Geography and Space in Recent Economic History

1 Introduction: Geography as a Bridge

Geography can be considered as a bridge between natural and social sciences, notably human-environment-interaction ^[1] geography, which describes and analyses spatial interactions between humans and the natural world. The discipline of economic history in turn has always felt the tension between natural and social sciences. Here the quest for scientific rigor in economics seems opposed to a tendency to broaden the scope of questions and methods in historical study. While this tension has been a marker of economic history since the beginnings of this discipline – see the various debates over methodology or Methodenstreite of the 1890s – it became palpable again in the 1990s, when economics experienced the “credibility revolution”, ^[2] while history once again turned the other way towards cultural studies and anthropology. ^[3]

This background might help to understand why Paul Krugman’s “New Economic Geography” (NEG) attracted so much attention in economics and, with a time lag, in economic history. His paper Increasing Returns and Economic Geography was published in 1991 in the Journal of Political Economy, the leading journal of the Chicago School. ^[4] It opened the door to new perspectives very much beyond the standard neoclassical framework in economics. Krugman had always been puzzled by the absence of space or geography in the field of international trade. In the tradition of David Ricardo and Eli Heckscher-Ohlin (extended by Paul Samuelson and others) trade occurred between abstract points, with abstract geographical characteristics that were assumed but not modelled, such as climate or resource abundance. More generally, in neoclassical economics, all agents (consumers, firms) were assumed to be atomistic, hence so small in size that their individual contribution could be ignored.

In his 1991 paper, Paul Krugman showed a way to think about real geographical objects, such as cities, regions and countries, by combining two older ideas to something genuinely new. The first idea was the concept of “monopolistic competition” of Avinash Dixit and Joseph Stiglitz that was used to model firms with a specific size that together could form a geographical cluster of a specific size. ^[5] Second, Krugman added transportation cost, using Paul Samuelson’s “iceberg” formulation, to avoid modeling a specific sector that provides transport services. ^[6] The interaction of the two concepts allows modelling long-run differences between geographical units as the outcome of purely economic forces, shaped by a small set of parameters. Crucially, for some range of parameters, this type of model features multiple equilibria, hence areas in the parameter-space where the long-run equilibrium of economic landscapes such as cities or regions is the endogenous result of small, possibly random perturbations, which get larger over time.

In Paul Krugman’s paper there are three economic forces at work. A home market effect and a price effect that work towards concentration, and a competition effect working against it. The home market effect means that an increase in demand leads to a more than proportionate increase in the production of a certain good. Paul Krugman and Anthony Venables modified and generalized this idea with reference to “backward linkages” stemming from demand and “forward linkages” that derive from supply, with similar effects. ^[7] Put differently, the NEG model implies mechanisms of cumulative causations, where some initial advantage of one place gets magnified over time as a result of interactions with other places.

Why are Paul Krugman’s ideas so interesting for economic history? First, they provide economists with tools to think about geography without leaving the academic mainstream. The modelling tools that Krugman used were already well established in the areas of industrial organization and international trade. This widening of the scope of the field was beneficial for economic history, and generally for the exchange of ideas between economics and history, because it could draw on long traditions in regional and urban economics and regional and urban history, which had been sidelined for some time.

Especially in Germany there was a tradition in spatial economics, beginning with Johann Heinrich von Thünen and continuing with Alfred Weber, Walter Christaller to Andreas Predöhl and Martin Beckmann. Interest declined after 1945 but waited to be rediscovered. German economic historians had been writing extensively about the rise of cities, settlements and regional developments, including Rudolf Kötzschke, Eberhard Gothein, or more recently Hubert Kiesewetter, Rainer Fremdling, Toni Pierenkemper, or Richard Tilly. ^[8] Paul Krugman’s work could connect to these older strands of research.

