財務金融研究

Abstract

本論文由兩篇探討法規與科技如何影響財務決策與資產定價的研究組成。第一篇論文檢視臺灣於2014年實施企業社會責任（CSR）報告強制揭露制度對自願性財務揭露行為的影響，發現企業管理者在投資人與分析師關注度高的情況下，顯著增加盈餘預測的發布。此結果顯示，管理者進行自願性揭露主要是為了解決代理問題，而非為了降低專有資訊洩漏風險或資訊不對稱問題，突顯強制性非財務揭露與自願性財務揭露之間的互補關係。第二篇論文探討機器學習技術是否能提升盈餘貝他值（earnings beta）的預測效果，該指標為評估系統性風險的替代方案。研究比較傳統迴歸方法與套索（Lasso）與彈性網（Elastic Net）等機器學習模型的表現，發現儘管後者在預測準確度上略有提升，其在橫斷面資產定價的經濟解釋力仍然有限。綜合而言，這兩篇論文針對公司揭露行為與資產定價文獻做出貢獻，展示法規變遷與運算技術進步如何共同影響市場透明度、風險評估與資本市場效率。

This dissertation comprises two essays that explore how regulation and technology shape financial decision-making and asset pricing. The first essay examines the impact of Taiwan’s 2014 CSR reporting mandate on voluntary financial disclosures, finding that managers significantly increase the issuance of earnings forecasts, particularly under heightened investor and analyst scrutiny. These findings suggest that voluntary disclosures serve primarily to address agency concerns rather than proprietary costs or adverse selection, highlighting the complementary relationship between mandatory nonfinancial and voluntary financial reporting. The second essay investigates whether machine learning techniques enhance the estimation of earnings betas—an alternative to market betas in capturing systematic risk. While methods such as Lasso and Elastic Net slightly improve forecast accuracy over traditional regressions, their economic significance in pricing cross-sectional returns remains limited. Together, these essays contribute to the literature on corporate disclosure and asset pricing by illustrating how regulatory frameworks and computational advancements jointly influence transparency, risk assessment, and market efficiency.