DESIGN OF FOOD COMMODITY PRICE PREDICTION FRAMEWORK IN INDONESIA USING THE NOWCASTING METHOD

• Main Author: Ammar Erdianto, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55995
• Institution: Institut Teknologi Bandung

The imbalance between supply and demand for food commodities causes price fluctuations. Commodity prices is one of the inputs in formulating government policies to deal with price fluctuations in food commodities. The government requires input in the form of data on estimates of future prices in addition to past prices. High uncertainty in the conditions of the COVID-19 pandemic requires the government to make real-time decisions, including forecasting food commodity prices. A short-term forecasting framework was designed for the prices of ten commodities using the nowcasting method based on these problems. The data used is the daily price of ten commodities on a national scale and the keyword search index from Google Trends. Commodity price data then goes through initial processing stages, such as missing data imputation and data aggregation. The predictive modeling of food commodity prices is carried out with the keyword search index predictor variables from Google Trends using four models: linear regression, random forest, gradient boosting, and regression tree. Modeling is also carried out to determine the lag value between predictor variables and target variables. Applying the proposed food commodity price prediction framework provides good forecasting results based on the performance measure of the MAPE value on the testing data, namely, seven commodities have values below 10%. In addition, three commodities have values between 10% and 20%. The proposed framework can be a new tool for the government in price forecasting. It can be used as additional input for policy formulation related to handling food commodity price fluctuations. The framework's application also results in better performance for commodities with high price fluctuations and approximately the same results for commodities with low fluctuations compared to the time-series method.