Table 2 reports pooled brand-channel regressions for log annual treatment price for our three country groups. Robust standard errors are clustered at the country level. Exponentiated coefficients (including a variance correction (Kennedy, 1981)) for indicator variables are in italics. Table 3 reports separate regressions by channel and brand status for the matched PCI-range countries. These channel-specific equations permit all coefficients to differ by brand and channel status and show effects of individual GPRM purchasers. Our discussion here is based mainly on the pooled regressions, with reference to the channel-specific regressions where relevant.
Brand and channel effects
For the all-countries sample, the price differentials, relative to retail originators, are −35.1% for retail generics, −71.2% for tendered originators, and −81.9% for tendered generics. For the matched PCI-range countries, the differentials relative to retail originators are −31.6 percent for retail generics (but not significant after controlling for volume), −42.4% for tendered originators, and −66.8% for tendered generics. Thus, although retail channel generics in MLICs on average charge 31.6% less than originators, the large variance implies statistical insignificance. By contrast, both originator and generic prices are significantly lower through procurement. These large procurement differentials plausibly reflect procurers' use of competitive tendering and the focus of competition on price, by requiring that suppliers meet WHO qualification to reduce quality risk. Originators may also be more willing to discount to the procurement channel due to reduced risk of price spillovers to high income purchasers in the same or other countries. Excluding volume does not significantly change these mean channel price differentials.
The average elasticity of price with respect to volume is −0.106 in the pooled estimates. The channel-specific estimates show slightly larger volume elasticities in procurement compared with retail pharmacy: −0.04 and −0.07 for retail generics and originators, respectively, and −0.067 and −0.145 for tendering generics and originators, respectively. Thus, contract volume contributes to lower procurement prices mainly for originators. However, these contract-specific volume elasticities do not capture the scale economies that may have resulted from the expansion of aggregate demand that was associated with procurement.
The income elasticity of prices with respect to average PCI is 0.267 for the full range of countries and only 0.100 in MLICs, or far less than required to maintain prices proportional to PCI. In the channel-specific MLIC regressions, the income elasticity of originator prices is insignificant in the retail channel but 0.32 in the procurement channel, suggesting that originators are more willing to discount to low-income countries in the protected procurement context. The income elasticity of generic prices is insignificant in the procurement channel, as expected if tendering forces generics to price at marginal cost, which varies little across countries. By contrast, in the retail channel, the income elasticity of generics is negative, consistent with the hypothesis that price competition is ineffective in MLIC retail markets.
Income skewness also appears to contribute to higher prices in retail pharmacies in MLICs, but this effect is eliminated by procurement.19 The channel-specific regressions for PCI-range countries show a significant, positive skewness effect for both originator (0.05) and generic prices (0.086) in retail pharmacies but no significant effect in the procurement channel.20
Originator prices are significantly inversely related to HIV prevalence in MLICs, but the effect is small, with larger effects in the pharmacy channel than the GPRM channel.
Overall, this evidence implies that drugs in retail pharmacies are least affordable in the lowest income countries, due to insignificant or perverse PCI-based price differentials for originators and generics and positive skewness effects for both generics and originators. Procurement reduces drug prices on average and eliminates the regressive mean PCI and skewness effects.
In the all-countries, pooled channels regressions, the marginal effect of an additional tendering generic is to reduce drug prices by 4.5%, compared with only −0.89% for an additional retail generic. The smaller marginal effect for retail generics may partially reflect the much larger mean number of retail generics than tendering generics. Further, our count of tendering competitors may be downward biased, because it omits any who tendered unsuccessfully, unless they chose to sell to retail pharmacies in that country-class-year.
The channel-specific regressions shed further light on the nature of competition. The marginal effect of tendering generics is significantly negative (−0.068) for originator prices but positive for generic prices in the retail sector. By contrast, additional retail generics tend to lower average retail generic prices but have no effect on originator prices. This supports the hypothesis that in the retail channel, tendering generics are more effective competitors for originators and can charge high prices relative to retail generics because tendering generics are perceived to be relatively high quality. In the procurement channel, number of competitors appears to have no effect on prices. This conclusion is tentative due to only observing successful bidders.
Markets with the originator present have 36–44% higher prices on average. The channel-specific regressions confirm that generic prices in the retail channel are 38.3% higher when the molecule originator is present, consistent with shadow pricing by generics. Procurement eliminates this effect.
