Forecasting Model of Indonesia's Oil & Gas and Non-Oil & Gas Export Value using Var and LSTM Methods

Abstract

As a country with abundant natural resources in the form of mineral and non-mineral products, Indonesia is characterized by its ability to fulfill domestic and foreign needs through export activities categorized into two commodities: oil and gas and non-oil and gas. Export activities are an indicator of the country's economic growth that often fluctuates in value, and these conditions are fundamentally caused by a decrease in production quantity and the instability of the global economic climate. The strategy to overcome these problems is to create a forecasting model. This research aims to develop a forecasting model using time series analysis methods, including vector autoregressive (VAR) and long short-term memory (LSTM) methods based on oil and non-oil and gas value parameters. The results of the Granger causality test stated that the values of oil and gas and non-oil and gas affect each other. The VAR model with the optimum lag produced by the Akaike Information Criterion (AIC) test obtained an accuracy value of MAPE oil gas and non-oil and gas of 18.4% and 32.1%, respectively. LSTM generates the best model with a MAPE value of 6,23% for oil gas and 8,18% for non-oil and gas.

Keywords

• forecasting model
• export value of oil and gas
• non-oil and gas export
• deep learning
• machine learning
• var method
• lstm method.