Raising the bar (5)

ABSTRACT

Raising the bar (5). Spatial Economic Analysis. This editorial summarizes and comments on the papers published in this issue 12(1) so as to raise the bar in applied spatial economic research and highlight new trends. The first paper examines the impact of the level of education on the decision to migrate and finds that it is approximately twice as large if both variables are modelled simultaneously. The second paper is one of the first papers to introduce a spatial component to models of international environmental agreements and to develop an exciting overlap with New Economic Geography. The third paper provides a tool, applied to Beijing, with which urban economic planners can investigate the role of variation and selection mechanisms in cluster development and identify possible paths of growth. The fourth paper contributes to the existing literature on retail geography by examining the role of consumption possibilities as an urban amenity. The fifth paper develops a Bayesian estimator of a linear regression model with spatial lags among the dependent variable, the explanatory variables and the disturbances. Finally, the sixth paper develops a semi-parametric generalized method of moments (GMM) estimator for a spatial autoregressive model with space-varying coefficients of the explanatory variables and a spatial autoregressive coefficient common to all units.

摘要

提高标准（五）。Spatial Economic Analysis。本编辑评论摘要并评析本期刊第十二期第一辑所发表的文章，以此提高应用空间经济研究的标准，并强调新的趋势。第一篇文章检视教育程度对于迁徙决定的影响，并发现若两个变项皆同时进行模式化的话，该影响约为两倍大。第二篇文章是最早将空间元素引进国际环境协议模型，并与新经济地理学发展出令人兴奋的叠合的文章之一。第三篇论文提供城市经济规划师能够藉此探讨变异与选择机制在集群发展中的角色，同时指认可能的成长路径之工具，并将之应用至北京。第四篇文章透过检视消费可能性作为城市环境的角色，对于零售地理学的既有文献做出贡献。第五篇论文建立一个在依变项上具有空间迟后的线性迴归模型的贝氏估计量。最后，第六篇文章为具有解释变数的空间变异係数以及所有单位共有的空间自迴归係数的空间自迴归模型，建立一个半一般化动差法（GMM）之估计量。

RÉSUMÉ

Relever la barre (5). Spatial Economic Analysis. Cet éditorial cherche à résumer et à commenter les articles publiés dans ce numéro 12(1) pour relever la barre dans le domaine de la recherche appliquée en économie spatiale et souligner les nouvelles tendances. Le premier article examine l’impact du niveau de scolarité sur la décision de migrer. Il s’avère environ deux fois plus important dans le cas où les deux variables seraient modélisées simultanément. Le deuxième article constitue l’un des premiers articles à intégrer une composante spatiale aux modèles des accords environnementaux internationaux et à développer un chevauchement intéressant avec la Nouvelle géographie économique. Le troisième article fournit un outil, appliqué à Beijing, à partir duquel les urbanistes économiques peuvent examiner le rôle des mécanismes de variation et de sélection dans le développement des clusters et identifier des sentiers de croissance éventuels. Le quatrième article contribue à la documentation actuelle au sujet de la géographie du commerce de détail en examinant le rôle des possibilités de consommation comme un actif urbain. Le cinquième article développe un estimateur bayésien d’un modèle de régression linéaire avec retards quant à la variable dépendante, aux variables explicatives et aux perturbations. Pour terminer, le sixième article développe un estimateur semi-paramétrique par la méthode des moments généralisés (MMG) pour un modèle autorégressif spatial comportant des coefficients des variables explicatives qui varient dans l’espace et un coefficient autorégressif spatial qui est un point commun.

RESUMEN

Elevar el listón (5). Spatial Economic Analysis. Este editorial es un resumen y una observación acerca de los artículos publicados en este número 12(1) con la finalidad de elevar el listón en la investigación económica espacial aplicada y resaltar las nuevas tendencias. En el primer artículo se analiza cómo repercute el nivel de formación a la hora de tomar la decisión de emigrar y se observa que esta repercusión es aproximadamente dos veces mayor si se modelan ambas variables a la vez. El segundo artículo es uno de los primeros en introducir un componente espacial a los modelos de los acuerdos internacionales sobre el medio ambiente y desarrollar un interesante solapamiento con la Nueva Geografía Económica. En el tercer artículo se ofrece una herramienta, aplicada a Pekín, con la que los planificadores de la economía urbana pueden investigar el papel de los mecanismos de variación y selección en el desarrollo de aglomeraciones e identificar posibles vías de crecimiento. El cuarto artículo contribuye a las publicaciones existentes sobre la geografía de comercios minoristas al examinar el papel de las posibilidades de consumo como una amenidad urbana. En el quinto artículo se desarrolla un estimador bayesiano de un modelo de regresión lineal con lagunas espaciales entre la variable dependiente, las variables explicativas y las perturbaciones. Para terminar, en el sexto artículo se desarrolla un estimador del método generalizado de momentos semiparamétrico para un modelo espacial autorregresivo con coeficientes de variación espacial de las variables explicativas y un coeficiente espacial autorregresivo común a todas las unidades.

