Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
On estimating the potential impact of meta-labelling
Thesis (MCom)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614095412297728 |
|---|---|
| access_status_str | Open Access |
| author | Van Niekerk, Jana |
| author2 | Visagie, S. E. |
| author_browse | Van Niekerk, Jana Visagie, S. E. |
| author_facet | Visagie, S. E. Van Niekerk, Jana |
| author_sort | Van Niekerk, Jana |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MCom)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134854 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:46:35.680Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134854 On estimating the potential impact of meta-labelling Van Niekerk, Jana Visagie, S. E. Stellenbosch University. Faculty of Economics and Management Sciences. Dept. of Logistics. Investments -- Mathematical models Portfolio management -- Mathematical models Decision making -- Statistical methods Capital market -- Mathematical models Stocks -- Prices -- Mathematical models UCTD Thesis (MCom)--Stellenbosch University, 2025. Van Niekerk, J. 2025. On estimating the potential impact of meta-labelling. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/dc7dabd9-a11a-499a-8e4f-b570cd7b471d ENGLISH SUMMARY: Traditional price prediction models often fail to capture the non-linear, noisy, and dynamic characteristics of financial time series. Meta-labelling, an additional layer that can be added to a trading system, has recently emerged as a promising approach to enhance trading strategy performance by filtering out false positive signals. It is necessary to know whether the use of a meta-labelling framework will improve performance enough to justify its implementation. A model exists that relates precision and the Sharpe ratio for trading strategies with fixed returns. In reality, the returns from trading strategies are rarely fixed. Thus, the question arises: Is it possible to express the Sharpe ratio as a function of precision for trading strategies with variable returns? A novel Sharpe ratio estimation model for variable returns is presented. The model quantifies expected improvements in risk-adjusted performance as a function of trading signal precision. In contrast to existing formulations that assume fixed and symmetric returns, the proposed approach incorporates the stochastic nature of trading outcomes by modelling trade returns as truncated normal distributions. The model is validated with three cases. In Case I, results from a real-world implementation using a random forest classifier show perfect alignment between actual and estimated Sharpe ratios (Δθa−d = 0.0000), confirming the accuracy of the model. Case II applies a simulation framework in which prediction outcomes are systematically altered by controlled flipping of true and false positive and negative signals across 121 scenarios. Average deviations between actual and estimated Sharpe ratios remain below 0.001, demonstrating robustness. In Case III, random flipping enforces specific positive prediction proportions independent of classification outcomes. The estimated Sharpe ratios remain closely aligned with observed values (Δθa−d = 0.000961), indicating reliability under stochastic prediction conditions. Together, these results provide empirical support for the model’s consistency and generalisability irrespective of secondary model choice. The underlying assumptions of the model are evaluated under non-normal return distributions, including skew-normal, log-normal, and Student’s t distributions. Results show that the model remains reliable for near-normal distributions with low skewness (|ˆγ| < 0.1272), low kurtosis (ˆκ < 0.0378), and a Shapiro-Wilk statistic (S ≥ 0.9989), achieving relative errors of less than 5%. However, accuracy diminishes under moderate to high skewness (|ˆγ| ≥ 0.6618) or kurtosis (ˆκ ≥ 0.4662), with complete failure for log-normal distributions due to their positive-only returns and extreme tails (ˆκ ≤ 78.0781). AFRIKAANSE OPSOMMING: Tradisionele prysvoorspellingsmodelle misluk baie keer om die nie-lineere en dinamiese eienskappe van finansiele tydreekse vas te vang. Meta-etikettering, ’n bykomende laag wat by ’n handelstelsel gevoeg kan word, het onlangs na vore gekom as ’n belowende benadering om die presisie van handelsstrategiee te verbeter deur vals positiewe seine uit te sif. Dit is nodig om te weet of die gebruik van ’n meta-etiketteringsraamwerk die handelstelsel se prestasie genoeg sal verbeter om die implementering daarvan te regverdig. Dus ontstaan die vraag: Is dit moontlik om die Sharpe-verhouding as ’n funksie van presisie vir handelsstrategiee met veranderlike opbrengste te bepaal? ’n Nuwe Sharpe-verhouding-benaderingsmodel vir veranderlike opbrengste word ontwikkel. Die model kwantifiseer verwagte verbeterings in risiko-aangepaste prestasie as ’n funksie van presisie. In teenstelling met bestaande formulerings wat vaste en simmetriese opbrengste aanneem, sluit die voorgestelde benadering die stogastiese aard van handelsuitkomste in deur handelsopbrengste as afgesnyde normale verdelings te modelleer. Die model word met drie gevalle gevalideer. In Geval I toon die resultate van ’n werklike implementering met ’n ewekansige woudklassifiseerder perfekte belyning tussen werklike en geskatte Sharpe-verhoudings (Δθa−d = 0.0000), wat akkuraatheid van die model bevestig. Geval II pas ’n simulasieraamwerk toe waarin voorspellingsuitkomste stelselmatig verander word deur beheerde omkering van ware en vals positiewe en negatiewe seine oor 121 scenario’s. Gemiddelde afwykings tussen werklike en geskatte Sharpe-verhoudings bly onder 0.001. In Geval III dwing ewekansige omkering spesifieke positiewe voorspellingsverhoudings af, onafhanklik van klassifikasie-uitkomste. Die geskatte Sharpe-verhoudings bly ooreenstem met waargenome waardes (Δθa−d = 0.000961), wat betroubaarheid onder voorspellingsomstandighede aandui. Saam bied hierdie resultate empiriese ondersteuning vir die model se konsekwentheid en veralgemeenbaarheid ongeag van ’n sekondere model. Die onderliggende aannames van die model word geevalueer onder nie-normale opbrengsverdelings, insluitend skeef-normale, log-normale en Student se t-verdelings. Resultate toon dat die model betroubaar bly vir byna-normale verdelings met lae skeefheid (|ˆγ| < 0.1272), lae kurtosis (ˆκ < 0.0378), en ’n Shapiro-Wilk-statistiek (S ≥ 0.9989), met relatiewe foute van minder as 5%. Akkuraatheid neem egter af onder matige tot hoe skeefheid (|ˆγ| ≥ 0.6618) of kurtosis (ˆκ ≥ 0.4662), met volledige mislukking vir log-normale verdelings as gevolg van hul slegspositiewe opbrengste en ekstreme sterte (ˆκ ≤ 78.0781). Masters 2026-01-13T05:52:44Z 2026-01-13T05:52:44Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134854 en Stellenbosch University xvi, 138 pages : illustrations, includes annexures application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Investments -- Mathematical models Portfolio management -- Mathematical models Decision making -- Statistical methods Capital market -- Mathematical models Stocks -- Prices -- Mathematical models UCTD Van Niekerk, Jana On estimating the potential impact of meta-labelling |
| title | On estimating the potential impact of meta-labelling |
| title_full | On estimating the potential impact of meta-labelling |
| title_fullStr | On estimating the potential impact of meta-labelling |
| title_full_unstemmed | On estimating the potential impact of meta-labelling |
| title_short | On estimating the potential impact of meta-labelling |
| title_sort | on estimating the potential impact of meta labelling |
| topic | Investments -- Mathematical models Portfolio management -- Mathematical models Decision making -- Statistical methods Capital market -- Mathematical models Stocks -- Prices -- Mathematical models UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/134854 |
| work_keys_str_mv | AT vanniekerkjana onestimatingthepotentialimpactofmetalabelling |