Pharmaceutical Regulation

• “Improving the FDA Approval Process,” by Anup Malani, Oliver Bembom, and Mark van der Laan. October 2011. SSRN #1945424.

Under current U.S. Food and Drug Administration policy, pharmaceuticals are approved for sale only if they are safe and effective relative to the current standard of care, or a placebo, for the entire treated population. While the agency understands that a drug may have positive effects only for certain subgroups, such as women or young people (and, in fact, clinical trials may be designed with subgroup analysis in mind before they start), no data analysis of subgroups to determine if such benefits exist is allowed after a trial has ended.

The reason for the prohibition is that the probability of a false positive result (the conclusion that a drug is efficacious when it really is not) is increased by such subgroup analysis. The probability of finding a false positive result among 10 subgroups (if you use the usual criterion of keeping the chance of a false positive in each subgroup to less than 5 percent) is 1 – 0.95ⁿ, where n is the number of subgroups. For example, if a trial has adults whose ages range from 20 to 70 and we divide the data into five-year age bins (10 subgroups), the cumulative probability of a false positive from among the 10 age subgroups would be 1 – 0.95¹⁰ = 40 percent.

The FDA could institute a statistical correction, called the Bonferroni correction, that would change the acceptable rate of false positive inferences in each subgroup to account for the existence of all the subgroups. For example, if the acceptable rate of false positive results is 0.05 (5 percent) and there are 10 subgroups, then the correction would be 0.05 ÷ 10 = 0.005. Thus in order to be 95 percent confident that the differences between the treated and controls in any subgroup were real rather than the result of chance, we would actually have to act as if we were 99.5 percent confident rather than 95 percent confident.

The FDA does not allow companies to self-report the number of subgroups investigated after the completion of trials because currently no way exists to verify the number. If a trial has 10 age subgroups as well as gender and ethnicity (white, black, Hispanic, other) subgroups, then that would mean there is a total of 10 × 2 × 4 = 80 subgroups. Under normal procedures without correction, the possibility of false positive findings from at least one subgroup would be 1 – 0.95⁸⁰ = 98 percent. Drug companies would have strong incentives to say they looked for positive results in only a few subgroups (thereby lowering their officially announced n) rather than admitting they looked at all 80 subgroups so as to increase the possibility of FDA approval of a drug for use in a subpopulation.

The authors propose a two-step solution to this credibility problem. First, an independent consulting firm identifies promising subgroups (groups whose treatment effects appear to be real in a small subsample of the trial data). Second, the sponsor of the trial would perform analysis of the health outcomes of the promising subgroups on a different and larger subsample of the trial data and implement the Bonferroni statistical correction to the subgroup results.

While this paper proposes a clever solution to problems of scientific inference in clinical trials, the most important problems in the current FDA regime may not be scientific. Instead, problems result from a lack of clarity between where science stops and values begin. The current FDA pharmaceutical approval process has two components: a genuine scientific enterprise in which clinical trials are conducted to generate knowledge about the safety and efficacy of new drugs relative to a current standard of care (or placebo), and a value-laden decision as to whether the safety and efficacy results merit permission to sell to consumers. In a more libertarian world, these two components and the role of government in each would be considered separately.

While a laissez-faire regime of knowledge generation and disclosure may be hard to imagine, it is not logically impossible. That is, firms might have to conduct trials and disclose the results in order to convince patients (or, more likely, the patients’ physicians) to use (recommend) their products. Or, more precisely, some patients would require such information to make informed decisions about use and some firms would generate and disclose while others would not (a separating equilibrium).

If a laissez-faire knowledge regime produces an unacceptably low level of information, an intervention that is more minimalist than current policy would mandate research and disclosure of results, but no organization would decide how the knowledge would guide decisions. That is, there would be no centralized decision masquerading as a scientific decision about the wisdom of the market availability of a drug because such decisions are not scientific.

I have always been troubled by members of scientific advisory committees voting through majority rule to advise the FDA commissioner as to whether a drug should be available to consumers. Acceptable risks and appropriate ratios of costs and benefits are not scientific questions; they are economic questions. So if experts are going to vote on acceptable risks, at least they ought to be economists rather than scientists or physicians. And most economists would argue that the acceptability of risk from using pharmaceuticals is not a collective decision and thus should not be determined centrally by government.

Viewing such decisions within an economic rather than scientific framework also would encourage the public to think of risk and safety more appropriately. Instead of thinking of risk in dichotomous terms (i.e., is a drug safe or not?), the public, if they internalized an economic perspective, would ask how much safety or risk does a drug pose at what cost. And this would encourage the public to realize no right answer exists as to whether a drug’s costs and benefits are worthwhile. Instead, many different answers exist for different individuals—and that is perfectly appropriate.

