apskritis Lithuania Stier et al. (1956) Amsterdam Groot-Amsterdam Netherlands King (1985), Stier et al. Dollinger (1966), Kinder (1956) and Hilgemann (1982), Spufford (2002) Deventer Zuidwest-Overjissel Netherlands King (1985), Stier et Dollinger (1966) al. (1956) Dordrecht Zuidoost-Zuid-Holland Netherlands King (1985), Stier et Dollinger (1966), al. (1956) Spufford (2002) Kampen Noord-Overjissel Netherlands King (1985) Dollinger (1966), Spufford (2002) Maastricht Zuid-Limburg Netherlands Stier et al. (1956) Rotterdam Groot-Rijnmond Netherlands Stier et al. (1956) 69 Utrecht Utrecht Netherlands Stier et al. (1956) Braniewo Elblaski Poland Stier et al. (1956) Dollinger (1966) Cracow Miasto Kraków Poland Davies and Moorhouse (2002), Ammann (1955), Kinder King (1985), Magocsi (2002), and Hilgemann (1982), Stier et al. (1956) Spufford (2002) Elblag Elblaski Poland Magocsi (2002), Stier Kinder and Hilgemann (1982) et al. (1956) Gdansk Gdanski Poland Davies and Moorhouse Ammann (1955), Dietze (2002), King (1985), Magocsi (1923), Dollinger (1966), (2002), Stier et al. (1956) Kinder and Hilgemann (1982), Spufford (2002) Malbork Starogardzki Poland King (1985) PiotrkówTrybunalski Piotrkowski Poland Davies and Moorhouse (2002) Plock Ciechanowsko-plocki Poland Magocsi (2002) Table A.3 – Continued Poznan Poznanski Poland Davies and Moorhouse Ammann (1955) (2002),Magocsi (2002), Stier et al. (1956) Torun Bydgosko-Torunski Poland Davies and Moorhouse (2002), Dollinger (1966), King (1985), Magocsi (2002), Spufford (2002) Stier et al. (1956) Warsaw Miasto Warszawa Poland Davies and Moorhouse (2002), Ammann (1955) and Kinder Magocsi (2002), Stier et al. (1956) and Hilgemann (1982) Wroclaw Miasto Wroclaw Poland Davies and Moorhouse Ammann (1955), Dietze (1923), (2002), King(1985), Magocsi Kinder and Hilgemann (1982), (2002), Stier et al. (1956) Spufford (2002) Stargard Szczeciński Poland Stier et al. (1956) Dollinger (1966) 70 Table A.4: Medieval Trade Cities and Regions with long-run trade activity

