Nighttime Lights and Economic Activity:

Evidence from Developed and Developing Countries

Yilin Chen

Nagoya University, Japan

Hussein Suleiman

Nagoya University, Japan

Nagoya University, Japan

Prepared for the 2024 (63rd) Annual Congress of the

European Regional Science Association (ERSA)

The story of this paper in 3 slides...

Based on 1557 first-level subnational regions over 2013-2019...

... (VIIRS) NTL luminosity predicts

better across economies than over time

NTL inequality correlates with GDP inequality in developing countries ...

... but this relationship fails in developed countries

The rest of this presentation

2. Data and methodology

3. Five stylized facts

Luminosity at night can be a proxy for economic activity

Source: NASA Earth Observatory

and NASA WorldView

Explore HERE

Validating the NTL-GDP relationship: A stylized overview

First wave:

(Elvidge et al. (1997), Chen & Nordhaus (2011), Henderson et al. (2012), among others)

• Data source: Defense Meteorological Satellite Program (DMSP)
• Focus: 1992-2013 period, multiple countries, subnational regions of single/few countries
  
• Problems: Calibration, geocoding errors, blurring, low resolution, top-coding
  

Key finding: NTL luminosity predicts GDP better across economies than over time

Validating the NTL-GDP relationship: A stylized overview

Better geo-technologies and new data

(Elvidge et al. (2013), Elvidge et al. (2017), Elvidge et al. (2021) )

From the Defense Meteorological Satellite Program (DMSP) to the Visible Infrared Imaging Radiometer Suite (VIIRS)

Second wave:

(Chen & Nordhaus (2015, 2019), Gibson & Boe-Gibson (2021), Zhang & Gibson (2022), among others)

• Data source: Visible Infrared Imaging Radiometer Suite (VIIRS)
• Focus: 2013- present , subnational regions of single countries
  
• Opportunities: Better temporal comparability, higher spatial resolution, no geocoding errors, much less blurring, no top-coding
  

Key finding: NTL luminosity predicts GDP better across economies than over time

Validating the NTL-GDP relationship: A stylized overview

Missing validation and comparisons:

• Subnational regions across multiple countries
  
• Subnational regions of developing countries
  
• Developed vs developing countries
  
• NTL-GDP in cross-section
  
• NTL-GDP in time series
• NTL-GDP in inequality
  

Research gap/challenge : Subnational GDP data for multiple developing countries

The rest of this presentation

1. NTL literature: Validation studies and new data

3. Five stylized facts

Data

GDP data:

NTL data:

• National GDP (Feenstra, 2015)
  
• 139 countries, 2013-2019 period
  
• 37 developed countries
  
• 102 developing countries
• Subnational GDP (Smits & Permanyer , 2019)
  
• 1557 subnational regions, 2013-2019 period
  
• 408 developed regions
  
• 1149 developing regions
  

• VIIRS (Elvidge et al., 2021)
  
• VIIRS masked (Elvidge et al., 2021)
  
• DMSP extended (Ghosh et al., 2021)
  
• DMSP like (Li et al., 2020)
  

New datasets and new research opportunities:

1. Study the NTL-GDP relationship using new subnational data from multiple developed and developing countries

2. Compare the performance of new NTL datasets (VIIRS vs DMSP)

Methods

A simple panel data framework :

(1)

(2)

Spatially weighted Lorenz curves and GINI indices for GDP and NTL

The rest of this presentation

1. NTL literature: Validation studies and new data

2. Data and methodology

Based on 1557 first-level subnational regions over 2013-2019...

... (VIIRS) NTL luminosity predicts

better across economies than over time

VIIRS and new DMSP products perform similarly ...

... only at the national level

VIIRS and new DMSP products perform similarly ...

... only at the national level

VIIRS excels DMSP for subnational analysis...

Caution!

For the 2013-2019 period, DMSP-like images were “simulated/derived” from VIIRS images

... only in developing countries

At the national level, NTL inequality correlates with GDP inequality in developing countries ...

... but this relationship fails in developed countries

For developing countries, at the subnational level,

VIIRS offers a more accurate view of economic inequality ...

... compared to new DMSP products

Concluding remarks

New technologies and better data, but old habits:

• Most of the economics literature keeps using outdated and imprecise NTL data: old DMPS data (1992-2013)
• Rapid advances and new data from the remote sensing literature:
• DMSP extended (Ghosh et al., 2021)
  
• DMSP like (Li et al., 2020)
  
• VIIRS (Elvidge et al., 2021)
• Many others
  

Opportunities for worldwide subnational studies:

• The dataset of Smits & Permanyer (2019) can foster the study of worldwide subnational development (at least at the first subnational level)
  

Concluding remarks

Summary of main results:

1. NTL luminosity better predicts economic differences between economies than changes over time
2. Both VIIRS and the new DMSP products perform similarly at the national level
   
3. VIIRS excels DMSP for subnational analysis in developing countries
4. At the national level, NTL inequality correlates with GDP inequality in developing countries, but this relationship fails to hold in developed countries.
5. Across subnational regions of developing countries, the VIIRS data offers a more accurate view of economic inequality compared to the new DMSP data.
   

Implications:

• Caution is needed when using NTL data: various new data products, cross-section vs time series analyses, and scale of analysis
• New geospatial technologies can advance our understanding of economic activity
  

Thank you very much!

Quantitative Regional and Computation Science Network

https://quarcs-lab.org

This research project was supported by JSPS KAKENHI (Grant Number 24K04884)