Review of Industrial Organization 19: 425–435, 2001. © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
425
The Impact of Generic Drug Competition on Brand Name Market Shares – Evidence from Micro Data THOMAS ARONSSON1, MATS A. BERGMAN2 and NIKLAS RUDHOLM1, 1 Department of Economics, Umeå University, SE-901 87 Umeå, Sweden 2 Swedish Competition Authority and The Research Institute of Industrial Economics, Box 5501,
SE-114 85 Stockholm, Sweden
Abstract. This paper analyses how market shares for brand name drugs are affected by generic competition. The analysis is based on micro data for twelve different original drugs, which are all subject to generic competition. For five of these drugs, we find that the price of the original relative to the average price of the generic substitutes significantly affects the market share of the original drug. In addition, the introduction of a so called “reference price” system appears to have had a significant impact on the market shares of five original drugs. Key words: Micro data, patent expiration, pharmaceutical industry. JEL Classifications: L65, I11.
I. Introduction Throughout the industrialized world, the cost for pharmaceutical drugs raise concern. Potentially, these costs could be reduced if government regulations could foster a more powerful competition between the original manufacturers and the manufacturers of generic substitutes. The Swedish pharmaceuticals market is regulated in the sense that the prices of prescription drugs, which are subsidized by the government, are determined by negotiations between the producers of these pharmaceutical products and the regulatory authority. Once the price has been set, the regulatory authority will not in general allow large price increases, unless the firm can show that its costs have increased by a similar magnitude. The prices of pharmaceutical products are, nevertheless, high during patent protection and can be expected to be relatively close to the monopoly price. When patents expire, and generic products enter a market, the reference price system becomes effective. The purpose of this system is to stimulate the producers of drugs, and particularly the producers of brand name drugs, to lower their prices in order to reduce the costs of medical subsidies facing The authors would like to thank Mårten Palme, Magnus Wikström and two anonymous referees
for helpful comments and suggestions. A research grant from the Swedish Competition Authority is also acknowledged.
426
THOMAS ARONSSON ET AL.
the National Social Insurance Board. The reference price system came into effect in 1993, and specifies that any costs exceeding the price of the least expensive generic substitute by more than 10 per cent must be borne by the patients.1 In the early 1990s, a number of studies suggested that generic competition affects brand name prices and market shares. Grabowski and Vernon (1992) find that generic producers often capture a relatively large market share very soon after patent expiration. Contrary to previous authors, they attribute this gain of market shares to the price differentials: The price of the generic substitutes are on average much lower than the price of the brand name product. However, the loss of market shares of brand name products also suggests that these producers may have an incentive to lower their prices, when generic products enter the market. This may also be the case according to Caves et al. (1991): Their results imply that the prices of brand name drugs decline as a consequence of generic entry. On the other hand, the more recent paper by Frank and Salkever (1997) suggests the opposite result. They find that the prices of brand name drugs increase after generic entry into the market, while the prices of already established generic products tend to decrease. Hudson (1992) uses European data to analyse the consequences of generic competition. His model is explicitly dynamic and is applied on data from the U.S., U.K., West Germany, and France. The results imply that the entry of new products into a therapeutic class reduces the price of the incumbent drugs significantly in Germany and France. The purpose of this paper is to study how the market shares of brand name products in Sweden are affected by generic competition. In particular, our concern will be to examine how the price of the brand name product, relative to the average price of the generic substitutes, affects the market share of the original product. This paper differs from almost all previous studies in at least three ways. First, we have access to very long time-series, with prices and quantities, for individual drugs (both original and generic). This makes it possible to perform separate estimations for each such individual drug. As we are about to show, the use of disaggregate data is very important in order to fully understand the consequences of generic competition, since there are considerable differences between the submarkets. Second, contrary to most previous studies we use European data, which provide the possibility of comparison with the pharmaceuticals markets in the U.S. and Canada. Finally, we are able to analyse how the introduction of a reference price system has affected the market shares of original drugs, as well as the price of the brand name drugs relative to the price of the generic substitutes.