KEYWORDS:

• migration
• clusters
• cooperation
• retail geography
• spatial econometrics

关键词:

• 迁徙
• 集群
• 合作
• 零售地理学
• 空间计量经济

MOTS-CLÉS:

• migration
• clusters
• coopération
• géographie du commerce de détail
• économétrie spatiale

PALABRAS CLAVES:

• migración
• aglomeraciones
• cooperación
• geografía de comercios minoristas
• factores econométricos espaciales

JEL:

• C
• I21
• L52
• L81
• Q57

Spatial Economic Analysis is a pioneering journal dedicated to the development of theory and methods in spatial economic analysis. This issue contains six papers contributing to these theoretical and empirical developments.

Migration keeps inspiring economic researchers. Recently, we paid attention to this trend by publishing a special virtual issue on migration (Jordan & Elhorst, 2016). The papers comprising this virtual special issue are freely downloadable until the end of March 2017 from the journal’s website. The first contribution to the current issue continues this trend by examining the causal impact of the level of education on migration within Finland’s 18 NUTS-3 regions. Haapanen and Böckerman (2016) use the gradual reform of vocational colleges into polytechnics, which took place in Finland in the 1990s, as an identification strategy to show that traditional estimates of the effect of education on migration have underestimated the true causal effect. For this purpose they use individual data on 233,839 graduates. This huge dataset also offers them the opportunity to show that the causal effect is heterogeneous across gender, field of study (business, technology and health) and region (the capital region of Uusimaa and other regions). From a methodological viewpoint, this paper offers a textbook example of how to deal with the endogeneity of regressors, an issue that does not always get the attention it requires. Instead of a standard binary logit model in which the decision to migrate is taken to depend on the level of education and a set of control variables, this study poses a multinomial treatment effects model in which the decision to migrate and the choice of education are jointly determined. It turns out that the marginal effect of polytechnic education on the decision to migrate is approximately twice as large when this simultaneity is accounted for. The empirical findings are consistent with research published previously in this journal by Venhorst, Van Dijk, and Van Wissen (2011).

The second paper by Alvarado-Quesada and Weikard (2017) is one of the first papers introducing a spatial component to models of international environmental agreements (IEAs). Countries are geographically placed along a circular model (Salop, 1979), and the distance between neighbours and the initial endowments are allowed to vary. The purpose is to show how IEAs can promote greater biodiversity when the benefits are shared both regionally and globally. Biodiversity between two countries is measured by ecosystem dissimilarity (ED), which is the number of species that appear in one country, but not in the other (Weitzman, 1992). The main question is whether there is stability in a large coalition of countries that coordinate on biodiversity and how much of the potential gains they can obtain from cooperation. The paper uses the standard notion of internal and external stability for the numerical analysis. Coalition formation occurs in a two-stage game. In the first stage countries choose whether or not to join the IEA. In second stage signatories to an agreement choose biodiversity to maximize their collective net benefits and those countries outside the agreement maximize individual net benefits.

The benefits from biodiversity are of three different types. First, there is a constant marginal benefit (λ) that is contained within the country providing the biodiversity. This implicitly assumes that each species is equally valuable. Second, there is a constant regional marginal benefit (ρ) which is a weighted sum of the biodiversity provided by other countries. Third, there is a constant global marginal benefit (γ) which is a pure public good, where γ is normalized to 1. The marginal cost of providing biodiversity is increasing at a rate c. Hence, we have a very common formulation in the IEA literature known as the linear-quadratic model.

