Editorial

The first three papers in this issue were initially presented at conference workshops organized by members of the editorial team, held in Glasgow and Bilbao. The first paper by Alvaro Angeriz, John McCombie & Mark Roberts (2009), entitled ‘Increasing Returns and the Growth of Industries in the EU Regions: Paradoxes and Conundrums’, significantly extends and summarizes the literature relating to Verdoorn's law,1 which has a long tradition as an alternative non-orthodox paradigm within the post-Keynesian School, standing opposed to the dominant neoclassical model as a means of explaining growth variations between countries and regions. At its simplest, neoclassical (Solow) growth theory assumes constant returns to scale and ‘catching-up’ by faster growing initially poorer regions, typically in the form of conditional convergence to different equilibrium development paths. In contrast, estimates of Verdoorn's law, suitably specified in dynamic form, consistently produce evidence pointing to the presence of increasing returns, meaning in essence that doubling the inputs used for industrial production within a region has the effect of more than doubling output. There are fundamental differences between the two paradigms in the role attributed to the constraints imposed by supply and demand conditions. In the neoclassical tradition, the supply of factors of production is what determines output growth, but in the non-orthodox approach demand is the causal influence determining growth and the supply of factors of production, inducing regional exports and hence output and productivity growth, and this in turn strengthens regional competitiveness and hence export growth in a process of circular causation. Verdoorn's law is embedded within this chain of causation, being the link between manufacturing output growth and productivity growth. This emphasis on increasing returns and the divergent tendency of regional economies has provided an important precursor to the emphasis given to circular and cumulative causation in new economic geography (NEG). There are several important connections made in the paper between Verdoorn's law and previously separate concepts and these are skilfully drawn together as a cohesive whole by the authors of the paper, although a clear distinction remains between the two schools of thought. Perhaps the development of the two traditions, drawn closer together by the advent of NEG and by the extension of neoclassical growth theory far beyond the basic Solow growth model (Acemoglu, 2008), is leading towards the same ultimate outcome. On one level at least this seems to be true, for, despite theoretical differences, in terms of empirical reduced forms it is apparent that the basic Verdoorn equation can be obtained from a model with micro-level assumptions deriving from the same Dixit–Stiglitz monopolistic competition market structure assumptions that typify the NEG (Ciccone & Hall, 1996; Fingleton, 2001).

Dermot Leahy & Catia Montagna (2009) explain why firms might choose to obtain their foreign inputs by outsourcing rather than engaging in foreign direct investment (FDI), using a theoretical model with a two-firm, two-region (North vs South) set-up. There are no data in the paper, the evidence being the force of logic in a ‘laboratory’ situation from which all the complexities of the real world have been eliminated. They show that identical firms can choose different ways of sourcing their inputs, either by FDI or by outsourcing, as equilibrium outcomes of a strategic game. When firms have ex ante asymmetries, meaning, for example, that they have different fixed costs and production costs at the outset, lower cost firms opt for FDI (vertical integration) while higher cost firms will prefer to outsource.

One of the intriguing elements of their analysis is the assumption of an oligopolistic market structure entailing strategic interaction between the firms. In contrast, the market structure in the standard NEG model, alluded to above, is monopolistic competition. Monopolistic competition is a necessary element of the basic NEG model, but theory is evolving to accommodate alternative assumptions.

The advantage of monopolistic competition is its simplicity, and this has led to widespread adoption as a workhorse market structure model, following the influential paper by Dixit & Stiglitz published in 1977 in the American Economic Review entitled ‘Monopolistic Competition, and Optimum Product Diversity’. This revolutionized model building in several fields of economics, trade theory, industrial organization, growth theory, geographical economics, and urban economics. It provided an elegant and simple way to model production at the firm level benefiting from internal economies of scale operating in a monopolistically competitive market.

Under monopolistic competition, there are very many sellers producing products with each producing a different variety, and these varieties are to some degree, depending on the precise assumptions of the model, substitutes. Moreover, there is no role for double guessing what rivals are up to—in other words, there is no strategic interaction between competing firms. With a large number of small firms, any action taken by an individual firm is of negligible consequence for other firms in the sector. Like perfect competition, there is free entry and exit. However, a monopolistically competitive firm has a downward sloping, not horizontal, demand curve. It cannot sell as much as it wants at the going price.

In contrast, under oligopoly there are only a few producers, each recognizing that its own price and market share depends on its own output and on the actions of competitors, so there is an element of strategic interaction between firms, for instance firms might attempt to erect barriers to entry to preserve their market share.

Peter Neary (2001) is critical of the monopolistic competition assumption underpinning the NEG, finding it unrealistic to assume that firms are myopic and lack strategies to stop competitors entering the market and eroding their profits. While oligopoly seems a reasonable alternative to consider, the development of NEG-type models embodying alternative market structures is in its infancy, but no doubt theory will emerge to accommodate alternative assumptions. Already we have seen some development of NEG models analysing the impact of Cournot competition and strategic interaction, but these are highly theoretical and it remains to be seen whether and how these might be confronted with data.