Trade City NUTS-3 Region country mentioned earliest by earliest period mentioned

Linz Linz-Wels Austria Humnicki and Borawska 9th century (eds.) (1969) Vienna Wien Austria Dietze (1923) before 14th century Antwerp Arr. Antwerpen Belgium Heyd (1897b) 14th century Bruges Arr. Brugge Belgium Heyd (1897b) 14th century Brno Jihomoravskýkraj Czech Republic Humnicki and Borawska 9th century (1969) Olmouc Olomouckýkraj Czech Republic Humnicki and Borawska 9th century (1969) Prague Hlavn´ımesto Praha Czech Republic Humnicki and Borawska 9th century 71 (1969) Avignon Vaucluse France Heyd (1879b) high medieval Bordeaux Gironde France Dollinger (1966) 15th century Limoges Haute-Vienne France Heyd (1879a) before 12th century Lyon Rhône France Dollinger (1966) 15th century Marseille Bouches-du-Rhône France Heyd (1879a) before 10th century Metz Moselle France Heyd (1879b) 14th century Montpellier Hérault France Heyd (1879a) before 12th century Narbonne Aude France Heyd (1879a) before 12th century Paris Paris France Dollinger (1966) 15th century Strasbourg Bas-Rhin France Dollinger (1966) before 1250 Troyes Aube France Dietze (1923) before 9th century Augsburg Augsburg, District-Free City Germany Dietze (1923) before 9th century Berlin Berlin Germany Dollinger (1966) 15th century Table A.4 – Continued Brunswick Braunschweig, District-Free City Germany Dietze (1923) before 9th century Bremen Bremen, District-Free City Germany Heyd (1879a) before 12th century Bremerhaven Bremerhaven, District-Free City Germany Heyd (1879a) before 12th century Cologne Cologne, District-Free City Germany Dietze (1923) before 9th century Constance Konstanz Germany Dietze (1923) before 9th century Erfurt Erfurt, District-Free City Germany Heyd (1879a) before 12th century Frankfurt (Oder) Frankfurt (Oder), District-Free City Germany Heyd (1879a) before 12th century Frankfurt (Main) Frankfurt am Main, District-Free City Germany Dietze (1923) before 9th century Görlitz Görlitz,District-Free City Germany Rutkowski (1980a) 14th century (1370) Greifswald Greifswald, District-Free City Germany Dietze(1923) before 14th century Hamburg Hamburg Germany Dollinger (1966) before 1250 Hildesheim Hildesheim Germany Dollinger (1966) 13th – 14th century Lübeck Lübeck, District-Free City Germany Heyd (1879a) Treaty of Smolensk (1229) 72 Lüneburg Lüneburg,District Germany Dollinger (1966) 13th – 14th century Magdeburg Magdeburg, District-Free City Germany Heyd (1879a) before 10th century Minden Minden-Lübbecke Germany Dollinger (1966) 13th – 14th century Münster Münster,District-Free City Germany Dollinger (1966) Treaty of Smolensk (1229) Nuremberg Nuremberg, District-Free City Germany Dietze (1923) before 9th century Osnabrück Osnabrück, District-Free City Germany Dollinger (1966) 13th – 14th century Paderborn Paderborn Germany Dollinger (1966) 13th – 14th century Regensburg Regensburg, District-Free City Germany Dietze (1923) before 9th century Rostock Rostock, District-Free City Germany Dollinger (1966) 13th – 14th century Soest Soest Germany Dollinger (1966) 13th – 14th century Stralsund Stralsund, District-Free City Germany Dietze (1923) before 14th century Ulm Ulm, Urban District Germany Dietze(1923) before 9th century Wismar Wismar, District-Free City Germany Dollinger (1966) 13th – 14th century Budapest Budapest Hungary Wojtowicz (1956) 14th century Table A.4 – Continued Ancona Ancona Italy Heyd (1879a) before 12th century Bari Bari Italy Heyd (1879a) before 12th century Bologna Bologna Italy Heyd (1879b) 14th century Florence Firenze Italy Heyd (1879b) 14th century Genoa Genova Italy Heyd (1879a) before 12th century Lucca Lucca Italy Heyd (1879a) before 13th century Milan Milano Italy Heyd (1879b) 14th century Naples Napoli Italy Heyd (1879b) before 12th century Parma Parma Italy Heyd (1879b) 14th century Pisa Pisa Italy Dietze (1923) before 14th century Rome Roma Italy Heyd (1879a) before 12th century Siena Siena Italy Heyd (1879b) 13th century (1209) Venice Venezia Italy Heyd (1879a) before 12th century 73 Kovno Kauno apskritis Lithuania Dollinger (1966) between 1350 and 1500 Cracow Miasto Kraków Poland Humnicki and Borawska 9th century (1969) Gdansk Gdanski Poland Dollinger (1966) 13th – 14th century Malbork Starogardzki Poland Wojtowicz (1956) 14th century PiotrkówTrybunalski Piotrkowski Poland Wojtowicz (1956) 14th century Plock Ciechanowsko-plocki Poland Wojtowicz (1956) 14th century Poznan Miasto Poznan Poland Wojtowicz (1956) 14th century Szczecin Miasto Szczecin Poland Wojtowicz (1956) 14th century Torun Bydgosko-Torunski Poland Dollinger (1966) 13th – 14th century Warsaw Miasto Warszawa Poland Wojtowicz (1956) 14th century Wroclaw Miasto Wroclaw Poland Dollinger (1966) 13th – 14th century B. Robustness Checks

Robustness to Inﬂuential Observations and Additional Controls

In this appendix we report the results of several robustness checks and additional results mentioned in the main text of the study. To be precise, in Table B.1 we re-run some specifications from Table 5 and 6 in the main text, including additional control variables (a dummy variable for medieval copper mining regions, an interaction term of latitude and longitude, the country-evel share of Catholics and a dummy for regions containing important medieval residence cities).In Table B.2 we look whether the results are sensitive to the exclusion of influential observations, identified by the DFITS statistics (see main text for a detailed description).