The paper presents numerical simulations for a 12-country world. This idea of 12 countries on a circle is strongly related to the literature on New Economic Geography (NEG), making this paper interesting for spatial economic researchers specialized in this area. Initially, each country has identical cost and benefit parameters and countries only differ in their location on the circle. The base model assumes specific values for the eight parameters in Alvarado-Quesada and Weikard’s (2017) Table 1. The simulations consider two spatial distributions. First, equal ED, which implies very little differentiation between pairs of countries. Second, increasing ED as countries get farther away from each other. This seems a more realistic scenario, but the equal ED is needed to understand the role of the spatial distribution as a benchmark. The measure of effectiveness of an IEA is the ‘closing the gap index’ (CGI) (Eyckmans & Finus, 2006) and shows the percentage of the difference between the full and no cooperation that the stable IEA can obtain.

The authors find that the largest stable IEA consists of only two of the 12 countries when there is equal ED in the baseline model. The CGI is about 11%, again a very small result given the potential gains from cooperation. The results change remarkably little when the eight other choices of higher or lower benefit and cost parameters and the number of costless species are presented in their Table 2. The simulations for increasing ED show very similar results. The largest stable IEA again consists of two countries and the CGI is between 11% and 14% for the various parameter values. For differences in the size of coalitions, we refer to Gelves and McGinty (2016).

Finally, the paper shows that transfers have no impact on the IEA results. This is because the countries are identical, other than their location. The main intuition is that higher regional benefits between countries with larger ED levels limit the potential stability and hence size of a stable IEA. The main policy conclusion is that multiple small biodiversity agreements are more likely to be successful than a single global agreement. These agreements should consist of countries with relatively small ED levels.

The spatial aspects of climate change and its impact has started gaining attention in the literature (e.g., Dall’Erba & Domínguez, 2016). The contribution in this paper is significant since it is one the first to explicitly address a spatial structure in IEAs and in developing a connection with NEG.

The third contribution to this issue, by Yang and Dunford (2016), analyses the relationship between regional structural change and industrial cluster life cycles. Methodologically, the paper uses exploratory factor analysis in combination with varimax rotation to derive factor loadings and, based on these factor loadings and the economic importance of different industries, to identify clusters. The major step forward in this paper is the way cluster change is analysed and described. To disclose changes in the composition and role of industries, the authors consider six stages of cluster evolution: emergence, stabilization, upgrading, restructuring, decomposition and disappearance. By applying this cluster life-cycle approach to Beijing, the authors provide a tool which urban economic planners can use to investigate the role of variation and selection mechanisms in cluster development, and to identify possible paths of growth.

Previous literature has analysed spatial linkages using input–output tables (e.g., Chung & Hewings, 2015; Noblet & Belgodere, 2016) as well as factor structures extracted by exploratory factor analysis (Bhattacharjee, Castro, & Marques, 2012). The major innovation in this paper is to put the two approaches together and then use the framework to understand temporal dynamics in input–output linkages. Further, the application to a planning context is novel.

The fourth paper in this issue is by Öner (2017). It sets out to determine whether consumers’ access to retail units in municipalities and/or wider regions in Sweden is relevant to the municipalities’ attractiveness, where attractiveness is proxied by investment in housing in a municipality. It concludes that even though such a relationship is present for urban municipalities, this is not the case for rural ones. The paper contributes to the existing literature on retail geography (e.g., Guy, 2013; Okeahalam, 2009) by examining the role of consumption possibilities in enhancing place attractiveness, a relationship which is often neglected in debates on local development and growth. It convincingly argues that the function of the retail sector extends beyond a simple supply-and-demand schedule. Finally, it produces a new measure of ‘accessibility to shops’ based on the ideas behind Tobin’s Q.

The fifth paper by Ramírez Hassan (2016) develops a traditional Bayesian estimation procedure to estimate the parameters of the general nesting spatial (GNS) model, a linear spatial econometric model with a full set of spatial interaction effects, namely among the dependent variable, the exogenous variables and among the disturbances. The number of studies on this model is small but growing (Anselin, 1988; Burridge, Elhorst, & Zigova, 2016; Elhorst, 2014; Halleck Vega & Elhorst, 2015; Lee, Liu, & Lin, 2010). As early as 1988, Anselin (1988, pp. 61–65) advocated this general model with all types of interaction effects, though without providing conditions under which the parameters of this model are identified. In addition to this, he allowed the disturbances to be heteroskedastic, an issue which is interesting with respect to the next paper to this issue. Lee et al. (2010) were the first to provide formal proofs and conditions under which the parameters of the GNS model. Importantly, their proofs are limited to a spatial weights matrix that is specified as an equally weighted group interaction matrix with a zero diagonal. This is a block diagonal matrix where each block represents a group of units that interact with each other but not with members of other groups.

