Ragdad Cani Miranti (p)

Paul Elhorst

Using a novel provincial-level dataset of regional growth, industrial agglomeration, and related determinants, this paper attempts to examine the spill-over effect of industrial agglomeration on growth across 34 provinces in Indonesia. By employing Spatial Durbin Model, we evaluate the influencing factors and spatial externalities which tend to lead regional economic growth in Indonesia. In particular, we also examine the role of spatial spillover to regional growth across provinces. The results show that there was positive spatial autocorrelation both on industrial agglomeration and economic growth during 2010-2021. However, spatial autocorrelation was negative for economic growth in 2010. Despite positive spatial association, the magnitude indicated a weakening pattern for industrial agglomeration in 2021. Another appealing finding shows that decreasing patterns of spatial dependence are significantly associated with increasing patterns of industrial agglomeration inequalitySecond, Moran’s I scatter plot and directional LISA indicate the existence of spatio-temporal patterns in both agglomeration and regional growth. Third, spatial effect of industrial agglomeration, Information and communication technology (ICT), and infrastructure level significantly leverage regional economies. Using Spatial Durbin Model, we find that spatial lags of some explanatory variables are significant, which are industrial agglomeration, ICT, and infrastructure level, suggesting significant effects on the growth performance in a given province. From a policy perspective, our findings alert policymakers to consider the possibility of augmenting industrial clusters and improving infrastructure-based policies with region-specific efficiency measures.

Paul Elhorst (p), Chang Tan, Michaela Kesina

Itsaso Lopetegui

This paper sets out the general nesting spatial econometric model with spatial lags in the regressand, regressors and the error term, as well as individual and time fixed effects for both large N fixed T and large N large T panel data settings. The error term is specified as a spatial moving average process and heteroscedastic disturbances. The spatial weight matrices are distance-based, parameterized and different for each spatial lag to achieve that the magnitude and spatial reach of the spillover effects caused by changes in the regressors or shocks to the error term take different values. The model parameters are estimated by an iterative two-stage quasi-maximum likelihood procedure. It is demonstrated that this modelling setup and estimation procedure significantly reduces theoretical and practical identification problems that have prevented broader application of this model in empirical research. Its finite sample properties are investigated by means of Monte Carlo simulation. GDP per growth data is used to illustrate the benefits in an empirical setting.

Itsaso Lopetegui (p), Ikerne del Valle

Vincenzo Nardelli

This paper investigates the relationships between diversity, risk, and prices in the fisheries sector in the EU, focusing on 23 distinct sub-ecosystems, each representing a fishing country. To measure biodiversity, we employ several indices, including the Berger-Parker Index, Simpson’s Index, Concentration Ratio, and Shannon Index, which provide insights into the sustainability and resilience of fisheries across the sub-ecosystems as each member-state operates within a unique marine sub-ecosystem composed of diverse fish species, which may also change over time. The study also incorporates an analysis of risk, using a financial left-tail risk measure to capture downside risk. This serves as a sub-ecosystem level proxy for the inherent risk in fishing activities, providing an empirical and probabilistic measure of worst-case losses, which is crucial for understanding the economic vulnerability of fisheries.

We explore price trends of fish landings within the sub-ecosystems to assess market behavior, and evaluate the efficiency of each sub-ecosystem by comparing output to input levels. To measure technical efficiency, we employ a Cobb-Douglas Stochastic Production Function, alongside a Bias-Corrected Data Envelopment Analysis (DEA) with environmental variables. The DEA framework utilizes a second algorithm for bias correction, estimating efficiency scores in both input- and output-oriented models while accounting for exogenous environmental factors. Additionally, the paper examines spatial autocorrelation to identify regional hotspots or coldspots in both biodiversity and market performance, uncovering patterns of geographic interdependence.

A key objective of the research is to explore the temporal and spatial dynamics of these indices. Specifically, we analyze whether diversity, risk, and prices change over time or vary across sub-ecosystems. To do so, we apply both parametric (e.g., ANOVA) and non-parametric (e.g., Kruskal-Wallis) statistical tests to identify significant differences between sub-ecosystems and over different time periods. This approach allows us to assess the stability or variability of these metrics, highlighting potential shifts in market conditions or ecological health.

Theoretically, lower diversity is associated with higher concentration, dominance, and dependency of the fishing sector on the dynamics of dominant fish species, thereby increasing the risk of potential collapse. Furthermore, efforts to enhance efficiency may reduce diversification, potentially leading to greater income variability and financial risk. Our findings reveal notable differences in diversity, risk exposure, and price trends across EU fisheries, with significant spatial variability. These results offer valuable insights for policymakers and stakeholders, providing a basis for more targeted and informed interventions aimed at ensuring long-term sustainability and economic stability in the fisheries sector.

Giuseppe Arbia, Vincenzo Nardelli (p), Niccolò Salvini

Katarzyna Kopczewska

In spatial data analysis, global measures such as Moran’s I, Geary’s C, and APLE coefficients provide insights into general spatial patterns but fail to capture local variations. Local Indicators of Spatial Association (LISA) address this limitation by identifying clusters and spatial outliers, offering valuable applications in public health, urban planning, and environmental monitoring, to name a few. However, multiple hypothesis testing in LISA dramatically increases the risk of Type I errors, especially in large datasets. Standard corrections, like Bonferroni and Benjamini-Hochberg, are based on the assumption of independence of observations and prove overly conservative in spatial contexts due to spatial autocorrelation. This paper proposes an adjustment that incorporates the notion of effective sample size, thus improving statistical inference while balancing false discovery control and statistical power in LISA-based analyses.

Katarzyna Kopczewska (p)

Ragdad Cani Miranti

This paper presents my book with Oxford University Press on 'Modelling spatial density'. It is an overview of spatial methods that use geo-located point data to detect, classify, and exploit global and local intensity. Geo-located point data are a natural source of information in urban and regional studies - they come from a variety of sources and are collected for a wide range of topics, from urban management and human mobility to health, environment, climate, natural resources, socio-economic development, business and industry, agriculture, infrastructure and culture. It is clear that geo-located point data are available for probably all areas of life, and rough estimates suggest that 80% of data may have a spatial dimension. Therefore, there is a need for good methodological overviews on how to perform quantitative analyses on these data.

The quantitative toolbox allows one to go beyond simple visualisation of data - one can detect high and low local density clusters, find density outliers, understand neighbourhoods of low and high density areas, measure the degree of agglomeration, track the spatio-temporal stability of observations and the relationships between their properties, and many others. This presentation covers a comprehensive set of methods based on four main statistical concepts: distances, circles, tesselations and grids. It goes far beyond the well-known Kernel Density Estimation plot or density clustering with DBSCAN. It is a combination of old fundamental concepts (such as entropy, kernel density estimation, graphical weighted regression, Voronoi polygons) mixed with recently popularised machine learning approaches (such as random forest, partitioning around medoids, k-means, artificial neural networks, time series clustering, etc.). The result is a unique, novel approach that balances spatial statistics, spatial econometrics and spatial machine learning. All examples on population and business location point data show how much potential there is in this information if it is well treated with appropriate quantitative methods.

An important element of an applied approach to quantitative methods is to understand what is being analysed for. This presentation links research questions within urban and regional science with specific methods and policy questions and implications that can arise. These include issues of cohesion, equity and efficiency in dealing with lagging places and their densities, the impact of depopulation, the importance of density for development, business location, innovation, saturation, overcrowding, elasticity, growth multipliers, over- and under-represented places, urban design and urban spillover policies, and diversity of densities.