Novel and Traditional Approaches to Pairs Trading in Equities

This is an ongoing project investigating traditional and more sophisticated approaches to statistical arbitrage, specifically pairs trading algorithms. I plan to first familiarize myself with econometric models for identifying pair relationships (validity, robustness, stability across time and so on) and test tradeability. I have conducted an initial literature review to identify more modern approaches and also various related ideas and methodology that appeared interesting to me, so in that vein, parts of this project will also be semi-replications of results in the literature.

I will of course present my results and output as I complete phases and subphases, similarly to how projects have been presented in the past. However, I also wanted to use this project to focus a little bit more on the scientific aspects of backtesting and strategy design, and this will play a larger role in at least the early parts of the project. I will lay out some of the ideas and issues under consideration in the following sections.

Available materials: Literature Data Trading cost models Backtesting design

Precursor

In my previous project generating alpha signals using factors, a few issues arose as I worked through the project that I wish to identify before I begin implementing and experimenting here. These seem to be very instructional to me.

To mitigate issues with multiple testing, I plan to lay out precisely the experimental design and trials before any of them are executed; I think this will first prevent more meta-selection bias that I felt was present in the previous project, and second, it will present the results more honestly. This will require a more meticulous documenting and presentation.
The trading model and backtester will now need to be event-driven, and model costs more thoroughly than previous projects (e.g. vectorized backtesting of an L/S portfolio). I will discuss this in detail in a later section.
Data considerations e.g. survivorship bias will also need to play a larger role here since these can have catastrophic consequences in this setting. I also discuss this more in the next section.

Progress

17 Aug 2026

Dataset cleaned and process documented

Identity, adjustment, membership; consistency, continuity, and correctness verified.

01 Aug 2026

Full membership, price and corporate actions dataset collected

Point-in-time membership, 724 price series from 2016-2026, corporate actions table with 20,460 events.

19 Jul 2026

Backtesting model

Signal on adjusted prices, execution on raw prices with explicit corporate actions and costs. Initial gross investment for simulated account.

13 Jul 2026

Trading cost model

Execution friction, explicit regulatory fees and short-borrow carry.

May–Jun 2026

Literature review

Researched classical econometric models (distance/cointegration) which I plan to implement as the baseline. Also obtained papers about modern and novel approaches (e.g. generalizations, baskets, ML, graph-based) which I thought were interesting to investigate.