{"id":{"repo_id":"stellenbosch","oai_identifier":"oai:scholar.sun.ac.za:10019.1/135739"},"canonical_url":"https://search.dev.ndltd.org/etd/stellenbosch/oai:scholar.sun.ac.za:10019.1/135739","repository":{"repo_id":"stellenbosch","name":"Stellenbosch University","base_url":"https://scholar.sun.ac.za/server/oai/request"},"display":{"title":"A generic optimisation framework for reinforcement learning in the foreign exchange market","abstract":"The application of algorithmic trading in financial markets has grown considerably in recent years, with deep reinforcement learning emerging as a prominent technique for developing autonomous trading agents. The success of such agents is, however, often hindered by the non-stationary nature of financial markets and the inherently opaque decision-making processes of deep reinforcement learning models. An abundance of research has been dedicated to applying deep reinforcement learning in finance, resulting in various frameworks for algorithmic trading. Most of these existing frameworks are, however, aimed at single facets of the trading problem, such as the application of a specific deep reinforcement learning algorithm or a narrow feature engineering approach. Generic, integrated frameworks that accommodate market non-stationarity and embed explainability techniques into the model development lifecycle are largely absent from the literature. A generic framework is proposed in this dissertation that facilitates a systematic approach towards the optimisation, comparative evaluation, and selection of deep reinforcement learning trading agents. The design of the framework accommodates market non-stationarity by following a non-parametric data processing pipeline, which utilises time series partitioning techniques and distribution-based clustering approaches to identify and classify distinct market regimes. These regime-labelled data are then utilised in an iterative process during which explainable artificial intelligence techniques are employed to facilitate data-driven state-space optimisation and agent transparency enhancement. The framework culminates in a robust evaluation methodology—comprising multi-seed training, rolling-window back-testing, and the execution of non-parametric statistical tests—in order to quantify and comparatively analyse the performance of deep reinforcement learning candidate agents, thereby ultimately guiding the user's selection of a suitable configuration for a given market context. A computerised instantiation of the framework is implemented as a proof of concept. The efficacy of its constituent components is first verified in respect of synthetic and historical market data, after which the practical applicability of the framework is demonstrated by applying the framework instantiation to two case studies comprising historical market data.","abstract_html":"The application of algorithmic trading in financial markets has grown considerably in recent years, with deep reinforcement learning emerging as a prominent technique for developing autonomous trading agents. The success of such agents is, however, often hindered by the non-stationary nature of financial markets and the inherently opaque decision-making processes of deep reinforcement learning models. An abundance of research has been dedicated to applying deep reinforcement learning in finance, resulting in various frameworks for algorithmic trading. Most of these existing frameworks are, however, aimed at single facets of the trading problem, such as the application of a specific deep reinforcement learning algorithm or a narrow feature engineering approach. Generic, integrated frameworks that accommodate market non-stationarity and embed explainability techniques into the model development lifecycle are largely absent from the literature. A generic framework is proposed in this dissertation that facilitates a systematic approach towards the optimisation, comparative evaluation, and selection of deep reinforcement learning trading agents. The design of the framework accommodates market non-stationarity by following a non-parametric data processing pipeline, which utilises time series partitioning techniques and distribution-based clustering approaches to identify and classify distinct market regimes. These regime-labelled data are then utilised in an iterative process during which explainable artificial intelligence techniques are employed to facilitate data-driven state-space optimisation and agent transparency enhancement. The framework culminates in a robust evaluation methodology—comprising multi-seed training, rolling-window back-testing, and the execution of non-parametric statistical tests—in order to quantify and comparatively analyse the performance of deep reinforcement learning candidate agents, thereby ultimately guiding the user's selection of a suitable configuration for a given market context. A computerised instantiation of the framework is implemented as a proof of concept. The efficacy of its constituent components is first verified in respect of synthetic and historical market data, after which the practical applicability of the framework is demonstrated by applying the framework instantiation to two case studies comprising historical market data.","abstract_has_math":false,"creators":["De Wit, John Spencer"],"institution":"Stellenbosch : Stellenbosch University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Van Vuuren, J. H."],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-03","date_published":"2026-03","updated_at":"2026-07-24T04:40:06Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholar.sun.ac.za/handle/10019.1/135739","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Van Vuuren, J. H."]},{"key":"dc:contributor.other","label":"Dc Contributor Other","values":["Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering."]},{"key":"dc:creator","label":"Author","values":["De Wit, John Spencer"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-04-09T06:55:32Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-04-09T06:55:32Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-03"]},{"key":"dc:publisher","label":"Institution","values":["Stellenbosch : Stellenbosch University"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholar.sun.ac.za/handle/10019.1/135739"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (PhD)--Stellenbosch University, 2026.","De Wit, J. S. 2026. A generic optimisation framework for reinforcement learning in the foreign exchange market. Unpublished doctoral dissertation. