Fgcasal.org

PRICE COMPETITION IN PHARMACEUTICALS: THE CASE A fundamental question in industrial organization regards the relationship between price and the number of sellers. This relationship has been particularly important in the pharmaceutical industry where legislative changes were specifically designed to foster competition. Previous research on the pharmaceutical industry has shown generic entry has a mixed impact; generic prices fall rapidly with generic entry, whereas branded prices tend to increase or decrease only slightly. Using more complete data, focused on one segment of the pharmaceutical industryÐanti-infectivesÐwe find that the relationship between pharmaceutical prices and the number of sellers is more like that found in other industries. (JEL L11, L65, D4) between 6 and 15, and fall by another 52% as sellers increase from the 6 to 15 range to more organization regards the relationship between than 40. These results contrast with previous price and the number of sellers (N ). The inter- work on pharmaceuticals using more limited est in this issue has been rekindled by substan- data and showing little impact of entry on tial recent work.1 This article uses an extensive branded prices.4 The results here indicate data set to provide an empirical analysis of the instead that the effect of an increase in number price±N relationship for antiinfective pharma- of sellers on prices in pharmaceuticals is similar ceutical products.2 The analysis shows that (1) prices fall rapidly moving from one seller to a few and (2) subsequent increases in the number attracted scrutiny from both policy makers of sellers continue to reduce prices, even when there are numerous sellers.3 Prices fall about have been a number of studies of pharmaceu- 83% as the number of sellers increases from 1 to tical pricing behavior in general and the impact of generic entry on branded and generic drug prices in particular. Most of the previous stud- *We thank Michael Baye, Dean Lueck, and seminar ies have concentrated on small sets of drugs participants at Chicago, LSU, Penn State, Texas A&M, that have faced patent expiration across a num- UCLA, the University of Texas, and the spring 1994 meet- ings of the Industrial Organization Group of the NBER for ber of therapeutic classes. Early studies focused helpful comments. Financial support from Merck & Co.
on reduced form or semireduced form regres- sion models that showed that branded prices Wiggins: Professor, Department of Economics, Texas and generic prices responded differently to A&M University, College Station, TX 77843-4228.
Phone 1-979-845-7351, Fax 1-979-847-8757, E-mail generic entry. In particular, these studies showed that branded prices responded little Maness: Senior Managing Economist, LECG, LLC, 2700 Earl Rudder Frwy., Suite 4800, College Station, TX 77845. Phone 1-979-694-5780, Fax 1-979-694-2442, 4. The analysis here builds on Caves et al. (1991) and Grabowski and Vernon (1992) but covers a much larger set of products, including all 98 in the anti-infective category.
1. See, for example, Applebaum (1982), Bresnahan In contrast, Caves and colleagues used 30 products and (1981), Bresnahan and Reiss (1991), Caves et al. (1991), Grabowski and Vernon used only 18. Later studies also Porter (1993), Reiss and Spiller (1989), and Suslow use limited numbers of products. Frank and Salkever (1986); portions of this literature are summarized in (1997) have a sample of 45 drugs; Reiffen and Ward (2002) uses 32 drugs. By using a larger number of products 2. Although our analysis concentrates on the relation- concentrated in a single therapeutic category, we hope to ship between price and the number of firms, there has also determine how prices are affected by the number of sellers been substantial work on the relationship between price for an entire group of products and avoid thorny problems associated with the heterogeneities in product use. The 3. It should be noted that there is little if any price effect analysis here also provides a much more in-depth evalua- when the second generic enters, but price effects become tion of the competition between branded and generics not significant as the third and fourth generics enter.
found in these earlier investigations.
# Western Economic Association International to generic entry, and in some studies even primarily treat acute conditions, and a single increased (see Caves et al., 1991; Grabowski prescription is usually enough to treat a condi- and Vernon 1992; 1996). Frank and Salkever tion, prescriptions make a natural measure of (1992) developed a theoretical model to explain quantity. However, anti-infectives are some- the anomaly of rising branded prices in the face what different from other therapeutic cate- of generic competition. Their model posited gories in that generic entry has historically a segmented market where there existed two been less costly than in other categories. The segment that continues to buy the established branded product after generic entry and a price-conscious segment. Frank and Salkever approval of generic products, had essentially show that because of the segmented nature of been in place for decades for anti-infectives. As the market, entry likely makes the demand a result, although other categories did not see facing the branded manufacturer less elastic substantial generic entry until after 1984, anti- infectives had faced numerous generic entrants and Salkever (1997) provide empirical tests and for some time. In addition, many anti-infective confirm that, consistent with the segmented products, especially older ones, tend to be pre- markets theory, branded prices rise and generic scribed by generic name instead of the brand prices fall in response to generic entry.
name, which hastens the acceptance of generic Finally, a number of studies have attempted to characterize the behavior of generic firms.
These factors have led some researchers to Scott Morton (1999; 2000) finds that revenue in predict that branded anti-infective prices may the years before patent expiration is the most respond differently from those in other cate- important determinant of how many generic gories. Early empirical research seemed to con- firms enter a given market. She also finds firm that result.7 Later empirical research has that generic firms tend to specialize in certain provided more mixed results, with some studies showing a pattern of rising branded anti- infective prices in the face of generic entry.8 response of generic prices to generic entry.
Our results are more in line with the earlier This article investigates how prices decline studies, showing significant impact on branded as the number of sellers rises, evaluating the ability of formal theoretical models to organize A second difference in our data is that they the data. The results show that although exist- measure activity at the retail level. The use of ing models organize some of the data regarding retail prices raises two issues for our analysis.
price declines, there are important discrepan- Previous research has pointed out that phar- cies between the data and existing models.
macists may have a financial incentive to favor We then characterize the differences between generic dispensing, especially since the advent the data and the models, providing a basis for of managed care (see Grabowski and Vernon continued theoretical and empirical work.
