Launch Decisions of Pharmaceutical Companies

Journal of Economics and Financial Analysis, Vol:1, No:1 (2017) 35-58

Launch Decisions of Pharmaceutical Companies

Abdülkadir CİVANa*, Michael MALONEYb

a Department of Economics, Clemson University, United States

b Department of Economics, Clemson University, United States


This paper models the launch decision of pharmaceutical companies in regard to new drugs and country markets. New drugs are launched with a delay or not launched at all in many countries. Considering that many of these new drugs would have created health benefits to the patients, there seems to be welfare loss. We use market characteristics to explain this phenomenon. We show that most of the estimated launch with a delay and no-launch decision is due to observable market characteristics. The model has an accuracy of 70 percent in explaining the no-launch decision. Intellectual property rights protection is especially important. The policy implication is that stronger property rights increase the likelihood and speed of new drug launch.

Keywords: Drug Launches; Intellectual Property Rights; Pharmaceutical R&D.

JEL Classification: I12, I18.

* Corresponding author.

E-mail addresses: (A.Civan), (M. Maloney)

Notes: Valuable comments were received from participants at the IO workshop, Clemson. Special thanks to Tom Mroz for help with the econometrics.


This paper studies the decision by pharmaceutical companies of whether and when to launch a new drug in a specific country-market. We expect that innovator firms would want to sell their new drugs in every country if they can price their products at more than marginal costs which include production, licensing and retail elements. However, the facts seem contrary. Out of the 1095 new chemical entities introduced since 1982 only 427 on average have been made available in OECD countries.

Market characteristics, such as potential demand, as well as the regulatory environment can influence the launch and timing decisions of pharmaceutical companies. We expect that innovator firms would want to sell their new drugs in every country assuming they can price their products more than marginal costs (including the costs associated with entering and operating a new market area). However Danzon et al. (2005) find that for a specified set of drugs among 25 industrialized countries almost no country had more than half of the potential country drug launches within 12 months of their first launch in the world.

The no-launch phenomenon is what strikes us as perplexing and inefficient. Chemical entities that are safe and effective in treating disease are a public good in the Demsetz (1970) sense. Efficiency considerations say they should be universally available. Health considerations put a fine edge on the efficiency argument. Lichtenberg (2001, 2005, 2007) has shown the improvements in health outcomes due to new drugs. Thus knowing why drugs are not launched everywhere is a step toward saving lives.

We wonder how much of the no-launch decision might be explained by intellectual property considerations. We examine the phenomenon of drug availability by looking at several factors that should affect drug launch. Potential market size, intellectual property rights protection, and other country specific regulatory effects should all play a role. We are able to explain a substantial part of the no-launch decision. A large majority of no-launch observation is due to quantifiable market characteristics. A major finding is that property-rights protection plays a strong role in the launch/no-launch decision. The important policy implication is that stricter protection of the intellectual property rights of pharmaceutical developers might improve health outcomes. However please note that we do not study the costs associated with stricter protection of the intellectual property rights. So we do not propose this policy for all countries. Each country should carefully review the costs and benefits associated with intellectual property rights.

These results are especially important considering the public debate about high drug prices in the United States (Danzon and Furukawa, 2008; Kanavos et al., 2013). When we look at the issue superficially, other developed countries that keep drug prices low, get the best of both worlds. Drug companies spend huge amount of R&D investments and create new and (sometimes) better drugs because they hope to make profits from the United States. Once these drugs are developed, other developed countries get access to them at discount prices. However, the story is not quite so simple. Civan and Maloney (2006, 2009) find that drug companies focus their R&D money on the diseases which are more prevalent in the United States. Although this does not eliminate free riding motives, it should diminish it. In this study along with others in the relevant literature we show that there are some other costs of reducing drug company profits. Drug companies delay or stop altogether the launches of their new drugs in less profitable countries. In the meantime members of those societies are deprived of potential benefits of those new drugs.

Background Literature

Most medical scientists and heath economists believe that many new drugs significantly improve the medical care and patients’ life quality. Obviously not all new drugs are wonder drugs or have substantial advances over the existing drugs. There are many “new” drugs that are very similar to the existing ones. The opinion about those drugs is more mixed. Many claim that they provide at least incremental benefits for some patients while others are in the opinion that they do not provide any clinical benefits. However even in the absence of clinical benefits, these new drugs might improve social welfare by reducing prices of existing drugs.