A second aspect why Krugman’s ideas matter for economic history is more subtle but possibly more important. A feature of all NEG models is the potential for non-ergodicity, which implies path dependency. This means that – in contrast to models in neoclassical economics – long-run equilibrium outcomes need not be pinned down by fundamentals such as endowments, preferences and technology, but can depend on accidents. While this sounds highly technical, it opens economic theory for the concept of human agency, which is central to historical sciences. Take for example the location of firms. NEG provides a theoretical concept for why the decision of a few or even one single firm for or against a particular location can lead to a cumulative self-reinforcing dynamic of industrial agglomeration. The clearest exposition of this idea stems from Paul Krugman himself, in a short paper published just a month before his aforementioned ground-breaking NEG contribution. ^[9] In this paper entitled History and Industry Location: The Case of the Manufacturing Belt, Krugman argues that the rise and persistence of the US manufacturing belt is the result of accidents combined with self-reinforcing agglomeration. According to Krugman, “If there is one single area of economics in which path dependence is unmistakable, it is in economic geography – the location of production in space”. ^[10]

A third reason for the key importance of Paul Krugman’s ideas to economic history is that patterns of economic geography are malleable, but they tend to change slowly. This implies that a long-run perspective, spanning over decades, better centuries, is required to observe substantial changes in economic geography. Hence, economic geographers will always have a tendency to refer to historical sources and expertise.

Nevertheless, it took over ten years before economic historians started to use NEG frameworks for historical research. The reason for this delay stems from the considerable challenges that NEG poses to empirical work. While self-reinforcing agglomerations are rather straightforward in theory (that is, after Krugman had demonstrated how), it is much harder to put them to an empirical test. This is because a key aspect of NEG theory is cumulative causation, or endogeneity: some locations develop better than others because they have some initial advantage. NEG theory states that the economic size of a location will increase with its market potential, which is reflected in its own economic size and the size of all other locations weighted by the cost to access them. Empirically, this type of circular argument is hard to test, at least if researchers aim to make causal claims, especially given that causality moved center stage in empirical work in economics in the 1990s only. A related challenge is to find empirical data that allows tracking economic outcomes and their possible determinants, such as fundamentals or market potential, at various locations over a sufficiently long period of time. In much of the literature, fundamentals such as natural resource endowments are referred to as first nature geography, while the NEG concept of market potential, which encompasses local economic size and factors that shape access to other markets, are called second nature geography.

Apart from Paul Krugman’s NEG, there were other developments in the field of economic geography during the 1990s that made this field attractive for economic historians. In particular, it was the discussion around the gravity model, its theoretical foundations and empirical applications, that inspired new work in economic history. The original gravity model had been introduced by Jan Tinbergen in 1962, who showed that trade flows between countries are roughly proportional to their economic size and inversely proportional to the distance between them, once all variables are expressed in logarithmic form. ^[11] The appeal of this approach lies in its simplicity, by providing a benchmark for trade flows that allows measuring factors that either foster or limit economic integration compared to this benchmark.

Among those applying the gravity model, the paper by John McCallum stands out. He estimated the effect of the political border between US states and Canadian provinces and found that trade between provinces within Canada was 22 (!) times larger than trade across the border. ^[12] This striking result sparked a wave of empirical studies, such as Volker Nitsch, who studied the role of national borders on European trade. ^[13] James Anderson and Eric van Wincoop derived gravity from a trade model based on monopolistic competition and trade costs (which were, after all the building blocks of Paul Krugman’s NEG), to explain some part of these border effects. ^[14] However, Jonathan Eaton and Samuel Kortum derived a very similar gravity structure from a generalized Ricardian trade model with trade costs, and heterogenous firms under perfect competition. ^[15] These theory-consistent structural gravity models imply a modification of the original Tinbergen-model to take time-varying but country-specific factors into account. A survey of the literature and a theoretical synthesis is provided by Keith Head and Thierry Mayer. ^[16]

Over the next pages, I will discuss the empirical literature that followed from both NEG and the gravity model, with a focus on applications in economic history after 1991 and here especially in Europe and Germany. While this does not provide the full picture, it discusses some of the first and most influential contributions in the field. I will organize this discussion around four themes: the starting point is the measurement of market potential and its components over the long-run. This includes attempts to estimate regional GDP as well as changes in the accessibility of markets due to changing transportation costs, changes in tariffs or political borders drawing on the gravity model. Next, I will discuss various attempts to measure the effect of market potential (or its components) on economic outcomes, ie. empirical tests of NEG models. Third, there are several papers that try to dig deeper and identify the underlying microeconomic mechanism behind market potential, notably localized externalities, spillover effects, and the interplay between changes in transport infrastructure and structural change. And finally, I will consider a strand of the literature that has examined the non-ergodicity implied by NEG theory, namely path dependence and multiple equilibria, which speaks to the fundamental question of agency in economic models.

