Applications of machine learning in portfolio management
Advances in predicting market returns using machine learning have seen tremendous growth which can be attributed to veritable data mining and computational capabilities. However, a large portion of this literature focuses on short-term effects by evaluating models solely on their prediction error ra...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project (FYP) |
| Language: | English |
| Published: |
Nanyang Technological University
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/153308 |
| _version_ | 1826111282585010176 |
|---|---|
| author | Numair Fazili |
| author2 | Yeo Chai Kiat |
| author_facet | Yeo Chai Kiat Numair Fazili |
| author_sort | Numair Fazili |
| collection | NTU |
| description | Advances in predicting market returns using machine learning have seen tremendous growth which can be attributed to veritable data mining and computational capabilities. However, a large portion of this literature focuses on short-term effects by evaluating models solely on their prediction error rates. Long-term consequences of these predictions across a portfolio are not assessed and as such the compounding effects remain unexplored.
We propose a novel framework towards implementing the Constant Proportion Portfolio Insurance (CPPI) strategy aided by predictive returns. This work is inspired by the advances in machine learning frameworks for financial markets in conjunction with gaps within the existing CPPI system.
Technical analysis was used to augment the standard asset price/volume data by introducing 14 additional features associated with momentum, trend, volume, and volatility measures. These features were used in training Support Vector Regression (SVR), Facebook Prophet (FBP), and Long Short Term Memory (LSTM) models. The evaluations were conducted for six years from 2015 to 2020 using RMSE, MAE, and MAPE as error metrics. The performance analysis indicates that LSTM network with a 7-day walk forward validation serves as the ideal architecture for attaining the lowest error rate.
Predictive returns generated from machine learning algorithms were subsequently used inside a modified CPPI framework. Our empirical analysis shows that this approach outperforms the existing CPPI towards achieving consistently high portfolio valuations. We observe that all three machine learning architectures outperform the standard CPPI with the LSTM model delivering optimum portfolio performance attributed to a 13.79% increase in Sharpe Ratio.
Keywords: CPPI, Machine Learning, Predictive returns, Technical Analysis, Sharpe Ratio. |
| first_indexed | 2024-10-01T02:48:08Z |
| format | Final Year Project (FYP) |
| id | ntu-10356/153308 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T02:48:08Z |
| publishDate | 2021 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1533082021-11-17T05:42:06Z Applications of machine learning in portfolio management Numair Fazili Yeo Chai Kiat School of Computer Science and Engineering ASCKYEO@ntu.edu.sg Engineering::Computer science and engineering Advances in predicting market returns using machine learning have seen tremendous growth which can be attributed to veritable data mining and computational capabilities. However, a large portion of this literature focuses on short-term effects by evaluating models solely on their prediction error rates. Long-term consequences of these predictions across a portfolio are not assessed and as such the compounding effects remain unexplored. We propose a novel framework towards implementing the Constant Proportion Portfolio Insurance (CPPI) strategy aided by predictive returns. This work is inspired by the advances in machine learning frameworks for financial markets in conjunction with gaps within the existing CPPI system. Technical analysis was used to augment the standard asset price/volume data by introducing 14 additional features associated with momentum, trend, volume, and volatility measures. These features were used in training Support Vector Regression (SVR), Facebook Prophet (FBP), and Long Short Term Memory (LSTM) models. The evaluations were conducted for six years from 2015 to 2020 using RMSE, MAE, and MAPE as error metrics. The performance analysis indicates that LSTM network with a 7-day walk forward validation serves as the ideal architecture for attaining the lowest error rate. Predictive returns generated from machine learning algorithms were subsequently used inside a modified CPPI framework. Our empirical analysis shows that this approach outperforms the existing CPPI towards achieving consistently high portfolio valuations. We observe that all three machine learning architectures outperform the standard CPPI with the LSTM model delivering optimum portfolio performance attributed to a 13.79% increase in Sharpe Ratio. Keywords: CPPI, Machine Learning, Predictive returns, Technical Analysis, Sharpe Ratio. Bachelor of Engineering (Computer Science) 2021-11-17T05:42:05Z 2021-11-17T05:42:05Z 2021 Final Year Project (FYP) Numair Fazili (2021). Applications of machine learning in portfolio management. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/153308 https://hdl.handle.net/10356/153308 en SCSE20-0798 application/pdf Nanyang Technological University |
| spellingShingle | Engineering::Computer science and engineering Numair Fazili Applications of machine learning in portfolio management |
| title | Applications of machine learning in portfolio management |
| title_full | Applications of machine learning in portfolio management |
| title_fullStr | Applications of machine learning in portfolio management |
| title_full_unstemmed | Applications of machine learning in portfolio management |
| title_short | Applications of machine learning in portfolio management |
| title_sort | applications of machine learning in portfolio management |
| topic | Engineering::Computer science and engineering |
| url | https://hdl.handle.net/10356/153308 |
| work_keys_str_mv | AT numairfazili applicationsofmachinelearninginportfoliomanagement |