Online Learning and Pattern Matching for Long-Short Investing: Generalised Pairs Trading Across South African Asset Classes

<p dir="ltr">This dataset and accompanying code archive form part of a 2025 BCom Honours project in the Department of Statistical Sciences at the University of Cape Town. The study investigates a generalised pairs trading framework using <b>online learning</b> and <b>nearest-neighbour pattern matching</b>, following the non-parametric portfolio selection methods of Györfi et al. and the extended agent-based formulation of Loonat & Gebbie.</p><p dir="ltr">The project develops an adaptive long–short trading strategy that combines:</p><ul><li>nearest-neighbour pattern matching,</li><li>multi-agent online learning,</li><li>Mutual Fund Separation (benchmark + active sleeves), and</li><li>tracking-error-controlled portfolio construction.</li></ul><p dir="ltr">The approach is tested across <b>three South African asset classes</b> - JSE equities, ZAR currency pairs, and SA government bonds - using both <b>weekly and monthly</b> data, along with a comprehensive suite of <b>synthetic datasets</b> (GBM, lognormal, ARMA, GARCH). The MATLAB implementation is fully modular and designed for reproducibility, allowing users to replicate results or extend the methodology to new universes.</p><p dir="ltr">The deposited files include:<br>• cleaned real-world datasets (JSE equities, FX crosses, SA bonds)<br>• synthetic datasets used for stress testing<br>• the complete MATLAB codebase for the pattern-matching portfolio class<br>• experiment scripts for generating results<br>• all output figures used in the thesis<br>• the final Honours thesis PDF</p><p dir="ltr">This item provides a reproducible research package for students and researchers interested in non-parametric portfolio selection, statistical arbitrage, pattern-matching methods, and online learning in financial markets.</p>