1996). However, our data end in 1990, when There are a number of important differences managed care was a much smaller factor than it between the data we use and those of previous is today. A second issue is to what extent retail researchers. First, unlike previous research prices can be used to analyze price competi- tion among manufacturers. Given the intense focused on a single therapeutic categoryÐ anti-infectives. This focus on a single therapeu- tic category provides a number of benefits, drugs. See Lu and Comanor (1998), Scott Morton (1999, particularly with regard to controlling for 6. Such acceptance may also occur because the U.S.
cost differences and demand differences.
Food certifies the manufacture of each batch of active Anti-infectives in particular is a useful category ingredient for anti-infectives, but not necessarily for because they are primarily used for acute con- ditions, thus making demand conditions more 7. See in particular, Schwartzman (1976, chapter 12).
See also Congressional Budget Office (1998, chapter 3).
uniform.5 In addition, because anti-infectives 8. Both Ellison (1998), and Griliches and Cockburn (1994) find that average branded anti-infective prices rise with generic entry, whereas Ellison et al. (1997) find 5. Several studies note that the demand and supply significant responsiveness between branded and generic conditions may be different for acute versus chronic WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS competition among retail pharmacies, and the data are used to construct an annual price per use of simple pricing rules that apply across prescription for each seller of each product most (if not all) drugs, price competition (detailed data appendix available on request).
among manufacturers is likely to be reflected Attention is restricted to anti-infectives to con- trol for units of measure, cost, and regulatory conditions across products. Doctors normally econometric analyses of product differentia- write an anti-infective prescription for a quan- tion, including identifying separate effects on tity of medication designed to cure a given prices from increases in both generic and brand infection. This practice makes the individual name competition. This analysis of competi- prescription the natural unit of measure for tion between branded and generic products anti-infectives. For other pharmaceuticals, in shows (1) that it is important to distinguish contrast, a pill or a daily dose might be more between branded products sold by other inno- relevant.11 The active ingredients for these pro- vative firms besides the pioneer and unbranded ducts are also commonly manufactured using entry by traditional generic manufacturers; (2) identical equipment, making costs more similar branded entry by other innovative firms has a for anti-infectives than across pharmaceuticals different effect on pioneer prices than generic as a whole. Still, the analysis that follows intro- entry by strictly off-brand firms; and (3) there duces several control variables to allow for cost appears to be significant competition between differences in a more sophisticated way than in generic and brand-name sellers, including the pioneer. Hence the results indicate that there are three distinctive product groups in pharma- review standards are also the same within this ceuticals: the pioneer, branded versions of the class of products, so that regulatory costs and same molecule sold by other innovative firms, standards are also similar. Hence cost differ- and ordinary generics, and there is significant ences are likely small, and we can control price competition within and between these adequately for different manufacturing tech- niques for different groups of products (see We briefly review some summary statistics, a complete review is available in the data appendix (available on request). The average The data consist of retail-level pharmacy price (1982±84 dollars) in the sample is $10.29 transaction data for all anti-infective products with a standard deviation of $21.68. The smal- over the 1984±90 period.10 The data provide lest price is 0 and the largest price is $391.46.13 yearly observations on total expenditures and quantity of prescriptions sold for the indi- vidual sellers of anti-infective products. These 11. It should be noted that there may be different dos- ing and different presentations, e.g. liquid versus tablet, but that regardless of the dosing or presentation the standard practice is to write a prescription to cure the infectionÐ 9. The IMS data we use for our empirical analysis do which is why physicians tenaciously encourage patients to not include rebates that are commonly paid by branded manufacturers to managed care. Congressional Budget 12. The analysis introduces dummy variables for Office (1998) notes that these rebates can continue and each class of productsÐfor example, tetracyclines and increase after generic entry, and thus not measuring penicillinsÐand also uses separate dummy variables for them can obscure an important source of price competition year of introduction that allows a flexible, nonparametric for branded manufacturers. Note, however, that wholesale estimation of possible trends or other changes in cost over level data from IMS possess the same weakness, and to the time, reflecting increased complexity due to more sophis- extent that such rebates are important, our results under- ticated molecules or changes in manufacturing techniques.
estimate the degree to which branded manufacturers These controls represent a significant improvement over previouswork(see,e.g.,Cavesetal.1991,whouseadummy 10. The data were obtained from the National Pre- variable by therapeutic class, which would be the same as scription Audit of IMS America. Total expenditures and imposing constant costs for all products in our sample).
quantity data are collected from a stratified random sample 13. The reader should note that there are seven obser- of pharmacies and then used to construct nationwide retail vations where total revenue is reported as zero and several sales estimates. In the present analysis, these monthly data observations where revenue is also extraordinarily high, have been aggregated into annual series, forming a panel both yielding outliers in pricing calculations. Reestimation data set where the unit of observations are annual data for without these observations does not qualitatively change an individual product sold by an individual company.
The average branded price (constant dollars) is $30.30 with a standard deviation of $45.77, and the average generic price (constant dollars) is $6.27 with a standard deviation of $6.91. The number of sellers for a given product varies from 1 (monopoly) to 61. The mean is 26.9 sellers. Our measure for the number of related sellers has a mean of 31.6 and ranges between 0 and 112. We also calculated the Herfindahl entity in each year. The mean HHI in the sam- ple is 3,677, and the standard deviation is 2,669.
brand-name products, 13% are pioneer pro- ducts. The number of observations is fairly evenly spread across the sample period and across IMS classifications. Erythromycins (14%) and cephalosporins (9.6%) contain the Almost half the observations (46%) are from drugs introduced before 1962. In other years, introductions vary from none in some years up to 8.4% of the sample in 1974. Thus the data set provides much detail and ample variation for Anti-infective products are broadly catego- rized according to the general molecular struc- ture of the central active chemical entity, such as penicillins, tetracyclines, erythromycins, further disaggregated into specific molecular entities or combinations, and the analysis here focuses on the number of sellers of these The first point on the horizontal axis corre- individual products. Our data include 98 sepa- sponds to drugs with one seller and the price rately identified compounds. The analysis also is averaged over all such products in the considers competition among sellers of closely sample, the second price is the average price related molecules. Ellison et al. (1997) examine for all two-seller drugs, and so forth.14 cross-product price competition in cephalo- Figure 2 presents a similar view using HHI sporins, and find mixed evidence of price com- petition across products for these closely The two panels of Figure 1 present two dif- related antibiotics. In contrast, Stern (1994) ferent looks at the data. A shows average prices finds that in two of his four categories there for all sellers, and B shows the average price for is significant intermolecular substitution.
