Three essays on business analytics : applications of data-driven decision-making in the context of finance

In my dissertation, I propose a general research framework of MAD---Monitoring, Analyzing, and Data Informed Decision-making---for financial decision-making. I present three essays which concentrate on two consequential aspects of decision-making for financial risk management. The first two essays focus on better monitoring and analyzing the risk, and the last one focuses on better data-informed decision-making based on the observation and analysis.

In the first essay, I study the modeling of joint mortality for the practice of life insurance and annuity pricing. Specifically, I develop a new mathematical model to describe the joint mortality for coupled dependent lives. This model can be used to guide the risk management strategy and the pricing policy for insurance and annuity products. It is shown that it improves the current methods for modeling financial decision-making related to dependent life structures (such as joint life insurance, last survivor annuities, and defined benefit plans for married couples).

In the second essay, I study the prediction of Bitcoin price movement and the relevant implications for business analytics. I exploit Bitcoin transaction networks and link network characteristics with the Bitcoin market exchange price. Based on this linkage and the data record, I construct predictive models for Bitcoin price movement. With the innovative use of Bitcoin transaction network data, the predictive models lead to more accurate results which outperform existing models. This methodological innovation also presents new managerial insights from network perspectives.

In the third essay, I focus on data-driven decision-making in contexts of the allocation of disaster relief funds. Specifically, I tackle methodological challenges in disaster management when data are extremely sparse and insufficient in the beginning of the disaster evolution, and slowly become more available and reliable as time unfolds. Here I propose an iterative learning method within the general MAD framework to estimate disaster damage losses using very limited and slowly obtained data. Results show that this iterative learning method leads to highly accurate results with fast convergence of the estimation error to a very low level. The framework and results of this essay can be further used for disaster management and resource allocation in various scenarios.