Retail originator prices are positively associated with the number of originator competitors, contrary to simple price competition models. This may reflect increased promotional spending by originators in crowded classes, which makes product-specific demand more inelastic; the estimates may also be upward biased for net prices if competition leads to increased discounting that is not fully captured by the IMS data. Although these estimates are also potentially upward biased by endogeneity, such bias is mitigated because originators face lengthy regulatory requirements and other delays in launching a new product, which could take at least 2 years. Entry decisions depend on expected prices over the multi-year life of a product, and rapid entry in response to a current price premium would be irrational if other competitors could easily enter and compete down the price.
Overall, this evidence suggests that competition on price is generally weak in retail pharmacies in MLICs. Having multiple originators in a class does not reduce retail prices on average. An additional retail generic has no significant effect on originator prices and a weak (−2.0%) effect on average generic prices. An additional tendering generic reduces originator prices by about 7%, but raises average generic prices, plausibly because tendering generics can charge relatively high prices due to their perceived higher quality. These results are consistent with models in which uncertain quality leads to competition on brand as a proxy for quality, rather than price competition.
- Tendering purchaser effects
The channel-specific regressions show that the Global Fund and IDA pay at least 20% higher prices for both generic and originator drugs, and Missionpharma pays a similar premium for generics, compared with smaller purchasers (the omitted category). Given molecule, year, and volume controls, these effects cannot be attributed solely to differences in volume or drugs purchased by different purchasers. These purchaser differentials may reflect their intentional policies to pay prices sufficiently high to continue to attract multiple bidders to this market, which would be strategically rational for large purchasers. It is also possible that our DDD-corrected prices and form indicator do not adequately control for pediatric formulations that may be purchased disproportionately by some payers.21
Comprehensive anti-infective and cardiovascular class results
As a robustness check on our results for HIV, malaria, and TB drugs, Table 4 reports regressions for the entire anti-infectives (J) and cardiovascular (C) classes, for retail originators and generics in the matched PCI-range countries.22 These drug categories are not centrally procured by NGOs, and hence no GPRM data are available. The dependent variable is log price per standard unit rather than log DDD-adjusted annual treatment price as we lacked DDD data for many drugs.
Table 4. Effects of income, competition, and procurement on drug prices in retail pharmacies: all anti-infective (J) and cardiovascular (C) drugs, per capita income-range countries
|OLS regressions of log price per standard unit, 2004–2008 IMS data|
| ||HIV/AIDS, malaria, and TB drugs||Entire J class (anti-infectives)||Entire C class (cardiovascular)|
|Log per capita income||−0.591*||0.126||−0.274||−0.944***||−0.269||−0.940***|
|Gini missing indicator||−0.54||0.145||−0.825||−2.052***||−0.804||−2.055***|
|HIV prevalence per 1K||−0.0108||−0.0199***||−0.0191**||−0.0242***||−0.0192**||−0.0244***|
|Generic firms in class||−0.0123***||−0.0134***||−0.00405***||−0.00227***||−0.00442***||−0.00203***|
|Originator firms in class||0.00177||0.179***||0.00715||−0.0362||−0.0161||−0.0566**|
|Originator present in molecule||0.165||N/A||−0.11||N/A||−0.0698||N/A|
|Generic present in molecule||N/A||−0.0657||N/A||−0.113||NA||−0.106|
|Non-oral solid flag||1.685***||1.501***||0.689***||0.621***||−0.488**||−0.166|
|Molecule and year fixed effects||X||X||X||X||X||X|
Income elasticities of prices in both classes are significantly negative for originator drugs, whereas Gini coefficients are significantly positive. These estimates are tentative due to significant correlation between log PCI and Gini for these countries. In similar regressions for originators and generics combined for the entire income range of countries, the income elasticity for J and C class pharmacy drugs is around 0.3 and the Gini is insignificant.23
An additional generic competitor reduces generic prices by 0.4% and originator prices by 0.2% in both the J and C classes. Competition from other originators in the class reduces originator prices significantly only for the C class.
Thus, overall, the conclusions from the HIV/TB/malaria drugs are confirmed by these two very large classes. The at best small (0.0–0.3) income elasticity of drug prices implies that prices are least affordable, relative to income, in low-income countries. Income skewness exacerbates these effects. In MLICs, despite multiple competitors, price competition does not appear to be strong in retail channels, plausibly because competitors compete on brand rather than price due to quality uncertainty.