Online Non-linear Prediction of Financial Time Series Patterns

Simple item page
dc.contributor.advisorGebbie, Timothy
dc.contributor.authorda Costa, Joel
dc.date.accessioned2020-09-11T09:28:42Z
dc.date.available2020-09-11T09:28:42Z
dc.date.issued2020
dc.date.updated2020-09-11T09:28:28Z
dc.description.abstractWe consider a mechanistic non-linear machine learning approach to learning signals in financial time series data. A modularised and decoupled algorithm framework is established and is proven on daily sampled closing time-series data for JSE equity markets. The input patterns are based on input data vectors of data windows preprocessed into a sequence of daily, weekly and monthly or quarterly sampled feature measurement changes (log feature fluctuations). The data processing is split into a batch processed step where features are learnt using a Stacked AutoEncoder (SAE) via unsupervised learning, and then both batch and online supervised learning are carried out on Feedforward Neural Networks (FNNs) using these features. The FNN output is a point prediction of measured time-series feature fluctuations (log differenced data) in the future (ex-post). Weight initializations for these networks are implemented with restricted Boltzmann machine pretraining, and variance based initializations. The validity of the FNN backtest results are shown under a rigorous assessment of backtest overfitting using both Combinatorially Symmetrical Cross Validation and Probabilistic and Deflated Sharpe Ratios. Results are further used to develop a view on the phenomenology of financial markets and the value of complex historical data under unstable dynamics.
dc.identifier.apacitationda Costa, J. (2020). <i>Online Non-linear Prediction of Financial Time Series Patterns</i>. (). ,Faculty of Science ,Department of Statistical Sciences. Retrieved from http://hdl.handle.net/11427/32221en_ZA
dc.identifier.chicagocitationda Costa, Joel. <i>"Online Non-linear Prediction of Financial Time Series Patterns."</i> ., ,Faculty of Science ,Department of Statistical Sciences, 2020. http://hdl.handle.net/11427/32221en_ZA
dc.identifier.citationda Costa, J. 2020. Online Non-linear Prediction of Financial Time Series Patterns. . ,Faculty of Science ,Department of Statistical Sciences. http://hdl.handle.net/11427/32221en_ZA
dc.identifier.risTY - Master Thesis AU - da Costa, Joel AB - We consider a mechanistic non-linear machine learning approach to learning signals in financial time series data. A modularised and decoupled algorithm framework is established and is proven on daily sampled closing time-series data for JSE equity markets. The input patterns are based on input data vectors of data windows preprocessed into a sequence of daily, weekly and monthly or quarterly sampled feature measurement changes (log feature fluctuations). The data processing is split into a batch processed step where features are learnt using a Stacked AutoEncoder (SAE) via unsupervised learning, and then both batch and online supervised learning are carried out on Feedforward Neural Networks (FNNs) using these features. The FNN output is a point prediction of measured time-series feature fluctuations (log differenced data) in the future (ex-post). Weight initializations for these networks are implemented with restricted Boltzmann machine pretraining, and variance based initializations. The validity of the FNN backtest results are shown under a rigorous assessment of backtest overfitting using both Combinatorially Symmetrical Cross Validation and Probabilistic and Deflated Sharpe Ratios. Results are further used to develop a view on the phenomenology of financial markets and the value of complex historical data under unstable dynamics. DA - 2020_ DB - OpenUCT DP - University of Cape Town KW - online learning KW - feedforward neural network KW - restricted Boltzmann machine KW - variance weight initialization KW - stacked autoencoder KW - pattern prediction KW - JSE KW - non-linear KW - financial time series KW - combinatorially symmetrical cross validation KW - backtest overfitting KW - deflated Sharpe ratio KW - probabilistic Sharpe ratio LK - https://open.uct.ac.za PY - 2020 T1 - Online Non-linear Prediction of Financial Time Series Patterns TI - Online Non-linear Prediction of Financial Time Series Patterns UR - http://hdl.handle.net/11427/32221 ER -en_ZA
dc.identifier.urihttp://hdl.handle.net/11427/32221
dc.identifier.vancouvercitationda Costa J. Online Non-linear Prediction of Financial Time Series Patterns. []. ,Faculty of Science ,Department of Statistical Sciences, 2020 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/32221en_ZA
dc.language.rfc3066eng
dc.publisher.departmentDepartment of Statistical Sciences
dc.publisher.facultyFaculty of Science
dc.subjectonline learning
dc.subjectfeedforward neural network
dc.subjectrestricted Boltzmann machine
dc.subjectvariance weight initialization
dc.subjectstacked autoencoder
dc.subjectpattern prediction
dc.subjectJSE
dc.subjectnon-linear
dc.subjectfinancial time series
dc.subjectcombinatorially symmetrical cross validation
dc.subjectbacktest overfitting
dc.subjectdeflated Sharpe ratio
dc.subjectprobabilistic Sharpe ratio
dc.titleOnline Non-linear Prediction of Financial Time Series Patterns
dc.typeMaster Thesis
dc.type.qualificationlevelMasters
dc.type.qualificationlevelMSc
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
thesis_sci_2020_da costa joel.pdf
Size:
6.19 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
0 B
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Masters