just the pioneersÐthe firms that originally Accordingly, in the econometrics we allow for intermolecular effects, though our results drop in prices with the entry of the second are more similar to Ellison et al. than to Stern.
through fifth sellers. Prices fall from the single seller price of more than $57 to $9.46 when metric specification, however, it is valuable there are five sellers. This rapid decline in prices to examine the simple relationship between continues throughout most of the relevant price and number of sellers. Figure 1 presents these data. Although the econometric analysis uses individual prices for each seller in a panel 14. With 98 product categories and 7 years of data, data set, for clarity Figure 1 simply presents there are 686 possible industry concentration outcomes.
average prices, taking the mean price of all For example, there are 178 monopolies and 45 duopolies (90 price observations) making up the first two points in drugs with a particular number of sellers.
WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS Prices of Products and Number of Competitors range. B, focusing just on the prices of the pio- continuing price decline in the reduced form neer developer, shows essentially the same that continues from only a few sellers to more pattern. Hence, these data show a decline in than 40. Focusing just on the pioneer pro- both overall average prices and prices of the We tried other specifications, and all the The continuing price decline is well illus- results are broadly similar. For instance, trated through a series of simple regressions of although our analysis concentrates on the the price±N relationship, with progressive price±N relationship, we also calculated truncation of the sample on N from the left- the HHI for each of our 98 chemicals. The hand side. These regressions permit one to mean HHI was 3,677 with a standard deviation assess the persistence of the inverse relation- of 2,669. The range was 1,082 to 10,000. After ship between price and N. The results show constructing the HHI, we then replicated the that there is a statistically significant impact regressions using the HHI in place of the num- of the number of sellers on price, including ber of firms. We ran a series of regressions, when the sample is restricted to more than 30 sellers.15 Hence there is a significant, 16. It should be noted that there is a single outlier for the pioneer products, which makes some difference in the 15. The coefficient on the number of sellers in the sim- results. There appear to be data problems with the prices ple price±N truncated regressions are as follows (t-statistics for this drug, Vibramycin (PFizer: the pioneer doxy- in parentheses): for regressions including all chemicals cycline). If Vibramycin is included, the effect of N is insig- (N b 0), b ˆ À0.47 (À21.42); for the regressions including nificant for N ! 20, and significant and positive for N ! 30.
only chemicals with five or more competitors (N ! 5), With Vibramycin removed, the result is a significant nega- b ˆ À0.15 (À14.213); for N ! 10, b ˆ À0.13 (À14.46); for tive coefficient for N ! 20 and negative but insignificant for N ! 20, b ˆ À0.15 (À13.38); N ! 30, b ˆ À0.06 (À3.81); N ! 30. We have not found other instances where Vibra- mycin swings results in this way or other similar outliers.
truncating the Herfindahl from above. The To formalize the Cournot prediction, sup- coefficient remained positive and significant, even when the regression was restricted to a P(Q) ˆ a À bQ with constant marginal cost, HHI of less than 2,000. Thus, decreasing con- c ` a, and where Q represents total market centration reduces price, even for chemicals output. With N identical sellers, the market price generated by the Cournot model takes It is unlikely that manufacturing cost differ- ences explain the observed differences in price.
Anti-infectives appear to exhibit constant marginal production costs, and manufactur- ing techniques are highly similar. There are differences in presentation, such as tablets, Note that the Cournot model predicts a linear capsules, and suspensions, but these differ- relationship between price and roughly the ences cut across products and groups of pro- could make a large, systematic difference is equation (1) when there is a linear demand for intravenous drugs, which we investigate and constant marginal cost. Constant cost is later. Hence manufacturing cost differences highly plausible for anti-infectives because the are unlikely to explain observed price differ- raw ingredients are manufactured in batches ences. Still it is important to control carefully that are small relative to total output and pill making is constant cost. Linearity of demand, might affect the observed price±N relation- however, is a stronger assumption suggesting ship, requiring a more detailed econometric that one can think of (1) as a Cournot-like out- come rather than a structural test (see later The Cournot model also assumes that firms are identical. For many of the chemicals in our analysis, especially those that have been off patent for a number of years, this assumption relationship between price and the number of is plausible. The industry uses batch produc- sellers, it is useful to rely on formal models to tion technology characterized by constant guide the analysis. The data summarized in returns to scale. For this reason, most of the Figure 1 show a continuous decline in price, previous literature in the field also uses ruling out a simple Bertrand model of price the number of competitors as the key variable competition. Recognizing this, two simple determining the competitiveness of prices alternative models of competition are the stan- (see Stern 1994; Ellison et al. 1997 for instance).
Though the results reported use the number of the entry threshold model of Bresnahan and competitors as the key variable, we also ran Reiss (1991). Standard Cournot oligopoly the- regressions using the HHI. The results are ory predicts that prices should initially fall broadly similar to those reported. Thus the quickly and then steadily approach marginal assumption of identical firms does not signifi- cost. A similar prediction is also provided by cantly affect the empirical results.
Bresnahan and Reiss's entry threshold ratio In our analysis N is the exogenous variable method, which predicts a steady fall in variable driving prices in equilibrium. We treat N as profit margins. We compare the predictions of this simple Cournot model to the alternative of and other factors in the previous period or per- a simple linear relationship between price and iods, and price is then determined by the pre- the number of sellers and to a more general determined number of sellers in a particular model that nests these alternatives.
consistent with the data and with previous 17. The coefficients on the HHI in the truncated regres- research on the determinants of generic entry sions are as follows (t-statistics in parentheses): for the (see Scott Morton 1999; 2000). For example, a whole sample, 0.003 (24.54); restricting the sample to regression of N on sales in the year before HHI ` 5,000, 0.001 (11.48); for HHI ` 3,000, 0.002 patent expiration (where data are available) (5.214); for HHI ` 2,000, 0.005 (6.97); for HHI ` 1,000, WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS 60% of variation of the number of sellers in the postpatent period. Hence ex ante market size is fixed effects provides a much richer set of con- the key variable driving variation in N.