Lichtenberg (2001, 2005, 2006) has published a series of articles on the values of new drugs using various data sources and methodologies. His studies confirm the proposition that new drugs are very useful in terms of better health outcomes. Civan and Koksal (2010) and Lichtenberg (2001) show that new drugs not only improve health care quality but reduce the total health care expenditures by reducing the need and demand for other types of health care such as hospital care and physician visitations. The effects of new drugs on the welfare of the individuals can be through two ways. First one is the improvements on the longevity.1 Moreover, increase in longevity can increase the incentives for human capital accumulation (mostly by increasing education and training) and can decrease the risky life choices (smoking, speeding and similar attitudes).2

1 See Rosen (1988), Topel and Murphy (2003), Becker et al. (2005) and Soares(2007) for different aspects and estimates of increase in longevity on welfare.

So it would seem like the real resources of the world are wasted when new drugs are not available for the majority of patients. In the dynamic sense this not only affects the societies of no-launch or late-launch countries but also early-launch societies as well. No-launch and late-launches decrease profit potentials of innovating pharmaceutical companies thus causing higher prices and/or reduced R&D investments (thus fewer new drugs).

Once a drug is developed and clinical trials are completed, the innovator firm applies for the approval of the drug by the regulatory agency of the country. For many countries there are some fees involved on that licensing process.3 In many countries after the approval of the drug, the innovator firm has to settle with Ministry of Health and/or Social Security Agency in order to be included in drug reimbursement lists. This process naturally takes time and money. Many countries’ public agencies (drug licensing and reimbursement agencies) are slow so that there are significant lags between first approval date of the drug in another country in the world and its launch to that specific country. If the profit potential (due to low prices, low potential sales and low intellectual property rights protection) of the new drug in specific country is not high, the innovator firms might not apply for the approval of the drug for those relatively unprofitable markets.4 On the other hand, sometimes drugs can obtain approval in one country and not in another because of different level of strictness of drug licensing agencies.

In the relevant literature two other potential reasons of drug companies’ unwillingness to market their new drugs are discussed: Parallel imports and reference pricing. (Lanjouw, 2005; Danzon and Epstein, 2008) Parallel imports are the importation of the drugs from cheaper countries to the more expensive ones. Since there are significant variations in the prices of same drugs across countries, there are profit potentials for the arbitrageurs by importing drugs from low price countries to high income countries. This creates a natural reservation for the drug companies to market drugs in low price countries. Although currently national legislations, policies of insurance companies, and the mood among health professionals and public at large prevent substantial volume of parallel imports, it has the potential to be a more significant cause of no-launch in the future.

2 See Oster and et all. (2012) for a recent empirical estimate of increase in longevity and increase in education and training and also decrease in smoking.

3 In the United States the fee is $2,169,000 for a new drug application which require

clinical trials.

4 See Acemoglu and Linn (2004), Giaggotto, Santerre and Vernon (2005), and Civan and Maloney (2006, 2009)

The bigger issue is that reference pricing of the drugs with the prices in other countries. Many countries determine the prices of the drugs by Ministries of Health and/or Social Security Reimbursement Agencies. An increasing number of countries use some form of external referencing system in the price determination process. Although there are various forms of the external reference pricing, the underlying concept is the same: the drug price is determined by the average or minimum price of the same drug in a set of countries selected by the government. Drug companies that fear the indirect price effect of low drug price countries on high price country markets, either do not launch or delay the launch of their drugs in those countries.

Several papers on the literature studied the issue formally and tried to determine factors affecting drug companies launch decisions. Danzon, Wang and Wang (2005) analyzed the launch in 25 major countries of 85 new chemical entities (NCEs) between 1994 and 1998. They use the average price of the existing drugs in the same therapy group as a proxy of the intensity of regulation. Similarly in order to estimate potential market size they use the average sales volume of the existing drugs. Higher prices and bigger potential of sales are found to be increasing the probability and speed of launch. Danzon and Epstein (2008) analyzed the launch decisions of all drugs in 12 therapeutic categories launched in 15 countries between 1992 and 2003. Similar to Danzon’s first paper on the topic they used the average price of established products on the therapeutic group of the drug as the proxy for the intensity of regulation. They found a small but statistically significant, negative effect of strict regulation on the launch of drugs.

However those methods do not explicitly control for the regulatory environment. Lanjouw (2005) used a direct variable to control the intensity of regulation in each country. She created dummy variables for several regulation forms such as existences product patents, process patents, price controls, and national drug formularies. She found that intensity of regulation in high income countries reduce access to new drugs though her evidence on poor and middle income countries is more mixed.

On the other hand in order to incorporate the regulatory environment of the countries Kyle (2007) used the general market competition index by Djankov et al. (2002) and a set of dummies for different types of price controls. However market competition index is a general one which has a single value for all sectors in the country. So it might not represent the characteristics of pharmaceutical markets of the countries. Nevertheless she concluded that price controls have statistically

negative effect on the launch not only for the price controlling country but other countries as well. Heuer et al. (2007) focused on drugs approved by the European Medicines Evaluation Agency centralized procedure. Drugs approved by European Medicines Evaluation Agency are automatically approved in member countries. This approach allowed them to separate the effects of national price and reimbursement regulations from the impact of the market authorization procedure. They found that among the direct price controls, only international price comparisons have a significantly negative impact on the launch timing.

