Identifying Drivers of Cyclical Returns in Semiconductor Equities: A Jump Model Approach

Abstract

The semiconductor industry is one that investors often struggle to invest in. Despite numerous iterations of cycles being observed in public markets historically, investors often fail to forecast future earnings and stock price movements properly. The central issue for many is understanding the cyclicality of the industry. Previous work has been done to forecast future cycles, but these models lack focus on equity markets specifically and may be outdated given the pace of change within the industry. In this work, we formulate a prediction model that utilizes both unsupervised and supervised learning techniques by combining the sparse jump model and random forest classification to predict future states. We demonstrate that a three-state model that corresponds to bear, neutral, and bull states offers the most intuitive explanation for the behavior of semiconductor equity returns. Using the feature selection and predictive capabilities of the combination of the two models, we are able to show adequate out-of-sample performance for predicting future states in semiconductor markets. Our model identifies industry-level, macroeconomic, financial, and technical variables that are key drivers for returns within the industry. To demonstrate the usefulness of these results, we develop a trading strategy that makes portfolio allocations as dictated by the future regime predictions of our model and achieve a Sharpe ratio that is more than 30% higher than relevant benchmarks. The results may prove useful for investors who are trying to better understand the current stage of the semiconductor cycle and what variables to focus on to understand future performance.