trols for cost and demand differences than the It is also important to control for other fac- existing literature. For example, Caves et al.
tors that might affect prices. One key factor is (1991) consider products across therapeutic possible competition from related products.
categories (such as cardiovasculars versus anti- Ellison et al. (1997) found little evidence to infectives) and simply use a single fixed effect support such competition, and Stern (1994) for anti-infectives, which is equivalent to finds such effects in some categories. To control imposing a restriction that all of the fixed for these effects we included the number of effects considered here are zero because the other sellers in the appropriate broader cate- entire data set consists of anti-infectives. Hence goryofanti-infectives,suchaspenicillins,tetra- the present study provides a large advance in cyclines, and so forth. Brand recognition is also the control of cost and demand factors in important, and we included a dummy variable studying pharmaceutical pricing relationships.
for versions of the same molecule sold by other We used weighted least squares because the innovative companies that had achieved brand cells for different numbers of sellers had sharp- ly different variances. We used as weights the extensive analysis of branded, pioneer, and inverses of the estimated standard deviation of generic products. In contrast, different presen- the residual for each value of the number of tations of drugs do not differ sharply, except competitors.20 The results from the estimation injectable products are systematically more are reported in Table 1. The results show expensive to manufacture and distribute. We general goodness of fit with a large F-statistic, introduced an IV dummy to control for these R2, and all of the key variables are highly The model derived from equation (1) does a differences that require fixed effects to control good job of explaining the empirical relation- for product age, product group, and year. Year ship between price and the number of sellers of of introduction fixed effects were included to a given product. The results show that price is control for differences in product age that closely related to the inverse of the number of could lead to differences in either demand sellers of individual products, as predicted by or cost. For example, physicians sometimes the Cournot model. The parameter estimate for use older products first in a course of therapy, 1/(N ‡ 1) is 28.70 with a t-statistic of 7.07. The potentially affecting demand and price. Alter- natively there may be a trend where manu- declines when a firm loses its (patent-protected) facturing costs of more recently developed products are higher due to the more advanced seller drops prices by nearly $5, the third by nature of the products. Individual year of intro- $2.39, the fourth by about $1.43, and the duction fixed effects provide a flexible proce- fifth by about $0.97. These price effects, more- dure for accounting for such differences, and over, are large relative to the mean price in the we will compare the results of using such fixed sample of $10.29. Hence, when there are more sellers in a market the price falls sharply, but In addition, products in different therapeu- these price effects moderate quickly. Still, a tic subcategories may differ in manufacturing significant though small effect continues far cost. Even though anti-infectives are generally into the sample; going from 25 to 35 competi- manufactured using similar techniques, there tors is predicted to reduce prices in the struc- could exist some differences across categories, tural model by $0.31, or about 7.5%.
such as tetracyclines or penicillins. Accord- It is worth noting several other results.
ingly, we include fixed effects for these groups.
Finally, fixed effects were included for individ- economically and statistically significant ual sample years to control for changes in 19. The data appendix provides summary statistics for 18. This information on brand names was obtained from Scientific American Medicine. See section 3 for a 20. This method implies that separate weights are cal- more complete investigation of our definition of brand culated for each regression reported in the article. The name and how brand name affects pricing and price results were not qualitatively affected by using a common set of weights, except as reported in the HHI regressions.
Regression Results Based on the Cournot Model Notes: Dependent variable: real price per prescription. N ˆ number of sellers. NR ˆ number of sellers of related products. HHI ˆ Herfindahl index. t-statistics are in parentheses. Real price is in 1982±84 dollars. Regressions are weighted OLS. Weights are calculated by number of competitors and in each regression are equal to the standard deviation of the residual for each number of competitors. An appendix, available from the authors, contains the fixed coefficient for sellers of related products is small, only 1.333, not quite significant at the price of generic products in the sample is 95% confidence level. Hence, these data suggest slightly more than $6, this variable indicates that price effects are driven primarily by changes in the number of sellers of the com- pound in question, where we concentrate the firmsÐover generics.21 The variable represent- ing intravenous products also shows a substan- Although we concentrate on sellers of indi- tial price premium; the coefficient indicates vidual chemicals, we do not interpret these that such products are priced at $13.11 more chemicals as being separate markets. Instead, than other products, ceteris paribus.
we take the weaker position that taking the number of sellers of related products into pound in question (N ) has a large effect on account makes little empirical difference.
price, the number of sellers of related com- Hence it is appropriate to focus primary atten- pounds (NR) has little effect. As noted, param- tion on sellers of individual products.
eters predict a price decline of nearly $5 moving Before turning to the other specifications it is important to consider the fixed effects.23 The number of sellers from two to three cuts prices by another $2.50, and so forth. In contrast, the 22. Results with other specifications, such as using dif- ferent functional forms for N and using the HHI in place of the number of firms, yield broadly similar results.
21. The analysis in section III contains a more complete 23. A complete tabulation of regression results, includ- ing the fixed effects, is available on request.
WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS results show that only 9 of the 24 individual indicate significant differences across subcate- year of introduction fixed effects are statisti- gories because a joint test restricting these cally significantly different from the omitted coefficients to zero is highly significant with median (1973) at the 5% confidence level. How- ever, an F-test reveals that these effects are Finally, only three of the year fixed effects jointly significant at the 1% level (F ˆ 14.38).
(1984±89) are significant at the 5% level.
