Thesis on household finance in selection markets

Abstract

The credit market and insurance market are two leading examples of selection markets in which consumers vary in their willingness to pay for the products as well as in the costs they incur to sellers. This thesis contains three essays studying behavioral household finance related to credit risk and health risk.

In the first essay, we investigate the value of borrower gaming behavior in assessing credit risk. Using a unique dataset that contains daily, individual-level information on video game activity and credit cards, we discover that past gaming activity predicts future credit card default rates. Default rates are higher among individuals who spend more, more frequently, and more erratically on video games and those who have more and more diverse games on their mobile devices. These results are less pronounced when game spending occurs on weekends and bad weather days. Differences in financial literacy, income, income variability, education, and demographics do not drive the results. Furthermore, intense gamers increase luxury, addictive, and impulsive expenditures more than others after receiving credit cards. The results are consistent with neurological and psychological studies stressing that excessive gaming is associated with impulsivity, addiction, and other self-control problems.

The second essay provides transaction-level evidence on the importance of high-quality payment data in credit screening. We transform the borrower’s “soft” payment flows into over 2,000 “hard” and quantitative behavioral features for screening. Leveraging a series of machine learning models with their strengths in both factor selection and expanding the information content, we show that even a small number of behavioral predictors can strongly outperform the sociodemographic information set in default prediction and help reduce the predicted default rate by 30%. In particular, some fundamental payment behaviors like the payment time or method preferences can well predict the borrower’s creditworthiness. We also find a set of novel behavioral traits with significant predictive power relative to others, such as gambling with lotteries, consuming addictive goods, digital tipping on social media, and donating to others. We discuss the economic interpretations of the prominent behavioral features.

The third essay examines how much the individual insurance decisions are shaped through information spillovers in the social network. Specifically, I combine a unique panel of consumer transactions on medical bills and health insurance policies with matched social network data to identify the effects of peer health risks on individual insurance decisions. Employing a social learning model, I formalize the established heuristic in psychology research that individuals often self-evaluate the underlying health risk by recalling and learning about recent health events among their acquaintances. I empirically test the model implications by exploiting the sharp timing of peer health events as the information shocks in one’s social circle. The event study design shows that a young peer’s hospital-visiting event causes her close friends’ insurance coverage to increase by 16% within 12 months after the peer event. I also find substantial heterogeneous peer effects in aspects of the social pair’s degree of closeness, the peer’s social influential power, and other network characteristics.