2 Measuring Market Potential and Regional Inequality in the Long-run

A first attempt to examine long-run patterns of the geographic location of economic activity inspired by NEG was made by Sukkoo Kim for US regions, covering the period from 1860 to 1987. Using employment data, Kim developed several indices of concentration and localization and discussed them in the light of theories on economic fundamentals (ie. Heckscher-Ohlin) as well as NEG with rather mixed results. ^[17] This was not taken up by European researchers, maybe because the empirical challenges (and the cost of expected data collection compared to academic return) were considered too high.

It was really the work of Nick Crafts during his time at the London School of Economics that started a wave of systematic research in economic history using NEG frameworks. Together with his then LSE-colleague Anthony Venables he co-authored a chapter in the conference volume Globalization in Historical Perspective. ^[18] According to the new geographic perspective on globalization “[…] locations derive some of their comparative advantage from scale, and ability to exploit scale is in turn limited by the extent of the market. In this approach firms seeking profitable locations will be drawn to locations with good market access and proximity to clusters of related activities, as well as locations with appropriate factor endowments.” ^[19] Crafts and Venables argue that a key role is played by the market potential (or market access) of a place, as well as positive feedbacks from economies of scale. More specifically, they suggest that the decline in transport costs between and within countries since the 18th century interacted with local characteristics and market structure to generate an increasingly heterogenous landscape of development. In consequence, the first globalization coincided with the emergence of megacities, industrial agglomerations, and rising inequality.

Over the following years, Nick Crafts developed an ambitious research agenda. His focus was initially on Victorian Britain, the cradle of industrialization, before he turned to a detailed analysis of the US manufacturing belt that had been initially suggested by Paul Krugman. A central objective of his research was the historical reconstruction of market potential, as the weighted sum of the size of a region and the sizes of all its neighbors, weighted by the cost of accessing them. Hence, the challenge was to estimate indicators of economic size (typically GDP) at various geographic scales (country, regions, counties) and trade costs between these units, going backwards in time.

Partly funded by an ESRC grant, Crafts collected data to estimate the GDP of regions within the UK, regional market potential, and revisited changes in British transport infrastructure. ^[20] At around the same time, several authors started to examine patterns of industrial location in other European countries, namely Spain and Poland. ^[21] A related strand of work has examined changes in the accessibility of markets (or market integration), either by looking at price dynamics ^[22] or frictions as measured by a gravity model applied to trade flows – or more specifically in the case of Europe, with a focus on the role of changing political borders.

Following James Anderson and Eric van Wincoop’s theory-based approach to re-estimate of the effect of the Canadian-US border on aggregate bilateral exports between US states and Canadian provinces, my study on the economic integration of Polish regions after 1918, was the first to apply the structural gravity model to a historical setting. The results for Interwar Poland suggest that the unification of Poland was economically successful. Similarly, compared to findings for the EU or US intra-national trade, ^[23] interwar Poland was a remarkably well-integrated economic area. The same approach has been applied to estimate the economic integration of Germany before and after 1914/1918 and to study the role of changing political borders in Central Europe following the First World War. ^[24] This latter study uses a variety of sources to construct sub-national trade flows between regions of Germany and its neighbors between 1885 and 1933. In contrast to nearly all empirical studies on the effect of political borders on trade, this study estimates the treatment effect of changing political borders based on a structural gravity model. In contrast to estimates of border effects from cross-sectional variation, the treatment effects are much smaller. This confirms a long-standing suspicion that national borders should not be considered as random shocks, but they follow other, deeper patterns, such as ethno-linguistic and religious heterogeneity between regions. While this might be obvious for historians of World War I, this important finding is often neglected. Another influential contribution in this area is the paper by David Jacks and colleagues, who use a structural gravity framework together with data on bilateral trade flows and GDP to estimate indices of trade costs that vary by country and over time. ^[25] This works well, because it allows the authors to quantify trade costs in a general way, including frictions stemming from geography (such as distance), policy (such as tariffs) or culture (such as ethno-linguistic or religious diversity).