74 Table B.1: Inclusion of Additional Control Variables

(1) (2) (3) (4) (5) (6) (7) (8) Dep. Var. ln(GDP per capita) Modiﬁed Speciﬁcation Table 6 Table 6 Table 5 Table 5 Table 6 Table 6 Table 6 Table 6 column (3) column (6) column (3) column (6) column (4) column (9) column (3) column (8)

Modiﬁcation Adding Dummy for Adding a interaction variable Adding share of Catholics Adding a dummy for important medieval mining regions of latitude and longitude in a country residence cities

Additional Variable No No Yes No Yes 75 signiﬁcant

Trade Center 0.181*** 0.264*** 0.13*** 0.181*** (0.029) (0.031) (0.027) (0.03) ln(Distance to Trade Center) -0.134** -0.291*** -0.138*** -0.135* (0.053) (0.055) (0.041) (0.053)

Obs. 839 839 839 839 518 518 839 839 Adj. R2 0.784 0.776 0.778 0.762 0.878 0.872 0.784 0.776 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. For the controls included in each specification consult the main text or the notes to the original tables mentioned in the third row. Each regression includes a constant not reported. Table B.2: Regressions of Table 5 Without Influential Observations

Dep. Var. ln(GDP per capita) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Trade Center 0.17*** 0.11*** 0.153*** 0.117*** 0.0794** (0.022) (0.024) (0.025) (0.026) (0.021) ln(Distance to Trade Center) -0.108*** -0.081** -0.111*** -0.12*** -0.064* (0.038) (0.039) (0.046) (0.043) (0.038)

Country Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes NUTS-2 Dummies Yes Yes Yes Yes No Yes Yes Yes Yes No Basic Geographic Controls Yes Yes Yes Yes No Yes Yes Yes Yes No Geographic Centrality Controls Yes No No No No Yes No No No No Region Characteristics No Yes No No No No Yes No No No Historical Region Characteristic No No Yes No No No No Yes No No Development Covariates No No No Yes No No No No Yes No All Robust Controls No No No No Yes No No No No Yes 76 No. of removed regions 40 45 40 41 47 40 45 41 43 44 Obs. 799 794 799 477 771 799 794 798 475 774 Adj. R2 0.844 0.891 0.829 0.911 0.901 0.837 0.887 0.816 0.904 0.899 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. The basic geographic controls include a region’s latitude, longitude and altitude. The geographic centrality controls include the ln distances of a region’s centroid to the nearest airport, railroad, road, border and coast point. Region characteristic controls include a dummies for regions in Germany that are district-free cities, for regions including a country’s capital, are classified as mountain regions, with ore or coal mines, located in the former GDR and located in an Eastern European post-communistic transition country. Furthermore it encompasses the ln of a regions area. The historical region characteristics consist of a dummy variables indicating regions with a university founded before 1500 AD, that adopted printing technology before 1500 AD, contain cities that were members of the Hanseatic League, with former imperial cities and were located on an imperial road. Moreover it includes the ln of the distance of a region’s centroid to Wittenberg. The growth covariates encompass a region’s unemployment rate, number of registered patents, average firm ln fixed capital stock, average worker compensation. Furthermore, it includes the share of people aged between 25-64 with tertiary education on NUTS-2 level, the quality of government index on NUTS-1/ NUTS-2 level and the ratio of an average workers compensation to a region’s GDP per capita as inequality measure. The set of all robust covariates encompasses altitude, the ln distances to airports and railroads, dummies for district free cities, capital cities, capital cities of autonomous regions, post-communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. A region is removed from the estimation if its DFITS value is above the cut-off of DFITSj > 2 k N (with k indicating the number of regressors and N denoting the number of observations in the sample). Each | | \ regression includes a constant not reported. p Results for Alternatively Coded Medieval Trade Variables

