Book Description
The rapid advancements of digital technologies have fostered transformation of various online and offline environments, which in turn created opportunities and challenges on modern management and decision-making strategies. In this dissertation, I examine three areas that are subject to the impact of such transformation, using evidence-based inference and optimization approaches to gain insights and enhance decision-making quality. The first essay examines individual investors' learning in crowdfunded supply chain finance (SCF) markets under the unique presence of loan guarantors in the financing process. My estimation results confirm the existence of investor learning. In addition, I observe that this latent perception has different moderating effects on investor responses to listing attributes, such as interest rate and loan duration. Counterfactual simulations suggest that enabling investor learning from correlated investment experience can help mitigate adverse selection and improve overall market efficiency; for supply chain members, optimizing the structure of loan listings could accelerate investor learning, which in turn can help simulate fundraising performance as a desirable outcome of reputation building. The second essay investigates the effect of scarcity-Induced demand on the crowdfunding market. Using a hierarchical Bayesian framework, this essay empirically validates the positive effect of the scarcity strategy in the crowdfunding market; Interestingly, my mechanism analysis reveals that individual backers are more susceptible to demand-induced scarcity: given the relative scarcity level, individuals are more attracted to invest in rewards that achieve this scarcity due to excess demand rather than limited supply. In my third essay, I use data-driven analytics to facilitate medical decision-making based on electronic medical records (EMRs). Specifically, I consider the problem of designing personalized treatment recommendations for patients with multiple myeloma, which is the second most common blood cancer in the United States. Using clinical data with patients' cytogenetics information, this essay formulates the treatment recommendation problem as a multilevel Bayesian contextual bandit, which sequentially selects treatments based on contextual information about patients and therapies, with the goal of maximizing overall survival outcomes. I also propose a causal offline evaluation framework which integrates the structural econometric model into bandit optimization and generates counterfactuals for policy evaluation. This novel framework provides reliable performance measures when field experiment or long log data are not available. Compared with clinical practices and benchmark strategies, my method suggests a rise in overall survival outcomes, with higher improvement for aging or high-risk patients with more complications.