This dissertation focuses on climate finance and explore how to incorporate machine learning techniques into financial research. In the first chapter, we focus on climate innovation. Through a novel design to link climate risk and the U.S. firm patents related to climate change mitigation technologies (CCMTs), we find that CCMT innovations generate significant economic value. These innovations are effective in mitigating firms’ carbon risk. We also find that adoption of a new patent classification scheme has promoted more CCMT innovations in the United States. However, we find mixed evidence on firms’ carbon risk and their CCMT innovation activities. Our work shows that the government can play a more prominent role in promoting activities in CCMTs. In the second chapter, we study how climate change risk affects the mutual fund industry. We construct a novel carbon risk measure to assess mutual funds’ carbon risk exposure based on mutual fund holding data. We find carbon risk negatively predicts fund future raw and risk-adjusted performance. Carbon risk correlates with different risk factors and higher carbon risk is associated with higher unexplained risk. We show that carbon risk adversely affects future fund flow, especially when social sentiment on climate change is high. Institutional investors pay more attention to climate change issues than retail investors do. Finally, we find that mutual funds’ carbon risk exposure is not fully captured in traditional Environmental, Social, and Governance (ESG) measures. The third chapter is a joint work with Tianyi Qiu. In this chapter, we utilize machine learning techniques to measure top executives’ personality traits of a large sample of U.S. public firms. We show that CEOs’ innate personality traits help explain corporate financial policies and investment decisions. We find that more extroverted CEOs tend to invest more aggressively but make less efficient decisions. They spend more on R&D expenses yet receive less favored outputs and conduct more M&A deals yet have lower average deal quality. The machine learning based personality measures suggest an alternative approach to directly assess top managers’ individual traits and reveal additional information on the association between CEOs’ personality traits and firm policies.