Time Inconsistency in the Credit Card Market

Thumbnail Image


umi-umd-2049.pdf (346.29 KB)
No. of downloads: 3018

Publication or External Link






Does consumer behavior exhibit time inconsistency? This is an essential, yet difficult question to answer. This dissertation attempts to answer this question based on a large-scale randomized experiment in the credit card market. Specifically, we apply both time consistent preferences (exponential) and time inconsistent preferences (hyperbolic) to study two puzzling phenomena in the experiment.

The two puzzling phenomena seem to suggest time inconsistency in consumer behavior. First, more consumers accept an introductory offer that has a lower interest rate with a shorter duration than a higher interest rate and a longer duration. However, ex post borrowing behavior reveals that the longer duration offer is better, because respondents keep on borrowing on the credit card after the introductory period. Second, consumers are reluctant to switch, and many of those consumers who have switched before fail to switch again later.

A multi-period model with complete information is studied analytically, which shows that standard exponential preferences cannot explain the observed behavior because they are time consistent. However, hyperbolic preferences that are time inconsistent come closer to rationalizing the observed behavior. In particular, two special cases of hyperbolic discounting are carefully examined, sophisticated and naive. Sophisticated consumers prefer the short offer because it serves as a self-commitment device. Naive consumers prefer the short offer because they underestimate their future debt.

To further explore the possibility of explaining observed behavior by exponential preferences, we study a dynamic model in which realistic random shocks are incorporated. Estimation results show that consumers have severe self-control problem with a present-bias factor (0.8). It is also shown that the average switching cost is $150. With the estimated parameters, the dynamic model can replicate quantitative features of the data.