1 Before 1993, the patient had to pay all cost below 120 Swedish crowns themselves, while costs
exceeding 120 crowns were borne by the National Social Insurance Board (NSIB). In addition, after paying a total of 1500 Swedish crowns, all costs were reimbursed by the NSIB.
THE IMPACT OF GENERIC DRUG COMPETITION ON BRAND NAME MARKET SHARES
427
II. The Model A physician has no direct pecuniary incentives to choose less expensive generic products or on the whole to inform himself/herself about generic alternatives. According to an estimate presented by Bleidt (1992), the pharmaceutical industry spends slightly more on marketing and promotion than on R&D. In many instances, the bulk of this effort promotes the brand-name product, increasing the likelihood that a physician will patronize the original manufacturer’s product.2 The physician may also feel loyalty towards the original manufacturer, or sheer inertia may stop him/her from changing prescription habits. However, out of consideration for the patients he/she may, nevertheless, react to price differences. To simplify the analysis, we assume that the market demand for a given drug (i.e., the combined demand for the original drug and its generic substitutes) is perfectly inelastic with respect to prices. The market shares of the competing producers are determined by relative prices. Hellerstein (1998) analyses a model, where the prescribing physician acts as an agent for the patients, which means that concern for the patients may provide an incentive to change prescription habit from the brand name product to less expensive generic substitutes. In Sweden, however, the consumer prices of pharmaceutical products are subsidized to a large extent, meaning that only a small fraction of the costs are borne directly by the patients. Instead, the tax payers will carry the main part of the burden. Therefore, consider a situation where the prescribing physician acts as an agent for the tax payers. To make this idea operational in a simple way, suppose that the physician faces a utility gain (loss) if his/her prescription habits contribute to decrease (increase) the costs of medical subsidies for the tax payers. Let ut be the total discounted change in expected utility of the physician if changing the prescription habits from the brand name product to a generic substitute in period t. We simplify by assuming that ut depends on the currently observed price of the brand name product, pto , relative to the currently observed price of the g generic substitute, pt , according to ut =
nη g (p o /pt − 1), (1 − δ) t
(1)
where n is the number of times the physician prescribes the pharmaceutical product per time period, η a preference parameter and δ the discount factor. This formulation means that the utility change is positive (negative) if the price of the original drug exceeds (falls short of) the price of the generic substitute. Considering that the main part of the marketing effort in the pharmaceutical industry is used to promote brand name products, the physician may also act as 2 Fridman et al. (1987) reports that only half of 245 surveyed physicians believed generic drugs to
be as reliable as trade name drugs. Consistent with this result, a relatively low fraction of physicians report to prescribe generics often, except in the case of antibiotics. In contrast, Kendall et al. (1991) reports that “generic substitution is highly acceptable to (patients)”; even more so if their costs are not fully reimbursed.
428
THOMAS ARONSSON ET AL.
an agent for, or at least feel loyalty towards, the producer of the brand name drug. Suppose that the physician incurs a switching cost, c, (in utility terms) if he/she changes prescription habits from the brand name drug to a generic substitute, and that this switching cost may differ among physicians. Given that the physician prescribed the original drug in period t − 1, he/she will switch to the generic drug in period t if ut − ct > 0,
(2)
i.e., if g
γ (pto /pt − 1) − ct > 0,
(3)
where γ = nη/(1 − δ). To be able to relate Equation (3) to the change of market share of the original drug, let the switching costs for the physicians that prescribe the more expensive original drug be uniformly distributed in all periods and independent of the market share of the original drug. That is, in every period, a new switching cost is drawn from a uniform distribution. Formally, let the distribution for switching costs in period t be defined over the interval [at , at + b]. If, at the end of period t − 1, a share st −1 of the physicians patronize the original drug, the g fraction out of that share for whom ct < γ (pto /pt − 1) is given by g
γ (pto /pt − 1) − at (γ + at ) γ o g st − st −1 + (pt /pt ) = =− − st −1 at + b − at b b
(4)
Rewriting Equation (4), we have st − st −1 g = αt + β(pto /pt ), st −1
(5)
where αt = (γ + at )/b and β = −γ /b. Equation (5) gives the relative change of market share for the original drug as a linear function of the price of the original drug relative to the price of the generic substitute. Another interpretation of Equation (5) is that the first difference in the market share of the original drug in period t depends on the market share of the brand name drug at the end of period t − 1 and the relative price. This relationship will serve as a starting point for the empirical analysis to be carried out below. III. The Empirical Analysis 1. DATA We have access to quarterly time-series data with respect to prices and quantities for each original product and its generic substitutes from 1972 to 1996, which have been provided by the Swedish Medical Product Agency. These medical substances
THE IMPACT OF GENERIC DRUG COMPETITION ON BRAND NAME MARKET SHARES
429
Table I. Market shares and relative pricesa Substance
M-share Brand
M-share Generic
Relative price (std. dev.)