Inspection of the coefficients shows that prior Furthermore, there is no distinct trend in the to 1973 all 11 are negative and after 1973 8 of 14 year coefficients, which indicates that the data are positive. Nevertheless, there does not exist a do not suggest a simple trend in price over the simple trend in that there is considerable var- iation in the individual year fixed effects over The fixed effects show that it is important to time, showing that this method provides a control for cost and demand conditions. The highly flexible method for accounting for pos- rich set of fixed effects shows substantial non- linear variation by year of introduction, and there are generally small but significant differ- included to control for vintage effects, such ences in prices across subcategories. These as increased quality for newer drugs, which fixed effects provide a significant improvement Lichtenberg (2001) demonstrates is an impor- tant attribute of newer drugs relative to older possible cost and demand differences. Hence ones. To quantify these effects in a simple way, the measured effect of competitors on price we also ran the regression using an age variable should not be the result of cost or demand in place of year of introduction dummies. This procedure also permits us to test whether our Turning to the other specifications, an alter- more general treatment of vintage effects mate- native to the highly nonlinear, inverse specifi- rially affected the results. Age is defined as cation of column (1) of Table 1 is a linear years since introduction of the chemical. We relationship. To test between these specifica- also include age squared to account for poten- tions, one can nest the two models including tial nonlinear vintage effects. This parametric both the linear and nonlinear terms, as is done treatment of age is more restrictive than the in column (2) of Table 1. The results for the nested model indicate that both terms are constrains age effects to a quadratic form. The significant, suggesting that prices fall faster results, however, are essentially unchanged than the simple Cournot-like model would pre- for all the key variables. These results also dict. Furthermore, although the linear terms ensure that the specification of age effects are significant, they are economically small.
does not account for any differences between For example, as the second firm enters, the the results here and those of other investiga- primary model predicts a price decline of tors. The estimate of the key coefficient on about $5, whereas the nested model predicts 1/(N ‡ 1) in this specification is 37.18 with a somewhat slower price decline of just over a t-statistic of 8.97. The coefficient for age $3; these differences are fairly small so that is À0.461 with a t-statistic of À7.14. The both models organize the data fairly well and squared term is small, 0.007, with a t-statistic of 5.70. For drugs that first entered the market The most important difference between the during our sample period, the average first- year price is $54.41. The quadratic age effects regards their prediction of price declines when thus predict a fall in price of little more than there are large numbers of firms. The nested 13% over 40 years ($7.14). This parametric model predicts a larger price decline when there approach suggests that systematic age effects capture better the empirical phenomenon of continuing price declines far into the sample.
fixed effects, we omitted the median fixed effect In fact, the linear term is roughly the same for trimethoprim (IMS class 15500). Of the size as the coefficient in the simple regressions 20 fixed effects for individual product groups, of price on N when the sample is restricted to only 10 are statistically significantly different more than 30 sellers; hence, the linear term in from this median. Even though the number of the nested model enables it to capture the con- significant coefficients is small, the data do tinuing price declines far into the sample.
Another difference occurs in the effect of with t ˆ 4.816). Thus, removing other HHI the number of sellers of related compounds.
from the primary regression does not alter In the inverse specification, these sellers have the results in any way. We also repeated the a marginally significant effect, whereas in the regressions in column (1) and (4) using the nested specification, these products have a log of price as the dependent variable.
statistically insignificant effect. The small size The results are qualitatively unchanged.25 of these effects and their sensitivity to the The price±N regressions and the price±HHI specification may help explain the disparate regression represent two competing linear findingsofStern(1994)andEllisonetal.(1997).
The third column of Table 1 shows the sim- ple linear model without the nonlinear term, tests to determine which model best explains and the results are similar. The model exhibits the data. The results were inconclusive. One large F-statistics and R2 as in the other two method to test these nonnested alternatives is specifications, and the coefficients of all to create an artificial nested modelthat includes explanatory variables are of the right sign both (HHI and other HHI) and (1/[1 ‡ N ] and and statistically significant. The key difference 1/[1 ‡ NR]) as independent variables along with is that the linear model does not capture the all the control variables and then conduct an sharp initial price decline that characterizes F-test on the HHI coefficients and the N coeffi- cients (see Greene 1993, 222±23). The results indicate that neither specification can be initial entry on prices of these various thera- rejected. The F-test that the coefficients on peutic compounds. This relationship is present, HHI and other HHI are both zero is 4.21. How- moreover, both in a simple reduced-form anal- ever, in this specification, the coefficient on ysis and in an in-depth econometric analysis, HHI is not statistically significant (0.00009, where control is introduced for product age, with a t-statistic of 1.19), and the joint signifi- potential manufacturing cost differences, the cance may be driven by the significance of the number of sellers of closely related products, (À0.0003, t-statistic ˆ À2.21). The test that We also ran these regressions with HHIs in 1/(1 ‡ N) and 1/(1 ‡ NR) are simultaneously zero is also easily rejected (F(2,3110) ˆ regression is reported in the final column of 21.69). We also conducted a J-test of the two Table 1. The results show a good fit, with an competing models to ensure that the results R2 of 0.808, slightly less than for the regressions were not affected by the control variables using number of sellers. The coefficient on the (see Greene 1993, 223). Using the J-test, the HHI for individual chemicals is positive and hypothesis that the HHI model does not add significant, indicating that decreasing concen- any explanatory power cannot be rejected, and tration does reduce price. The coefficient on the HHI for the other chemicals in the IMS model is easily rejected. Thus, the J-test class is negative and significant, with a coeffi- seems to favor using the inverse of the number cient equal in size to that of its own chemical of competitors as an independent variable over entity, indicating that decreased concentration in the broader class results in higher prices.
It is interesting to note the similarities and differences between these results and those in HHI regression (column [4]) with other HHI prior work. Perhaps the most closely related removed. The results were almost identical work in the literature on the price±N relation- to those reported in Table 1, column (4) (the ship is Bresnahan and Reiss (1991). They exam- coefficient in this regression was also 0.0004, ine prices for dentists, auto repair shops, and the like in geographically isolated county seats. Their key finding was that the largest competitive impact occurs moving from two 24. One limitation of the approach is that it imposes the Cournot model. A second alternative is to construct a gen- to three firms, another large impact moving eral conjectural variations model which nests the Cournot conjecture and let the data determine if the Cournot con- jecture can be rejected. We constructed such a model and found (using nonlinear least squares) that the Cournot con- 25. An appendix containing the results of this regres- WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS from three to four, and smaller subsequent price impacts. They also argue that their inferred prices in relatively concentrated pricing by recognizing the differentiated nature of generic and brand-name products. Brand- The analysis here shows a sharp price drop as the number of sellers increases from one to consciouscustomers,andgenericfirmscompete four or five and much smaller price effects over price. Competition within and between thereafter, similar to their finding. Bresnahan these segments can provide insight into pricing and Reiss also find a significant difference patterns and permits testing of specific models of competition between these groups.
