Pricing in the US pharmaceutical industry.
Although the worldwide slowdown in productivity growth dates back about 20 years, we know surprisingly little even today about its causes. For some time, we have been suspicious of the reliability of official government statistics on growth in output and productivity, and we have conjectured that some of the reported slowdown may reflect difficult measurement problems. In particular, while government officials typically are able to obtain reasonably reliable data on sales, decomposing changes in sales into their price and quantity components is a rather difficult task. That is particularly true when the number of new goods increases, when forms of retailing change (for example, to increased mail order purchases), and when quality changes.
Given the fact that the value of sales equals a price index times a quantity index, any errors in calculating a price index imply corresponding errors in the quantity index. If inflation is overestimated, then growth in real output must be understated. But, how might we assess the accuracy of widely used price indexes, such as the Producer Price Index (PPI)?
We decided to audit a single industry in considerable detail. Although some of our earlier work focused on adjusting for quality the price indexes for personal computers,(1) we chose the prescription pharmaceutical industry this time (NBER Faculty Research Fellow Joshua Rosett was also a participant). We selected that industry in part because it apparently experienced a sharp slowdown in productivity growth in the 1970s. It also has seen considerable technological change (and therefore there are possibilities for measurement error); its pricing policies have been the focus of considerable public attention; and it has other interesting attributes (for example, it is heavily engaged in research and development, and enjoys protection under the U.S. patent system). Moreover, we were able to obtain microdata on a confidential basis from four major U.S. pharmaceutical companies.
Soon, though, we encountered a mystery. From January 1984 through December 1989, the Bureau of Labor Statistics (BLS) price index for prescription pharmaceutical preparations grew at an annual rate of 9.09 percent. We compared this official data to monthly price and quantity sales data on all 2090 prescription products sold by four major U.S. pharmaceutical manufacturers, accounting for about 24 percent of total domestic industry sales in 1989. Using procedures that mimicked those employed by the BLS (weighting a select number of products by the same factor over time, known as a Laspeyres index), we found that over the same time period, the four-company price index increased at only 6.68 percent per year. Moreover, when we used an index procedure that changed over time reflecting evolving market shares (a Divisia index), the aggregate four-firm price index grew at only 6.03 percent per year. Thus, the focus of our initial research was why the BLS price index grew approximately 50 percent more rapidly than our lower estimate.(2)
We were most concerned about the representativeness of our four-company sample. In cooperation with BLS officials, we obtained the four companies' records of the prices they had initially reported for the products sampled by the BLS. We learned that the rates of price increase for those products grew at an annual rate of 8.94 percent, virtually the same as the reported overall growth in the official PPI. Hence the discrepancy apparently was not caused by these four companies being unrepresentative.
Moreover, when we obtained data from a private sector source (IMS America, Inc.) covering virtually the entire population of one subclass of pharmaceutical products (systemic anti-infectives), we found that while the official PPI for this subclass grew at an annual rate of 6.26 percent, the growth rate based on IMS data grew at only 2.63 percent per year.
After eliminating "list versus transactions prices" and "company weights versus BLS weights" as explanations, we discovered that the age distribution of products sampled by the BLS tended to be much more concentrated than the age distribution for products at our four companies. Moreover, the BLS tended to undersample newer drugs (from zero to four years old), and considerably oversample medium-aged drugs (four to ten years old).
We also found that while prices of younger drugs tended to grow about 3.5 percent slower per year than older (over age 25) drugs, prices of medium-age drugs tended to grow about 2.5 percent faster than older products. Hence, we discovered that a major determinant of the discrepancy in growth rates between the four-company and the BLS price indexes was the fact that the BLS undersampled younger drugs with below-average price increases, and oversampled medium-aged drugs with above-average price rises.
We also examined the impact of very new drugs on aggregate price measures. Differences between the average annual growth rates of price indexes based on the fixed-weight price index and the evolving market share index were rather modest when computations were undertaken over an identical set of goods. However, when the set of goods include in the computations immediately incorporated new and relatively young products, the resulting market share indexes grew at a considerably lower rate than did the fixed-weight index that excluded these new goods. We concluded that in the U.S. pharmaceutical industry, new and relatively young goods promptly into the price index calculations results in substantial overestimates.
Our study entirely neglected one very important issue: the role of generic drugs. In the context of pharmaceuticals, the issue is how to link a post-patent generic drug to its patented antecedent. The U.S. Food and Drug Administration (FDA) certifies a variety of a previously existing drug as being "(bio-)equivalent" to the previously available patented version. The generic differs only in packaging, including the inert matter enclosing the active ingredients, in labeling, and in source. Thus, from the point of view of the FDA, generics and branded drugs are perfectly substitutable.
