BooKang Seol

Infrastructure investments generate widely varying returns across rural communities, yet the sources of this heterogeneity remain poorly understood. This paper argues that the definition of "effective infrastructure" is endogenous to the economic strategy that a community's social structure fosters. Using a triple-difference design that leverages the 1973 construction of the Namhae Bridge in Korea, I show that identical market access catalysed two distinct, yet equally successful, development paths. Cohesive clan-based villages leveraged their capacity for collective action to modernise coordination-intensive rice production, achieving 21% productivity gains. Socially fractionalised villages, facing coordination challenges, instead pursued diversified portfolios of cash crops. Both strategies yielded comparable household income gains, though diversifying villages achieved superior population retention. These findings demonstrate that infrastructure creates value not through uniform mechanisms, but by relieving the specific constraints that bind each community's feasible development path.

Why do firms in developing countries struggle to scale? Two leading explanations — resource misallocation and market access constraints — point to different policy responses, yet disentangling them empirically has proved difficult. This paper leverages the unique market access expansion experienced by Indian IT firms during the Y2K period to test which constraint bound firms before the sector's rapid take-off. The worldwide scramble to remediate legacy software in the late 1990s generated a massive, temporary spike in outsourcing demand, compelling Western firms to form new partnerships with Indian providers and dramatically lowering the sunk costs of first export matches. Employing semantic-based digitisation using generative AI, I extract firm-level data from Indian IT company directories and combine them with panel data on listed firms. I find that significant misallocation existed in the Indian IT sector prior to the Y2K event, and that firms exposed to the Y2K-induced market access expansion demonstrated substantial catch-up in productivity. Many firms maintained their improved productivity levels even after the Y2K shock had abruptly ended, suggesting a persistent shift to a higher productivity frontier. I explore potential mechanisms behind this sustained growth, including overcoming sunk costs of partner acquisition, achieving economies of scale, and integrating into global value chains. These findings suggest that market access expansion, albeit temporary, can catalyse durable industrial transformation in developing economies — not by correcting misallocation directly, but by clearing the fixed-cost hurdles that kept productive firms from reaching the market.

Does building more schools improve the quality of education? I study this question using India's Rashtriya Madhyamik Shiksha Abhiyan (RMSA), a national programme that constructed over ten thousand secondary schools between 2009 and 2016. Combining administrative data on the near-universe of Indian secondary schools with variation in the timing and location of new construction, I find that new schools increase aggregate enrolment but do not improve learning outcomes. While the absolute number of students taking and passing the nationally administered board exam rises mechanically, pass rates in existing schools do not improve — and may decline. The enrolment gains are driven primarily by public schools, while new private schools generate larger substitution effects, drawing students away from existing schools without expanding overall access. These results suggest that school construction alone is insufficient to raise educational quality, and that expanding access without complementary investments in teaching and resources may dilute rather than strengthen the system.

We construct the first comprehensive subnational economic database covering South Korea's high-growth period (1960–1989). The project integrates three complementary data streams: (1) approximately two million pages of municipal statistical yearbooks — published annually by every county government but never systematically compiled — digitised using a custom AI-OCR pipeline for mixed Hangul/Hanja archival tables and harmonised into a township-year panel of ~3,500 units covering demographics, agriculture, industry, infrastructure, public finance, and education; (2) topographic maps and aerial photographs, linked to the panel on a 1km×1km standardised grid that resolves the boundary changes which have made consistent spatial analysis of this period impossible; and (3) oral history and local encyclopedia materials, processed via LLM-based text analysis to integrate qualitative accounts with the quantitative record. The resulting database will be released as a public resource with full documentation, an interactive data explorer, and replication code, with the goal of lowering the barriers to empirical research on one of modern history's most important development episodes.

Environmental degradation is often viewed as an inevitable cost of early industrialisation. This paper challenges that assumption by examining how South Korea restored 2.8 million hectares of degraded forests while growing GDP per capita 20-fold between 1962 and 1987. We hypothesise that reforestation created complementary infrastructure that amplified returns to other investments through three channels: disaster mitigation that enabled adoption of higher-risk, higher-return technologies; watershed stabilisation that increased agricultural productivity; and reduced fuelwood collection burdens that improved school attendance. Using quasi-random variation in reforestation timing across townships driven by bureaucratic constraints in the availability of forestry-trained personnel, we provide causal evidence that environmental restoration can accelerate rather than constrain structural transformation.

Subnational data for South Korea's high-growth period barely exists. Municipal statistical yearbooks were published annually by every county government, but surviving copies are fragmented across provincial archives, recorded in mixed Hangul and classical Chinese scripts, and scrambled by two major boundary reorganisations (1963, 1973) and dozens of smaller changes. We digitise approximately two million pages of these archival tables using a custom AI-OCR pipeline fine-tuned for mixed Hangul/Hanja scripts — documents that conventional OCR cannot process. The pipeline combines context-aware layout parsing, structured output extraction, and row-level cross-validation. We then harmonise variable definitions across editions and municipalities and assign time-consistent geographic identifiers (miracle_id) that track every township through the boundary changes, allowing researchers to follow the same unit across three decades.

The yearbooks form the backbone, covering demographics, agriculture, industry, infrastructure, public finance, and education. Additional archival layers extend the panel into domains the yearbooks do not reach: Korea Forest Service spatial archives (from 1910), the 1972 New Village Survey (새마을총람, ~35,000 villages), foreign loan records, and expressway construction logs. The data infrastructure is designed to generalise to other growth miracle economies with comparable subnational statistical traditions, including Taiwan and post-war Japan. Pilot data release in late 2026 covers the Gyeongbu Expressway corridor (~400 townships) with Core Keys, Demographics, and Agriculture modules. A diagnostic framework under development with KDI will use the historical relationship between institutional endowments and programme returns to support applied policy assessment.