(county seats) and unconcentrated (urban) markets. Their finding is consistent with our are two reasonable alternatives. One way is results, which show a continuing price decline to categorize the original patent-protected product, defined hereafter as the pioneer, as In contrast, the analysis here shows different results from those found in previous research products marketed by other innovative compa- on pharmaceuticals. Studies by Caves et al.
nies that are separately promoted and have achieved significant brand recognition even though they are not the pioneer. We evaluate entry. The results here show that pharmaceu- both alternatives. To separate products with tical prices respond to an increase in the num- significant brand recognition, we categorized ber of sellers similar to price responses found in products specifically listed by brand name in other markets.There are several possible expla- the Scientific American Textbook of Medicine.
nations for these differences in results.
Theseproductsaregenerallydetailedtodoctors Perhaps the most important is that the sam- in varying degrees and may or may not differ ple here is much larger than that used in these from more traditional generic entry.
prior studies and contains much more variation in the number of sellers. In addition, we do not names, and generics leads to several reduced restrict attention to a short period following form relationships between price and N. We are patent expiration but instead consider a variety particularly interested in: (1) average brand- of products with sharply different numbers of name prices and the total number of sellers, sellers. We also control more tightly for cost (2) average brand name prices and the number and demand differences. A different type of of brand name sellers, (3) average prices of possibility is that the analysis here focuses on pioneer products and the total number of sell- anti-infectives. Such a focus contributes to bet- ers, and (4) average generic prices and the total ter control for cost and demand differences, but number of sellers. These relationships are illus- comes at the cost of a more narrow focus. For trated in the four panels of Figure 3. Panel A example, by concentrating on anti-infectives, shows the relationship between average brand- the analysis focuses on products for which a name prices and the total number of sellers of single prescription is designed to cure, usage a chemical. A shows that the price of brand- patterns are similar, and underlying cost con- name drugs falls steadily with the number of ditions are likely more similar. These factors competitors until there are roughly 25 to improve the econometrics but of course mean 30 competitors; afterward the point estimate that extrapolation to other therapeutic cate- indicates continued price declines, but the decline is no longer statistically significant.
Panel B of Figure 3 provides a different look analyses of the price±N relationship and at brand-name prices by showing how the aver- prior studies of pharmaceutical pricing. The age price of brand-name products is related to omission is that it does not take into account chemical.26 The data show a considerable but unbranded products. We now turn to an expli- irregular decline in price as the number of cit analysis of the differentiated nature of pharmaceutical products and the effect of 26. Note that care must be used in interpreting these figures. For instance, the first point in B does not represent brand name sellers rises. Panel C shows that These data show that increases in the num- prices of pioneers steadily decline as the num- ber of sellers of individual chemicals leads to ber of sellers of the chemical rises. This panel is price decreases over a large range, but the pat- a repeat of Figure 1B. Note that the price tern of decrease and average price levels differ decline exhibited here is much like that of significantly for branded and generic products.
Figure 1A. Finally, D presents data on average For generics significant price effects continue generic prices as related to the total number of far into the sample, whereas prices stabilize for competitors. The results show that for generic brand names when there are large numbers of products prices fall steadily until there are more metric analysis of the competition between an average monopoly price because there may still be com- petition from generic sellers even if there are no other brand 27. The coefficients for the number of competitors is as It is common in the industry to view brand- follows (t-statistics in parentheses): For brand-name name products as being differentiated from prices: N b 0, b ˆ À0.95 (À8.93), N ! 10, b ˆ À0.13 their generic competitors. Pricing patterns (À3.84), N ! 20, b ˆ À0.12 (À2.62), N ! 30, b ˆ À0.03 after patent expiration and generic entry (À0.38), N ! 40, b ˆ À0.02 (À0.13). For generic prices: N b 0, À0.14 (À16.71); N ! 5, À0.12 (À14.17); N ! 10, show that the price of the pioneer product À0.12 (À13.78); N ! 20, À0.15 (À13.84); N ! 30 À0.07 remains much higher than the level of generic (À4.40); N ! 40, À0.01 (À0.38). For the HHI regressions, competitors for substantial periods of time. In the coefficients are as follows. For brand names: overall, b ˆ 0.006 (11.34); HHI ` 5,000, b ˆ 0.001 (2.88); fact, previous empirical studies have found that HHI ` 4,000, b ˆ 0.001 (1.873), HHI ` 3,000, b ˆ 0.002 prices of pioneer products tend to rise after (1.584); HHI ` 2,000, b ˆ 0.004 (1.13). For generic prices: overall, b ˆ 0.0006 (9.22), HHI ` 5,000, b ˆ 0.001 (10.72), HHI ` 4,000, b ˆ 0.002 (9.20), HHI ` 3,000, b ˆ 0.002 (5.34), HHI ` 2,000, b ˆ 0.006 (7.23).
way to approach competition between branded WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS Notes: NG ˆ number of generic competitors. HHI ˆ Herfindahl index. t-statistics are in parentheses. Dependent variable is real price is in 1982±84 dollars. Regressions are weighted OLS. Weights are calculated by number of competitors and in each regression are equal to the standard deviation of the residual for each number of competitors.