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/3ff15aa9-fe84-4dbc-a9cb-3ddc75eab4cf"]},{"key":"dc:description.abstract","label":"Abstract","values":["The application of algorithmic trading in financial markets has grown considerably in recent years, with deep reinforcement learning emerging as a prominent technique for developing autonomous trading agents. The success of such agents is, however, often hindered by the non-stationary nature of financial markets and the inherently opaque decision-making processes of deep reinforcement learning models. An abundance of research has been dedicated to applying deep reinforcement learning in finance, resulting in various frameworks for algorithmic trading. Most of these existing frameworks are, however, aimed at single facets of the trading problem, such as the application of a specific deep reinforcement learning algorithm or a narrow feature engineering approach. Generic, integrated frameworks that accommodate market non-stationarity and embed explainability techniques into the model development lifecycle are largely absent from the literature. A generic framework is proposed in this dissertation that facilitates a systematic approach towards the optimisation, comparative evaluation, and selection of deep reinforcement learning trading agents. The design of the framework accommodates market non-stationarity by following a non-parametric data processing pipeline, which utilises time series partitioning techniques and distribution-based clustering approaches to identify and classify distinct market regimes. These regime-labelled data are then utilised in an iterative process during which explainable artificial intelligence techniques are employed to facilitate data-driven state-space optimisation and agent transparency enhancement. The framework culminates in a robust evaluation methodology—comprising multi-seed training, rolling-window back-testing, and the execution of non-parametric statistical tests—in order to quantify and comparatively analyse the performance of deep reinforcement learning candidate agents, thereby ultimately guiding the user's selection of a suitable configuration for a given market context. A computerised instantiation of the framework is implemented as a proof of concept. The efficacy of its constituent components is first verified in respect of synthetic and historical market data, after which the practical applicability of the framework is demonstrated by applying the framework instantiation to two case studies comprising historical market data."]},{"key":"dc:title","label":"Title","values":["A generic optimisation framework for reinforcement learning in the foreign exchange market"]}]}],"canonical_facts":{"dc:contributor.advisor":["Van Vuuren, J. H."],"dc:contributor.other":["Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering."],"dc:creator":["De Wit, John Spencer"],"dc:date.accessioned":["2026-04-09T06:55:32Z"],"dc:date.available":["2026-04-09T06:55:32Z"],"dc:date.issued":["2026-03"],"dc:description":["Thesis (PhD)--Stellenbosch University, 2026.","De Wit, J. S. 2026. A generic optimisation framework for reinforcement learning in the foreign exchange market. Unpublished doctoral dissertation. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/3ff15aa9-fe84-4dbc-a9cb-3ddc75eab4cf"],"dc:description.abstract":["The application of algorithmic trading in financial markets has grown considerably in recent years, with deep reinforcement learning emerging as a prominent technique for developing autonomous trading agents. The success of such agents is, however, often hindered by the non-stationary nature of financial markets and the inherently opaque decision-making processes of deep reinforcement learning models. An abundance of research has been dedicated to applying deep reinforcement learning in finance, resulting in various frameworks for algorithmic trading. Most of these existing frameworks are, however, aimed at single facets of the trading problem, such as the application of a specific deep reinforcement learning algorithm or a narrow feature engineering approach. Generic, integrated frameworks that accommodate market non-stationarity and embed explainability techniques into the model development lifecycle are largely absent from the literature. A generic framework is proposed in this dissertation that facilitates a systematic approach towards the optimisation, comparative evaluation, and selection of deep reinforcement learning trading agents. The design of the framework accommodates market non-stationarity by following a non-parametric data processing pipeline, which utilises time series partitioning techniques and distribution-based clustering approaches to identify and classify distinct market regimes. These regime-labelled data are then utilised in an iterative process during which explainable artificial intelligence techniques are employed to facilitate data-driven state-space optimisation and agent transparency enhancement. The framework culminates in a robust evaluation methodology—comprising multi-seed training, rolling-window back-testing, and the execution of non-parametric statistical tests—in order to quantify and comparatively analyse the performance of deep reinforcement learning candidate agents, thereby ultimately guiding the user's selection of a suitable configuration for a given market context. A computerised instantiation of the framework is implemented as a proof of concept. The efficacy of its constituent components is first verified in respect of synthetic and historical market data, after which the practical applicability of the framework is demonstrated by applying the framework instantiation to two case studies comprising historical market data."],"dc:identifier.uri":["https://scholar.sun.ac.za/handle/10019.1/135739"],"dc:language.iso":["en"],"dc:publisher":["Stellenbosch : Stellenbosch University"],"dc:title":["A generic optimisation framework for reinforcement learning in the foreign exchange market"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T04:40:06Z"}