کد مقاله کد نشریه سال انتشار مقاله انگلیسی نسخه تمام متن
494961 862810 2015 14 صفحه PDF دانلود رایگان
عنوان انگلیسی مقاله ISI
MGP-INTACTSKY: Multitree Genetic Programming-based learning of INTerpretable and ACcurate TSK sYstems for dynamic portfolio trading
موضوعات مرتبط
مهندسی و علوم پایه مهندسی کامپیوتر نرم افزارهای علوم کامپیوتر
پیش نمایش صفحه اول مقاله
MGP-INTACTSKY: Multitree Genetic Programming-based learning of INTerpretable and ACcurate TSK sYstems for dynamic portfolio trading
چکیده انگلیسی


• MGP-INTACTSKY is a fuzzy rule based system for dynamic portfolio trading.
• Multitree Genetic Programming (MGP) is applied to learn the TSK fuzzy rule bases.
• The new TSK structure leads to a more interpretable and accurate system.
• Input variables are technical indices selected by stepwise regression analysis.
• The results are based on testing of the model on one emerging and two mature markets.

In this paper, a Multitree Genetic Programming-based method is developed to learn an INTerpretable and ACcurate Takagi-Sugeno-Kang (TSK) fuzzy rule based sYstem (MGP-INTACTSKY) for dynamic portfolio trading. The MGP-INTACTSKY utilizes a TSK model with a new structure to develop a more interpretable and accurate system for dynamic portfolio trading. In the new structure of TSK, disjunctive normal form rules with variable structured consequent parts are developed in which the absence of some input variables is allowed. Input variables are the most influential technical indices which are selected by stepwise regression analysis. The technical indices are computed using wavelet transformed stock price series to eliminate the noise. The proposed system directly induces the preferred portfolio weights from the stock's technical indices through time. Here, genetic programming with the multitree structure is applied to learn the TSK fuzzy rule bases with the Pittsburgh approach. With this approach, the correlation of different stocks is properly considered during the evolutionary process. To evaluate the performance of the MGP-INTACTSKY for portfolio trading, the proposed model is implemented on the Tehran Stock Exchange as an emerging market as well as Toronto and Frankfurt Stock Exchanges as two mature markets. The experimental results show that the proposed model outperforms other methods such as the momentum strategy, the multitree genetic programming-based crisp system, the genetic algorithm-based first order TSK system, the buy and hold approach and the market's main index in terms of accuracy and interpretability.

Figure optionsDownload as PowerPoint slide

ناشر
Database: Elsevier - ScienceDirect (ساینس دایرکت)
Journal: Applied Soft Computing - Volume 34, September 2015, Pages 449–462
نویسندگان
, , ,