半導體資本市場泡沫的兩階段機器學習預測方法

Abstract

金融領域的過往研究中，資產泡沫一直是一個關注的焦點，相關的子領域，如泡沫的形成、泡沫理論和泡沫檢測等，持續受到學術界的討論。本文探討了機器學習在金融領域的新應用，通過兩階段的機器學習方法，預測了半導體在資本市場中可能出現的泡沫現象。其中選擇費城半導體指數作為產業的代表。實證結果表明，透過總體經濟指標、產業供需指標和財務指標的綜合應用，機器學習模型能夠成功地預測泡沫的出現。此外，本文嘗試對模型中具有經濟意義的變數進行解釋，藉此擴充了對泡沫現象的研究。透過結合模型解釋的完整研究，不僅拓展了當前學術界在這一領域的實證分析，同時也有助於市場參與者更好地理解市場的動態，識別潛在的市場泡沫，進而提升其決策能力和市場反應能力。

Asset bubbles have long been a focal point of concern in financial research, with subfields such as bubble formation, bubble theories, and bubble detection continuously debated within academia. This paper proposes a novel machine learning approach to predict potential asset bubbles in the semiconductor sector. Using the Philadelphia Semiconductor Index, the study demonstrates that machine learning models, trained on macroeconomic indicators, industry data, and financial metrics, can successfully identify bubble formation. Furthermore, the research goes beyond prediction by analyzing the model's internal workings to pinpoint economically meaningful variables that contribute to bubble emergence. This combined approach not only advances academic understanding of bubbles but also equips market participants with valuable tools to navigate market dynamics and make informed investment decisions.