Penerapan Algoritma Bayesian Regulation untuk Estimasi Posisi Cadangan Devisa Indonesia

Abstract

Foreign exchange reserves need to be predicted because it is a significant monetary indicator to show the strength or weakness of a country's economic fundamentals. Therefore, the purpose of this study is to estimate the position of Indonesia's foreign exchange reserves at the end of 2022 and 2023 so that the government has benchmarks and information in determining the right economic policy so that the position of foreign exchange reserves remains stable. The estimation algorithm used in this study is the Bayesian Regulation algorithm, one of the Artificial Neural Network algorithms. The research data used is data on the position of Indonesia's foreign exchange reserves (US$ million) obtained from the economic reports of Bank Indonesia. This research will be analyzed using three network architecture models 4-9-1, 4-18-1, and 4-27-1. Based on the analysis of the three models used, the results show that the 4-27-1 model is the best because it has a lower Mean Square Error (MSE) value of 0.00203297. Another benchmark is seen from more minor epochs (iterations) and faster times than the other two models, even though they both produce a 100% accuracy rate. Thus, it can be concluded that the Bayesian Regulation algorithm is good enough to estimate the position of foreign exchange reserves using the 4-27-1 model. Based on the prediction results, the part of Indonesia's foreign exchange reserves at the end of 2022 and 2023 slightly decreased compared to 2021.