Title: “Pairs Trading - Disentangling Components’ Contribution to the System”
Dominice Goodwin, Lund University.
This paper offers an study on the individual importance (effects) of a set of assumptions, or characteristics, that is central to a pairs trading system and are shared among pairs trading models suggested and employed in the academic literature. Pairs trading systems employ a systematized way in which assets form pairs (the pair formation decision) and how pairs are traded (the trading rules). Suggested design for these sub-system varies in the literature but generally builds on either purely statistical methods, forming pairs of assets that share common risk factors (e.g. shares of companies belonging to the same industry, or are of a similar size, etc.), and more recently on machine learning methods. By employing the pairs trading framework suggested in the pioneering and widely cited paper Gatev et al. (2006), this paper explores the individual effect and importance of separate components (sub-systems) of the trading system, as well as commonly employed restrictions to the pair formation, by industry, size, leverage and profitability. Using data for publicly traded US firms over the period 1999 to 2019, the result show that the pair formation method have a significant impact on accumulated returns and Fama-French 3-factor alpha in isolation, while the trading rules only adds a significant effect when combined with the pair formation method. The results also show that when employing restrictions to the pair formation, by industry, size, leverage or profitability only the industry restriction have a significant effect, and that the effect increase with the precision of the industry grouping (i.e. grouping bases on SIC 4-digits give better results than grouping on SIC 12 industry).