On Draft Circular A4, Behavioral Biases, and Discount Rates

Download this comment (PDF)

Introduction

I am Emeritus Professor of Strategic Management and Public Policy at the George Washington University School of Business, where I have been since 1988, and a Senior Scholar in the Regulatory Studies Center at GW. I received my Ph.D. in Economics from the University of Chicago in 1978, and I graduated magna cum laude and Phi Beta Kappa from Georgetown University in 1972. From 2001 through 2004, I served as the Director of the Bureau of Consumer Protection at the Federal Trade Commission (“FTC”). In 1987-1988, I was Chief of the Human Resources and Housing Branch in the Office of Information and Regulatory Affairs. From 1981 to 1987, I served as the FTC’s Bureau of Consumer Protection Associate Director for Bureau of Consumer Protection Associate Director for Policy and Evaluation, Acting Deputy Director, and Assistant to the Director. I was also an economist with the Bureau of Economics from 1977 to 1981.

I am submitting this comment as an academic with a long career participating in regulatory decision-making and academic analysis of regulatory policy, particularly consumer protection policy and information regulation. It is not submitted on behalf of any other person or organization.

As is well recognized in the cost-benefit community, cost benefit analysis informs decisions. It does not make them. Rather, it identifies the objective consequences of potential decisions, both good and bad, using the best possible economic analysis. There are always other legitimate policy considerations and consequences that cannot be quantified reliably that are relevant to the decision itself. The task of cost benefit analysis is to highlight those choices, so policy makers can be held accountable for their decisions.

The utility of the cost benefit approach is reflected in the growth and persistence of its use since the early days of the Carter Administration. Unfortunately, some of the changes proposed in Circular A4 seem more designed to embed policy choices in the analysis itself, thereby greatly reducing its utility. A purportedly scientific approach that cloaks policy judgments in seemingly technical modeling choices does little to inform policy decisions. Instead, it seeks to justify them, making sure the analysis comes out “right.” That approach is likely to lead to repeated revisions to the A4 as the political winds blow, a consequence that can only undermine the utility of the analytic approach.

The proposed revisions to Circular A4 raise numerous issues that, given the time limitations, I have chosen not to address. I confine my attention to three important issues: the draft’s consideration of behavioral biases both as a rationale for regulation and as a basis for “adjusting” evidence-based assessments of the value of benefits and costs, distributional considerations, and the determination of the discount rate. I do not intend to suggest that I agree with other changes that are not discussed.

Behavioral Biases

One of the most problematic aspects of the proposed revisions to A4 is the draft’s treatment of behavioral biases. The draft would recognize behavioral biases as a new rationale for regulatory intervention that is allegedly consistent with the principles of the long-standing executive order establishing regulatory philosophy. It would also sanction “adjustments” to values used in a cost benefit analysis based on behavioral biases. Both changes should be rejected.

Many economists who have proposed regulatory interventions to address behavioral biases have recommended careful cost benefit analysis to make sure that proposed solutions in fact make consumers better off. The draft circular stands that advice on it head: It proposes instead to “adjust” the cost benefit analysis based on behavioral biases.

As the draft circular notes, addressing behavioral biases requires abandoning the central assumption of cost-benefit analysis, that consumers maximize their own well being subject to the constraints they face. If agencies can replace actual effects with adjustments based on asserted biases, little remains of objective analysis of effects.

Nowhere is this clearer than in the draft’s example that “where there is evidence that manipulative, rather than informational, aspects of advertising influence” on prices, “the observed or measured [values] should accordingly be adjusted” (p. 30). Decades of effort, however, have not produced any convincing decomposition of advertising into “manipulative” and “informational” aspects.

The informational impact of advertising is clear in the empirical evidence that advertising lowers prices, encourages product innovation, and narrows the disparities between different demographic groups.[1] Evidence for “manipulative” impacts is in the eye of the beholder. Consumers have numerous subjective preferences for products where there is no objective difference in relevant objective characteristics, whether it is products that are kosher, or organic, or free range, or Made in USA. Of course, providing truthful information may persuade consumers to make different choices, which is the rationale for most disclosure requirements. But that does not constitute manipulation and offers no objective basis for “adjusting” consumer preferences as revealed in the marketplace. Other “biases” are no different. Similarly, we may have evidence that some other bias exists, but there is no generally accepted methodology for decomposing observed values into “true” values and biases.

At the very least, the circular should demand evidence that when agencies rely on alleged biases either as the basis for intervention or to adjust cost benefit analyses, they produce evidence that the bias actually affects market outcomes or valuations. The mere fact that a bias can be shown in experimental settings is not enough to justify either intervention or adjustment, because the market itself includes checks on the impact of biases. Few would contend that all consumers always make perfectly rational choices, but it does not follow that market outcomes are distorted, or even affected.

