Performance analysis of pairs trading using firm characteristics via clustering methods in the Korean stock market

Abstract

This paper analyzes the performance of pairs trading constructed via three unsupervised learnings in the Korean market. In addition to traditional pairs trading, it incorporates firm characteristics to identify more robust pairs. The result reveals that two of the three clustering algorithms tend to outperform the benchmark KOSPI even after accounting for the risk. Long-short equally weighted portfolios via k-means clustering performs best in the Korean market with an annualized mean return of 34% and Sharpe Ratio of 0.667. Moreover, utilizing firm characteristics enhance the performance of the strategy by improving return and reducing volatility at the same time. However, in the robustness check, it reveals some limitations such as profitability differs variously according to the selection of hyperparameters.