In this paper we use the similar techniques and data sources to study the launch decisions of drug companies. However we believe our analysis extends the existing literature in several important aspects. Most papers use either the average price of the existing drugs or ad hoc dummies for specific regulations as a proxy for the strictness of the regulation.5 However, a low price can be attributed to the either low demand or strict market regulations. Demand depends on the number of potential consumers (patients) and their willingness to pay. So the studies which used average price as the proxy for intensity of regulation are arguably mixing regulation and demand conditions. In this study we use new data compiled by Liu and La Croix (2008) that gives a Pharmaceutical Intellectual Property Protection Index for 154 countries. This index is better for our purposes than general market competition index of Djankov et al. (2002), and complements ad hoc specific dummies for certain types of market control instruments and average price of existing prices. We discuss the Liu-LaCroix index in some detail later.

The second contribution of our paper is that we specifically consider the potential market size for new drugs. In earlier papers some researchers proxy potential market size by sales volume of existing drugs and/or number of existing drugs in the same therapy group. Although these measures are useful we believe that they are not ideal. Even if they treat the same disease, no two drugs are exactly the same. Many new drugs will have improvements over the existing drugs at least on some margin. So new drugs not only steal market share from exiting drugs but also can expand the market size of the whole therapeutic group. In other words the sales volume of existing drugs is not a perfect proxy for the potential market size for new drugs. In previous papers we have hypothesized that if there are still patients suffering due to certain diseases, existing drugs are not doing a good job of treating all patients. Maybe some patients cannot be treated with the existing drugs because of side effects or they are not accessible by all patients due their high prices. In the extreme case where nobody is dying due to a disease, the implication is that existing drugs can treat all patients and potential market for the new drug is zero6.

5 See Garattini and Ghislandi (2007) for a short critique of the existing literature.

On the other extreme case where lots of people are dying due to a disease, it implies that existing drugs or other therapies are not effective and a good and an effective drug would have lots of potential patients (customers). Thus, in this study we use the “years lost” to disease as the proxy for potential market size for new drugs. Obviously many diseases do not kill people but reduce their quality of life. In earlier papers we used both mortality (potential years lost) and morbidity as the predictor for potential markets for new drugs. The results were very similar. In this study due to data limitations we use mortality (morbidity data is not available in the panel form for the countries we use in our analysis).7

Data and Methodology

Information on drug launches is obtained from IMS New Product Focus dataset maintained by consulting firm: IMS Health8. The dataset includes the trade name of the drug, active ingredients of the drug, the target disease, and launch date of the drug in each country for the most major drug markets in the world between 1982 and 2008.9 Our main variable of the interest is launch decisions of the drug companies. We check for each new chemical entity (NCE) whether it has been launched in each country and if it has how long it did take to launch the drug in the country since the first introduction of the drug anywhere in the world.10

The IMS New Product Focus dataset has the information about the target disease for each NCE. We want to measure the potential market demand for each NCE in each country. A measure of market demand is the prevalence of the specific disease treated by the NCE in each country. Our hypothesis is that if there is significant number of people dying at a relatively early age due to the disease, it means that existing drugs or alternative treatments do not work very well. Thus there is a high sale potential for the drug.

6 Of course new drugs can compete with the old ones on the price and obtain some market share

7 See Civan and Maloney (2006, 2009) for more thorough discussion

8 IMS Health is a consulting company which provides data to the investors, companies, policy makers and academics.

9 Unfortunately, we do not have data on China.

10 Some drugs have multiple active ingredients. In that respect we did not treat multi- ingredient drugs differently from single-ingredient drugs. For multiple ingredient drugs

there is only one new chemical entity introduced first time to the country with that specific drug.

The disease prevalence statistic that we use is the Potential Years of Life Lost (PYLL) obtained from OECD Health Data 2010 dataset. OECD Health Data 2010 dataset’s definition and method of calculation are given as follows:11

"Potential Years of Life Lost (PYLL) is a summary measure of premature mortality which provides an explicit way of weighting deaths occurring at younger ages, which are, a priori, preventable. The calculation of PYLL involves summing up deaths occurring at each age and multiplying this with the number of remaining years to live up to a selected age limit.

The limit of 70 years has been chosen for the calculations in OECD Health Data. In order to assure cross-country and trend comparison, the PYLL are standardized, for each country i and each year t as follows:


where a stands for age, l is the upper age limit chosen for the measure (70 years old in OECD Health Data), dat is the number of deaths at age a, pat refers to the number of persons aged a in country i at time t, Pa refers to the number of persons aged a in the reference population, and Pn refers to the total number of persons in the reference population."

