Using Unsupervised Learning to Find Pairs Trading Candidates

Summary

The introduction presents a machine-learning framework for selecting securities for pairs trading. It frames pair discovery as a search-space problem: limiting candidates to securities in the same sector may exclude useful relationships, while searching every possible combination can be impractical and leave traders competing for opportunities already identified by others.

The proposed process first uses unsupervised learning to define a manageable search space, then clusters relevant securities, and finally searches within those groups for promising pairs. The approach allows candidate pairs to cross sector boundaries. The document offers a conceptual outline and an example figure, but gives no algorithmic details, data description, trading rules, or measured results. It therefore explains the pair-selection workflow rather than demonstrating that selected pairs are profitable or robust after costs. The introduction also notes that the fuller description follows a cited book and that the software module requires additional machine-learning packages.

Key ideas

• Pairs trading depends on identifying suitable security pairs.
• Restricting pair searches by sector can be too limiting, while unrestricted searches can be too broad.
• The framework uses unsupervised learning to define a search space and cluster securities.
• Pair candidates are identified within clusters and need not belong to the same sector.
• The introduction supplies no performance evidence or detailed implementation methods.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.