In Tables B.3 and B.6 we conduct the OLS, IV and mediation analysis estimations with alternatively coded medieval trade variables, i.e. alternative samples of medieval trade cities. Here, Table B.3 show the estimation results with when we only consider trade cities mentioned in more than one of the sources. In Table B.4 we redo this estimations this time excluding cities for which the actual importance in trade is in doubt. To continue, in Table B.5 we repeat this, using the original sample and include additional cities for which we think they might be important, albeit they are not mentioned by our main sources. At last, in Table B.6 we show the results for a sample of trade cities that only includes cities for which historical sources indicate long-run trade activities (i.e. cities that are important trade cities around 1500 AD and that were important also in the period before). An overview over these cities the earliest period in which trade activities are reported and the source mentioned the respective city are depicted in Table A.4.

77 Table B.3: Results for Alternative Trade Center Dummy – Without Regions Mentioned by Only One Source

Dep. Var. ln(GDP per capita) ln(City Growth) ln(Relative GDP Density) ln(GDP per capita) (1) (2) (3) (4) (5) (6)

Method OLS LIML IV Lewbel (2012) OLS Mediation Analysis Estimated Equation Equation (6) Equation (7)

Estimated Speciﬁcation Table 5 Table 6 Table 6 Table 8 Table 9 Table 9 Column (5) Column (1) Column (2) Column (1) Column (4) Column (4)

Trade Center 0.0543** 0.363*** 0.0613** 0.479** 0.3267*** -0.00912 (0.0225) (0.133) (0.0260) (0.232) (0.071) (0.0181) ln(Relative GDP Density) 0.203*** (0.0109) 78 Obs. 818 818 818 86 818 818 Centered R2 R2 0.877 0.534 0.629 0.344 0.938 0.924 \ ACME 0.0654 Direct Effect -0.0085 Total Effect 0.0569 % of total mediated 112.7 Underidentification Test 14.45 173.3 p-value 0.000 0.000 Overidentification Test 0.000 61.09 p-value 1.000 0.236 AP F-statistic of 8.27 46.53 excluded IV’s p-value 0.000 0.000 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. For the controls included in each specification consult the main text or the notes to the original tables mentioned in the third row. In columns (1) and (2) the adjusted R2 is reported. In column (3) and (4) the centered R2 is shown and in columns (5) and (6) the R2 . In column (3) the results of the first stage are omitted but available from the author. Each regression includes a constant not reported. Table B.4: Results for Alternative Trade Center Dummy – Cities with Uncertain Importance Removed

Dep. Var. ln(GDP per capita) ln(City Growth) ln(Relative GDP Density) ln(GDP per capita) (1) (2) (3) (4) (5) (6)

Method OLS LIML IV Lewbel (2012) OLS Mediation Analysis Estimated Equation Equation (6) Equation (7)

Estimated Speciﬁcation Table 5 Table 6 Table 6 Table 8 Table 9 Table 9 Column (5) Column (1) Column (2) Column (1) Column (4) Column (4)

Trade Center 0.0670*** 0.375*** 0.073*** 0.468* 0.3621*** -0.0036 (0.023) (0.14) (0.027) (0.235) (0.074) (0.02) ln(Relative GDP Density) 0.203*** (0.011) 79 Obs. 818 818 818 86 818 818 Centered R2 R2 0.877 0.544 0.621 0.342 0.939 0.919 \ ACME 0.0724*** Direct Effect -0.0029 Total Effect 0.0694** % of total mediated 102.8** Underidentification Test 15.18 160.2 p-value 0.000 0.000 Overidentification Test 0.008 58.41 p-value 0.93 0.317 AP F-statistic of 8.57 43.92 excluded IV’s p-value 0.000 0.000 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. For the controls included in each specification consult the main text or the notes to the original tables mentioned in the third row. In columns (1) and (2) the adjusted R2 is reported. In column (3) and (4) the centered R2 is shown and in columns (5) and (6) the R2 . In column (3) the results of the first stage are omitted but available from the author. Each regression includes a constant not reported. Table B.5: Results for Alternative Trade Center Dummy – Cities with Uncertain Importance Added

Dep. Var. ln(GDP per capita) ln(City Growth) ln(Relative GDP Density) ln(GDP per capita) (1) (2) (3) (4) (5) (6)

