Empirical analysis of macroeconomic data using machine learning methods

• Main Authors: Lee, Raynor Shang Ze, Seah, Kendrick, Ho, Spencer Choon Hooi
• Other Authors: Wang Wenjie
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Social sciences::Economic theory::Macroeconomics; Social sciences::Economic theory::Public finance; Business::General::Economic and business aspects
• Online Access: https://hdl.handle.net/10356/162548
• Institution: Nanyang Technological University

Motivated by existing economic literature on applying machine learning methods to forecast highdimensional macroeconomic data, we endeavoured to apply these methodologies and models on data from Singapore and contribute to the lacking research in this area. Collecting data from multiple government open-data sources, we utilised 196 monthly macroeconomic time series across 9 macroeconomic categories from 2000 to 2020. The main goal of our study was to investigate the forecasting performance and accuracy of linear models and neural networks. We utilised the vector autoregression and diffusion index models alongside the neural network models. Our results indicate that amongst the linear class of models, the DILASSO model outperforms other linear models across all forecast horizons. For the non-linear class of models, our study found the neural network autoregression model to be the best overall model amongst not only against other neural network models, but across the linear models as well. Overall, the excellent forecasting performance and accuracy of models explored in our study adds to the growing possibility of augmenting current forecasting techniques in Singapore with machine learning methods.