The other ingredient to market potential – historical GDP estimates – are available from the Maddison database, at least for country-level GDP. ^[26] However, data for geographical units below the level of nation-states was not available for years prior to the 1980s, with the exception of the UK. Starting in 2006, Joan Rosés and myself conducted a collaborative project involving over 20 researchers that was funded by the European Science Foundation to estimate GDP data at the level of NUTS-2 regions for every decade back until at least 1900. ^[27] For most countries regional GDP was estimated top-down based on regional employment and wage data. The approach of the project followed Frank Geary and Tom Stark, who attempted to estimate the GDP of Ireland (as part of the UK GDP) before 1918. ^[28] Joan Rosés and I describe the project results for regional GDP estimates. We highlight a remarkable feature regarding the evolution of regional inequality since 1900. ^[29] While older literature had suggested that inequality would follow an inverted U-shape both for personal income inequality ^[30] and income inequality between regions, ^[31] the new data suggests an increase in all dimensions of inequality starting around the 1980s. For personal income inequality this reversal has been documented by Facundo Alvaredo and colleagues in a global approach or Charlotte Bartels for Germany. ^[32] The work on historical regional GDP and regional inequality is currently revising older estimates, going further back in time, and looks at other parts of Europe. ^[33]

Stephen Redding and Anthony Venables made a fundamental contribution to this strand of research by combining work on structural gravity models with that on NEG to estimate a market potential in a two-stage procedure. ^[34] In a first step, they use a structural gravity model following James Anderson and Eric van Wincoop to estimate the cost of market access. In a second step they use the results together with GDP data to calculate market potential. Moreover, they were the first to suggest using the inverse of a region’s distance matrix as an instrument to deal with the empirical problem of endogeneity (reverse causation and omitted variable bias). An early application of this approach to economic history is my study on the location of industry in Interwar Poland, which will be discussed in the following sections.

3 The Effects of Market Potential on Location and Growth

Arguably the first empirical papers in economic history that took up the challenge to test hypotheses from NEG theory in a historical context were those of Daniel Tirado and colleagues, as well as Joan Rosés, both examining Spain. ^[35] Rosés studied Spain’s industrialization over the long 19th century (from 1797 to 1910), drawing on a framework suggested by Donald Davis and David Weinstein to identify the economic effect of one particular element of market access: the home market. ^[36] He first describes long-run patterns of industrial location and then examines to what extent geographical fundamentals – as in neoclassical economic theory or NEG forces in the form of a home market effect – can account for this pattern in a cross-section around 1861. Very much in line with Davis and Weinstein, Rosés finds that a combination of fundamentals and NEG forces matter, with substantial variation across industries.

A downside of the approach by Davis and Weinstein is that it captures only some aspects of NEG theory and cannot be easily applied to a panel with time, region and industry variation, being based on rather strong identifying assumptions. A more flexible approach has been suggested by Karen Midelfart-Knarvik and colleagues. ^[37] Their approach develops a multi-region, multi-industry general equilibrium framework that includes both NEG and neoclassical forces, that can also be empirically estimated. The first attempt to do this was made by Nick Crafts and Abay Mulatu for the case of Victorian Britain. The main finding of this new research was that over the period of 1871 to 1911, both, forces of neoclassical location theory, such as the endowment with natural resources (coal) and skilled labor, as well as forces of new economic geography, are needed to explain the location of British industry. There is some evidence that the role of market potential may have increased over time, but this effect was generally modest for most regions and industries in Victorian Britain. ^[38] Using the same framework, I examined the case of industry location for Poland after 1918. ^[39] I estimated market potential based on the approach by Stephen Redding and Anthony Venables, which also deals with endogeneity. ^[40] The paper concluded that for Poland, changes in market potential explain about one third of the overall variation in industrial location.

Stephen Redding and Daniel Sturm started a series of papers that analyze the division and unification of Germany with a focus on economic geography. A central idea in their work is to exploit the change in Germany’s political borders as a quasi-experimental setting that provides researchers with credibly exogenous variation in market potential across regions. In particular, Redding and Sturm examine the development of West German cities, distinguishing between those close to the new intra-German border and cities further away. They find strong evidence that the loss of market potential causally led to a decline of treated cities relative to others. Moreover, they develop a simple variant of an NEG-model (following Elhanan Helpman) and show that this can account for the quantitative magnitude of their findings while providing additional evidence against alternative possible explanations. ^[41]

Florian Ploeckl refers to the Redding/Sturm framework to analyze the effects of changing economic integration relative to local fundamentals on the distribution dynamics of population in villages and cities across Saxony between 1550 and 1834. ^[42] Several other papers have taken up these ideas and tested them in different contexts, as discussed in a survey article by Julio Martinez-Galarraga and colleagues. ^[43] Overall, the findings of this research suggest that the location of industry – or more generally of economic activity – is determined by a combination of neoclassical fundamentals such as natural resource endowment or the abundance of skilled labor, as well as NEG forces.

