with Hanhyung Pyo

This paper examines how the quality of government spending changes at the end of the fiscal year, using unique data from small and medium-sized enterprise (SME) financing programs in Korea. The quality of SME finance spending is assessed based on the average sales growth rates of financed firms, measured at the program-month level. A theoretical model suggests that when budgets cannot be carried over to the next fiscal year and early spending requirements are enforced, the government may shift toward lower-risk investments at year-end—leading to lower returns or quality. Empirical results support the model's predictions: not only does a risk-return trade-off exist in SME finance spending, but the level of risk, measured by the variance in sales growth rates of financed firms, also declines toward the end of the fiscal year.

This paper investigates overconfidence among civil servants. Using survey data from civil servants and citizens in Korea, overconfidence is measured in two ways: (1) the gap between civil servants’ perception of their own productivity and their perception of their organization’s productivity (overplacement) and (2) the gap between civil servants’ perception of their local citizens’ customer satisfaction level and the citizens’ actual customer satisfaction level (overestimation). On average, civil servants exhibit overconfidence in both overplacement and overestimation. Moreover, public service motivation (PSM) is positively associated with overconfidence. To the extent that overconfidence is an undesirable cognitive bias, this finding suggests a potential dark side of PSM. Lastly, overconfident civil servants are more likely to prefer indirect policy instruments, such as vouchers, over direct instruments, such as regulations. Thus, civil servants’ overconfidence can have significant consequences for policy design and outcomes, particularly in the context of the New Public Management (NPM) reforms.

with Saebom Lee

The COVID-19 pandemic served as a global stress test for public administration, revealing substantial disparities in local government capacity. While national strategies often dominate scholarly and public discourse, the operational effectiveness of crisis management depends critically on local implementation. This study investigates the determinants of crisis management effectiveness across 208 South Korean municipalities during the early phase of the pandemic (February–November 2020). To address endogeneity concerns inherent in COVID-19 case data, we introduce a novel behavioral proxy: the growth rate of varicella (chickenpox).

Methodologically, we propose a machine learning–guided specification search, a hybrid framework that integrates high-dimensional variable selection with econometric interpretability. We first apply Lasso regularization to screen more than 80 administrative, fiscal, demographic, and political variables. We then use Random Forest and Gradient Boosting models as predictive benchmarks to diagnose nonlinearity and interaction effects overlooked by linear specifications. High-importance features identified from these benchmarks are incorporated into a refined Ordinary Least Squares (OLS) model through targeted interaction terms.

The analysis yields two main findings. First, we identify an administrative surge effect: the number of civil servants has no statistically significant effect in isolation but becomes  effective as COVID-19 cases accumulate, providing empirical support for Landau’s redundancy theory. Second, we detect a spatial friction effect, whereby administrative effectiveness diminishes in geographically expansive jurisdictions. Together, these findings challenge the “lean government” paradigm and suggest that crisis resilience requires elastic surge capacity and differentiated governance strategies to address spatial diseconomies.