USULAN METODE PREDIKSI HARGA MINYAK MENTAH WTI, BRENT, DAN DUBAI MENGGUNAKAN NOWCASTING

• Main Author: Magareta, Vabila
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68765
• Institution: Institut Teknologi Bandung

Indonesians still rely on crude oil as their main energy source on a daily basis. The annual crude oil consumption in Indonesia keeps growing, meanwhile domestic production continues to decline and forces Indonesia to change status to net importer. As a net importer country, Indonesia bought tons of crude oil from abroad, thus it must continue to monitor the prices of various types of crude oil on the market. This is because crude oil price acts as one of the inputs for policy making, such as the amount of subsidies for fuel and electricity allocated in the state budget, national income and expenses, and anticipation for inflation. So far, Indonesian government only uses existing crude oil price to make those policies. When in fact, the prediction of future commodity prices is a crucial input in the decision-making process of a country. Therefore, this study aims to develop a method of predicting crude oil prices, particularly WTI, Brent, and Dubai benchmark, using nowcasting. Nowcasting can provide agile predictions for quick decision-making purposes, especially amid conditions with great uncertainty, such as the COVID-19 pandemic. The research was conducted by collecting price data and predictor data in the form of Google Trends search index and USDX from January 1, 2017 to December 31, 2021. The prediction models used were multiple linear regression, decision tree, random forest, and gradient boost. The performance of the prediction model is measured using MAPE, where the model with the smallest MAPE will be selected as the best model for each crude oil benchmark. The results showed that the best regression model for WTI and Dubai benchmark is random forest with MAPE of 10.20% and 17.17%, respectively. Meanwhile, Brent has a MAPE value of 4.75% using the gradient boost model.