Publicação
Neural networks applied to financial forecast
| Resumo: | The realm of bond yield forecasting holds substantial importance, influencing critical economic decisions. Recent research has focused on harnessing the predictive capabilities of advanced deep neural network architectures for this purpose, yet the integration of macroeconomic data remains relatively unexplored. This study aims to evaluate the performance of LSTM models in bond yield forecasting using macroeconomic data and to discern the most pivotal features within the constructed macroeconomic dataset. The experimental findings unveil five preeminent macroeconomic variables—namely interest rates, inflation, gross domestic product, total debt, money supply, and the intricate interrelation among bond yields of varying maturities—as key contributors to bond yield forecasting. The performance evaluation of the LSTM models demonstrates their aptitude for generating precise one-step predictions, particularly when fed solely with the target’s historical time-series data. Nonetheless, a noteworthy limitation surfaces concerning the multi-step forecasting approach employed within this study. This methodology struggles to encapsulate forthcoming volatility within the target variables. Consequently, an alternative methodology is proposed as a stepping stone for future research, aiming to overcome this limitation and enhance predictive accuracy. |
|---|---|
| Autores principais: | Galvão, Matheus Rigoni |
| Assunto: | Bond yields Forecasting LSTM Macroeconomics |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Instituto Politécnico de Bragança |
| Idioma: | inglês |
| Origem: | Biblioteca Digital do IPB |
| Resumo: | The realm of bond yield forecasting holds substantial importance, influencing critical economic decisions. Recent research has focused on harnessing the predictive capabilities of advanced deep neural network architectures for this purpose, yet the integration of macroeconomic data remains relatively unexplored. This study aims to evaluate the performance of LSTM models in bond yield forecasting using macroeconomic data and to discern the most pivotal features within the constructed macroeconomic dataset. The experimental findings unveil five preeminent macroeconomic variables—namely interest rates, inflation, gross domestic product, total debt, money supply, and the intricate interrelation among bond yields of varying maturities—as key contributors to bond yield forecasting. The performance evaluation of the LSTM models demonstrates their aptitude for generating precise one-step predictions, particularly when fed solely with the target’s historical time-series data. Nonetheless, a noteworthy limitation surfaces concerning the multi-step forecasting approach employed within this study. This methodology struggles to encapsulate forthcoming volatility within the target variables. Consequently, an alternative methodology is proposed as a stepping stone for future research, aiming to overcome this limitation and enhance predictive accuracy. |
|---|