ATCcode
Cimetedine Furosemide Atenolol Pindolol Propranolol Indomethacine Naproxen Allopurinol Paracetamol/Codeine Diazepam Clomipramine Timolol
64.57% 47.40% 85.51% 97.63% 96.98% 63.56% 44.30% 93.03% 92.85% 41.68% 39.53% 82.08%
35.43% 52.60% 14.49% 2.37% 3.02% 36.44% 55.70% 6.97% 7.15% 58.32% 60.47% 17.92%
1.62 (0.66) 1.57 (0.45) 1.58 (0.44) 1.36 (0.18) 1.87 (0.55) 1.15 (0.10) 1.44 (0.37) 1.33 (0.22) 1.23 (0.10) 1.17 (0.12) 1.44 (0.19) 1.75 (0.40)
A02B C03C C07A C07A C07A M01A M01A M04A N02B N05B N06A S01E
a The figures in the table refer to average market shares and relative prices (po /pg ) during the estimation period.
come from five of 14 different fields of use, as defined by the Swedish Medical Products Agency (MPA). All of the substances in our sample have a minimum sale of ten thousand packages each quarter for the chosen package size. Our data then refers to twelve different substances, and the information about prices and sold quantities is available both for each original product and for the generic substitutes. The price of generic products to be used in the estimations is the average price of generic substitutes in each submarket, which is measured as a quantity weighted average. In addition, the products used in the study refer to the dose and package size with the largest registered sales. Table I contains information about the average market share (measured in quantity) for each original substance and its generic substitutes during the estimation periods. We also present means and standard deviations for the relative prices (i.e., the price of the original product relative to the average price of the generic substitutes) corresponding to each such substance. 2. R ESULTS FOR A BASIC M ODEL In this section, we present estimation results for a “basic” model, which is based on Equation (5). The basic regression model sit − sit −1 g = αi0 + αi1 T + βi (pito /pit ) + uit , sit −1
(6)
where uit is a random term, which is assumed to be i.i.d. across substances. The term T represents a time trend, the purpose of which is to capture possible time
430
THOMAS ARONSSON ET AL.
Table II. Estimation results for the basic modelb Substance
αi0
αi1
βi
R2
Paracetamol/Codeine
0.027 (1.01) 0.026 (3.25) −0.025 (−1.53) 0.048 (1.77) 0.040 (5.54) −0.29E−2 (−0.45) 0.014 (0.42) 1.400 (1.13) 0.160 (2.90) 0.92E−2 (0.77) 0.014 3.17 −0.031 (−1.96)
−0.13E−3 (−0.87) −0.35E−3 (−1.81) 0.18E−4 (0.01) 0.63E−4 (0.56) −0.18E−3 (−2.32) −0.74E−4 (−1.19) 0.63E−4 (0.64) −0.018 (−0.59) −0.12E−2 (−1.12) −0.93E−4 (−0.72) −0.23E−3 (−3.67) 0.14E−2 (4.44) −0.78E−4 (−0.97)
−0.022 (−1.17) −0.017 (−4.79) 0.300 (0.49) −0.055 (−2.17) −0.039 (−8.74) 0.11E−2 (0.27) −0.019 (−0.65) −0.930 (−1.35) −0.121 (−4.55) −0.63E−2 (−0.86) −0.053 (−3.32) 0.64E−3 (0.087) −0.021 (−4.80)
0.21
Atenolol Cimetedine Diazepam Furosemide Allopurinol Indomethacine Clomipramine Naproxen Pindolol Propranolol Timolol All substances
0.39 0.06 0.16 0.46 0.18 0.43 0.44 0.52 0.14 0.38 0.40 0.14
b t-Values are given within parentheses.
