class: center, middle, inverse, title-slide # Regional Income Disparities, Distributional Convergence, and Spatial Effects: ## Evidence from Indonesia ### Carlos Mendez
Associate Professor
Graduate School of International Development Nagoya University
Anang Budi Gunawan
Doctoral Student
Graduate School of International Development, Nagoya University
Felipe Santos-Marquez
Master Student
Graduate School of International Development Nagoya University ### Prepared for the Applied Regional Science Conference (ARSC) 2019 Saga, 23-24 November 2019
[ Slides available at:
https://quarcs-lab.rbind.io
] --- ## Motivation: - Large regional differences in income per capita despite several policy efforts (Akita 1988; Garcia and Soelistianingsih 1998; Kataoka 2012) - Spatial effects play a small role in provincial convergence in Indonesia (Vidyattama 2013, 2014) ## Research Objective: - Study the spatio-temporal dynamics of income per capita accross Indonesian regions using **_a novel district-level dataset_** constructed for the 2000-2017 period ## Methods: - Classical convergence (Barro and Sala-i-Martin 1992) - Distributional convergence (Quah 1996; Hyndman et. al 1996) - Spatial autocorrelation (Moran 1948, Anselin et. al 2006) - Spatial decomposition (Getis 1995; Fischer and Stumpner 2010) ## Data: - 514 districts (34 provinces) over the 2000-2017 period. --- class: middle ## Main Results: 1. **_On average_,** convergence at the provincial and district levels 2. **_Beyond the average_,** lack of distributional convergence - Mostly relative stagnation with some signs of divergence at the tails 3. **Significant and increasing spatial autocorrelation:** ONLY at the district level 4. **Spatial effects play a role in the distribution dynamics** - Spatial dependece helps reduce extreme disparities - It helps avoid further income polarization --- class: middle # Outline of this presentation 1. **Some Stylized Facts ** - **_On average_,** convergence at the provincial and district levels - **Significant and increasing spatial autocorrelation:** ONLY at the district level 2. **Distributional convergence and spatial decomposition frameworks** - Distributional convergence framework (intuition) - Spatial decomposition framework (intution) 3. **Main Results:** - **_Beyond the average_,** lack of distributional convergence - **Spatial dependece helps reduce extreme disparities** --- class: center, middle # (1) Some Stylized Facts **_On average_,** convergence at the provincial and district levels **Significant and increasing spatial autocorrelation:** ONLY at the district level --- class: center, middle ## On average, is there convergence at the provincial level? <img src="figs/beta_province.png" style="width: 80%" /> --- class: center, middle ## On average, is there convergence at the district level? <img src="figs/beta_districts.png" style="width: 80%" /> --- class: center, middle ## Is spatial dependency _at the district level_ statistically significant? `$$I = \frac{\sum_i\sum_j w_{ij} z_i.z_j}{\sum_i z_i^2} = \frac{\sum_i (z_i \times \sum_j w_{ij} z_j)}{\sum_i z_i^2}.$$` <img src="figs/stam.png" style="width: 70%" /> Yes, moreover it is increasing over time. --- class: center, middle # How do we compute the spatial weights matrix? We want to avoid the use of an arbitrary distance band. Original feature: We use the locations of capital cities to estimate Thiessen polygons and recover a contiguity matrix. ![](figs/contiguity-polygons.jpg) --- class: center, middle # (2) Distributional convergence Let's study convergence BEYOND the average --- class: middle, center # The distribution dynamics framework <img src="figs/dynt.png" style="width: 80%" /> --- class: middle, center # Some illustrative patterns of stagnation, convergence, and divergence <img src="figs/Intra-distribution_dynamics.jpg" style="width: 60%" /> --- class: center, middle # Spatial decomposition framework: The Getis filter `$$x_i^* = \frac{xi(W_i)}{(n-1)G_i(d_m)}$$` ![](figs/decomposition.jpg) --- class: center, middle # (3) Main Results (1) **_Beyond the average_,** lack of distributional convergence (2) **Spatial dependece helps reduce extreme disparities** --- class: middle, center # Beyond the average, lack of distributional convergence ## Mostly relative stagnation and divergence at the extremes ![](figs/rawVariable.jpg) --- class: middle, center # Beyond the average, lack of distributional convergence (further polarization) ## Distribution dynamics of the non-spatial component ![](figs/non-spatial.jpg) --- #Concluding Remarks ## Classical convergence VS Distributional convergence - Classical: On average, convergence - Distributional: Beyond the average, lack of convergence ## On the role of space - Increasing spatial autocorrelation - Geographic neighbors helped reduce some extreme disparities ## Implications - Beyond the average progress, regional inequality is still an issue - For further research : - Using district-level data, what is the role of geographical neighbors in accelerating the speed of convergence? - Using the time series of the district-level data, are there convergence clubs? - Integration of spatial and dynamic clusters --- class: center, middle # Thank you very much for your attention You can find this presentation on our QuaRCS lab website https://quarcs-lab.rbind.io <img src="figs/logo2.png" style="width: 20%" /> **Quantitative Regional and Computational Science Lab**