Teachable Machine을 이용한 쿠즈네츠 역U자 가설의 검증: 지방재정격차를 중심으로

This study measures the fiscal disparities of basic local governments using the Gini coefficient and tests the Kuznets inverted U-shaped hypothesis through a machine learning approach (Teachable Machine). By dividing revenue sources into local taxes, non-tax revenue, self-finance, grants, subsidies, and total revenue, the researchers calculated the Gini coefficients for each source and classified them into five patterns: the traditional Kuznets pattern (k1), the opposite pattern (k3), and three intermediate patterns (k2a, k2b, k2c). The results show diverse patterns across different local governments and revenue sources, with more than half in the “initial phase of the Kuznets curve (k2a), where inequality increases along with fiscal growth.” This finding indicates that one single inverted U-shaped curve cannot capture all of Korea’s local government finances, as each government is at a different stage of development. While some go through all phases of the Kuznets pattern (k1), there is also a group (around 10%) that experiences a resurgence of inequality in the later stages (k3). The study’s significance lies in highlighting the coexistence of multiple patterns rather than simply focusing on whether the Kuznets hypothesis is valid or not. Furthermore, it introduces the potential of leveraging CNNs (Convolutional Neural Networks) to analyze images instead of numerical data, opening new avenues of research in artificial intelligence. Lastly, it points to Piketty’s (2014) proposition that, in the long run, an “inverted S-shaped” pattern may emerge, suggesting a new research direction for future studies.