Forecasting the Philippines’ GDP growth using long short-term memory neural network regression and mixed-data sampling regression models

• Main Authors: Arsua, Andre Millard M, Azucena, Raphael Matthew D
• Format: text
• Language: English
• Published: Animo Repository 2022
• Subjects: Gross domestic product > Philippines; Forecasting; Mathematics
• Online Access: https://animorepository.dlsu.edu.ph/etdb_math/3; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1003&context=etdb_math
• Institution: De La Salle University

Having better forecasts is crucial in the Philippines’ state of economic recovery. The study intends to forecast the Philippines’ GDP growth rate from 2011 to 2021 using two emerging methods used for mixed-frequency data: Mixed-Data Sampling (MIDAS) Regression and Long Short-Term Memory (LSTM) Neural Network Regression. These models were applied to the recommended new set of Leading Economic Indicators (LEI) for forecasting the state of the Philippine economy, which were obtained from PSA and BSP. The results were compared using RMSE, MAE, MAPE, SMAPE, and other metrics for forecasting accuracy to determine the better model. Among the MIDAS models, the Exponential Almon Weight MIDAS performed best in all fit statistics in the study while for LSTM models, a model with an Adam-based optimization function and a Median-based fill function for missing values performed the best in overall forecast performance and ability to follow trend. It was found that based on the RMSE criterion, MIDAS and LSTM were able to outperform the currently used Dynamic Factor Model (DFM), and the Exponential Almon Weight MIDAS model is the superior model for forecasting the Philippines’ GDP growth rate in a pandemic given the new set of macroeconomic indicators for LEI.