Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Machine Learning with FEARS index: does the inclusion of investor sentiment improve a machine learning model's ability to predict volatility?
The aim of this study is to determine whether the inclusion of investor sentiment allows machine learning methods to produce improved predictions of volatility in equity markets. Specifically, the investor sentiment measure is constructed as an index by using search volume data of different search t...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Department of Finance and Tax
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | The aim of this study is to determine whether the inclusion of investor sentiment allows machine learning methods to produce improved predictions of volatility in equity markets. Specifically, the investor sentiment measure is constructed as an index by using search volume data of different search terms obtained from Google Trends. The resulting Financial and Economic Attitudes Revealed by Search (FEARS) index is then utilised as a feature to forecast volatility via three different machine learning (ML) techniques, namely the Random Forest, Artificial Neural Network (ANN), and Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) methods. A consolidated dataset, where all G7 countries were combined into a single series, as well as an individualised dataset, where each individual country is analysed independently, were used to test the different ML methods' volatility forecasting ability. Our results show that, for the consolidated dataset, the inclusion of the FEARS index does not provide significant additional predictive power. However, through the individualised dataset, the FEARS index was shown in certain cases to provide greater predictive accuracy. Furthermore, it was observed that the LSTM-RNN outperformed the ANN and Random Forest methods, which indicates that our volatility prediction indeed benefits from elements of prior periods' volatilities as feature variables. |
|---|