國際資本流動的時變特徵與產業黏性：基於臺灣股市的實證研究

Time-Varying Characteristics of International Capital Flows and Industrial Stickiness: An Empirical Study of the Taiwan Stock Market

臺灣股市呈現與經濟基本面背離的“牛長熊短”特徵，既有文獻多聚焦於線性均衡分析，對其傳導機制的時變性與非線性特徵著墨有限。本文突破參數恆定與線性假設，以國際資本存量主導論為核心，採用2003年至2023年季度數據，構建時變參數向量自迴歸（TVP-VAR）模型與隨機森林（Random Forest）算法，系統考察國際資本流動對臺灣加權股價指數（TWSE）的動態傳導機制。研究發現：第一，國際資本存量的影響力具有顯著的時變性，2008年全球金融危機與2018年中美貿易摩擦等結構性事件深刻重塑了匯率與資本流動的傳導路徑。第二，外資存在顯著的“二元結構”，半導體產業因其“技術鎖入效應”形成“黏性資本”，在2020年新冠疫情期間展現出強大的市場緩衝功能，其減持率（2.1%）遠低於金融業（18.7%），成為臺股韌性的微觀基礎。第三，隨機森林模型首次實證識別出國際資本存量對股市波動率的“J型”非線性影響及政策利率的“微笑曲線”效應，並揭示了二者間的強交互作用。本文承接了VECM研究的長期均衡結論，進一步從時變與非線性視角深化了對“存量主導”與“產業黏性錨定”新範式的理解，為新興市場實施分級資本流動管理與精細化貨幣政策提供了量化依據。

The Taiwan stock market exhibits a pattern of prolonged bull markets with short bear markets, diverging from its macroeconomic fundamentals. Existing literature largely focuses on linear equilibrium analysis, paying limited attention to the time-varying and nonlinear characteristics of its transmission mechanisms. Moving beyond the assumptions of parameter constancy and linearity, this study takes the dominant role of international capital stock as its core theoretical premise. Utilizing quarterly data from 2003 to 2023, it constructs a Time-Varying Parameter Vector Autoregression (TVP-VAR) model and employs the Random Forest algorithm to investigate the dynamic transmission mechanisms of international capital flows on the Taiwan Weighted Stock Index (TWSE). The findings reveal three key results. First, the influence of international capital stock exhibits significant time variation; structural events such as the 2008 Global Financial Crisis and the 2018 U.S.-China trade friction reshaped the transmission pathways between exchange rates and capital flows. Second, foreign capital demonstrates a distinct dual structure; the semiconductor industry, characterized by a technology lock-in effect, constitutes sticky capital. During the COVID-19 pandemic in 2020, this sticky capital demonstrated a strong market stabilization function, with a divestment rate (2.1%) far lower than that of the financial industry (18.7%), serving as a micro-foundation for the Taiwan stock market’s resilience. Third, the Random Forest model empirically identifies a J-shaped nonlinear impact of international capital stock on stock market volatility and a smile curve effect of the policy interest rate, while also revealing strong interactions between these two variables. Building upon the long-run equilibrium findings of previous VECM research, this study deepens the understanding of the new paradigm of stock dominance and industry stickiness anchoring from time-varying and nonlinear perspectives, providing quantitative foundations for implementing tiered capital flow management and refined monetary policies in emerging markets.