Exploring neural network models in understanding bilateral trade in APEC: A review of history and concepts

Machine Learning algorithms, such as the artificial neural networks, are acknowledged to outperform several econometric procedures in predictions. Machine learning becomes popular for doing operations that practically require more efficiency and accuracy, derived basically from the algorithm's powerful learning capability that can be applied in many fields, particularly in anticipating future event. Decision- and policy-makers have always been concerned in predicting the future in order to implement appropriate decision and policies. This necessitates the need to improve the accuracy of prediction and utilize more superior algorithms. This paper seeks to understand certain frameworks that can be used to improve the analysis and prediction of trade flows in APEC using neural networks. Discussions include the history of neural network development, the biological neuron, the artificial neuron, and the potential use of neural networks in trade analysis. This paper also compares the different estimation procedures of the gravity model - OLS, PPML, and GPML - with the neural network. The neural network estimation of the gravity equation is found superior over the other procedures in terms of explaining the variability of the dependent variable (export) around its mean, and in terms of the accuracy of predictions.