Nick Crafts and myself examine why the English cotton textile industry was so concentrated in Lancashire, ultimately with the aim to learn something about the driving forces behind the First Industrial Revolution. ^[44] Earlier results had suggested that external economies of scale mattered for the productivity of the Lancashire cotton industry. ^[45] We use data from the Factory Return, which documented the location and scale of operation of all the textile mills and factories in England, Wales and Scotland as of 1838. ^[46] Explaining the economic geography of one single industry could not rely on the workhorse model by Karen Midelfart-Knarvik and colleagues, ^[47] but required a different approach. Instead, we followed Paulo Guimaraes and colleagues, and estimated the empirical probability to observe a cotton mill in one place based on various characteristics of this place, using a simple Poisson estimator. Drawing on a large literature on the factors that might explain the location of the British cotton industry, we collected data ranging from humidity, access to coal and waterpower, to market potential. The major finding is that several factors, including neoclassical location factors such as the availability of waterpower and coal interacted with forces of NEG. Moreover, and in contrast to Nick Crafts earlier work, ^[48] we find that for this particular industry, and as early as 1838, access to markets was decisive. Running various counterfactuals, we find that a 10 percent increase in market potential of a place had a larger effect than a 10 percent increase in any of the other variables (such as access to coal or waterpower). Paul Krugman positively commented our work by writing: “The two Nicks, Crafts and Wolf, have a piece right up my alley: they argue that the cutting edge of Britain’s Industrial Revolution, the cotton textile industry, benefited hugely from agglomeration.“ ^[49]

A directly related contribution on German economic history was made by Theresa Gutberlet who focused on the transition from water to steam power, to explain concentration in industrial employment in the late 19th century. Her findings strongly suggest that the transition to steam power contributed to industrial concentration. In an accompanying paper, she tried to distinguish between the direct effect of closeness to coal and indirect agglomeration effects in the spirit of NEG, which might have magnified her findings, taking spatial spillover and endogeneity effects into account. She concluded that both effects were present, but for most industries the direct effect of access to coal mattered more. ^[50]

In another series of papers, Alex Klein and Nick Crafts took up Paul Krugman’s early suggestion to examine the dynamics of the US manufacturing belt from a historical perspective. Based on the work of Karen Midelfart-Knarvik and colleagues, ^[51] but using more sophisticated estimation techniques than in their earlier papers, Klein and Crafts established that NEG was essential to explain the location of US manufacturing. The remarkable concentration of industrial activity in the manufacturing belt can be approximately described by a parallelogram with corners at Green Bay, St. Louis, Baltimore and Portland (Maine), that produced about 80 percent of US manufacturing output in 1900. ^[52] Klein and Crafts show that the persistent dominance of the US manufacturing belt was in large part due to the forces of market potential. Moreover, they suggest that linkage effects, especially forward linkages between industries and their use of intermediate inputs, are key to understand the logic of this massive industrial agglomeration. In a next step, Klein and Crafts analysed how industrial agglomeration mattered for US productivity. Several papers had established that agglomeration, or density of economic activity enhances productivity, likely due to a combination of size effects, knowledge spillovers, access to qualified labor and other inputs, as well as final demand. ^[53] Klein and Crafts suggest distinguishing between Marshallian and Jacobs externalities, where the first are meant to capture external economies from specialization, while the latter are meant to describe external economies form diversity. They find that specialization had indeed a strong positive effect on labor productivity, while the effect of diversity is less robust. ^[54] Finally, Crafts and Klein traced the spatial concentration of manufacturing in the US over the entire 20th century, from 1880 to 2007, and conclude that concentration followed a secular decline. ^[55] This finding stands in contrast to earlier results by Sukkoo Kim, and it is also different from the pattern observed in Europe. ^[56]