Each NCE is a unique active ingredient. The data have information about the target diseases of each NCE. However matching the specific drugs with the potential years of life lost is very difficult. For many drugs the target disease is not unique. For example many drugs are used in the treatment of different types of cancers. Moreover the mortality statistics (potential years of life lost) might not be accurate at the cause of the death at the disease level. So we grouped the drugs and mortality statistics into broad health problem categories such as endocrine, nutritional and metabolic diseases, diseases of the blood and blood- forming organs, and malignant neoplasm, etc. There are 11 health problem categories. These are: Endocrine, nutritional and metabolic diseases, Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism, Diseases of the circulatory system, Diseases of the skin and subcutaneous tissue, Diseases of the genitourinary system, Certain infectious and parasitic diseases, Malignant neoplasm, Diseases of the musculoskeletal system and connective tissue, Diseases of the nervous system, Diseases of the respiratory system, Diseases of the eye and adnexa, ear. Table 1 also shows the number of NCEs that we were able to match to one of these health problem categories.

11 (accesses on 12/08/2010).

Per capita income, per capita health expenditures and population data are also obtained from OECD Health Data 2010 dataset at country, year level. At the health-problem, country, and year level per capita drug expenditure is also obtained from the same dataset.

One of our main explanatory variables is the intellectual property rights environment of the country. Our hypothesis is that drug companies would not launch their drugs in “unfriendly” countries where intellectual property rights of pharmaceutical innovator firms are not protected rigorously. There are many indexes which measure the intensity of general intellectual property protection of the countries. However the countries which are similar on general intellectual property protection level can have very different protection for pharmaceutical drugs. So we used a drug specific index. Liu and La Croix (2008) created Pharmaceutical Intellectual Property Protection (PIPP) Index and report its value for every five years from 1960 to 2005 for 154 countries. Their index consists of three parts: Pharmaceutical Patent Rent Appropriation (PPRA) index, Pharmaceutical Patent International Agreements (PPIA) index, and Pharmaceutical Patent Enforcement (PPE) index. PPRA measures whether a country has established six types of intellectual property rights in pharmaceuticals;12 PPE looks for seven statutory measures of enforcement;13 and PPIA controls for the whether the country has signed main international agreements on intellectual rights protection.14 Their index has values from 0 to 5. Higher numbers suggest better protection of pharmaceutical intellectual property rights.15

12 These are product patents, process patents, new medical indication patents, formulation patents, pediatric use patents and data exclusivity.

13 Preliminary injunctions, contributory infringement pleadings, burden-of-proof reversals,

national exhaustion, working requirements, compulsory licensing and revocation of patents for nonworking.

14 This is measured by whether the country is a signatory of Paris Convention, Patent

Cooperation Treaty

(PCT) of 1970 and the Trade-Related Aspects of Intellectual Property Rights (TRIPS) Agreement of 1995.

15 An alternative index for property rights protection is one developed by Ginarte, Juan C.,

Walter G. Park. 1997. “Determinants of Patent Rights: A Cross-National Study.” Research Policy 26(3): 283-301. We use this index as well in some specifications. The estimated effects are little changed. The G-P appears to build in more demand determinants and

Table 1. Summary Statistics




Std. Dev.









Delay Time






I.P. Protection












GDP per capita






Years Lost to Disease






Health Care Exp./GDP






Pharmaceutical Sales






Ref. Pricing Int’l






Ref. Pricing Th.






New Drugs in Class






Total Drugs in Class






Notes: Launch takes value of one when the drug is launched in a country. Delay Time is years from initial launch world wide to launch in a given country. Population in thousands. Years Lost takes account of the age at death and the life expectancy by country by ten different disease codes. Pharmaceutical sales are by disease codes. I.P. Enforcement is intellectual property enforcement for pharmaceuticals reported every five years starting in 1980. We extrapolate between years. Drug Class is the therapeutic grouping of each drug. New drugs are the number (other than the observation) approved in country j in the given year. Total is the sum of drugs approved in country j in the past. Pharmaceutical sales are by disease class. Ref. Pricing Int’l is regulatory pricing based on international reference; Ref. Pricing Th. is regulatory pricing based on domestic therapeutic references.

It is reasonable to wonder how this property-rights index is distributed across countries. We look at OECD countries and one might speculate that there is little variation in the property rights index across these countries. There has been a substantial improvement in property rights protection from 1980 to 2005. Although generally more advanced countries have higher index values, there is substantial variation in the index across countries at each end of the spectrum. For example US, Canada, UK, and UK can be considered as a comparable development levels their index numbers of 2005 are consecutively: 4.78, 3.09, 3.64, 3.30. Hungary on the other hand has the index value of 3.98. So property- rights index does not just correlate with the development level of the country and indeed measures some other aspects of the regulatory environment.

therefore is not a useful in disentangling these alternative effects in terms of the drug launch decision. Liu and La Croix state that their index values for many countries are substantially different that Ginarte-Park values and overall correlation coefficient for two indexes is 0.65.

