Testing Cointegration-Based Pairs Trading in Thematic ETF Markets

Abstract

Thematic exchange-traded funds (ETFs) have grown rapidly as investors seek concentrated exposure to narratives such as artificial intelligence, clean energy, and genomics. These funds are frequently criticized as difficult long-only investments: their holdings are narrative-driven, composition shifts are common, and a single dominant factor explains the majority of return variance within each theme. This thesis asks whether cointegration-based pairs trading, a market-neutral relative-value strategy, can exploit within-theme price relationships more effectively than long-only positioning. Using daily price data for six thematic ETF groups, the analysis applies a walk-forward framework combining Engle–Granger cointegration tests, Benjamini–Hochberg false discovery rate control, half-life screening, and out-of-sample trade execution across multiple formation and trading horizons. Cointegrated pairs are detectable across all six themes, but the opportunity operates as a three-stage funnel in which statistical detection, portfolio-level activation, and net profitability after costs each impose an independent filter. Shorter formation windows outperform longer ones across both dimensions, driven by composition drift at six-to-twelve-month timescales. Portfolio feasibility does not guarantee economic viability: Blockchain's high activation rate does not translate into positive returns, while Cloud Computing and Genomics are the strongest positive cases, though gains remain narrow and cost-sensitive. The results characterize thematic ETF pairs trading as a conditional, theme-specific opportunity. What makes a theme compelling as a narrative investment is largely what makes it difficult to arbitrage.