Taken together, our results are consistent with models in which consumers have limited and reactive attention to add-ons like overdrafts, and suppliers respond by shrouding add-on costs. Specifically, it seems that overdraft costs and availability are not at the top of consumers’ mind, and even when brought closer to top of mind they do not stay there for long. As such, recent behavioral models of add-on pricing, marketing, and usage capture key aspects of reality, with consumers that tend to underestimate add-on costs and react strongly but temporarily when their attention is drawn to add-on, and with firms that lack incentives to unshroud or compete on add-on costs.
Classically rational models can explain at most a subset of our results. In principle, one could rationalize the finding that a large price discount decreases overdraft demand with a wealth effect that is big enough to counteract the standard price effect. In practice, such a wealth effect seems implausible. First, the wealth effect would likely operate among those who actually overdraft and have the most wealth to gain, yet we find effects on the extensive margin.
Second, the demand reduction does not persist after the promotions stop, which implies that any wealth effect would have to be transient and begs the question of why, since the overdraft discount is conditional on usage and hence does not alleviate liquidity constraints. Third, a wealth effect should operate through nonoverdraft discounts as well, yet we do not find any evidence that debit card or automated bill payment discounts alone reduce overdrafting. Fourth, a wealth effect does not explain why overdraft availability messaging also affects demand.
Another potential explanation for the overdraft price discount backfiring, particularly in light of Johnson, Meier and Toubia (2015), is that consumers view the offer as “too good to be true,” that is, they mistrust YK. But mistrust would not readily explain our other key results on availability and lack of persistence. First, it is silent on why availability increases demand and begs the question of why discount-driven mistrust would dissipate almost immediately after the campaign ends. Second, it is not clear why consumers would mistrust the overdraft interest discount but not other deep discounts that prevail in equilibrium, like “free” checking and teaser rates on credit cards. Third, a mistrust channel need not be distinct from the behavioral mechanisms described above; indeed, Johnson, Meier, and Toubia (2015) find that some “households expect there to be hidden fees and cumbersome processes that are not compensated by the attractiveness of the offer.” Fourth, it is not clear why our sample would respond by decreasing demand for overdrafts rather than simply ignoring the offer: do consumers assume that hidden costs exceed the value of the discount? Fifth, if YK’s clients did think that YK was trying to trick them, we might expect them to reduce their demand for other YK services. Yet, we find no such evidence (see Appendix Table A2), and the point estimates of the effect of the overdraft discount on the number of active YK accounts are actually positive rather than negative. Sixth, several institutional differences between our experiment and Johnson, Meier, and Toubia’s (2015) make mistrust more important in their context. In particular, the offer in Johnson, Meier, and Toubia (2015) was too good to be true in the sense of being government-subsidized, and that offer was made by a mortgage servicer at a time when the servicing industry was known to be mistreating and scamming customers (e.g., Consumer Financial Protection Bureau (2013)).
Rational inattention, and/or high search costs as in Ellison (2005), could explain our results, but under assumptions that strike us as antithetical to those sorts of models. For example, instead of remembering new information, consumers quickly forget it. Perhaps more critically, consumers systemically underestimate costs and credit lines instead of having accurate (if noisy) perceptions. But does it make sense to think of consumers as rational if they hold biased perceptions of contract terms in equilibrium? Much of behavioral economics answers this question with an emphatic no—indeed, it draws the line between rational versus behavioral based on a distinction between mean-zero versus biased deviations from classical assumptions about decision-making.
Relatedly, one could rationalize the bundled discount results as due to reduced attention costs more broadly, rather than to increased attention to add-on costs in particular, as the key factor mediating consumer choices. Specifically, inducing use of autopay or debit cards could increase the customer’s engagement with the bank, thereby lowering costs of monitoring her cash flows and reducing overdrafts.18 But messages that discount only the bundled products and do not mention overdraft do not reduce overdraft usage. Thus, the mechanism seems unlikely to hinge only on a (rational) increase in attention to the checking account more broadly. Some heightened awareness of add-on costs is likely key.
Notwithstanding the above, we are not dismissing rational or near-rational explanations for our results. Rather, we are merely speculating that behavioral models of limited attention, memory, and shrouding have great potential to explain the full picture.
