Essays on FinTech and empirical asset pricing

Abstract

The accelerated digitization of global capital markets has fundamentally reconfigured the research paradigms of modern finance. Financial technology (FinTech) innovations-spanning decentralized cryptographic assets to machine learning-driven textual analysis-provide not only new type of financial markets but also new research methodologies in empirical asset pricing. This thesis addresses these paradigm shifts through two chapters that explore how cutting-edge fintech technologies shape empirical asset pricing methodologies and investment strategies.

The first chapter studies the financialization of cryptocurrency. Through the lens of cryptocurrency financialization, we show that change in Grayscale Bitcoin Trust premium is the most significant predictor of Bitcoin daily return. Using K-means clustering and LDA analysis, we find that this predictability is especially significant when there is a large variation in bullish and bearish market sentiment, innovation regarding CBDC, and regulations on crypto exchanges, but not when there is innovation regarding blockchain technology or Bitcoin mining. These findings suggest that indexing serves as a channel for information transmission, and Bitcoin prices react with a delay to the information contained in the sentiment of traditional investors.

The second chapter shifts focus to AI-driven investment strategies, specifically examining the capabilities of large language models (LLMs) in generating financial investment strategies. LLMs, including ChatGPT, leverage their reasoning capabilities to create portfolio recommendations beyond the reach of conventional textual analysis. By refining the LLM model with specific training data and adjusting ChatGPT's parameters for enhanced output flexibility, there is potential for ChatGPT to craft portfolios that outperform market benchmarks in out-of-sample tests. Utilizing two distinct types of textual data in different languages, articles from the Wall Street Journal in the U.S. and policy announcements from the Chinese government, we show that ChatGPT can produce portfolios with a monthly three-factor alpha of up to 3\%, particularly in response to policy-related news. When comparing these outcomes to those generated by traditional textual analysis methods, we observe that the conventional methods fail to produce portfolios with positive alpha.

Collectively, the two constituent studies of this thesis provide a systematic exploration of FinTech’s dual impact on empirical asset pricing—both as a catalyst for novel pricing anomalies and as a driver of methodological innovation. This thesis establishes a scaffold for future inquiries at the FinTech-empirical finance nexus. Methodologically, they delineate protocols for integrating alternative data streams (e.g., on-chain analytics, NLP-processed financial text) into multifactor pricing models. Theoretically, they motivate a re-examination of market efficiency definitions in hybrid ecosystems where algorithmic agents and human investors co-evolve. For policymakers and practitioners, this work provides actionable insights into risk management and regulatory design in increasingly automated, data-intensive markets.