Publication Date

5-2022

Date of Final Oral Examination (Defense)

5-7-2022

Type of Culminating Activity

Thesis

Degree Title

Master of Science in Economics

Department

Economics

Supervisory Committee Chair

Michail Fragkias, Ph.D.

Supervisory Committee Member

Jayash Paudel, Ph.D.

Supervisory Committee Member

Samia Islam, Ph.D.

Abstract

Gross metropolitan product (GMP) is one the most critical indicators for determining a metropolitan area’s economic performance. While GMP data currently exists for major cities in the US and OECD countries, the rest of the world is a blind spot. This study aims at estimating the GMP of 1289 cities in non-US and OECD countries, where no official city-level statistics are produced. We perform this estimation through multiple machine learning models, using night-time lights satellite imagery, and other publicly available data. We analyze eight spatial databases and four cross-sectional datasets and derive a feature vector of covariates through various techniques, i.e., downscaling and bootstrap. We specify OLS, Ridge, Lasso, Elastic Net, and Random Forest models, out of which Random Forest generated the most accurate results with 0.3 RMSE for out-of-sample predictions. With this methodology, we produced the first existing data set that groups the 1298 cities into 20 quantiles, with the first quantile denoting the lowest five percent regarding estimated income and the twentieth quantile denoting the highest five percent regarding the estimated economic product.

DOI

https://doi.org/10.18122/td.1934.boisestate

Share

COinS