<!-- class: inverse, center, title-slide, middle --> class: center, middle <style> g { color: rgb(0,130,155) } r { color: rgb(174,77,41) } y { color: rgb(177,148,40) } </style> # Lecture 10: Geospatial Data Sciences # and Economic Spatial Models ## <img src="figs/bse_primary_logo.png" style="width: 35%" /><br><br>Bruno Conte ## 11-12/Mar/2026 --- # Geospatial Data and Spatial Models: Schedule 1. ~~Introduction to (spatial) data and programming in `R`~~ **[08/Jan/2026]** 2. ~~Week 2-4: Vector spatial data~~ **[14 - 29/Jan/2026]** 3. ~~Week 5-7: Raster spatial data + (basic) interactive tools~~ **[05 - 19/Feb/2026]** 4. Week 8-10: Spatial models and applications with data **[25/Feb - 12 Mar/2026]** - ~~Week 8: Introduction to economic spatial models~~ - ~~Week 9: Linking models to (spatial) data~~ - Week 10: Models, data, and counterfactual simulations<br> <br> 5. <span style="color: rgb(177,148,40)">Take-home exam</span> **[27/Mar/2026]** --- # Spatial Data, Spatial Models, and Counterfactuals: Roadmap .pull-left[ 1. <u>Exam</u> - Any <r>final questions</r>? ] .pull-right[ 2\. <u>Using spatial to measure (spatial) shocks</u> - Flooding in coastal cities - Pollution in industrialized towns - Past and future climate change effects - Lake shrinking in rural economies - Global warming 3\. <u>Counterfactual effects</u> - Spatial shocks in partial equilibrium ] --- # Geospatial Data and Spatial Models: Feedback - <u>Recall</u>: BSE runs a quality survey to assess students' feedback on the courses - Make sure to answer it: important for <g>improving</g> next iterations - I am interested in <r>remarks/comments</r> on - Course pace? - Technical level (too basic/difficult)? - Course approach (concepts + hands-on)? - Data + theory? - <y>Anything else?</y> --- class: center, middle # Counterfactual Simulations with # Spatial Models and Spatial Data --- # Geospatial Data and Spatial Models: Counterfactual Experiments - <u>Spatial models</u>: economic mechanisms behind the <g>spatial distribution</g> of economic activity - Where does economic activity gets concentrated? - What drives location/migration choices of economic agents? - How are these choices <r>affected by changes</r> in the (spatial) aspects of the economy? - <u>This class</u>: examples of studies that answer these <y>hypothetical questions</y> - Projects that combine spatial data with spatial models - For that: creative and clever implementation of spatial tools --- # Geospatial Data and Spatial Models: Counterfactual Experiments To get started, recall how spatial models characterize the <g>geography of the economy</g> - Canonial model: `\(\mathcal{G} = \{ A_n, B_n, H_n, d_{ni} \}_{n,i \in S} \equiv\)` geographical features of the economy - Intuition of <r>counterfactual experiments</r>: what would happen if elements of `\(\mathcal{G}\)` are affected by policies or shocks? Elements (and shocks to them): - `\(A_n \equiv\)` fundamental productivity in locations `\(n\)` (climate change) - `\(B_n \equiv\)` level of amenities in locations `\(n\)` (pollution/smog) - `\(H_n \equiv\)` level of amenities in locations `\(n\)` (floods) - `\(d_{ni} \equiv\)` trade costs between `\(n\)` and `\(i\)` (expansion of railroads; Donaldson, 2018) - <u>Next</u>: examples with shocks to the first three! --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>First example</u>: worldwide consequences of <g>global warming</g> (Conte et al., 2021) - Idea: climate change as a sector-specific <r>productivity shock</r> to `\(A_n\)` - Why? Different sectors `\(\rightarrow\)` `\(\neq\)` sensitivity to changing temperatures - Recall that, in the canonical spatial model, total agents choosing to live in `\(n\)`, `\(L_n\)`, is `$$L_n = \frac{B_n (v_n / \mathcal{P}_n)^\epsilon}{\sum_i B_i (v_i / \mathcal{P}_i)^\epsilon} \mathcal{L}, \text{ where}$$` `$$\mathcal{P}_n = \gamma \left[ \sum_{i \in S} A_i (w_i d_{ni})^{-\theta} \right]^{-\alpha/\theta} \times r_n^{1-\alpha}$$` - In partial equilibrium, `\(d L_n / d A_n >0\)` (i.e., more productive regions `\(\rightarrow\)` more population) - What about the effects of `\(\Delta A_i\)` in other `\(i\)` locations? --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Challenge</u>: how to <g>map changes</g> in (global) temperature to (sectoral) productivities `\(A_n\)`? .pull-left[ <u>Step 1</u>: measure sectoral <r>damage functions</r> based on temperatures - Idea: each sector has an optimal temperature for productivity - Deviating from it decreases `\(A_n\)` - Hence, climate change effects can positive or negative (depends on initial temperatures) ] .pull-right[ <img src="figs/lssww_1.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 2</u>: map the damage function to a fine-grain spatial resolution (and model): .center[ Figure: changes in temperature following a 1 degree increase in global temperature <img src="figs/lssww_2.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2010 <img src="figs/lssww_map_smooth_10.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2020 <img src="figs/lssww_map_smooth_20.