Forecasting Financial Time Series Movements with Rough Sets and Fuzzy Rough Sets

Mariusz Podsiadło

Abstract

The feasibility of rough sets and fuzzy rough sets in building trend prediction models for financial time series is experimentally investigated. An adaptive time-weighted rule voting method is proposed, where the rule voting weight depends from the age of supporting data. The efficiency of the proposed models is verified and compared with the one of rough sets with equal-weighted rule voting algorithm, as well as support vector machines models. Aside of the standard classification accuracy measures, financial profit and loss backtesting using a sample market timing strategy was performed, and compared with the performance of the buy and hold strategy based on market data of multiple well known indices S&P500, DAX, and HSI. The experiments show that the proposed models using rough sets enhanced with the adaptive time-weighted rule voting, as well as fuzzy rough sets, delivered on pair or better performance than the used benchmark models, and the buy and hold strategy.
Diploma typeDoctor of Philosophy
Author Mariusz Podsiadło
Mariusz Podsiadło,,
-
Title in EnglishForecasting Financial Time Series Movements with Rough Sets and Fuzzy Rough Sets
Languageen angielski
Certifying UnitFaculty of Electronics and Information Technology (FEIT)
Disciplineinformation science / (technology domain) / (technological sciences)
Start date26-04-2016
Defense Date17-01-2017
Supervisor Henryk Rybiński (FEIT / IN)
Henryk Rybiński,,
- The Institute of Computer Science

External reviewers Andrzej Skowron
Andrzej Skowron,,
-

Alicja Wakulicz-Deja
Alicja Wakulicz-Deja,,
-
Pages122
Keywords in Englishartificial intelligence, rough sets, fuzzy rough sets, financial time series prediction
Abstract in EnglishThe feasibility of rough sets and fuzzy rough sets in building trend prediction models for financial time series is experimentally investigated. An adaptive time-weighted rule voting method is proposed, where the rule voting weight depends from the age of supporting data. The efficiency of the proposed models is verified and compared with the one of rough sets with equal-weighted rule voting algorithm, as well as support vector machines models. Aside of the standard classification accuracy measures, financial profit and loss backtesting using a sample market timing strategy was performed, and compared with the performance of the buy and hold strategy based on market data of multiple well known indices S&P500, DAX, and HSI. The experiments show that the proposed models using rough sets enhanced with the adaptive time-weighted rule voting, as well as fuzzy rough sets, delivered on pair or better performance than the used benchmark models, and the buy and hold strategy.
PKT classification4100
KBN classification28-Informatyka
EU classification80-30
Thesis file
Podsiadło_doktorat.pdf 1.53 MB
Reviews
Recenzja pracy Mariusza Podsiadło wykonana przez prof. dr hab. Andrzeja Skowrona of 03-10-2016
317.92 KB
Recenzja pracy Mateusza Podsiadło wykonana przez prof. dr hab. inż. Alicję Wakulicz-Deję of 01-09-2016
2.04 MB
Other files
streszczenie-Podsiadły.pdf 1.1 MB

Get link to the record

Back
Confirmation
Are you sure?