In their paper ‘On the Specification of the Gravity Model of Trade: Zeros, Excess Zeros and Zero-inflated Estimation’, Martijn Burger, Frank Van Oort & Gert-Jan Linders (2009) compare how economists have used the ‘gravity model’ to analyse trade with the work of geographers and regional scientists, arguing that there is much scope for improvement in the way this is done in the international trade literature. This call for better communication between geographers, regional scientists and economists is not the first. Alan Wilson (1970), in his classic Entropy in Urban and Regional Modelling, considered the approach of economists to modelling trade flows, or commodity flows as he calls them, and pointed out some deficiencies in the way economists were applying gravity models (as recognized at that point in time). He emphasized the need for a merging of the geographic and economic approaches to trade analysis, and in particular showed how the gravity model and input–output approaches could be integrated using entropy-maximizing methods. Burger, Van Oort & Linders focus, in effect, on how the gravity model can be treated as a member of the family of generalized linear models (GLMs; Nelder & Wedderburn, 1972) with specific distributional assumptions and specific link functions. Typically, the basic linear regression model can be presented as a GLM with a normal response and identity link function, but such a model would be most inappropriate for trade flows, unlike other members of the family of GLMs. The most obvious GLM is one assuming a Poisson response and loglinear link, or better still a generalized Poisson response (such as the negative binomial) and loglinear link in order to handle over-dispersion. Each variant has particular properties and assumptions, for instance a Poisson response with logarithmic link function cannot predict negative trade flows. However the data may possess too many zeros compared with what is predicted by the Poisson or generalized Poisson assumptions, which may ‘masquerade’ as over-dispersion, but in reality that should be explained as something other than the outcome of a Poisson-like process, for instance trade between a pair of countries may have zero probability, which needs to be explained, or it may have a non-zero probability but just happen to be zero, as is typical of a Poisson distribution. Or, to put it more bluntly, ‘there is a difference between scientists who do not write any papers and hence do not receive citations, and scientists who do write papers but are still not cited’. To summarize, Burger, Van Oort & Linders give a state-of-the-art review of the contemporary literature, drawing on the contributions from both quantitative geography and economics, in particular highlighting the failings of the log-normal specifications that still holds sway in the economics trade literature, pointing out that ‘despite repeated warnings from the related fields of quantitative geography and regional science … international economics has only recently begun to take this issue seriously’.

In their paper on ‘Spatial aspects of contagion among emerging economies’, George Hondroyiannis, Harry Kelejian & George Tavlas show how spatial econometrics can provide insights into real economic crises involving currencies, such as the attack on the Mexican peso in December 1994. The paper develops a panel data model in which exchange market pressures have direct effects on other currencies, and which also takes account of interactions between economic fundamentals (real GDP growth, exports, net capital flows, etc.) and exchange rates. The analysis focuses on 25 emerging market economies and six currency crises, ranging from the aforementioned Mexican peso crisis to the Argentinean crisis of 2002. The paper looks at two main sources of contagion, via trade and as a result of spatial proximity, and shows that spatial proximity per se is not a significant factor, whereas trade between countries does evidently determine how events within one currency area are transmitted to other currency areas. Of course, this is not a problem, as spatial econometrics does not require space to be geographic space, but can be a space in which distances are measured according to diverse metrics. Hondroyiannis, Kelejian and Tavlas carefully distinguish between direct or immediate effects and final effects that take account of all spillovers throughout the system of equations and which are the solution of the equation system. An estimated direct effect, for instance, would give the impact that an attack on Turkey's foreign exchange market would have on the foreign exchange market in, say, Brazil. Likewise, the model allows complex interaction effects involving fundamentals to be estimated, such as the impact of real GDP growth in Mexico on the crisis index in, say, Russia.

One very useful aspect of a spatial econometric approach to financial modelling in general is the way in which global financial electronic interaction is in effect instantaneous and multilateral, so that although an emission in the form of a piece of information may have an initial source, it will in effect rebound around numerous locations with such rapidity that it only makes sense to treat the net outcome as an instantaneous mutual interaction process involving essentially all locations. The way we model this, typically as an autoregressive ‘spatial lag’ specification where the value of the endogenous variable both depends on and determines the value of the endogenous variable elsewhere, can be considered to be the equilibrium outcome of these complex interactions (Anselin et al., 2008). When financial transactions are travelling at the speed of light, as they are when shocks to currencies reverberate across the globe, then in effect they are, we believe, instantaneous.

Large-scale modelling went out of fashion long ago. Alan Wilson (2000) laments this loss, observing that ‘the hopes of the modelling communities in the 1960s have not been fulfilled … [and] from about the mid-1970s, interest in public planning declined, a process well-documented for the modelling context by Batty (1989, 1994)’. However he notes that ‘fashions change … and the efficacy of modelling in areas where it has been applied may well increase levels of interest in other areas. There may be grounds again, therefore, for optimism.’ It is on this note of optimism that we summarize the paper by Qisheng Pan, Harry Richardson, Peter Gordon & James Moore. Unfortunately, and somewhat paradoxically, the subject matter is not a very optimistic one in that it concerns ‘The Economic Impacts of a Terrorist Attack on the Downtown Los Angeles Financial District’. In this paper we see Regional Science in action, with the emphasis on application of a comprehensive large-scale model to a real problem. The paper summarizes the outcomes of the latest version of regional model building from Garin–Lowry-style models, which at their simplest can be described as the equilibrium outcome of a basic (export) and non-basic (population-dependent services) economic structure in each sub-region, with reallocations via feedback loops of population, employment, economic activity and land use using simple gravity models. In the present paper, Pan and co-authors report the data produced by models that are much more elaborate, combining input–output methods and endogenizing previously exogenous elements (travel behaviour). Put simply, the input–output model generates induced impacts of the attack which are then spatially allocated via the Garin–Lowry-style model. The results are plain to see—the effects of an attack are differentiated across space and time. With this information, planners may want to take account of possible repercussions. However, and hopefully someone has thought of this, terrorists who are planning attacks may (possibly do) also have access to ‘models’ showing differentiated impacts of attacks, so perhaps we need a new generation of models which take account of the terrorist response to the likely outcome of an attack. In which case they may also adapt their ‘model’ to take account of the adaptation by planners and the authorities, and so on. Perhaps it is at this point that game theory meets regional modelling.

Notes

1. At its simplest, Verdoorn's law relates manufacturing output growth to manufacturing productivity growth.