Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI
Our paper investigates the performance of two machine learning models, namely Support Vector Machine (SVM) and Artificial Neural Network (ANN) in predicting the direction of movement of Stock Exchange of Thailand (SET) index and Philippines Stock Exchange Composite Inde...
| Main Author: | |
|---|---|
| Format: | Dissertation (University of Nottingham only) |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | https://eprints.nottingham.ac.uk/63254/ |
| _version_ | 1848800008543928320 |
|---|---|
| author | Chin, Jern Tat |
| author_facet | Chin, Jern Tat |
| author_sort | Chin, Jern Tat |
| building | Nottingham Research Data Repository |
| collection | Online Access |
| description | Our paper investigates the performance of two machine learning models, namely Support Vector Machine (SVM) and Artificial Neural Network (ANN) in predicting the direction of movement of Stock Exchange of Thailand (SET) index and Philippines Stock Exchange Composite Index (PSEI). We investigate the predictive abilities of two methods of forecasting stock prices, namely fundamental and technical analysis. Therefore, for technical analysis, ten technical indicators of various input window length are selected a s input features for our model. Input window length are the number of periods in the past used to calculate technical indicators. For fundamental analysis, 12 fundamental variables are selected as input features.
The dataset covers a period from Jan/Feb entire datapoints) and test dataset ( ≈ 2008 until Dec 2018 and is divided into training ( ≈ 60% of 40% of entire datapoints). The study is carried out using R programming. We used several performance measures to address our research questions: prediction accu racy, onetailed ttest, fscore, PesaranTimmerman test, total return, unpaired ttest and binomial test. Research results show that prediction accuracy generally exceeds 50% which argues against random walk theory and that up ward and down , which states that stock price movement ward is random and unpredictable movements of stock prices are equally likely .
We find that optimal input window length for technical indicators are market specific and that shorter ,term indicators have better predictive ability f or PSEI while longer Based on Pesaran-- term indicators are better for SET. Timmerman test, we find that our models are generally capable of predicting stock price direction which demonstrate the usefulness of machine learning in time series forecasting. We also find that our models generally outperform buy and hold strategy which demonstrates its market timing and yield enhancement potential for investors.
Based on f algoscore, we find that SVM outperforms ANN in forecasting ability. This could be due its rithm which seeks to minimize an upper bound of generalisation error while ANN seeks to minimise training error. Therefore, SVM can generalise better on test dataset. that fundamental variables have better predictive ability than tech financial and economic variables have more information content than past price Additionally, we find nical indicators. This implies that s and trading volumes. We also conducted binomial test on percentage of buy/sell signals generating profitable trades and find that buy signal s generated are generally useful in predicting price movement while sell signals are not useful. Finally, we identify and discuss the most important predictor variables for explaining our outcome variables in ANN model. |
| first_indexed | 2025-11-14T20:44:44Z |
| format | Dissertation (University of Nottingham only) |
| id | nottingham-63254 |
| institution | University of Nottingham Malaysia Campus |
| institution_category | Local University |
| language | English |
| last_indexed | 2025-11-14T20:44:44Z |
| publishDate | 2021 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | nottingham-632542021-04-16T03:41:52Z https://eprints.nottingham.ac.uk/63254/ Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI Chin, Jern Tat Our paper investigates the performance of two machine learning models, namely Support Vector Machine (SVM) and Artificial Neural Network (ANN) in predicting the direction of movement of Stock Exchange of Thailand (SET) index and Philippines Stock Exchange Composite Index (PSEI). We investigate the predictive abilities of two methods of forecasting stock prices, namely fundamental and technical analysis. Therefore, for technical analysis, ten technical indicators of various input window length are selected a s input features for our model. Input window length are the number of periods in the past used to calculate technical indicators. For fundamental analysis, 12 fundamental variables are selected as input features. The dataset covers a period from Jan/Feb entire datapoints) and test dataset ( ≈ 2008 until Dec 2018 and is divided into training ( ≈ 60% of 40% of entire datapoints). The study is carried out using R programming. We used several performance measures to address our research questions: prediction accu racy, onetailed ttest, fscore, PesaranTimmerman test, total return, unpaired ttest and binomial test. Research results show that prediction accuracy generally exceeds 50% which argues against random walk theory and that up ward and down , which states that stock price movement ward is random and unpredictable movements of stock prices are equally likely . We find that optimal input window length for technical indicators are market specific and that shorter ,term indicators have better predictive ability f or PSEI while longer Based on Pesaran-- term indicators are better for SET. Timmerman test, we find that our models are generally capable of predicting stock price direction which demonstrate the usefulness of machine learning in time series forecasting. We also find that our models generally outperform buy and hold strategy which demonstrates its market timing and yield enhancement potential for investors. Based on f algoscore, we find that SVM outperforms ANN in forecasting ability. This could be due its rithm which seeks to minimize an upper bound of generalisation error while ANN seeks to minimise training error. Therefore, SVM can generalise better on test dataset. that fundamental variables have better predictive ability than tech financial and economic variables have more information content than past price Additionally, we find nical indicators. This implies that s and trading volumes. We also conducted binomial test on percentage of buy/sell signals generating profitable trades and find that buy signal s generated are generally useful in predicting price movement while sell signals are not useful. Finally, we identify and discuss the most important predictor variables for explaining our outcome variables in ANN model. 2021-02-24 Dissertation (University of Nottingham only) NonPeerReviewed application/pdf en https://eprints.nottingham.ac.uk/63254/1/MACHINE%20LEARNING%20TECHNIQUES%20%28SUPPORT%20VECTOR%20MACHINE%20AND%20ARTIFICIAL%20NEURAL%20NETWORK%29%20FOR%20FUNDAMENTAL%20AND%20TECHNICAL%20ANALYSIS%20A%20STUDY%20ON%20SET%20AND%20PSEI.pdf Chin, Jern Tat (2021) Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI. [Dissertation (University of Nottingham only)] support vector machine artificial neural network fundamental analysis technical analysis SET PSEI |
| spellingShingle | support vector machine artificial neural network fundamental analysis technical analysis SET PSEI Chin, Jern Tat Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI |
| title | Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI |
| title_full | Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI |
| title_fullStr | Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI |
| title_full_unstemmed | Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI |
| title_short | Machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on SET and PSEI |
| title_sort | machine learning techniques (support vector machine and artificial neural network) for fundamental and technical analysis: a study on set and psei |
| topic | support vector machine artificial neural network fundamental analysis technical analysis SET PSEI |
| url | https://eprints.nottingham.ac.uk/63254/ |