Table 1 shows summary statistics for the variables that we will use in the analysis. The unit of observation is a drug, country, year. There are 1095 drugs in our sample, 31 countries, and 27 years of data for most countries (1982-2008). Observations begin for each country in the year that a drug is first introduced anywhere in the world. Observations end when a country launches the drug or at the end of the sample period, 2008.

For the drug-country pairs, for which we have data on 397,944, there are 13,315 drugs launches, or about 3 percent of the sample of drug|country|year observations. The delay time to launch averages about nine years but the median is around six. Total national healthcare expenditures vary from 1.6 percent of GDP (Turkey) to 16 percent (United States).

We include two variables in Table 1 that proxy for unobservable characteristics of the drug approval regime in each country. These are the number of drugs in a drug class that have been approved in the past and the number that are approved in a given year.16 The NCE data identify various categories for drugs: the 1095 drugs in the sample are placed into 221 categories. We count the drug launches by these categories. At the maximum there were 9 other new drugs in the same category launched in the same country in the same year. At the time of a drug launch, there were on average 2.9 existing drugs in the category.

The entire intrigue of our observed phenomenon would be deflated if drug- launch were just a block-buster effect. That is, if many or most drugs are launched everywhere because they are the best and the rest just pop up here and there, we don’t have an interesting story. To address this issue we construct Table 2 which shows the frequency of drugs launched across countries. The table shows that there is more of a dart-board effect than a block-buster one. Only a few drugs are launched everywhere; on the other hand, a substantial minority of drugs are launched in only one country.

It is somewhat striking that so many drugs are launched in only one country, and it does raise a concern that these drugs may be of little importance and their adoption is not driven by issues of intellectual property right protection. We investigate this by excluding those drugs only launched in one or two countries. We report these result along with other alternative specifications in Appendix.

We use a hazard model to capture the effects of regulation and market size on the launch decision for each drug in each country. The decision to launch is {0,1} for a drug, country, year. We model this as a probit function with delay time, intellectual property rights enforcement, and various other independent variables. Delay time is zero for the country that first launches the drug.

16 In the regression, when we use the variable for number approved in a given year, we subtract one so that it measures the number of other drugs in the class approved in that year.

Table 2. Frequency of Drugs Launched across Countries

Number of Countries Adopting

Number of Drugs































































Notes: Drugs launched in OECD countries totals 1066.

We want to explain the no-launch phenomenon. We do this by explaining the launch decision, and then by inference, the no-launch result. We use a probit specification to estimate the probability of launch. Launch may occur immediately or with delay. The model accounts for this. After some point if launch has not occurred, we use the model to declare a no-launch. This estimation method accounts for the cases where the drug is never launched in a country.

We choose ten years as the focus point; that is, we look at the probability of a drug launching or not by ten years after it is first launched somewhere in the world. The choice of ten years is arbitrary, and the results are not sensitive to this choice.17 This statement can be expressed in the following form using the probit regression estimates:


where Pnl10 is the probability that a drug is not launched for ten years, t is the delay

time from the first launch and x(t) are the other independent variables. The probit model estimates control for the factors that predict a drug launch. The probit specification allows for the launch to occur immediately or later, and allows for changes through time in the independent variables that affect this outcome. It gives an estimate of what causes drugs to get launched and then by probabilistic expression explains why drugs do not get launched.18

The independent variables in addition to delay time and property rights enforcement include population, years lost to disease in the disease category treated by the drug, health expenditures relative to GDP, the number of other drugs in that drug category launched in a country in prior years and in the given

17 Ten years is about the time that a drug will enjoy patent protection in the United States after it gains approval.

18 We considered and some readers have suggested that we use a selection-style model

with an equation for the launch decision and an equation for the delay. However, to be effective the selection model requires that the researcher has information on characteristics that separately affect launch and delay. In our case, it is hard to imagine variables that affect one and not the other. Moreover, the selection model is really a special case of the hazard model, which directly focuses on our point of interest. Consider the following problem. Launch can occur anytime after the chemical compound is ready for market. If we estimate a selection model of launch what demographic variables in time are we to use for a given country—those in place at the time the drug is first launched world-wide; those existing five years later; those ten years later; the average? Strong assumptions must be made in the selection-model approach while the hazard model naturally incorporates the time-evolution of the data in a meaningful way.

year, and the date of the world-wide launch. We include income per capita, pharmaceutical sales by health category, and indicators for reference pricing in alternative specifications. Population, years lost, health expenditures, and per capita incomes are all proxy measures of the size of the market for the drug. The number of similar drugs ever launched in a country is intended to pick up idiosyncratic characteristics of each country’s regulatory process. The date of first launch is included as a trend effect.