Another strand of research has questioned how market access affected economic dynamics and growth. Notable contributions here are by David Jacks and Dennis Novy, Michael Peters, and Caruana Galizia and colleagues. In his paper on market size and spatial growth, Michael Peters studied how the expulsion of eight million ethnic Germans after 1945 from Eastern Europe to West Germany affected market potential and how this in turn contributed to economic growth. ^[57] Using variation across West German counties, he showed that the population influx had large and persistent effects on the size of the local population, manufacturing employment, and income per capita. His quantitative model implies that refugee settlement increased aggregate income per capita by about 12 percent after 25 years and triggered a process of industrialization in rural areas. David Jacks and Dennis Novy studied the relationship between market potential that can be derived from a structural gravity model and economic growth for a sample of 51 countries over the entire 20th century. Using an approach similar to Stephen Redding and Anthony Venables, ^[58] they provide evidence that market potential had a significant causal role in driving global income growth over this period. Finally, Paul Caruana Galizia and colleagues summarize the literature on the role of “first and second nature geography”, for economic growth at a global scale since 1870. ^[59] They first discuss the effects of “first nature”, such as the prevalence of malaria or resource abundance on growth, next show evidence from case studies on Europe and Japan on the role of market potential (hence “second nature”), before summarizing how both types of geographical forces interacted to shape economic growth since the late 19th century.

4 Digging Deeper: On the Microeconomics Behind Agglomeration Economies

The early types of NEG models were highly stylized and remained rather unspecific about the scale of geography and the microeconomic logic of underlying mechanisms. The seminal contribution by Paul Krugman for example, could be applied to cities or regions, and relied on price and demand-side effects to generate agglomeration forces that were too stylized to be directly tested.

The search for specific microeconomic mechanisms was led by researchers of urban economics, which could draw on a wealth of previous theoretical work and many detailed case studies. At the heart of this research was the question why many types of economic activity are so concentrated in space, and even within cities. To quote Gilles Duranton and Diego Puga in their handbook chapter on the Micro-Foundations of Urban Agglomeration Economies: “One cannot make sense of this sort of numbers, of the extent to which people cluster together in cities and towns, without considering some form of agglomeration economies or localised aggregate increasing returns.” ^[60] Duranton and Puga suggest to distinguish between three types of micro-foundations for agglomeration economies, based on sharing, matching, and learning mechanisms. In the same Handbook series Stuart Rosenthal and William Strange discuss the available empirical evidence and conclude with a call for more data to deal with the difficult problems of identification. ^[61]

A paper that is relevant in the context of Germany and beyond is that by Gabriel Ahlfeldt and colleagues, which makes three important contributions to the literature. ^[62] First, the authors develop a quantitative model of internal city structure that features agglomeration and dispersion forces derived from first principles and an arbitrary number of heterogeneous city blocks. The model remains tractable and amenable to empirical analysis because of stochastic shocks to commuting decisions, which yield a gravity equation for commuting flows. The paper thus combines the stylized urban model by Robert Lucas and Esteban Rossi-Hansberg with the probabilistic approach to Ricardian trade theory as introduced by Jonathan Eaton and Samuel Kortum. ^[63] Second, to structurally estimate agglomeration and dispersion forces, Ahlfeldt and colleagues use data on thousands of city blocks in Berlin for the years 1936, 1986 and 2006, as well as exogenous variation from the city's division and reunification. Hence the paper also contributes to understanding the specific historical context of the division and unification of Berlin. Third, the authors estimate specific forces that can account for agglomeration, namely substantial and highly localized production and residential externalities. Their model with estimated agglomeration parameters can account both qualitatively and quantitatively for the observed changes in city structure. The framework used by Alhfeldt and colleagues has become the canonical structural model in urban economics, because it can be applied to any city and can be used to undertake counterfactuals for changes in the organization of economic activity within cities in response, for example, to changes in the transport network.

On a more macroeconomic level, Dave Donaldson and Richard Hornbeck developed a framework to examine how changes in transport infrastructure translate into changes in regional market access, which in turn affects land prices as a summary measure of productivity. ^[64] Notably, the authors did not base their work on Paul Krugman’s model, but on Jonathan Eaton and Samuel Kortum. ^[65] They focused on quantifying the aggregate impact on the agricultural sector in 1890 by constructing a network database of railroads and waterways to calculate lowest-cost county-to-county freight routes. This allows them to estimate how increases in county market access – due to the expansion of the railroad network from 1870 to 1890 – affected the value of land. In contrast to the well-known work by Robert Fogel, ^[66] they show that railroads mattered a lot: removing the effects of all railroads in 1890 is estimated to decrease the total value of US agricultural land by 60 percent, with limited potential for mitigating these losses through extensions to the water canal network or improvements to country roads. Richard Hornbeck and Martin Rotemberg extended this research by considering the effects on manufacturing and aggregate productivity. ^[67] They find very large effects of transport-induced changes in market access on national level aggregate productivity. However, Hornbeck and Rotemberg avoided any direct reference to NEG models, but instead argue that improved infrastructure can have large effects in the presence of micro-level “distortions”, which can derive from firm-level markups (as they would arise in any NEG model with monopolistic competition), credit-constraints or some combination thereof.

