Skip to main content

Shizuka Inoue

Ph.D. Candidate
Columbia University

RESEARCH

Work in Progress

2025

Rural Convergence, Structural Transformation, and Immigration: Evidence from the Post-WWII Japanese Repatriation

This paper examines the impact of the post-WWII Japanese repatriation on rural convergence and structural transformation.

2025

Martyrs, Morale, and Militarism: The Political Impact of Devastation and Slaughter

with David Weinstein and Atsushi Yamagishi

This paper examines the impact of the war casualties on the political behavior of the Japanese population.

BACKGROUND

2021—Present

Columbia University

Ph.D. in Economics

2019—2021

Research Assistant - Yale University

2019

Columbia University

B.A. in Economics

PROJECTS

Quantitative Spatial Model (QSM) for Japan

An implementation of the Allen and Arkolakis (2014) Quantitative Spatial Model for analyzing spatial economic patterns across Japanese cities from 1975 to 2010. This project examines population dynamics, wage patterns, and the spatial distribution of economic activity in Japan.

GitHub

Key Features

  • Implemented the Allen-Arkolakis model for Japanese city-level data
  • Calibrated model parameters using historical economic data
  • Visualized spatial economic patterns and their evolution over time
  • Conducted counterfactual analyses to assess policy impacts

Methodology

Based on the equilibrium conditions from Allen and Arkolakis (2014), this model processes city-level data to calculate productivity and amenity values. The analysis provides insights into spatial economic patterns and their changes over time in Japan.

Japanese Occupation Sector Classification

A machine learning system that classifies Japanese occupation titles into economic sectors (Primary, Secondary, Tertiary) using a combination of pattern matching and BERT models. This tool helps researchers standardize occupational data for economic analysis.

GitHub

Key Features

  • Multi-level classification into main sectors, subcategories, and detailed occupation groups
  • Specialized handling of Japanese characters and text normalization
  • Pattern-based initial classification with hierarchical category assignment
  • BERT model fine-tuning with cross-validation and ensemble learning
  • Confidence scoring and unknown category handling

Methodology

The classification system uses a two-stage approach: first applying pattern matching for initial categorization, then employing a fine-tuned Japanese BERT model (cl-tohoku/bert-base-japanese-v3) with 5-fold cross-validation for more nuanced classification. Confidence scores are calculated through ensemble averaging across folds.

Japanese Administrative Boundary Crosswalk

A comprehensive dataset and tools for tracking changes in Japanese administrative boundaries over time. This project creates geographic crosswalks between different years of Japanese administrative boundaries

GitHub

Key Features

  • Creates crosswalks between different years of administrative boundaries
  • Calculates area-based weights for accurate data transformation
  • Handles boundary changes and mergers systematically

Methodology

Following Eckert et al. (2020), the tool creates weights based on geographic area overlap between historical and reference boundaries. For each historical year, weights are calculated as the ratio of intersection area to total historical city area. These weights enable accurate transformation of historical data to modern administrative units.