Examining actual market outcomes is particularly important because there is no theory of which the 90+ biases listed in Wikipedia might be relevant in a particular context, or how best to correct them. For example, behavioral economists have sometimes argued for cooling off periods to offset myopia or hyperbolic discounting by giving consumers time to think about their decision. But one might also argue that a cooling off period reduces the perceived risk of purchase, and that once the purchase is initiated, the endowment effect suggests that consumers will not change their decision. Empirically, most frauds offer money back guarantees (albeit seldom honored), which would seem to suggest that the perceived risk effect is more important than additional time to reconsider a decision. But there is no a priori theory that predicts this result.

Similarly, some have argued that behavioral biases justify restrictions on various features of credit cards. Some have argued that rewards cards will lead to more revolving, and will increase revolving over time. In fact, we found less revolving on rewards cards compared to other cards, and less revolving over time. Some have argued that biases will lead to more revolving on cards with no annual fee, but in fact there is less. The evidence is consistent with rational choice, not predictions based on behavioral biases.[2] Clearly, evidence of actual market effects of a perceived bias is essential.

At least three factors limit the impact of individual biases on market outcomes.[3] First, and most important, is competition itself. For example, framing effects are well known – how a choice is presented may affect how consumers make that choice. But firms have strong incentives to frame choices in ways that are as attractive to consumers as possible. No doubt Miller Brewing benefited from framing when it positioned its then-new beer as “Lite,” rather than the equally accurate “lower alcohol” beer. The framing of “lean finely textured ground beef” as “pink slime” was spectacularly successful in reducing sales, albeit temporarily.[4] Similarly, competitors in the market for hot dogs may frame their offerings as “low fat,” “turkey,” “all beef,” or “kosher.” There is little reason to believe, and no evidence, that these competitive framings of the choice in any way lead to suboptimal consumer choices.[5]

Another well known behavioral bias is choice overload – presented with too many choices, consumers may decide not to choose at all. The phenomenon has clear support in experimental results. Nevertheless, supermarkets and superstores offer tens or hundreds of thousands of SKUs, without filling up with paralyzed shoppers. These retailers, as well as other sellers, have strong incentives to organize the options in ways that facilitate consumer choice. It seems apparent that they are successful.

Second, market outcomes are driven by the marginal consumer, not the average consumer. If the marginal consumer is unbiased, the market outcome will be exactly the same as if the bias did not exist, even though many inframarginal consumers display the bias. The issue is perhaps clearest with imperfect information, where it has long been recognized that as long as there are enough informed consumers to be worth competing for, competitive prices will prevail even though many, or even most, consumers are uninformed.[6] Even in standard form contracts, the marginal informed consumer drives the contract terms that are offered to all consumers.[7] Theoretical models of behavioral biases also reach similar conclusions, where competition for “sophisticated” consumers results in contracts or prices that also benefit “naïve” consumers who display the bias.[8] At the very least, such possibilities demand a careful look at actual market outcomes.

Third, market participants learn. When consumers make choices that do not maximize well being, they suffer losses, thus creating incentives to make better choices. Consumers will learn from their mistakes, particularly when they suffer actual losses (as opposed to opportunity losses).[9] Consumers on average choose the right credit contract, and are more likely to change when they make larger mistakes.[10] They learn from the experience of paying late fees to avoid future fees,[11] and learn rapidly to make optimal decisions about which telephone pricing scheme to choose.[12] For example, pricing plans that are too complex create incentives to offer and promote simplified plans.[13] In trading card markets where some have found endowment effects, “individual behavior converges to the neoclassical prediction as market experience increases.”[14]

Behavioral economics has much to teach about how to regulate most effectively. Indeed, behavioral considerations imported from the marketing literature, particularly regarding information processing, have been key to the FTC’s consideration of disclosure requirements since at least 1980.[15] But it does not provide a foundation for regulatory intervention on its own. At the very least, the circular should demand evidence to justify regulation or to “adjust” observed valuations.

Distributional Considerations

The distributional consequences of regulatory decisions are an important policy consideration that are all too often overlooked. No one wants to use regulatory decisions to take money from lower income people to produce benefits for the wealthy, although some regulations may do exactly that.[16] Ideally, cost benefit analysis should highlight such choices. The revised circular’s emphasis on conducting a distributional analysis when feasible and appropriate is therefore useful and appropriate.

The same cannot be said about the “suggestion” that agencies consider a weighted analysis, giving more or less weigh to different income groups. The draft argues that weighting may be appropriate because of diminishing marginal utility of income. For each individual, diminishing marginal utility is a reasonable assumption, but using it to weight different income groups violates a fundamental principle of economic analysis: it makes interpersonal utility comparisons, for which there is no scientific basis.

The logical fallacy is most apparent when applied in a different context. We know that marginal costs are generally increasing, so it must be the case that larger firms have higher marginal costs than smaller ones. Both logically and empirically, however, we know that is not the case. The fallacy arises because although each cost function exhibits increasing marginal costs, different firms have different cost functions.