Fixed effects dummy coefficients are not reported. Regressions only include cases where market has a single brand-name seller and one or more generic sellers. Regressions are based on 565 observations meeting these conditions.
and generic products is to view them as differ- than generic prices. Column (1) of Table 2 entiated products. In this setting, additional reports the results of the first specification.
generic entry should have a larger impact on The control variable coefficients are qualita- tively similar to those in section II. The because entry is taking place closer to existing F-statistics and R2 are large. The primary coef- generic competitors in product space.
ficients of interest in column (1) are the coeffi- Table 2 presents several specifications of the cient for the inverse of the number of sellers, relationship between price declines for branded and the variable interacting this variable with and generic drugs. Because generic drugs are brand name. The inverse of the number of sell- likely to compete more closely with each other ers has the incorrect sign but is not statistically than with branded products, one might expect significant. In contrast, the results show that branded prices to fall more slowly than generic generic entry lowers brand-name prices.
prices when additional generics enter. Such a The second column in Table 2 reports a loga- possibility is also consistent with some of the rithmic specification, and the third reports a empirical works described. To address this linear specification. The final column in issue in detail, we postulate several specifica- Table 2 uses HHI in place of a functional tions where one variable captures the overall form based on the number of competitors.
price decline and a separate variable captures All four specifications show a significant effect the differential decline in the price of branded of generic entry on brand-name prices but little effect of generic entry on generic prices. The The first column in Table 2 is derived from result is that the analysis is that the data indi- a formal model of spatial competition (see cate faster price declines for brand names than Wiggins and Maness 1998). The specification generics, contrary to the conclusions of pre- is very similar to the Cournot analysis. The other specifications relax the highly specific In addition, we examine several nonstruc- functional form to test the qualitative result tural alternatives. These models also permit that brand-name prices should fall more slowly Nested Nonspatial Models of Product Differentiation Notes: Dependent variable ˆ price. NG ˆ number of generic competitors. NB ˆ number of brand-name competitors.
t-statistics are in parentheses. Dependent variable is real price is in 1982±84 dollars.
branded sellers affect prices, in addition to the impact of generics on branded prices, sell the incremental ``branded'' products (so- called branded generics) in the sample. To the extent that branded entry causes larger model for a ``segmented market' in pharma- price effects, the results indicate that branding ceuticals with two groups, a loyal group of leads to more effective price competition by price-inelastic brand-name consumers and a incrementally reducing prices in the branded relatively price sensitive group of consumers.28 They argue there is little competition between these segments. In such a case, the prices of regressions investigating these hypotheses. A brand-name products will be unaffected by pooling test was conducted to see if the three the number of generic competitors and, sym- metrically, the number of brand-name sellers respect to the effect of the number of sellers, ought not affect generic prices. An alternative and the results reject pooling. Subsequent pool- view is that all products besides the original ing tests were conducted for pooling pioneers pioneer are generic. This view is implicit in and other brand names and for generics and much of the empirical literature, which gener- brand names, and in each case pooling was ally examines only the pioneer's prices, implic- rejected. Accordingly the econometric results itly treating all entry as generic.29 This will analyze the separate effects of branded and approach implies that pioneer products repre- The regression results permit an in-depth analysis of the nature of competition between Our approach, in contrast, separates prod- brand-name and generic sellers and the effect on prices. In the brand-name price regression, both brand name and generic entry affects 28. See also Grabowski and Vernon (1992). Frank and Salkever (1997) provide empirical support for the segmen- ted market hypothesis by demonstrating that branded 30. Due todatalimitations,theunreportedfixedeffects prices tend to rise with generic entry while generic prices discussed earlier are pooled across all three types, pioneer, other brand name, and generic. To the extent that these 29. See, for example, Grabowski and Vernon (1992), fixed effects represent cost and demand differences among Caves et al. (1991), and Frank and Salkever (1997).
chemicals, this specification is correct.
WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS prices, but the coefficient on brand-name products is about ten times as large as the coef- ficient on generic products and is the incorrect This article has provided an empirical inves- sign. Nevertheless, the results show significant tigation of the relationship between price and competition between these product groups, the number of sellers in pharmaceuticals. The analysis used a data set covering all anti- market hypothesis and with the notion that infective products and showed that initial all sellers other than the pioneer should be entry led to sharp price reductions, with prices falling from the range of more than $60 per Turning to the generic price equation, the prescription for single sellers, $30 when there results clearly indicate that generic firms do are two or three sellers, and less than $20 when respond to brand-name entry. Note, however, there are four or more sellers. The results also there are important differences in how generic show that prices continued to decline with prices respond to other generic as opposed to additional entry, eventually approaching $4 brand-name entry. The linear term for generic for products with more than 40 sellers.
entry is large and significant, whereas the The results show that increases in the num- inverse term is small and not significant. For ber of competitors significantly reduce prices, brand-name entry the response is similar, but even when there are numerous sellers. These the effect of brand-name entry on generic prices results tie in nicely with those of Bresnahan is substantially larger (À0.88 per brand-name and Reiss (1991). Using small, isolated county entrant versus À0.16 for an additional generic seats, they find that the competitive effects of entrant). The results once again show signifi- entry diminish after the third or fourth entrant cant competition between branded products but that prices stabilize above those in uncon- centrated urban markets. The results here show The results for pioneer productsÐproducts a similar, rapid initial price decline, and then go sold by the original developerÐalso contrast on to show a continuing steady decline as the with the segmented market hypothesis.
number of sellers rises from a few to many.
Generic entry has a large and statistically sig- nificant effect on pioneer prices, both for the linear and nonlinear terms. Hence pioneer observed pricing pattern. The analysis showed products do indeed lower prices quite signifi- prices broadly consistent with Cournot quan- cantly in the face of additional generic com- tity setting, though prices declined more with petition.31 Furthermore, there is suggestive large N than that model would predict.
The analysis also ties into the emerging work entry also effects pioneer prices in that the on pharmaceutical prices. The analysis here has linear term is quite large and statistically sig- extended that work in several directions. One nificant, although the inverse of the number of direction is to provide a much more compre- brand-name sellers has the wrong sign.
hensive analysis of pricing in a specific, impor- Hence the results support the argument that tant therapeutic category. Our results show there is significant competitive interaction substantial price sensitivity and stand in contrast to results found by Caves et al. and firms. These results indicate considerable com- Grabowski and Vernon. There are several pos- petition and show that increases in the number sible reasons for these differences. One is that of sellers in any segment generally reduce prices the analysis here relied on all anti-infectives, in the remaining segments. This competition not restricting attention to the period closely following patent expiration. The greater varia- tion in the number of sellers also provides a richer data set particularly because the analysis used data exclusively since 1984, when there were large numbers of generic sellers. This 31. There are several possible reasons for these differ- ences from Grabowski and Vernon. Perhaps the most important is that we have a much larger sampleÐthey con- Waxman-Hatch Act, which eased the burdens sidered only 18 products. It is also possible that their results of generic entry. A second possible reason is are confounded due to difficult to control for cost differ- ences, or that our results are special due to the differential that anti-infectives may be more price-sensitive than other segments of the pharmaceutical industry. This possibility, of course, means Brookings Papers on Economic Activity: Microeco- that one must be careful in drawing inferences from this analysis to pharmaceuticals more Congressional Budget Office. How Increased Competition from Generic Drugs Has Affected Pricesand Returns in the Pharmaceutical Industry. July 1998.