Method OLS LIML IV Lewbel (2012) OLS Mediation Analysis Estimated Equation Equation (6) Equation (7)

Estimated Speciﬁcation Table 5 Table 6 Table 6 Table 8 Table 9 Table 9 Column (5) Column (1) Column (2) Column (1) Column (4) Column (4)

Trade Center 0.0686*** 0.324*** 0.0765*** 0.544** 0.3268*** 0.0038 (0.021) (0.114) (0.024) (0.225) (0.62) (0.016) ln(Relative GDP Density) 0.202*** (0.011) 80 Obs. 818 818 818 86 818 818 Centered R2 R2 0.878 0.552 0.623 0.358 0.939 0.919 \ ACME 0.0652*** Direct Effect 0.0044 Total Effect 0.0696*** % of total mediated 93.3*** Underidentification Test 13.03 203.9 p-value 0.001 0.000 Overidentification Test 0.192 70.56 p-value 0.661 0.065 AP F-statistic of 7.60 56.93 excluded IV’s p-value 0.001 0.000 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. For the controls included in each specification consult the main text or the notes to the original tables mentioned in the third row. In columns (1) and (2) the adjusted R2 is reported. In column (3) and (4) the centered R2 is shown and in columns (5) and (6) the R2 . In column (3) the results of the first stage are omitted but available from the author. Each regression includes a constant not reported. Table B.6: Results for Alternative Trade Center Dummy – Only Cities with Long-Run Trade Activity

Dep. Var. ln(GDP per capita) ln(City Growth) ln(Relative GDP Density) ln(GDP per capita) (1) (2) (3) (4) (5) (6)

Method OLS LIML IV Lewbel (2012) OLS Mediation Analysis Estimated Equation Equation (6) Equation (7)

Estimated Speciﬁcation Table 5 Table 6 Table 6 Table 8 Table 9 Table 9 Column (5) Column (1) Column (2) Column (1) Column (4) Column (4)

Trade Center 0.0568** 0.320*** 0.0743** 0.123 0.3218*** -0.0061 (0.027) (0.112) (0.032) (0.253) (0.088) (0.024) ln(Relative GDP Density) 0.202*** (0.011) 81 Obs. 818 818 818 86 818 818 Centered R2 R2 0.877 0.574 0.62 0.305 0.938 0.919 \ ACME 0.0641*** Direct Effect 0.0053 Total Effect 0.0588** % of total mediated 105.0** Underidentification Test 14.84 140.80 p-value 0.001 0.000 Overidentification Test 0.406 65.7 p-value 0.524 0.132 AP F-statistic of 9.16 78.54 excluded IV’s p-value 0.001 0.000 Notes. Standard errors adjusted for two-way clustering within NUTS-1 and NUTS-2 regions are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. For the controls included in each specification consult the main text or the notes to the original tables mentioned in the third row. In columns (1) and (2) the adjusted R2 is reported. In column (3) and (4) the centered R2 is shown and in columns (5) and (6) the R2 . In column (3) the results of the first stage are omitted but available from the author. Each regression includes a constant not reported. Description and Sources of the Additional Variables

Residence city. Binary variable that represents important residence cities (of Dukes, Kings . . . ) in the Holy Roman Empire or the German Reich (after 1871). The coding follows a wikipedia list at http://de.wikipedia.org/wiki/Residenzstadt (accessed February, 24th 2013) and Köbler(1988). It also includes residences of electors (“Kurfürsten”)and prince-bishoprics. Furthermore, it represents the capitals or residence cities of Italian duchies, kingdoms and republics (like Venice, Lombardy, Sardinia, Parma, Modena, Tuscany, Naples or the Kingdom of the two Sicilies). For all other countries it marked the capitals of pre-existing states or kingdoms, duchies etc. (e.g. in Poland it includes the residence of the kings of the Kingdom of Poland, in Lithuania the residence of the grand duke of Lithuania. . . ). The coding here follows the author’s information or different versions of Putzgers historical atlas (Bruckmüller(eds.) 2011 and Baldamus et al. (eds.) 1914). Share of Catholics. The share of people with Roman Catholic denomination (in percent of total population) in a country is taken from “The World Religion Dataset, 1945 - 2010” (Zeev and Henderson 2013) available from the “Correlates of War” project website (http://www.correlatesofwar.org/COW2%20Data/Religion/WRD_national.csv; accessed at May, 8th 2013). As always, we took the values from 2009.