dependence (other than via the random term) of the distribution for switching costs facing prescribing physicians. For each substance, this equation is estimated using a Cochrane–Orcutt technique to control for serial correlation. To facilitate comparison with previous studies, we also estimate a version of the model where the data for all drugs are pooled together using fixed effects to control for differences across substances. The results are presented in Table II. Let us begin with the estimation results for the ‘pooled’ model, which are found at the bottom of Table II. Contrary to many previous studies referred to above, we find that relative prices has a significant effect on the change of market share of the original product. The higher the price of the original product relative to the
THE IMPACT OF GENERIC DRUG COMPETITION ON BRAND NAME MARKET SHARES
431
average price of the generic substitutes, the larger the decrease of market share of the original product. To gain further insight into the relationship between the change of market share for the original product and the relative price, let us turn to the estimation results corresponding to the different substances. We find that in five out of twelve cases, the relative price has a negative and significant effect on the change of market share of the original product. These five substances are Atenolol, Diazepam, Furosemide, Naproxen and Propranolol. In addition, the impact of generic competition seems to differ substantially across markets, which makes the use of disaggregate data very important in order to understand how the market shares of brand name drugs are affected by generic competition. Note finally that the ability of the basic model to explain the change of market share for an original product differs considerably across substances. The latter means that the model set out in the previous section may not always provide a suitable description of what factors determine the change of market share for the brand name product.3 3. E XTENSIONS OF THE M ODEL The reference price system briefly described in Section I may affect the equation for the change of market share via several distinct channels. First, the reference price system is likely to influence the price of the brand name drug relative to the average price of the generic substitutes, which is already implicit in Equation (6). However, it may also affect the distribution of switching costs facing the prescribing physicians, since the reference price system contributes to make price differences more “visible”. If this argument is correct, consistent estimation of the parameter βi requires explicit consideration of the reference price system in the estimation. The basic model is extended in the following way: sit − sit −1 g = αi0 + αi1 T + αi2 D + (βi0 + βi1 D)(pito /pit ) + uit sit −1
(7)
where D is a dummy variable taking the value one after the introduction of the reference price system and zero before the introduction of the reference price system. However, Equation (7) is not unproblematic to estimate. For several substances, the relative price does not change much between 1993 and 1996 (the period during which the reference price system is active in the data), which makes it difficult to 3 A potential problem is that the relative prices may be endogenous, i.e., the relative price may correlate with the error term of the equation for the change of market share. We have reestimated the model using two-stage least squares. Lagged values of relative prices were used as instruments. These instruments turned out to have a significant effect on relative prices, and the explanatory power in the first stage was found to be high. Due to serial correlation, these estimations were not performed for Diazepam and Furosemide. For the remaining substances, the results in the second stage were very similar to those reported in Table II.
432
THOMAS ARONSSON ET AL.
