Stock Market Volatility Forecasting Using Ensemble Models
MetadataShow full item record
- Master Thesis 
Extensive research has been done within the field of finance to better predict future volatility and anticipate changes in financial market uncertainty. The advent of more advanced machine learning methods, such as artificial neural networks, has led to ground-breaking improvements to modeling capabilities across many fields and industries, including finance and volatility forecasting. These advances have led to rendering some of the previous state of the art models obsolete. Even though it has been established that artificial neural networks are capable of outperforming traditional finance forecasting models when it comes to volatility forecasting, it remains an open question whether a more advanced machine learning algorithm can benefit from incorporating the strengths of specialized volatility forecasting models. In this study, we seek to uncover whether traditional finance volatility forecasting models, such as GARCH type models, contain unique information that when combined with artificial neural networks can lead to more capable models and improved prediction accuracy. We will explore these effects by looking into S&P 500 one-day-ahead volatility using GARCH type models to generate volatility forecasts and include those into different artificial neural networks to measure improvements in forecasting capabilities. GARCH forecasts will be added into the different artificial neural networks in the form of two different types of ensemble models. One approach being a stacked ensemble, and the other an averaging ensemble. We find evidence to suggest that even though the GARCH type models consistently underperform compared to artificial neural networks, there is sufficient grounds to conclude that there is great potential in combining different volatility forecasting models to attain better volatility predictions.