An overview over the additional variables used for the robustness checks is provided in Table B.6 above: Table B.7: Descriptive Overview over the Additional Variables

Variable Obs Mean Std. Dev. Min Max Latitude*Longitude 839 507.123 253.213 -197.378 1401.973 Residence City 839 0.067 0.25 0 1 Share of Catholics 839 49.623 22.29 26.85 89.15

82 C. Additional Results

Here the result of the estimation of Table 9 using the ln population density of a NUTS-3 region as mediating agglomeration measure is shown. The results are almost identical to that obtained with the relative GDP density. However, the probably biggest difference between both estimations is that the average ACME using the population density is clearly lower. Neverthless, since it is always significant and on average around three quarters of the effect of medieval trade on ln GDP per capita is mediated by the ln population density our main conclusion does hold. Furthermore we report the results of estimating Table 8 using the Index of Commercial Importance instead of the trade city dummy (Table C.2). We see that the result are a little bit weaker (especially concerning the results for city growth between 1200 and 1500 AD). Nevertheless, the overall results and therefore also the general implications of the results do stay the same.

83 Table C.1: Medieval Trade, Population Density and Regional Economic Development

(1) (2) (3) (4) (5) (6)

Method OLS Mediation Analysis City Growth from to 1200–1500 1300–1500 1400–1500 Equation (7) Dep. Var. ln(Population Density) ln(GDP per capita)

P opulation1500 0.337*** 0.178*** 0.172*** P opulationt (0.105) (0.067) (0.062) ln(Population Density) 0.135*** 0.139*** 0.137*** (0.015) (0.015) (0.015) Trade Center 0.0308 (0.019) ln(Distance to Trade Center) -0.007 (0.027) Commercial Importance 0.0067 (0.008)

R2 0.964 0.955 0.947 0.889 0.888 0.888 ACME 0.0405*** -0.0605*** 0.0178*** Direct Eﬀect 0.0314 -0.0062 0.0067 Total Eﬀect 0.0719*** -0.0667** 0.0247*** % of total mediated 55.7*** 90.0** 70.8***

Equation (6) ln(Relative GDP Density)

Trade Center 0.3043*** (0.053) ln(Distance to Trade Center) -0.4313*** (0.108) Commercial Importance 0.1318*** (0.019)

Country Dummies Yes Yes Yes Yes Yes Yes NUTS-1 Dummies Yes Yes Yes Yes Yes Yes All Robust Controls Yes Yes Yes Yes Yes Yes

Obs. 85 179 197 818 818 818 R2 0.867 0.87 0.87 Notes. Robust standard errors are reported in parentheses. Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a NUTS-3 region. The set of all robust covariates encompasses altitude, the ln distances to airports and railroads, dummies for district free cities, capital cities, capital cities of autonomous regions, post-communistic transition countries, Eastern Germany, the ln of a region’s area, the share of people with tertiary education, the inequality measure and the printing press before 1500 AD dummy. Each regression includes a constant not reported. ACME is the “Average Causal Mediation Effect” and means how much of the effect of medieval trade is mediate, i.e. works indirectly through the relative GDP density.

84 Table C.2: Medieval Trade Activity and City Growth - Estimations using the Index of Commercial Importance

Dep. Var. ln( P opulation1500 ) ln( P opulation1500 ) ln( P opulation1500 ) ln(Population) ln(∆ Population) P opulation1200 P opulation1300 P opulation1400 (1) (2) (3) (4) (5) Method OLS RE

Commercial Importance 0.301* 0.266*** 0.105 0.394*** 0.156*** (0.155) (0.084) (0.093) (0.065) (0.052) ln(Population 1200 AD) -0.605*** (0.148) ln(Population 1300 AD) -0.607*** (0.069)

85 ln(Population 1400 AD) -0.362*** (0.076) ln(Populationt 1) -0.416*** − (0.05)