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2030 <img src="figs/lssww_map_smooth_30.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2040 <img src="figs/lssww_map_smooth_40.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2050 <img src="figs/lssww_map_smooth_50.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2100 <img src="figs/lssww_map_smooth_100.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (1/3) <u>Step 3</u>: quantify the GE of the quantified `\(\{ \Delta A_n \}\)` in terms of (sectoral) `\(\Delta L_n\)` .center[ Figure: changes in agricultural employment in 2200 <img src="figs/lssww_map_smooth_200.png" style="width: 80%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (2/3) <u>Second example</u>: how did the industrial revolution affected inequality within UK cities? - Idea: smog as an <r>amenity shock</r> to `\(B_n\)` in the `\(n\)` affected (Eastern) locations - Analogously, `\(L_n \equiv\)` total choices to live in `\(n\)`, in spatial equilibrium: `$$L_n = \frac{B_n (v_n / \mathcal{P}_n)^\epsilon}{\sum_i B_i (v_i / \mathcal{P}_i)^\epsilon} \mathcal{L}$$` - Clearly, `\(d L_n / d B_n >0\)` (i.e., higher amenities `\(\rightarrow\)` more population) - What about the effects of `\(\Delta B_i\)` in other `\(i\)` locations? - <u>Question</u>: can real-world data provide evidence for this "East side story"? - <u>Challenge</u>: how to quantify this different pattern (eastwards) of smog diffusion? --- # Geospatial Data, Models, and Counterfactual Examples (2/3) .center[ <img src="figs/yanos_1.png" style="width: 85%" /> ] --- .center[ <img src="figs/yanos_2.png" style="width: 85%" /> ] --- # Geospatial Data, Models, and Counterfactual Examples (3/3) <u>Third example</u>: how would <g>coastal flooding</g> affect Jakarta (Hsiao, 2023) - Idea: flooded areas reduce <r>land availability</r> `\(H_n\)` - Flood-prone areas: capital damages `\(\rightarrow\)` lower potential for real estate development - Recall that, in the canonical spatial model, total agents choosing to live in `\(n\)`, `\(L_n\)`, is `$$L_n = \frac{B_n (v_n / \mathcal{P}_n)^\epsilon}{\sum_i B_i (v_i / \mathcal{P}_i)^\epsilon} \mathcal{L}, \text{ where}$$` `$$\mathcal{P}_n = \gamma \left[ \sum_{i \in S} A_i (w_i d_{ni})^{-\theta} \right]^{-\alpha/\theta} \times r_n^{1-\alpha}$$` - In <g>general equilibrium</g>, `\(d L_n / d H_n >0\)` because of GE effects in land prices `\(r_n\)` (why?) --- # Geospatial Data, Models, and Counterfactual Examples (3/3) <u>Challenge</u>: in practice, flood risk depend on <r>several (spatial) features</r> - Elevation (compared to sea level) - Terrain characteristics - Rainfall patterns - Closeness to water bodies (e.g., rivers and canals) <u>Solution</u>: data-science approach `\(\rightarrow\)` <r>machine learning model</r> that predicts flood risk - Idea: interaction of these (and other) features is complex - Using observed flooding data, train a random tree ML model - Identifies most important features and avoids <y>overfitting</y> --- # Geospatial Data, Models, and Counterfactual Examples (3/3) .pull-left[ <img src="figs/hsiao_1.png" style="width: 100%" /> ] .pull-right[ <img src="figs/hsiao_2.png" style="width: 80%" /> ] - Then, (future) changes to spatial inputs `\(\rightarrow \Delta H_n \equiv\)` spatial shock - GE effects of future flood + counterfactual scenarios (e.g., with or without sea walls) --- # Geospatial Data and Spatial Models: Takeaways <u>This course</u>: how to use state-of-art tools to work with spatial data - The last 2-3 classes: basics on how to integrate these tools inside spatial models Important (course) **takeaways**: - Each empirical exercise (regardless of the spatial dimension) is a <r>challenge on its own</r> - Avoid copy-pasting/ChatGPTing what you do not know! - Main source of mistakes: adapting black box-type of codes - My goal with <g>this course</g> was to teach you the architechture behind spatial data - Once you understand it, operations become simple implementation of basic concepts - If you do not understand it: do not go back to the codes - Review the material that link concepts to tools (codes) and then implement them! --- class: center, middle # The course is over - you have made it! <img src="https://memesgifs.com/wp-content/uploads/World-Cup-Success-gif-FIFA.gif" style="width: 60%" /> # Thank you so much! :) --- # References - Conte, B., Desmet, K., Nagy, D.K. and Rossi-Hansberg, E., 2021. Local sectoral specialization in a warming world. *Journal of Economic Geography*, 21(4), pp.493-530. - Fernihough, A. and O'Rourke, K.H., 2021. Coal and the European industrial revolution. *The Economic Journal*, 131(635), pp.1135-1149. - Giua, M., 2017. Spatial discontinuity for the impact assessment of the EU regional policy: The case of Italian objective 1 regions. *Journal of Regional Science*, 57(1), pp.109-131. - Heblich, S., Trew, A. and Zylberberg, Y., 2021. East-side story: Historical pollution and persistent neighborhood sorting. *Journal of Political Economy*, 129(5), pp.1508-1552. - Hsiao, A., 2023. Sea Level Rise and Urban Adaptation in Jakarta. Technical Report.