Table III. Estimation results for the extended model Substance
αi0
αi1
αi2
βi0
R2
Paracetamol/Codeine
0.62E−2 (0.19) 0.017 (1.21) −0.015 (−1.62) 0.048 (1.25) 0.068 (12.65) −0.031 (−3.25) 0.39E−4 (0.00) 1.401 (1.88) 0.400 (5.97) 0.57E−2 (0.27) 0.020 (3.28) −0.032 (−1.53)
−0.15E−3 (−0.99) −0.70E−3 (−1.43) −0.99E−3 (−4.78) 0.63E−4 (0.54) 0.49E−3 (4.44) −0.28E−3 (−3.63) −0.18E−4 (−0.13) −0.52E−2 (−0.19) 0.15E−2 (1.40) −0.11E−3 (−0.68) −0.24E−3 (−3.97) 0.14E−2 (1.54) −0.32E−4 (−0.32)
0.41E−2 (1.03) 0.010 (0.77) 0.044 (7.48) −0.82E−4 (−0.01) −0.062 (−7.02) 0.018 (3.57) 0.92E−2 (0.83) −0.500 (−5.04) −0.170 (−4.67) 0.14E−2 (0.21) −0.33E−2 (−1.42) 0.21E−2 (0.10) −0.40E−2 (−0.68)
−0.62E−2 (−0.25) −0.011 (−1.30) 0.49E−2 (1.42) −0.055 (−1.61) −0.072 (−12.65) 0.023 (3.28) −0.47E−2 (−0.14) −0.660 (−1.75) −0.280 (−7.08) −0.38E−2 (−0.26) −0.80E−2 (−3.23) 0.18E−2 (0.13) −0.024 (−4.27)
0.24
Atenolol Cimetedine Diazepam Furosemide Allopurinol Indomethacine Clomipramine Naproxen Pindolol Propranolol Timolol All substances
0.40 0.52 0.16 0.64 0.35 0.44 0.83 0.72 0.14 0.40 0.40 0.16
simultaneously identify αi2 and βi1 . We have estimated two versions of Equation (7); one in which αi2 = 0 and the other where βi1 = 0. Note that, if the relative price is roughly constant from 1993 and onwards, the data used in the estimation of the interaction dummy variable resembles a linear transformation of the data used when estimating the intercept dummy variable. As a consequence, we would expect the two versions of the extended model to provide qualitatively similar results. This is also what happens: the intercept dummy variable is significant in the same cases as the interaction dummy variable, and the estimated relative price effects are more or less the same in both versions of the model. Therefore, we only present the results corresponding to the version of the extended model, where the interaction dummy variable is set equal to zero. The results are presented in Table III.
THE IMPACT OF GENERIC DRUG COMPETITION ON BRAND NAME MARKET SHARES
433
Starting once again with the case where the data for all substances are pooled together, the point estimate of βi0 is only slightly different from that of Table II, and the estimate of αi2 is not significant. Therefore, when the data for all substances are pooled together, the results suggest that the extension of the model discussed above does not provide any significant improvement over the basic model, and that the reference price system does not affect the change of market share of the brand name drug other than via (a possible) change in the relative price. However, by looking at the estimation results for each individual substance, we find that the reference price system significantly affects the change of market share for five original products. For two of these substances, Allopurinol and Cimetedine, the reference price dummy variable has a positive sign. For Furosemide, Clomipramine and Naproxen, on the other hand, the effect of D is negative as expected, and the estimates are highly significant. The estimates of βi0 are negative and significant for Furosemide, Naproxen and Propranolol, while the estimate for Allopurinol is positive and significant. In addition, by comparing Tables II and III, we find substantial differences in the point estimates of βi0 corresponding to Furosemide, Naproxen and Propranolol. This suggests that the basic model, which neglects the direct effect of the reference price system, is not always a suitable basis for identifying the effect of relative prices on the change of market share for the original product. 4. R ELATIVE P RICES AND THE I MPACT OF THE R EFERENCE P RICE S YSTEM As we mentioned above, if the reference price system has a direct effect on the change of market share of a brand name drug, there is a missing variable problem in the basic model. This is further emphasized by the possibility that the reference price system influences the price of a brand name drug relative to the price of a generic substitute, which would imply that the estimates of βi in the basic model actually represents a mixture of “pure” relative price effects and effects of introducing the reference price system. To study this issue a bit further, let us examine how the relative price has been affected by the reference price system. Suppose that the relative price is determined by the following equation: pito g = ρi0 + ρ1 T + ρ2 D + λGit + it , pit
(8)
where Git is the number of generic substitutes in market i at time t. Equation (8) is estimated by using the data for all substances, and we use fixed effects to control for differences across substances. The results presented in Table IV suggest that the introduction of the reference price system tends to decrease the price of the original relative to the price of the generics. The negative effect of the reference price system appears to be reasonable, since the introduction of this system may have provided strong incentives for manufacturers of brand name products to lower their prices. Note also that the number of generic competitors has a positive and
434
THOMAS ARONSSON ET AL.