Obs. 86 199 180 826 390 Adj. R2 overall R2 0.346 0.381 0.173 0.344 0.26 \ Number of Clusters 361 194 Notes. Robust standard errors are reported in parentheses in columns (1) - (3). Standard errors clustered at city level are reported in parentheses in columns (4) and (5). Coefficient is statistically different from zero at the ***1 %, **5 % and *10 % level. The unit of observation is a city. The set of covariates encompasses the ln distances of a city to the next river or coast, dummies indicating cities that were residence of a bishop before 1000 AD, had the status of an imperial city, were located at a main imperial road, were member of the Hanseatic League or are classified as a mountain region by the EU regional statistics. Furthermore, we control for a city’s latitude and longitude and include country fixed effects. In columns (4) and (4) we additionally include year fixed effects. Each regression includes a constant not reported. References

Ammann, H. (1955): Vom geographischen Wissen einer deutschen Handelsstadt des Sp¨atmittelalters. Ulm und Oberschwaben: Zeitschrift f¨urGeschichte, Kunst und Kul- tur 34, pp. 39–65.

Baldamus, A., Schwabe, E., Koch, J. (eds.) (1914): F. W. Putzgers Historische Schu- latlas zur alten, mittleren und neuen Geschichte. Bielefeld and Berlin, Verlag von Velhagen & Klasing. url: http://www.maproom.org/00/04/index.php.

Becker, S., Boeckh, Katrin, Hainz, C., Woessmann, L. (2011): The Empire is Dead, Long Live the Empire! Long-run Persistence of Trust and Corruption in the Bureaucracy. CEPR Discussion Paper No. DP8288.

Benzing, J. (1982): Die Buchdrucker des 16. und 17. Jahrhunderts im Deutschen Sprachgebiet. Otto Hassowitz Verlag, Wiesbaden.

Bruckmuller¨ , E. (ed.) (2011): Putzger historischer Weltaltas. Berlin, Cornelsen Verlag.

Cantoni, D., Yuchtman, N. (2012): Medieval Universities, Legal Institutions, and the Commercial Revolution. NBER Working Paper No. 17979.

Clair, C. (1976): A History of European Printing. Academic Press, New York, NY.

Darby, H. C., H., Fullard (eds.) (1978): The New Cambrigde Economic History Atlas. Cambrigde, UK, Cambridge University Press.

Davies, N., Moorhouse, R. (2002): Microcosm: Portrait of a Central European City. Jonathan Cape, London.

Dietze, Hugo (1923): Geschichte des deutschen Handels. Gloeckners Handels-B¨ucherei. G. A. Gloeckner Verlagsbuchhandlung.

Dollinger, P. (1966): Die Hanse. Alfred Kr¨onerVerlag, Stuttgart.

Elbl, M. M. (2007): “From Venice to the Tuat: Trans-Saharan Copper Trade and Francesco di Marco Datini of Prato”. In: Armstrong, L, Elbl, M. (eds.). Money, Markets and Trade in Late Medieval Europe, pp. 411 – 459. Essays in Honor of John H. A. Muno. Leiden, Netherlands, Koninklijke Brill, pp. 411 –459.

Eulenburg, F. (1994): Die Frequenz der deutschen Universit¨atenvon ihrer Gr¨undung bis zur Gegenwart. Akademie Verlag, Berlin.

86 Eurostat (2007): Regions in the European Union. Nomenclature of Territorial Units for Statistics - NUTS 2006 /EU-27. Eurostat Methodologies and working papers. Dataset Downloaded from: http://epp.eurostat.ec.europa.eu/cache/ITY_OFFPUB/KS- RA-07-020/EN/KS-RA-07-020-EN.PDF; accessed at October 8th, 2012.

Heyd, W. (1879a): Geschichte des Levantehandels im Mittelalter. Vol. 1. Stuttgart, Verlag der J. G. Cotta’schen Buchhandlung.

Heyd, W. (1879b): Geschichte des Levantehandels im Mittelalter. Vol. 2. Stuttgart, Verlag der J. G. Cotta’schen Buchhandlung.

