A Comparative Export Prediction Study of Korea and Japan Using Semi-parametric Panel Neural network
Roh Jaewhak · 무역금융보험연구(구 무역보험연구) · 2021
Purpose : To compare the predictability of gravity model based on panel data, the two methodologies are used and compared. One comes from the traditional panel individual fixed effects and the alternative methodology comes from semi-parametric panel neural network. Research design, data, methodology : The exporting cases of Korea and Japan, for 30 years of data with 247 countries of export partners, were analyzed. Two scenarios dividing the learning period into 20 years with 10 years’ prediction period and the 25 year learning period with 5 years’ prediction periods are performed and compared. Results : The semi-parametric panel artificial neural network outperformed than the traditional panel fixed effect method in 7 out of 8 cases in terms of MSE. Conclusions : The results showed the superiority of semi-parametric neural networks in the analysis and forecasting based on panel data compared to traditional fixed effect panel analysis.