Generic drugs enter the market after the patent on a drug expires, and typically sell at a considerable discount relative to their patented antecedent. Moreover, it is not uncommon for branded drugs to increase in price after their patent has expired. On a prescription basis, generic drugs today constitute a very substantial and growing portion of the marketplace. But the fact that generic drugs and branded drugs coexist in the marketplace, selling at different prices, suggests that from an economic point of view they are not viewed as perfect substitutes by all purchasers.
Although the FDA treats generic and antecedent patented drugs as being bioequivalent, the BLS treats a generic drug as being entirely unrelated to its patented antecedent. Therefore the BLS does not incorporate into its overall price index the effects of the substantial growth over time in the market share of lower-priced generics. But precisely how one should incorporate generics into the price index, and by how much the resulting index would differ from the one based on current BLS procedures, is not at all clear.
Griliches and NBER Faculty Research Fellow lain Cockburn have examined the empirical implications of incorporating generic drugs into the aggregate price index in several different ways.(3) They stress that traditional index number theory is not particularly informative in this context because it is based on the notion of a representative consumer, while the substantial but less-than-complete market penetration by generics suggests that this market is segmented into at least two groups, one considerably more price-sensitive than the other. Making a variety of assumptions concerning the heterogeneity of consumers' preferences, they examine case studies of two anti-infective drugs: cephalexin and cephradine. They find that in the case of cephalexin "BLS approach" yields an increase in the producer price index of about 14 percent over the 45 months observed in the data. The "FDA approach," in contrast, based on the assumption that branded and generic versions are perfect substitutes, yields a price decline of 53 percent. Using an "adjusted Paasche" index that the authors prefer generates a price index that falls by 48 percent.
Griliches and Cockburn conclude by urging official statistical agencies to move quickly toward a more current sampling of new products, and to incorporate generic drugs into official price indexes by adopting some form of compromise linking generic drugs with their patented antecedents. In fact, the BLS has announced that it will be implementing several of these recommendations on an experimental basis beginning in 1994.
A number of other projects dealing with related aspects of the pharmaceutical industry are currently underway. Berndt and Stan Finkelstein (of the MIT Program on the Pharmaceutical Industry) are examining possible uses of hedonic price analysis to control drug prices for quality change, using data on selected antihypertensive drugs. NBER Faculty Research Fellow Valerie Y. Suslow also is assessing quality-adjusted prices, using data from the antiulcer drug market.
Sara Fisher Ellison, who is working with Griliches, Cockburn, and NBER Research Associate Jerry A. Hausman, is attempting to model and measure own- and cross-price elasticities for a number of chemically distinct but therapeutically equivalent drugs--drugs that could be prescribed for the same condition (as distinguished from generic drugs that are chemically equivalent to their patented antecedents). In related work, Ellison is analyzing a consumer demand model with heterogeneous consumers. She also is analyzing the information content of advertising and marketing for pharmaceutical products in the market for antiulcer drugs.
Finally, Judy Hellerstein currently is modeling physicians' decisions to prescribe generic versus branded drugs. She looks at both patient and physician characteristics, and the probability that a physician is aware of the presence of generic substitutes.
As can be seen, the NBER research on the pharmaceutical industry has taken on a life of its own, even though its original stimulation came from an interest in the reliability of official price deflator data. In a future article, we hope to be able to report not only on the results of various demand-related issues in the pharmaceutical market, but we also hope to return to our original focus and report implications of these various studies for the measurement of output and productivity growth in technologically dynamic industries such as the U.S. pharmaceutical industry. It is possible that the U.S. economy was considerably more productive in the last decade than is now thought.
(1) See E. Berndt and Z. Griliches, "Price Indexes for Microcomputers: An Exploratory Study," in Price Measurements and Their Uses, M. Foss, M. Manser, and A. F. Young, eds. Chicago: University of Chicago Press, 1993, pp. 63-93.
(2) See E. Berndt, Z. Griliches, and J. Rosett, "Auditing the Producer Price Index: Micro Evidence from Prescription Pharmaceutical Preparations," Journal of Business and Economic Statistics 11, 3 (July 99 3), pp. 251-264.
|Printer friendly Cite/link Email Feedback|
|Author:||Brendt, Ernst R.; Griliches, Zvi|
|Date:||Jun 22, 1993|
|Previous Article:||Working under different rules.|
|Next Article:||Fogel wins Nobel Prize: headed NBER program on Development of the American Economy.|