NETWORK ANALYSIS OF STOCK MARKETS

Abstract

This dissertation consists of three essays on the application of network methods in finance study at the country-level, firm-level, and fund-level. In the first essay, we extend the analysis of globalization from the market factor to the rest of Fama-French factors and the Carhart momentum factor. The findings show that most of the sample local factors are significantly globalized, with the degree of globalization varying substantially across factors. Specifically, the market factor is the most globalized factor on average, followed by the momentum, size, value, profitability, and investment factors. In addition, we show that the impact of financial globalization has been imputed in the local factors, which explains the intriguing finding of integrated international asset pricing. That is, the local Fama-French factors outperform the global counterparts in pricing stocks, seemingly suggesting that stocks are priced as if financial markets were segmented despite the evident globalization. Our results indicate that this puzzle is attributable to the globalization of local factors.

In the second essay, we propose a system-wide approach to the study of the firm-specific connections, which capture the distinct relatedness between firms through unique features, conditional on the U.S. market. The proposed approach provides a new system-wide and factor-free measurement of market integration. We find that the degree of the firm-specific connections has decreased over time, and industry and style attributes significantly positively affect these connections. By applying these connections, investors can consistently gain through holding relatively few stocks randomly chosen across communities clustered based on these connections. Moreover, this consistency of gains has increased substantially over time, pointing to the importance of considering the firm-specific connections in risk diversification.

In the third essay, we use holding-linked network of mutual funds, measured by the similarity between funds' portfolios, to examine the network predictability of fund performance and flows. Using the new network method, we find evidence of significant predictability between funds with similar holdings. The predictability persists three to six months for alternative performance measures and at least twelve months for fund flows. In addition, a long-short strategy based on these holding links yields a significant annual alpha of about 4.5%. These findings reflect the similar underlying drivers of funds' portfolio holdings and show the persistent prediction of fund performance and flows by the holding linked network.