The analysis also provided an econometric Dunne, Timothy, and Mark J. Roberts. ``Costs, Demand, investigation of differentiated product models and Imperfect Competition as Determinants of Plant- by treating brand-name and generic products Level Output Prices.'' Manuscript, Department of as differentiated. The problem is to explain Economics,PennsylvaniaStateUniversity,December the persistent price difference between brand- Ellison,SaraF. ``WhatPricesCanTellUsabouttheMarket name products and their generic competitors.
for Antibiotics.'' Working Paper, MIT, July 1998.
The results indicate that a general spatial Ellison, Sara F., Iain Cockburn, Zvi Griliches, and Jerry model of product differentiation does not Hausman. ``Characteristics of the Demand for Phar- adequately explain pricing behavior in the maceutical: An Examination of Four Cephalospor- ins.'' Rand Journal of Economics, 28(3), 1997, 426±46.
pharmaceutical industry because brand-name Frank, Richard G., and David S. Salkever. ``Pricing, Patent products respond aggressively to generic entry.
Loss and the Market for Pharmaceuticals.'' Southern The results, however, also reject the segmented Economic Journal, October 1992, 165±79.
market hypothesis, showing instead that there ÐÐÐ. ``Generic Entry and the Pricing of Pharmaceuti- is important competition between the pioneer, cals.'' Journal of Economics & Management Strategy, brand name, and generic segments. Multiple Grabowski, Henry G., and John M. Vernon. ``Brand Loy- alty, Entry and Price Competition in Pharmaceuticals there are important differences in the compet- afterthe1984DrugAct.''JournalofLaw&Economics, itive effects of additional branded entry com- pared to the effects of incremental generic ÐÐÐ. ``Longer Patents for Increased Generic Competi- Decade.'' PharmacoEconomics, 10(suppl 2), 1996, firms, there is a significant inverse relationship between price and the number of competitors, Greene, William H. Econometric Analysis, 2nd ed. New whether those competitors are brand name or Griliches, Zvi, and Iain Cockburn. ``Generics and New Goods in Pharmaceutical Price Indexes.'' American The broad implication of the analysis is that Economic Review, 84(5), 1994, 1213±32.
entry of additional sellers reduces prices much Lichtenberg, Frank R. ``Are the Benefits of Newer Drugs more substantially than previous work would Worth Their Cost? Evidence from the 1996 MEPS.'' suggest. The extent to which these results carry Health Affairs, September/October 2001, 241±51.
over to other therapeutic classes remains an Lu, Z. John, and William Comanor. ``Strategic Pricing of New Pharmaceuticals.'' Review of Economics and Statistics, 80, February 1998, 108±18.
Masson, Alison, and Robert L. Steiner. Generic Substitu- tion and Prescription Drug Prices: Economic Effects of State Product Selection Laws. Washington, DC: Fed- Applebaum, E. ``The Estimationof theDegree ofOligopoly Power.'' Journal of Econometrics, 19, 1982, 287±99.
Porter, R. ``A Study of Cartel Stability: The Joint Executive Bresnahan, T. F. ``Departures from Marginal-Cost Pricing Committee, 1880±1886.'' Bell Journal of Economics, in the American Automobile Industry: Estimates for 1977±1978.''JournalofEconometrics,11,1981,201±27.
Reiffen, David, and Michael R.Ward. ``Generic Drug ÐÐÐ. ``Empirical Studies of Industries with Market IndustryDynamics.'' WorkingPaper, FTC,February Power,'' in The Handbook of Industrial Organization, edited by R. Schmalensee and R. Willig. North- Reiss, Peter C., and Pablo T. Spiller. ``Competition and Entry in Small Airline Markets.'' Journal of Law & Bresnahan, T. F., and P. C. Reiss. ``Entry and Competition Economics, 32, October 1989, S179±S202.
in Concentrated Markets.'' Journal of Political Eco- Schmalensee, Richard. ``Entry Deterrence in the Ready-to- Eat Breakfast Cereal Industry.'' Bell Journal of Caves, Richard E., Michael D. Whinston, and Mark A. Hurwitz. ``Patent Expiration, Entry and Schwartzman, David. Innovation in the Pharmaceutical Competition in the U.S. Pharmaceutical Industry.'' Industry. Baltimore, MD: Johns Hopkins University Scientific American Medicine. Edited by Edward Rubin- 32. Reasons for such caution include possible cost and stein and Daniel Federman. New York: Scientific regulatory differences, the fact that a single prescription is often part of a maintenance program of therapy in other Scott Morton, Fiona M. ``Entry Decisions in the Generic pharmaceutical areas, and such repeated use may lead to PharmaceuticalIndustry.''RandJournalofEconomics, differences in brand loyalty and competition.
WIGGINS & MANESS: PRICE COMPETITION IN PHARMACEUTICALS Scott Morton, Fiona M. ``Barriers to Entry, Brand Adver- Suslow, V. ``Estimating Monopoly Behavior with Compe- tising, and Generic Entry in the US Pharmaceutical titive Recycling: An Application to Alcoa.'' Rand Industry.'' International Journal of Industrial Organi- Journal of Economics, 17(3), 1986, 389±403.
Wiggins, Steven N., and Robert Maness. ``Price Competi- Stern, Scott. ``Product Demand in Pharmaceutical tion in Pharmaceutical Markets: The Case of Anti- Markets.'' Mimeo, Sloan School of Management, infectives.'' Working Paper, Texas A&M University,

Source: http://www.fgcasal.org/politicafarmaceutica/docs/Steven_Wiggins_and_Robert_Maness.PDF