What drives drug procurement prices?

An arrow made of white pills points upward on a blue background

The factors that determine the prices paid by different public agencies for goods and services have first order implications for government finances and for the quantity and quality of public sector delivery, given that a significant share of public spending is devoted to public procurement.

In BSE Working Paper 1413, “Drivers of Public Procurement Prices: Evidence from Pharmaceutical Markets,” Claudia Allende, Juan Pablo Atal, Rodrigo Carril, José Ignacio Cuesta and Andrés González-Lira examine the determinants of public procurement prices using comprehensive data on pharmaceutical purchases by the Chilean public sector. 


Buyers pay very different procurement prices for the same product, but differences may be overstated by imperfect measurement

Figure 1 below shows the estimated distributions of buyer fixed effects on log prices based on two different product definitions. The first product definition is at the drug level (blue line), a coarser definition, and the second is at the product level (red line), more granular product definition.

Panel (a) displays the raw estimates of the densities and panel (b) displays the densities after applying empirical Bayes methods to account for noise when estimating the buyer fixed effects on procurement prices. 

This figure displays the density of buyer effects in log drug prices, estimates at the drug (blue) and product level (red). Panel (a) displays raw estimates, whereas Panel (b) displays results after shrinkage using empirical Bayes methods. The brackets on top of the densities indicate the 10th and 90th percentiles of each distribution.
Figure 1: Distribution of buyer effects on log(price)

Three patterns emerge from these results:

  1. There is substantial dispersion in the prices different buyers pay for the same products. Regardless of the specification, the estimated distributions of buyer effects display a large dispersion.
  2. By comparing distributions within each panel, the distribution of buyer effects is compressed when the product definition is more granular, as expected.
  3. By comparing across panels, we see that shrinking buyer effects indeed reduces the dispersion of estimates. In particular, the estimated difference between the 90th and 10th percentiles would be 18.5% without the shrinkage correction.

Overall, accounting for more accurate product definitions and estimation noise reduces the dispersion of buyer effects by 44.4%.

This variation is systematically related to observable characteristics of the buyer and the institutions that govern the procurement process

The authors project the estimated buyer fixed effects on a set of buyer observables that are grouped into institutional, geographic and size-related drivers. When considered jointly, all three sets of variables matter to explain buyer effects. Moreover, geography matters relatively less than institutional and size-related covariates after controlling for other drivers.

Market structure appears to be strongly correlated with procurement prices

Using a comprehensive descriptive regression analysis, the authors find that a higher number of available drug vendors is associated with lower prices. Their findings from this regression analysis also suggest that market structure plays a relevant role in explaining dispersion in procurement prices independent of procurement officers’ ability or behavior. Furthermore, the authors decompose how much of the variation in log prices is explained by buyer effects, market structure and other characteristics and find that 50% of the model’s explanatory power can be attributed to the number of vendors in the market.

Case study: changes in market structure due to patent expiration

To provide a more causal interpretation to the relationship between market structure and procurement prices, the authors leverage the expiration of drug patents as a natural experiment that induces changes in market structure.

A patent grants an innovator an exclusive right to sell products based on the patented molecule. Hence, this gives innovators a monopoly in upstream markets, curtailing product diversity and limiting the number of vendors. Once the patent has expired, generic manufacturers can enter the market and sell the drug, which is the source of variation that the authors exploit in this analysis.

Figure 2 below shows that the number of products and vendors increase after patent expiration, and Figure 3 displays the impact of patent expiration on procurement prices.

Average procurement prices decrease steadily after patent expiration, with the total decrease reaching almost 30 per cent four years after patent expiration (Figure 3). This decrease in average paid prices is not driven solely by lower-priced entrants: the authors estimate a slightly smaller price decrease on innovator prices. These results suggest that the increased numbers of products and vendors in the market have strong competitive effects.

This figure displays results from a drug-quarter-level regression for the number of products in the market.
Figure 2: The effect of patent expiration on market structure
This figure plots the point estimates and confidence intervals of βk in equation (3). An observation is a drug-quarter. The outcome variable is the log average procurement price of all products (blue) and the innovator only (red). Standard errors are clustered at the drug level.
Figure 3: The effect of patent expiration on prices: Innovators vs All products

Pharmaceutical market structure drives procurement prices more than buyer effects

In sum, the authors separately estimate the extent to which buyer effects and market structure explain procurement prices. The results highlight that market structure is a crucial driver of procurement prices. Through two separate analyses, the authors show that supply-side drivers of procurement prices explain more of the variation in prices than demand-side drivers, which have been the focus of much of the previous literature.