Results and Discussions

The main regressions are shown in Table 3, the estimated marginal effects are shown in Table 4, and assessment of the model accuracy is shown in Table 5. Five specifications are shown in Table 3. The standard errors are corrected for heteroskedasticity and clustered by country. Correlations between population, disease incidence, and health expenditures mean that not all variables can be included in each regression. The signs in the reported specifications are consistent with theory. Market size measured in different ways increases the probability of launch. Importantly, intellectual property protection is always positively and significantly related to launch. Delay has a negative coefficient in the probit estimates. This means that the probability of launch is highest in the beginning and then declines year by year. The cumulative probability of launch goes up, but at a decreasing rate.

The models presented in Table 3 estimate the probability of launch in one year. Our interest has a longer time frame, which is what we account for in Table 5. The control variable measuring the number of other drugs in the drug category that were approved in a country in a given year is significant and positive. The coefficient says that there seem to be waves of launch of the same sort of drugs; the likelihood of launch increases when other drugs of the same sort are launched. The total number of drugs launched in the past is negative and statistically significant in specification (1). The negative sign says that a new drug in a category is less likely to be launched; the more drugs of that sort are already on the market in a given country. These variables could be accounting for unobserved regulatory effects that vary across countries and we categorize them as unobservable market characteristics. Time of world-wide launch acts as a time fixed-effect and says that drug launches have been increasing over time. In the regressions shown in Table 3, we use disease incidence, population, health expenditures, and pharmaceutical sales as direct measures of market size. The coefficients are not always estimated with precision based on the clustered

standard errors. Even so the effects are consistent with theory, and taken together, the larger the market size, the higher the probability of launch. In those specifications where the regulatory proxies (number of other drugs approved by drug class) are excluded, the market size effects are stronger. It is possible that the number of drugs approved in each drug class is picking up market size as opposed to or in addition to regulatory idiosyncrasies. In this case it makes sense that omitting these variables would increase the significance of the other market sizes measures. It is worth noting that in specification (5), pharmaceutical sales is statistically significant. It measures the sales volume of drugs by disease class, so it varies by drug and country. Unfortunately, we have limited data coverage for this variable. Other variables of interest and other specifications are shown in Tables A1 and B2 in the appendix. Income appears to measure about the same thing as health expenditures and not as precisely. It has the wrong sign when included in addition to health expenditures (not shown). We also re-estimate specifications

(1) and (2) omitting those drugs that are only launched in one country (recall Table 2). The results shown in Table 3 are unaltered. These results are reported in Table A1.

Existence of reference pricing can influence the profitability and thus launch decisions. Reference pricing means that regulatory authorities peg the price that drug companies can charge based on the prices of similar drugs (therapeutic reference pricing) or on the price that the company charges for the drug in other countries (international reference pricing). Unfortunately we do not have a time series of these regulatory patterns but only an indicator of this for the most recent period. Moreover, maybe a more important issue for the drug companies is not whether the country has reference pricing system but whether the country is in the basket of reference countries for other countries with significant market potentials. It would be especially relevant if the drug prices in this country affect the prices in a bigger and profitable country (i.e. this country to be in the basket of reference countries in the bigger country) For example drug companies would be reluctant to launch their new drugs at Belgium at a cheap price if the prices in Belgium is used to determine the prices in Germany or France. Naturally drug companies would delay their launches in Belgium until their drugs are launched and established in France and Germany. Unfortunately keeping track of these baskets of reference countries is even more difficult. In any case reference pricing variables are not statistically significant and the inclusion of them does not change appreciably the magnitudes of the other variables.

Table 3.


Notes: Standard errors below coefficients are corrected for heteroskedasticity and clustered by country. I.P. protection, population, health expenditures and disease incidence in logs. Health expenditures divided by GDP. World-wide launch year is divided by 10. Total drugs in class is divided by 10. Significance levels: (a) 1 percent; (b) 5 percent; (c) 10 percent.

A major concern is that our model does not capture all of the dimensions of the demand for new drugs. This creates an omitted variable problem and one that may be correlated to the main variable of interest in our study: intellectual property right protection. It may be that countries with poorly measured demand for new drugs are also likely to be countries that have limited demand for IP protection. While we think that we have done a good job of measuring drug demand and market size, table A2 shows the probit regression results including fixed effects for country, country interacted with the time trend, and country interacted with time and with the property rights variable. There is basically no change from the results presented in Table 3. The estimated coefficient on the property rights variable is very similar to that in Table 3 when only country fixed effects are used. It is larger when country fixed effects are interacted with the time trend, and larger still on average when country fixed effects are interacted with time and with the property rights variable. In this last specification, there is a good deal of variation across countries in the magnitude of the estimated coefficient on the property rights variable. However, only four countries have an estimated effect of opposite sign to the average, and of these only one is statistically significant (Israel, at the 0.05 level).19 Overall, the results reported in Table 3 are robust to various specifications, and across countries and time.