Table IV. Estimation results for Equations (8) and (9) Dependent variable g
o /p pit it
o /p o pit i0
ρ1
ρ2
λ
R2
−0.187E−3 (−0.06)
−0.466 (−11.53)
0.071 (2.81)
0.90
κ1
κ2
π
R2
−0.476 (−13.60)
0.794E−2 (0.35)
0.95
−0.232 (0.53)
significant impact on the relative price. A possible explanation is that new entrants may choose lower prices than already established generic manufacturers, meaning that the average price of generics falls and the relative price increases. It is also interesting to study whether the relative price responses following the introduction of the reference price system originate from price responses of brand name or generic products. Therefore, as a complement to the results in Table IV, we also present results from the estimation of the following equation: pito o = κi0 + κ1 T + κ2 D + π G + νit , pi0
(9)
o is the “initial price” of the original (or brand name) product. In a way where pi0 similar to Equation (8), we also estimate Equation (9) by using the data for all substances and use fixed effects to control for differences across substances. The results are presented in Table IV. As expected, the results show that the brand name manufacturers have lowered their prices in order to conform to the reference price system.
IV. Summary This paper concerns the impact of generic competition on the market share of brand name drugs. We estimate a model, where the relative change of market share of the original drug depends on the price of the original relative to the price of the generic substitutes. The most important results are summarized below: • In the basic model, a higher price of the original product, relative to the average price of the generic substitutes, significantly decreases the market share of Atenolol, Diazepam, Furosemide, Naproxen and Propranolol. • The introduction of the reference price system seems to have decreased the market shares of Furosemide, Clomipramine and Naproxen. In addition, it appears to be important to control for the introduction of the reference price system in order to identify how the change of market share of the original
THE IMPACT OF GENERIC DRUG COMPETITION ON BRAND NAME MARKET SHARES
435
product is affected by the relative price. The differences in results between the substances also underline the importance of using disaggregate data. • Our results also suggest that the introduction of the reference price system is an important determinant of the price paths. Therefore, if we neglect the reference price system in the estimation, as in the basic model, estimated relative price effects on the changes of market shares most likely reflect a mixture of “pure” relative price effects and effects caused by the reference price system. References Bleidt, B. (1992) ‘Recent Issues and Concerns about Pharmaceutical Industry Promotional Efforts’, Journal of Drug Issues, 22, 407–415. Caves, R. E., M. D. Whinston, and M. A. Hurwitz (1991) Patent Expiration, Entry, and Competition in the U.S. Pharmaceutical Industry. Brookings Papers on Economic Activity: Microeconomics. Washington, DC: Brookings Institution. Frank, R. G., and D. S. Salkever (1997) ‘Generic Entry and the Pricing of Pharmaceuticals, Journal of Economics and Management Strategy, 6, 75–90. Fridman, D., A. Jaffe, and S. Steinhardt (1987) ‘Physicians’ Attitudes toward and Knowledge About Generic Drug Substitution, New York State Journal of Medicine, 87, 539–542. Grabowski, H. G., and J. M. Vernon (1992) ‘Brand Loyalty, Entry and Price Competition in Pharmaceuticals after the 1984 Drug Act’, Journal of Law and Economics, 35, 331–350. Hellerstein, J. K. (1998) ‘The Importance of the Physician in the Generic versus Trade-Name Prescription Decision’, RAND Journal of Economics, 29, 108–136. Hudson, J. (1992) ‘Pricing Dynamics in the Pharmaceutical Industry’, Applied Economics, 24, 103– 112. Hurwitz, M. A., and R. E. Caves (1988) ‘Persuasion or Information? Promotion and the Market Shares of Brand Name and Generic Pharmaceuticals’, Journal of Law and Economics, 31, 299– 320. Kendall, K. W., S. Ng, and B. Schoner (1991) ‘Consumer Responses to Generic/Chemically Equivalent Drugs’, Journal of Public Policy and Marketing, 10, 182–201.