Wage Inequality

• “Can the Rapid Growth in the Cost of Employer-Provided Health Benefits Explain the Observed Increase in Earnings Inequality?” by Mark J. Warshawsky. September 2011. SSRN #1932381.
• “Measuring the Impact of Health Insurance on Levels and Trends in Inequality,” by Richard V. Burkhauser and Kosali I. Simon. March 2010. NBER #15811.

Increasing inequality in the distribution of earnings has become one of those stylized facts that everyone “knows.” The nightly news reminds viewers that ordinary workers have not fared well in the labor market over the last 25 years, while corporate executives have. Many professional economists and a recent CBO report have supported this view as well.

While it is true that the cash explicitly paid to employees has become more unequal over the last generation, the implication that labor markets are not working well and that government should alter labor market outcomes does not necessarily follow. A more benign explanation for the change in cash compensation over a generation is the dramatic increase in health insurance costs. Employers may be paying all their employees a more or less equivalent increase on a percentage basis, but for lower-paid workers much of that pay is not showing up in cash. Thus, if this view is correct, inequality in the cash component of compensation has increased while inequality in total compensation has not increased because the fixed costs of health insurance are a much larger percentage of the total compensation of lower-earnings workers.

Burkhauser and Simon explore this explanation. They add the value of employer-provided health insurance as well as Medicaid and Medicare to the pre-tax, post-cash-transfer household income data and find that the bottom three income deciles actually exhibit higher growth than the top seven deciles from 1995 to 2008. If one analyzes data on only working-age individuals (age 25–61), inflation-adjusted real pre-tax, post-cash-transfer money income grew 1.9 percent and 10.5 percent respectively for the first (poorest) and 10th (richest) deciles from 1995 to 2008. But if one adds the value of health insurance, the first (poorest) decile grew 12.3 percent while the top decile grew 11.7 percent.

Warshawsky makes a similar discovery. Using unpublished BLS total compensation data, including employer health insurance expenditures, from 1999 to 2006, he finds that the growth in compensation by earnings decile (from the 30^th to the 99^th) averages 35 percent, with 41 percent growth at the 30^th percentile (workers earning $10–$14 an hour) and only 35.8 percent growth at the 99^th percentile (workers earning $59–$80 an hour).

Because expenditures on health care are increasing so rapidly and because so much of the cost of health care is paid for by employers or government, discussions about rising inequality that only consider cash income provide a misleading view of trends in inequality. When health insurance expenditures are added to household cash income, the increases in inequality from 1995 to 2008 are completely offset.

Canadian vs. U.S. Banking

• “Why Didn’t Canada Have a Banking Crisis in 2008 (or in 1930, or 1907, or …)?” by Michael D. Bordo, Angela Redish, and Hugh Rockoff. August 2011. NBER #17312.

In the quest for explanations of the U.S. financial crisis of 2008, the most glaring omission is the absence of any discussion of Canada. As the title of the paper by Michael Bordo and his colleagues suggests, Canada seems immune to financial crises. Why is this so?

The answer they suggest lies in the political and regulatory history of financial markets. This is an interesting answer for an economist because of the long-running debate over whether institutions are simply endogenous and efficient or exogenous and sticky. For those not familiar with this debate, the question is what happens to legal and cultural practices that inhibit efficient market adaptation. One answer is that if efficiency and institutions conflict, then institutions change to align themselves with efficient adaptation. The other answer is that institutions persist beyond their “sell-by” dates and thus historical paths matter.

For much of its history, the United States had a fragmented banking system because of an 1839 Supreme Court case that permitted states to exclude the branches of banks from other states. The fragmented banking system that resulted was ill-suited to the needs of national corporations and industrialization, which instead were served by unregulated financial and commercial paper markets. According to Bordo, the United States has always had something like the shadow banking system described by Yale finance professor Gary Gorton in his recent papers, which I’ve discussed in previous columns.

In contrast, Canadian banks were chartered nationally, like the First Bank of the United States, but Canada did not stop with one bank. Instead, Canada developed a national oligopolistic banking system with limited entry protected by the national government. In contrast, the Canadian broker-dealer and securities market system remained much smaller because the banks were national in character and capable of providing the financing for industrial development.

Another important difference between the U.S. and Canadian banking systems was their response to the inflation of the late 1960s and early 1970s. U.S. banks had interest rate controls while Canadian banks did not. Thus deposits stayed within the Canadian banking system while they fled the U.S. system for money market mutual funds. By the 2000s, more U.S. debt was financed outside than inside the traditional banking system, with much of that financing coming from short-term “deposits” that could flee the alternative shadow system if investors lost confidence.