The same is true for consumers. Different people care about different things. Each may experience diminishing marginal utility of income, but it does not follow that the marginal utility of higher income consumers is lower than for those with lower incomes. There is simply no objective or scientific basis for making such a comparison. Favoring one group over another may be a legitimate political choice, but it should not be concealed in a supposedly objective analysis. OMB should encourage distributional analyses to illuminate that choice, but it should discourage approaches such as weighting that purport to find an objective basis for the comparison.

The draft should also address the question of appropriate income measures for use in distributional analysis. The Census Bureau measures income by counting only cash payments, excluding, for example, employer-paid health insurance premiums (regarded as in-kind, not cash), government payments for Medicare and Medicaid (in-kind), payments under the Supplemental Nutrition Assistance Program (food stamps; regarded as in kind because the debit card can be used only for food), and even cash payments received under the earned income tax credit (treated as a negative tax).[17] These payments are substantial, particularly for lower income households, and are surely part of the “income” relevant to the marginal utility of income. The difference is literally an order of magnitude: in the lowest income quintile, average earned income in 2017 was $4,908, but income after taxes and transfers was $49,613.[18] The circular should be clear that agencies conducting a distributional analysis should use a more comprehensive concept of “income” than does the Census Bureau. Separately, OMB should use its authority over statistical policy to develop a more reasonable measure of income.

Finally, conducting a distributional analysis requires honesty about who bears the costs. There is hardly any prediction in economics that is less ambiguous or uncertain than the conclusion that increases in production costs will lead to increases in prices. Production cost increases are not a tax on residual income that may be shared by workers, consumers, and shareholders, they are a direct, price-determining factor. In a competitive industry with constant returns to scales, all of the increase in costs will be paid by consumers in the form of higher prices. The draft’s suggestion that a “manufacturer may be able to pass on a portion of those costs to its customers” (p. 64) is therefore wildly inappropriate. Consumers will pay increases in production costs, and a distributional analysis that assumes away that simple fact distorts differential impacts and undermines the rationale for distributional analysis in the first place.

The Discount Rate

It is not my purpose to address discounting comprehensively, but some narrower points are important and should be addressed in a final version of A4.

First, whatever the merits of the 30 year time horizon OMB adopted in 2003, the last 30 years are a seriously distorted base for estimating the social rate of time preference. For nearly half of that period, since 2008, short term interest rates have been near zero, a fact that says a great deal about the Federal Reserve’s monetary policy, but nothing at all about the social rate of time preference. In any historical context, these near zero rates are anomalies.

Low short term rates inevitably depress long term rates as well, especially as the Fed shifted to purchases of long term securities precisely to drive down long term rates. The influence of monetary policy on long term rates is clearest in the real rates on 10 year TIPS securities, which are negative for portions of the period. Plainly, however, a negative discount rate is nonsensical. Rates during this period are a good reflection of the Fed’s balance sheet, but they are not a good estimate of the social rate of time preference. There may have been a trend toward declining real rates over time that existed prior to 2008, but an estimate that relies heavily on the years since 2008 is likely to overestimate any actual decline.

Second, OMB uses the consumer price index for all urban consumers to calculate the real interest rates from market data on nominal rates. The CPI, however, is inappropriate for this purpose. The CPI uses a fixed basket of goods, adjusted only at relatively long intervals. It therefore overstates inflation, because consumers in fact change the bundle of goods they purchase in response to price changes. The substitution reduces consumer costs, but the CPI ignores that effect.

Chained price indexes avoid this substitution bias, essentially by adjusting the weights for different price changes each period rather than waiting. The chained CPI is only available since 1999, and is used to adjust tax brackets. The personal consumption expenditure index is a very similar index available over a substantially longer period. The PCEI is also the Federal Reserve’s preferred inflation measure, used as its policy target.

Over the period relevant to OMB’s calculation of real rates, the difference between the two indexes is substantial. From 1993 to 2002, the CPI-U shows an increase of 27.3%.[19] In contrast, the PCE index rose 17.8%.[20] Thus, on average over the relevant period, annual inflation was one percentage point higher measured by the CPI. Using the more appropriate PCE index would therefore increase the estimated real interest rate by roughly one percentage point.

The TIPS securities that OMB uses from 2003 on are also underestimates of the real rate because the principal is indexed to the CPI-U. (Interest is paid on the adjusted principal each period.) The advantage of using the TIPS is that it incorporates actual market expectations of future inflation, but it continues to use an inappropriate inflation measure. Because the CPI-U overstates inflation, the real principal is increasing over time using a more appropriate index, and the real value of interest payments increases as well. Investors will of course pay more for a payment stream that increases in real value over time, leading to lower yields than would result from a more appropriate price index. Again, the difference is substantial: From 203-2022, the PCEI increased 49%, while the CPI-U increased 59%.[21]

OMB should recalculate real rates using a more appropriate price index. And it should either discount the recent zero interest rate experience, or use a much longer time period to reduce the influence of what are plainly anomalous observations.