With all the talk about the CFPB, Elizabeth Warren has been in the news lately. The blogs too. Most of the discussion has been about whether or not Timothy Geithner is a friend or foe to the Democrats’ preferred option of getting Warren nominated as the first chief of the CFPB. Today, Megan McArdle started on what is a less interesting political topic, but a more interesting one for this blog with a long and detailed post on Elizabeth Warren taking on then-Professor Warren’s use of theory and data in the Two Income Trap and her controversial work on medical bankruptcies. McArdle later doubled-up with a be re-posting Todd Zywicki’s WSJ op-ed pointing out the odd manner in which tax data are presented in Two Income Trap. Put directly, Zywicki provides some evidence that the presentation (made in an attempt to show the increasing burdens of mortgage, car and health obligations) presents the data percentage terms in order to obfuscate the fact that changes in tax obligations play a much larger role in the economic burden facing the middle class than convenient for the story told in the book.
Larry’s post responding to an earlier blog post from Professor Warren relates not exactly to empirical skills, but empirical teaching and (do read the post) a view on law, theory and facts that reveals what I take to be a fairly deep misunderstanding of economic theory and methodology. David Evans and I make a similar point on the use and abuse of the behavioral economics literature to bolster the case for a consumer protection agency. Larry’s post also reminding me that as long as we are reprising old posts, I have one directly on topic that provides an example that I think gets at what McArdle, Zywicki and Ribstein are each discussing. Frankly, I also think it is a bit more interesting than both the tax and medical bankruptcy examples for the purposes of discussing the CFPB because I believe Warren’s actual data skills matter less than how she interprets data for legal and policy applications because she will have a large staff of competent economists to do data analysis. I do not suspect she will be running many regressions on her own.
The exercise is a fairly simple one. Warren authored a blog post pointing to an empirical study by Agarwal, Liu, Souleses, and Chomsisengphet (“ALSC”) which examines consumer credit card selection in a natural experiment setting in which a card company offers two cards to consumers: (1) a high interest rate, no annual fee card and (2) a low rate card with an annual fee. A bullet point summary of some of their findings:
- About 60% of consumers get the decision right with the benefit of hindsight
- 40% do not make initially select the right card
- Many of these initial errors are subsequently corrected as a result of consumer card switching, while ALSC report that “a small minority of consumers persists in holding substantially sub-optimal contracts without switching.”
So how are we to interpret these data? Going back to Larry’s post, recall the importance of economic theory as a lens through which to view “facts.” Warren asks whether “these data support the notion legal policy can be shaped by the presumption of economic rationality, or do the data support a call for more regulation?”
Its a good question.
Warren’s answer: is more regulation in light of what she describes as the “staggering” 40% error rate. Professor Warren writes:
Would it help to frame the policy question is from the provider angle? What’s the point of offering two different products, except to hope that the number of consumer who get it wrong will exceed in dollar volume the number who get it right. Or, from an informed consumers’ perspective, perhaps the optimal system is one in which they make good decisions and hope for cross-subsidization from less-clever consumers who help keep credit cards highly profitable and easy to use in a variety of settings (e.g., grocery stores, cabs, pizza deliveries, etc.). I realize it is heresy in many circles to ask if consumers should have fewer choices. But at some point the empirical studies about high error rates bring into question the assumptions that underlie the claim that more choice is always good.
Its a simple answer. High error rate implies irrationality. And irrationality implies regulation. Its important to note this particular answer is not without an economic model of its own to interpret facts. And it is not unlike the leap from evidence of irrationality to conclusions about market failure that Commissioner Rosch has made in the antitrust context. But I think its the wrong model (because it commits the Nirvana Fallacy). More importantly for our purposes, this is a great example where interpreting the data carefully can lead to vastly different policy conclusions which I discussed in my earlier post. And perhaps most importantly, this is an example involving interpreting data in the credit card market, something I suspect the head of the CFPB will be called upon to do frequently in forming policy. From here on out, I’m going to copy the earlier post verbatim:
While the burden of proof is on Warren and others advocating more regulation here to demonstrate that less choice would improve consumer welfare, not only does this study not satisfy the burden, I think a reasonable interpretation of the results cuts the other way. The results suggest that consumers making credit card contract decisions behave rationally, the initial error rate is not strong evidence of consumer irrationality in light of relative costs and benefits of card switching, and the error costs are very small. A little context is necessary to make the case for this interpretation of the data, as well as the reporting of some key results in the ALSC paper that Warren does not discuss in her post but shed light on the question of consumer rationality in the credit card market. In light of these findings, discussed below the fold, I think it is pretty clear that these findings support a standard economic model of credit card borrowing.