Table 4 shows the magnitudes of the estimated effects. To do this, we calculate the probability of launch in the first ten years as discussed above. The first row of Table 4 gives the probability of launch within the first ten years across all countries for a drug with a world-wide launch date of 1999. Obviously, the model fits the data at the mean. For the average country, the probability of launch is slightly less than 40 percent. Row two shows the model evaluated for the variables measured at the U.S. values. The probability of launch is around twenty percentage points higher.

The following rows in Table 4 give the marginal effects of the variables. Here we ask how much does a one standard deviation increase in each variable from the world-wide average change the probability of launch in the first ten years. Intellectual property protection increases the probability of launch by around 7 percentage points for a one standard deviation increase. Population and disease incidence have important effects, though disease incidence is smaller; health expenditure is also substantial. The effect of IP protection is consistently estimated across the two specifications shown. Standard errors derived from bootstrap estimation are shown in parentheses. The marginal effects for health expenditures, population, and disease incidence are not estimated with precision. However their combined effects in the two specifications are statistically significant at the 5 percent level.

19 These coefficients include the estimated coefficient of the property rights variable standing alone (-0.14) plus the average value of the log of the time trend, which is 3.12, times the estimated coefficient of the property rights variable for each country.

Table 4. Probability of Drug Launching by Year 10

Independent Variables Evaluated




for year 1999



for US 1999 values



Marginal Individual Effects




I.P. Protection









Disease incidence




Health Exp. / GDP





Percent of Probability of Launch

Made up by Observable Effects



Notes: Individual effects are the increase in the probability of launch by year 10 for a one standard deviation increase in each independent variable holding others constant. Bootstrap standard errors in parentheses. Independent variables are evaluated for year 1999 by averaging from 1995-2004. Joint test of significance of diseased incidence and health expenditures and of population and health expenditures significant at the 5 percent level.

In the probit specification we can compare the impact of the observable characteristics (I.P. protection, population, disease incidence, and health expenditures) to the unobservable (time trend and number of other drugs in the drug category) by multiplying the estimated coefficients by the means of the independent variables. That is, let xi, i=1,k be the independent variables associated with observable market characteristics and i=k+1,n be the variables associated with unobservable characteristics. Then, is the proportion of the cumulative distribution of the probability of launch that is attributable to observable market characteristics, where the β’s are the estimated coefficients and the x’s are evaluated at their means.20 This calculation is reported in Table 4. The values are greater than 70 percent. This says that market size and market friendliness explain a large majority of the estimated launch/no-launch decision.


Finally, Table 5 gives an assessment of the goodness-of-fit for the model. We focus on specification (2) from Table 3 because it encompasses the largest number of observations. The other models give similar results. Goodness-of-fit is assessed in a two-by-two table of predictions versus outcomes. We look at the drug, country pairs. We calculate the probability of launch within the first ten years and classify the model as predicting launch if that probability is greater than 0.5; the prediction is no-launch otherwise.21 The observed outcomes are whether the drug is actually launched or not in a given country. There are 29895 observed drug, country events. Of these, 12189 are drug launches and 17706 are no- launches. The table shows the number of observations and the percentage of the total sample in each of the four cells.

Table 5. Model Accuracy


Events (29895 total)

Model Accuracy


Not Launched







No Launch










Notes: Events are drug, country pairs. Percent of total sample in parentheses below subsample sizes. Model Accuracy is based on predicted value of 0.5 for ten years from specification (2) from Table 3; that is, the model is said to be correct if the predicted value is greater than 0.5 and the event occurs. Model Accuracy is the percent of time the model predicts a launch and it happens (first row); or the percent of time the model predict no-launch and there is no launch (second row). The number of observed drug launches includes drug launched in 2009, and so differs from the number used in Table 3.

20 This is not a perfect measure because it is depend on the estimated constant term, but it is indicative of the relation between observable and unobservable characteristics. The value of delay is set to zero.

21 For simplicity of calculation we use the x values for each country in the year of world-

wide launch.

The results shown in Table 5 are fairly striking. The model is quite accurate in explaining the puzzle that we posed in the introduction. The model has 70 percent accuracy in explaining the number of drugs that are never launched across countries. That is, the model predicts no-launch in 24,731 cases (7531+17200) and is correct 17200 times. The model is even more accurate in correctly predicting launches. The major error of the model is in under-predicting the number of drugs that are launched. But the fact that the model is good at predicting the no-launch event is important. The model explains 70 percent of the no-launch decision and of this, 90+ percent is attributable to directly observable market-size and market- friendly conditions.