There is currently no comparable study that would have estimated the frameworks provided by Dave Donaldson and Richard Hornbeck or Hornbeck and Martin Rotemberg for Europe or Germany. ^[68] However, several authors have analyzed very similar questions. An early but still important contribution was made by Rainer Fremdling, who estimated how the expansion of railroads mattered for German industrial growth during the early period of industrialization between 1840 and 1860. ^[69] Fremdling provides strong evidence for the presence of backward linkages from railroads to input suppliers (iron production and engineering), which can be interpreted as evidence for NEG mechanisms in the framework of Paul Krugman and Anthony Venables. A more recent contribution in this tradition came from Erik Hornung, who shows that the construction of railroads had a causal effect on urban population growth in Prussia after 1840, but also contributed positively to manufacturing employment and average firm size. ^[70] Michael Kopsidis and myself in turn discuss the interplay between the growth of urban markets and agricultural development in Prussia, following a Heinrich von Thünen framework of land use, extending a formulation originally proposed by Martin Beckmann. ^[71] We use data on one large cross-section of agricultural development across Prussian counties as of 1861, which allowed us to distinguish between types of land use and to directly observe agricultural land value (Grundsteuerreinertrag). Using mining as an instrument, we show that access to urban markets had a large and causal effect on agricultural land values, by shifting agricultural production towards higher value-added products, such as meat and dairy products. Hence, in contrast to conventional accounts that consider a rise in agricultural productivity as a conditio sine qua non for industrialization, ^[72] we argue that in the case of Prussia causality (also) could go the other way.

Kopsidis and Wolf are quite close to Dave Donaldson and Richard Hornbeck with their focus on the role of market access for agricultural development, only that we consider urban growth due to mining as a driver to local market size instead of changes in transport infrastructure. This has recently found justification in the work of Alan Fernihough and Kevin O’Rourke, who document how the local availability of coal was indeed a major causal factor behind the growth of European cities after 1750. ^[73] It would certainly be worthwhile to combine these various ideas and research approaches to understand better how coal mining and railroad construction affected the growth of agriculture and industry in Europe.

5 Non-ergodicity: Path Dependence and Multiple Equilibria

Let us return to some of the implications of NEG, which matter at a most fundamental level for the study of economic history: path dependence and multiple equilibria. The positive feedback mechanisms that stand at the center of NEG, imply that long-run outcomes need not be determined by the typical ingredients of neoclassical models, namely endowments, technology and preferences. Instead, there is scope for historical accidents, which can also be interpreted as agency or strategic decision space. A few players, such as large companies or political decision-makers can have major and lasting effects. Put differently, there is path dependence and/or coordination failure with a scope for multiple equilibria. As mentioned earlier, Paul Krugman discussed this in his paper on the rise of the US manufacturing belt, which was later supported by the work of Nick Crafts and Alex Klein. ^[74] However, all empirical work on path dependence suffers from potential omitted variable bias, insofar as it is hard to exclude the possibility that observed outcomes are not due to some other (short-term or long-run) determinants that are not part of the data set. What is more, if the observed outcome is the result of positive feedback effects, only some very large shock could possibly change such an equilibrium. However, such a shock would also have to be temporary, because otherwise it would effectively be a change of locational fundamentals.

An important contribution to this area of NEG comes from Donald Davis and David Weinstein, who suggested exploiting a dramatic yet temporary shock to economic location as a quasi-experiment: the Allied bombing of Japanese cities in World War II. ^[75] They ran variants of a difference-in-difference analysis, comparing the growth of cities before and after bombing, depending on the intensity of the shock as measured by the number of dead and missing residents. Surprisingly, Davis and Weinstein find that city populations recovered very quickly from the wartime shock, and cities returned to their pre-war growth path within less than twenty years. So, if even wartime devastation of cities, as observed in Japan, could not alter the economy between multiple spatial configurations of economic activity, this appears to suggest an overwhelming role for fundamentals in determining the location of economic activity. ^[76] Steven Brakman and colleagues replicated Davis and Weinstein’s idea in the context of Germany during World War II, and basically find the same results: bombing had a significant but only temporary impact on post-war city growth in Germany. Edward Miguel and Gérald Roland have done similar research on Vietnam in the 1960s and 1970s with similar results. ^[77]