First, and consistent with standard economic theory, the consumer error rate decreases in the cost of the error as well as the number of times a consumer makes the decision. In other words, consumers correct bad decisions with repeat play and perhaps most importantly, make fewer errors when stakes are higher. It is difficult to square these findings with models of irrational consumer behavior. As an aside, economic theory does not suggest that consumers are immune to errors! At the very least, an error rate that decreases in the cost of error is inconsistent with the simple behavioral/ consumer irrationality-based models of consumer behavior that frequent the legal literature. By the way, another empirical study using micro-level data on this and related questions (Brown & Plache (Paying with Plastic, 73 U. Chi. L. Rev. 63 (2006)) reaches very similar results concerning consumers’ abilities to select credit card contracts optimally and is a paper very much worth reading for those interested in this topic.
Second, what about the “staggering” magnitude of the social cost involved with the initial errors? Warren does not report that ALSC report that these error costs are generally bounded in magnitude by the size of the typically small annual fee (see Table 3, the median fee is $25). The most common “behavioral” call for regulation of the credit card market is the claim that unsophisticated users will be seduced by cards with low annual fees and higher interest rates, unknowing that this decision is sub-optimal ex post and incurring large chunks of debt at higher interests rates as a result. However, ALSC find that of those consumers in their panel that do not pay annual fees, the net annual error costs exceed $200 for only 225 out of over 64,000 no-fee accounts.
If one knew the credit card market only by reading the legal literature, the most staggering feature of the ALSC (and Brown & Plache) results would be shock at how often consumers are selecting contracts optimally, switching cards, minimizing error costs. To be sure, there are a very small margin of consumers who make persistent errors. But what are we to make of this group in the context of a decision where the costs of getting the decision “right” are, on average, bounded by the magnitude of the annual fee? How large are these costs relative to the costs of switching cards, or better yet, of regulation? Of the consequences (intended and otherwise) that regulation might have on these consumers?
So, do the ALSC findings support the inference that credit card consumers are irrational and in need of regulation or less choice? It appears not. After all, most consumers here are selecting cards optimally: i.e. they are better off because of their decision! Further, the cost of failing to do so is low, and yes, reducing consumer choice is likely to reduce welfare in this setting. If those advocating regulation here have a more nuanced view of these results that supports the view that prohibiting a menu of contracts would increase welfare, I would like to hear it. Note: my naive cost/benefit analysis also does not account for other obvious (and likely enormous) benefits of credit card spending versus cash or other payment forms for the large proportion of users that do not revolve debt.
Third, the relatively small error costs (and the fact that the majority of credit card users appear not to revolve balances at all) recast the 40% initial error rate in a different light in my view. If the expected benefit of card switching is small relative to search costs, it is not especially surprising that many consumers do not incur these costs. These findings, to me, suggest caution about overstating the magnitude of the effects of “consumer irrationality” in the credit card market.
As to Warren’s question of whether it would be helpful to reframe the policy question as something like: “why WOULD the provider offer multiple products but for the exploitation of consumer error?” No. I don’t think this reframing is helpful. It is the wrong policy question to be asking. Yes, we care about an explanation of why providers offer multiple products — but isn’t the dispositive question here whether regulation would improve welfare relative to the status quo net of the costs of regulation? Further, I can think of one obvious reason why the provider would offer multiple products just like multitudes of suppliers in other markets: consumer heterogeneity in demand for different credit card attributes.
Consumers who regularly revolve debt might want a different type of credit card than convenience users that pay off their debt every month and value different card attributes. Providers offer different products to different types of consumers because the consumers value different card attributes and consumers select cards accordingly. The ALSC results suggest that they do so pretty well, and it is well known that this sort of competitive price discrimination is generally welfare-enhancing.
In sum, these findings suggest that consumers generally behave rationally in the credit card market and select the optimal card. When they don’t do so initially, the probability of error in subsequent decisions decreases in the costs of error. In the context of a decision where the costs of errors are low, one should not necessarily be surprised at a high error rate. Persistent errors may well consist of “rational ignorance,” and it is a significant leap from high error rates to, as Warren puts it, “bring[ing] into question the assumptions that underlie the claim that more choice is always good.” The leap becomes much more daunting in light of the ALSC findings Warren does not report: error rates decrease in the size of the error and in experience with the decision. Sounds like fairly rational behavior to me.
There are a number of interesting things to comment on here, but I will limit myself to two for now.
The first is that a close, careful reading of the evidence does not “call into question the assumptions that underlie the claim that more choice is always good.”
The second is that contra the post Larry Ribstein responds to, here is an example where economic theory is not a substitute for empirical facts in the lawyer or regulator production function but rather economic theory helps one see the facts clearly. Oddly, the important data suggesting rationality and limits to the social costs of consumer mistakes are ignored. That is important. And not unlike Zywicki’s tax obligation example. But I think the most important point is that if one does not understand the fundamental economic concept that the optimal error rate in the credit card market is positive because of information and switching costs, one observes an error rate above zero and concludes that there is market failure. An economist would ask the “compared to what” question, i.e. about the optimal error rate. An economist might also look closely at the magnitude of the social harms discussed, as well as evidence of learning which reduces social costs, and compare those costs to the perceived benefits of a proposed regulatory solution. But the fundamental point is economics is not a substitute for facts, it helps understand them in a legal and policy relevant sense.