A third important difference is that Canadian mortgages have fixed rates for a maximum of only five years, thus eliminating the problems inherent in linking shorter-term deposits with longer-term loans.

The combined effect of interest rate controls, long-term fixed rate mortgages, and lack of national branching in the United States necessitated the development of mechanisms to link capital markets directly with housing debt through mortgage securities. None of this occurred in Canada. Mortgages and the deposits backing them stayed within the conservative banking system while our lending shifted to the shadow banking system.

Bordo et al. argue that the crisis of 2008 strongly paralleled the crises of the 1800s as argued by Gorton. The Canadian system was a five-firm oligopoly regulated by a single entity that preserved the profits of traditional banking and prevented unstable lower-cost shadow banking from developing.

Rural Telephone Subsidies

• “The Universal Service Fund: What Do High-Cost Subsidies Subsidize?” by Scott Wallsten. September 2011. SSRN #1927933.

Telecommunications regulation has always been accompanied by cross-subsidy schemes. That is, some services (primarily long distance) have been “taxed” to transfer resources to other services (mostly local-loop access in rural areas). When entry was restricted, the tax-and-transfer scheme was hidden in the excessive price of long distance service. Once long distance deregulation occurred, the scheme became an explicit tax-and-transfer scheme so that local-loop providers would not collapse.

Economists have long criticized this subsidy as an inefficient method of transferring resources because the demand for the taxed service was elastic while the demand for the subsidized service was inelastic. The result has been large deadweight losses probably equal to or more than the resources transferred.

In this paper, Scott Wallsten analyzes another aspect of the program: what do the subsidies buy? He finds that as customers have dropped landlines, subsidies have remained relatively constant. In a regression explaining general and administrative expenses, he finds that, controlling for firm and year fixed effects, each dollar of subsidy is associated with an increase of 59 cents in general and administrative expenses. That is, instead of paying for access loops, the subsidies pay for office staff. Instead of observing economies of scale, Wallsten finds diseconomies of scale: overhead expenses increase with firm size.

Rather than eliminate the program, the Federal Communications Commission recently announced changes to the Universal Service Fund. The agency proposed to end funding of phone services and shift subsidies toward high-speed broadband access. In addition, the agency capped the budget for the subsidies and proposed a competitive bidding scheme to allocate the funds. The latter may alter the incentives for extra administrative expenses

Fuel Tax Holidays

• “Fuel Tax Incidence and Supply Conditions,” by Justin Marion and Erich Muehlegger. March 2011. NBER #16863.

When gasoline prices rise, politicians are under pressure to alter policy to provide consumers with relief. A fuel-tax holiday is one such policy response. Economic theory argues that taxes are fully passed thorough to consumers and thus a gas tax holiday would reduce fuel prices to consumers.

Remarkably, little empirical work has been conducted to verify this prediction and examine whether the pass-through to consumers is altered by changes in supply elasticity. During the summer of 2008 when oil prices reached over $140 a barrel and top-tier presidential candidates Hillary Clinton and John McCain proposed suspension of the federal gasoline tax, economists were quoted in the press as arguing gas tax reductions would not result in lower prices for consumers because supplies in the summer were inelastic.

Marion and Muehlegger examine 20 years of monthly price data and the changes in the taxation of gasoline and diesel fuel and conclude that taxes on gasoline were fully passed through to consumers even when refinery utilization was over 95 percent. This result would appear to contradict economic theory, but only because the authors assume that high refinery utilization implies fixed supply. But in fact, because of the differential taxation of diesel (low tax) and gasoline (high tax) in Europe and the fairly fixed ratio of the two when crude is refined (in the absence of catalytic cracking), Europe has excess gasoline that is a source of elastic supply for the U.S. market in the summer. U.S. demand is very inelastic in the summer, so gasoline tax changes are passed through to consumers even in the summer.

Diesel fuel reacts differently because of its untaxed use as heating oil. In the winter, this untaxed alternative use has the effect of increasing the supply elasticity of diesel as taxed motor fuel (the sellers of heating oil get to keep more money so they divert product away from the taxed diesel to the untaxed heating oil market in response to a tax increase). As a result, the pass-through of diesel tax changes actually increases in states with greater heating oil demand. But for states without much heating oil use, tax changes are not fully passed through to consumers.

Diesel also reacts differently to refinery utilization. Utilization greater than 95 percent reduces the pass-through to less than half the change in the tax. This is consistent with domestic refinery utilization being a better measure of supply constraint for diesel compared to gasoline because of full European utilization of their diesel supply.