정규화 이동 창 회귀를 활용한 무역 개방도와 경제성장의 시변적 관계 분석

초록

This study applies an adaptive shrinkage estimator combining L1 and L2 penalties within a ten-year rolling-window framework to analyze the time-varying relationship between trade openness and economic growth. Adaptive weights derived from a first-stage fit limit overfitting in short and noisy windows. Estimating 5,519 country-period coefficients from a World Bank WDI panel of 126 countries over 64 years (1961–2024), we find that the trade-growth relationship varies substantially across periods and countries: the mean coefficient rises markedly from the 2010s onward, with pronounced heterogeneity across continents and individual economies. Given the single-regressor specification, the coefficients are conditional associations rather than causal effects. The findings show that combining regularized machine learning with rolling-window estimation effectively extracts time-varying relationships from large-scale panel data.

키워드