Humnicki, J, Borawska, D. (1969): “Slowianszczyzna Zachodnia w IX w”. In: Lowmi- anski, H. (ed.). Historia Polski: Tom 1 do roku 1764, czesc III - Chronologia, Bib- liograﬁa, Indexy, Tablice Genealogiczne, Mapy. Panstwowe Wydawnictwo Naukowe.

Hunt, E. S., Murray, J. M. (1999): A History of Business in Medieval Europe 1200 - 1500. Cambridge University Press, Cambridge UK.

Kinder, H., Hilgemann, W. (1970): Atlas zur Weltgeschichte. Von den Anf¨angenbis zur Gegenwart. R. Piper & Co. Verlag, Mnchen.

King, D. (1985): “Currents of Trade. Industries, Merchants and Money”. In: Evans, Joan (ed.). The Flowering Middle Ages. Thames and Hudson, London.

Kobler¨ , G. (1988): Historisches Lexikon der deuschen L¨ander.Die deutschen Ter- ritorien und reichsunmittelbaren Geschlechter vom Mittelalter bis zur Gegenwart. Mnchen, C. H. Beck’sche Verlagsbuchhandlung.

Kuhn¨ , C. (2005): Die Via Imperii als Pilgerstraße. Unterwegs im Zeichen der Muschel. Rundbrief der Fr¨ankischenSt. Jakobus-Gesellschaft W¨urzburg 52, pp. 13 –14.

Lendl, E., Wagner, W. (eds.) (1963): F.W. Putzger historischer Weltaltas zur allgemeinen und ¨osterreichischen Geschichte. Vienna, H¨older-Pichler-Tempsky / Osterreichischer¨ Bundesverlag.

Lopez´ , R. S., Raymond, I. W. (1951): Medieval Trade in the Mediterranean World. Illustrative Documents. New York, NY, Columbia University Press.

Magocsi, P. R. (2002): Historical Atlas of Central Europe. University of Washington Press, Seattle, Washington.

87 Oestreich, G., Holzer, E. (1970a): “Deutsche Kirchenprovinzen und Bist¨umer”.In: Grundmann, H. (ed.). Gebhardt - Handbuch der Deutschen Geschichte. Vol. 2: Von der Reformation bis zum Ende des Absolutismus. Union Verlag, Stuttgart.

Oestreich, G., Holzer, E. (1970b): “Ubersicht¨ ¨uber die Reichsst¨ande”.In: Grund- mann, H. (ed.). Gebhardt - Handbuch der Deutschen Geschichte. Vol. 2: Von der Reformation bis zum Ende des Absolutismus. Union Verlag, Stuttgart.

Postan, M., Rich, E.E. (eds.) (1952): The Cambridge Economic History of Europe. Vol. 2: Trade and Industry in the Middle Ages. Cambrigde, UK, Cambridge Univer- sity Press.

Roos, M.W.M. (2005): How Important is Geography for Agglomeration? Journal of Economic Geography 5(5), pp. 605–620.

Ruegg¨ , W. (ed.) (1993): Geschichte der Universit¨atin Europa. Vol. 1: Middle Ages. Munich, C.H. Beck’sche Verlagsbuchhandlung (Oscar Beck).

Rutkowski, H. (1980): “Polska za Kazimierza Wielkiego (1370)”. In: Tazbir, Janusz (ed.). Zarys Historii Polski. Warswa, Panstwowy Instytut Wydawniczy.

Schulte, A. (1966): Geschichte des mittelalterlichen Handels und Verkehrs zwischen Westdeutschland und Italien mit Ausschluss von Venedig. Ed. by Badische His- torische Kommission. Vol. 2: Urkunden. Duncker & Humblot, Berlin.

Spufford, P. (2002): Power and Proﬁt. The Merchant in Medieval Europe. London, UK, Thames and Hudson.

Wojtowicz, P. (1956): “Zycie Gospodarczo-Spoleczne Dawnej Polski”. In: Paszkiewicz, H (ed.). Polska i jej Dorobek Dziejowy. London, Ksiegarnia Pol- ska Orbis Polonia.

Zeev, M., Henderson, E. A. (2013): The World Religion Dataset, 1945 - 2010: Logic, Estimates and Trends. International Interactions 39.