머신러닝 방법을 이용한 수출의 조업일수 산정

초록

This study proposes a methodology to estimate the special-day effect in Korea’s 10-day export statistics. Conventional approaches assigning fixed weights to weekdays, Saturdays, and holidays fail to capture time-varying factors and working-day composition. In this study, special-day effects are estimated using regression and Random Forest models after detrending log-transformed exports. The analysis shows that holiday effects (excluding the Lunar New Year and Chuseok) differ from those of Sundays, and that the Saturday effect is not equal to half of the Sunday effect. Intra-month seasonality, with exports concentrated near month-end, is also statistically significant. The proposed model decomposes special-day effects into fixed and time-varying components: regression captures fixed effects, while Random Forest identifies interactions among special days and temporal variations. The results indicate that the negative effects of Saturdays and holidays have strengthened over time. This study enhances the precision of export analysis and provides an empirical basis for improving conventional adjustments.

키워드