In contrast, Stephen Redding and colleagues argue that shocks have an even more limited effect. If location decisions are forward looking, one must consider transport and commercial networks, but also property rights, which survive shocks. ^[78] The authors suggest to examine the effects of German division and unification as a large-scale shock that also erased property rights. They focus on the location of airports as a sector that features very strong positive feedback mechanisms. Their paper is entitled History and Industry Location, referring back to Paul Krugman’s original proposition about the US manufacturing belt. They argue that the aviation industry relies on hub-and-spoke networks with regional hubs that require large sunk costs. These make the location of the hub likely to be prone to multiple location equilibria, in the sense that once the sunk costs of creating the hub have been incurred, there is no incentive to relocate. Second, the existence of multiple location equilibria may be reinforced by network externalities, which imply that the profitability of operating a connection to an airport is likely to increase with the number of other connections to that airport. Redding and colleagues developed a simple theoretical framework and test its implications using data from all German airports between 1926 and 2002. They show that the relocation of Germany’s aviation hub from Berlin to Frankfurt was in fact due to the temporary shock of division, while Frankfurt’s rise to Germany’s postwar hub cannot be predicted from fundamentals or pre-war trends. Moreover, they show why unification did not undo the shock of division, given that the predicted changes in the net present values profits across alternative potential locations for Germany’s air hub are small relative to the sunk costs of creating the hub.

Another perspective when looking at locational path dependence can be coined geo-economics. Here, a limited set of political agents is basically assumed to have strategic decision space. This field has recently attracted a lot of attention, due to an increase in political and military conflicts over territories and resources. One example of geo-economics during peacetime is the study by Thilo Huning and myself on the formation of the German tariff union (Zollverein) in 1834. ^[79] Extending earlier work by Wolfgang Keller and Carol Shiue, as well as Florian Ploeckl, ^[80] we show how the exogenous change in borders after the Congress of Vienna in 1814/1815 enabled Prussia to control the profitable trade routes between southern and central Germany and the coastline. Since the cost of trade depended on trade routes passing through other states, this gave rise to a “coordination game” between Prussia and other German states of tariff policy and custom unions, with a large number of possible equilibria. Prussia used its beneficial strategic position to ultimately force all southern and central German states into the first customs union of the world, the Zollverein. The paper shows that the change in borders was a necessary (but not sufficient) condition. Under counterfactual borders the Zollverein would not have been formed.

6 Conclusion and Outlook

I opened this survey by saying that geography can be considered as a bridge between disciplines. Over the preceding pages, I provided many examples for the case of economics and history as being brought together by geographic considerations. They all make for strong evidence that geography matters for location, structural change and growth, not only via first nature features but clearly also via NEG-type second nature geography. Changes in market potential are essential to understanding economic development in the long-run, which is what economic historians are trying to do.

Arguably, this survey is selective and suffers from a focus on empirical studies that I know well (which include my own). There are many other relevant papers, and I would like to again refer the interested scholar to another recent survey by Julio Martinez-Galarraga and colleagues for further references. ^[81]

What becomes obvious when looking at both of these research surveys, is that while much has been achieved in the combination of NEG and economic history, there are still some large gaps in this literature that invite more work. At the European level, the data is still too patchy to examine how local market potential has changed over time, and to what extent this mattered for economic outcomes. Moreover, in the European context it would be essential to take changes in political borders and their effect on the cost of trade and factor mobility into account. Even more can be done on city structures and structural change within states. Recent work on address books will help to analyze cities at a more granular level, at least for the 20th and parts of the 19th century. ^[82] Improved data availability and identification methods allow following the lead of Nick Crafts’ work on Victorian Britain to study the developments in many other countries and further back in time.

But such research does not need to be quantitative. In the preceding pages, I have tried to argue that some of the theoretical ideas of Paul Krugman and his followers could also guide more qualitative studies on firm location, or the potential implications of large-scale policy interventions. At the most general level I think that a broader geographical perspective – inspired by NEG or any other type of human-environment-interaction geography – could open new paths for economic historians.