Financial activities have been central to the survival and prosperity of human societies. In modern days, we humans are carrying out financial activities on a daily basis, from setting up retirement funds to purchasing a cup of coffee from the local store. Every financial decision we made is in some way contributing to the invisible hand that propels the market forward.
The studying of financial behavior has been a long and hard battle. Starting from the realization that participants in economic activities are human and thus not fully rational, researchers from many fields have tried to shed lights on the financial behavior of individuals, ranging from psychology, neuroscience and behavioral economics. However, for many years, due to the sensitive nature of financial information,
and the resulting lack of large-scale data, the studies of financial behavior are typically in the forms of small-scale lab experiments, an environment that is quite different from real-world economic settings.
Fortunately, with the digital age, financial activities are increasingly moving online, exemplified by the proliferation of e-commerce and digital wallets. Research opportunities arise from the accompanying emergence of large-scale datasets about users' online financial activities in the wild.
Yet challenges still remain as to how to effectively utilize such datasets to truly understand the financial behaviors. The first challenge is the scale of data. With millions or even billions of users involved, it is important to find the appropriate methodology to make sense of such large-scale data. Second, the depth of understanding. Observational studies are limited by what is observable while a full understanding requires one to touch upon the hidden drivers behind the behavior. The third challenge is in how to transform the understanding into actionable knowledge.
In this thesis, we tackle the challenges one by one, measuring and understanding users online financial behavior and finally using our understanding in making financial decisions. We begin with a serious of measurement studies that explore the different dimensions
in online financial behavior, going from collaborative and friendly social payments to the competitive auction biddings. We show that the relationship between users plays a key role in forming their behavioral patterns and that financial behaviors are highly distinct even within the same system, potentially driven by different sets of motivations.
Next, we deepen our understanding by exploring the underlying motivations behind user behaviors. Leveraging existing works in social science and behavioral economics, we develop hypotheses regarding users financial behavior and verify our hypotheses using surveys, interviews as well as empirical measurement.
We uncover various aspects that shape financial behaviors such as the cultural differences in user behaviors during the adoption of
digital wallets and how auction mechanisms affect users reactions to competitions.
Finally, we explore how to use our models to guide financial decisions. By building a faithful model of bid-by-bid behavior during penny auctions, we have essentially built a testing platform to experiments with both prediction-based and learning-based decision strategies.
Using such models, we are able to identify a prisoners' dilemma embedded in the penny auction environment when multiple bidders adopt the same strategy.
In summary, this thesis presents how a combination of data-driven approach and behavioral theories can help understand and guide financial behaviors. We have developed methods to identify diverse sets of behavioral patterns in financial systems, introduced ways incorporate existing theories into the study of online financial behavior, and show how developed models can help navigate the vast decision space in the financial systems.