Elhorst (2014, pp. 27–32) applies the GNS model to Anselin’s (1988) cross-sectional dataset of 49 neighbourhoods in Columbus, Ohio, to explain the crime rate as a function of household income and housing values if the spatial weight matrix is specified as a row-normalized binary contiguity matrix. Similarly, Halleck Vega and Elhorst (2015) apply the GNS model to Baltagi and Li’s (2004) panel dataset of 46 US states over the period 1963–92 to explain cigarette consumption as a function of the price of cigarettes and real per capita disposable income, again if the spatial weight matrix is specified as a row-normalized binary contiguity matrix. This dataset is also used by Ramírez Hassan (2016). Finally, Burridge et al. (2016) test the feasibility, empirical implications and relevance of a group interaction model with a full set of interaction effects, as well as the extensions with group fixed effects as set forth in Lee et al. (2010) and heteroskedastic disturbances as proposed by Anselin (1988). The results from these empirical applications are illuminating. In many cases, including the group-interaction model, the model turns out to be over-parameterized, leading to weak identification of the spatial interaction effects especially among the dependent variable and among the error terms. Considering them both, as in the GNS model, has the effect that the significance levels of all variables go down, as a result of which it provides no additional information over simpler models, especially the popular spatial Durbin model, which contains no spatial interaction among the error terms, and the spatial Durbin error model, which contains no spatial interaction among the dependent variables. Moreover, if the spatial weight matrix takes a different form from the group interaction matrix, there is no formal proof yet available that the unknown parameters of the GNS model are identified. One obstacle to this proof is that the parameter estimates of the spatially lagged dependent variable (ρ) and the spatially autocorrelated error term (λ) in Monte Carlo simulation experiments are sometimes interchanged. This problem also seems to affect Ramírez Hassan’s (2016) simulation experiment reported in Table 1 of his paper. Although the author states that the results are very similar and close to the population parameter, this does not seem to hold for the coefficient λ of the spatially lagged dependent variable (WY) and the coefficient ρ of the spatially autocorrelated error term. Whereas λ = 0.9 and ρ = 0.7, the author finds simulated values of 0.858 and 0.503 when applying his Bayesian estimation procedure, a bias of respectively 4.7% and 28.1%. An explanation for these biases is not provided, but we believe that they might be caused by Halleck Vega and Elhorst’s (2015) finding that λ and ρ are sometimes interchanged. Either way, it is a positive development that different estimation methods are now available to estimate the parameters of a general spatial nesting model.

The final contribution, by Wei and Sun (2016), is an econometric–theoretical paper dealing with space-varying coefficients of the explanatory variables in a spatial autoregressive model, and a spatial autoregressive coefficient which is common to all units. It fits within a renewed interest in heterogeneous spatial econometric models (e.g., Zhang, 2013), previously advocated by Anselin (1988). In this respect it is to be noted that Aquaro, Bailey, and Pesaran (2015) recently developed a maximum likelihood estimator of a spatial autoregressive model with space-varying coefficients for all variables, including the spatially lagged dependent variable, and that LeSage and Chih (2016) derived the direct and indirect effects of this model. The latter find that, in contrast to homogenous spatial econometric models, a distinction needs to made between spill-in and spill-out effects when determining the indirect effects. Wei and Sun’s (2016) contribution builds upon previous studies in that they also allow for heteroskedasticity of an unknown form, thereby following Kelejian (2016) who explicitly states that we should not model heteroskedasticity, and in that they develop a two-step procedure to estimate the response parameters of their model. In the first step the space-varying coefficients of the explanatory variables are estimated by a non-parametric local linear regression technique, where the common spatial autoregressive parameter is treated as if it were known, and in the second stage the spatial autoregressive parameter is estimated by a generalized method of moments (GMM) estimator, given the space-varying coefficients of the explanatory variables obtained in the first step. A mathematical proof and a Monte Carlo simulation experiment are provided to show that this semi-parametric GMM estimator is consistent and that its finite sample performance is acceptable.