Pair Trading Analysis of the Semiconductor Stock Market

This study investigates the strength of the co-relationships between major semiconductor manufacturers and large technology companies through pair analysis and predicts their relevant future performance using machine learning. In recent years, unpredictable global events—including the COVID-19 pandemic and international trade conflicts—have initiated significant fluctuations in the semiconductor industry’s stability. The rapid emergence of generative artificial intelligence (AI) has increased demand for semiconductors, further disrupting the semiconductor supply chain and accelerating structural economic transformation.

We hypothesize that this economic instability is reflected in the performance relationships among semiconductor-related firms over a long-term period. To examine this hypothesis, we select ten major global technology companies highly relevant to the semiconductor industry, find a strong relationship, and analyze their future performance from 2019 to 2024. Four econometric pair analysis methods are applied to identify optimal pair combinations, and future pair performance is subsequently estimated using two machine-learning models.

The results indicate that, among the methods considered, only the Johansen cointegration method successfully captures long-term equilibrium relationships between the companies under study at the 95% confidence level. The decision tree and random forest models demonstrate the ability to predict the next-day movement, achieving an average accuracy exceeding 50% based on the forecasting z-score spread trends from the optimized company pairs. These findings suggest that conducting long-term pair analysis is feasible for the semiconductor industry. Although short-term movement projections may yield strong performance outcomes, they may not fully capture broader structural effects, including prolonged global disruptions.