This paper investigates the conundrum of why new drugs are not launched everywhere in the world. On average new chemical entities are launched in less than half the countries of the world. This is odd; we imagine that drug companies would want to launch every drug everywhere. Our model explains a substantial part of this phenomenon.

We find that market size and property rights enforcement account for a large part of the reason why drugs are not launched everywhere. For instance, comparing the United States to the rest of the OECD countries on these margins, the probability of a drug launch in the first ten years is higher by 20 percentage points. Moreover, we find that intellectual property protection has a large impact. Increasing IP enforcement one standard deviation adds around 7 percentage points to the likelihood of launch. Our model explains 70 percent of the no-launch events, and over 70 percent of these are explained by observable characteristics of market size and intellectual property protection. The major policy implication of our research is that stronger intellectual property protection could significantly reduce the launch delays.

On the other hand we want to emphasize that, we have not analyzed the benefits of price controls and loose intellectual property rights protection for drug companies. They might create substantial benefits to the patients and government budgets through lower drug prices. In order to provide clear policy implications both costs and benefits of price controls and intellectual property rights protection have to be assessed.


Acemoglu, D., and Linn, J. (2004). Market Size in Innovation: Theory and Evidence from the Pharmaceutical Industry. Quarterly Journal of Economics, 119 (August), 1049-1090.

Becker, G. S., Philipson, T. J. & Soares, R. R. (2005). The quantity and quality of life and the evolution of world inequality. American Economic Review, 95 (1), 277–291.

Civan, A., and Maloney, M. (2006). The determinants of pharmaceutical research & development investments. Contributions to Economic Analysis & Policy, 5 (1)

Civan, A., and Maloney, M. (2009). The effect of price on pharmaceutical R&D. Contributions to Economic Analysis & Policy, 9 (1), 1-24.

Civan, A., and Koksal, B. (2010). The effect of newer drugs on health spending: Do they really increase the costs? Health Economics, 19 (5), 581–595.

Danzon, P., Wang, R., and Wang, L. (2005). The impact of price regulation on the launch delay of new drugs: evidence from 25 major markets in the1990s. Health Economics, 14 (3), 269–292.

Danzon, P., and Epstein, A. (2008). Effects of regulation on drug launch and pricing in interdependent markets. NBER Working Paper 14041.

Danzon, P., and Furukawa, M. (2008). International Prices and Availability Of Pharmaceuticals In 2005. Health Affairs, 27 (1), 221-233.

Demsetz, H. (1970). The Private Production of Public Goods. Journal of Law and Economics, 8 (2), 293-306.

Djankov, S., La Porta, R., Lopez-de-Silanes, F., and Shleifer, A. (2002). The Regulation of Entry. Quarterly Journal of Economics, 117 (1), 1–38.

Garattini, L., and Ghislandi, S. (2007). Should we really worry about "launch delays" of new drugs in OECD countries? The European Journal of Health Economics, 8 (1), 1-3.

Giaccotto, C., Santerre, R., & Vernon, J.A. (2005). Pricing and R&D Investment Behavior in the Pharmaceutical Industry. Journal of Law and Economics, 48 (1), 195-214.

Heuer, A., Mejer, M., & Neuhaus, J. (2007). The national regulation of pharmaceutical markets and the timing of newdrug launches in Europe. Kiel Advanced Studies Working paper 437. Kiel, Germany: Kiel Institute for the World Economy.

Kanavos, P., Ferrario, A., Vandoros, S., and Anderson, G. (2013). Higher US Branded Drug Prices And Spending Compared To Other Countries May Stem Partly From Quick Uptake Of New Drugs. Health Affairs, 32 (4), 753-761.

Kyle, M. (2007). Pharmaceutical price controls and entry strategies. Review of Economics and Statistics, 89 (1), 88-99.

Lanjouw, J.O. (2005). Patents, price controls and access to new drugs: How policy affects global market entry. NBER working paper No.11321.

Lichtenberg, F. (2001). Are the benefits of newer drugs worth their cost? Evidence from the 1996 MEPS. Health Affairs, 20 (5), 241-51.

Lichtenberg, F. (2005). The Impact of New Drug Launches on Longevity: Evidence from Longitudinal, Disease-Level Data from 52 Countries, 1982-2001. International Journal of Health Care Finance and Economics, 5 (1), 47-73.

Lichtenberg, F. (2006). The impact of increased utilization of HIV drugs on longevity and medical expenditure: An assessment based on aggregate U.S. time-series data. Expert Review of Pharmacoeconomics and Outcomes Research, 6 (4), 425-436.

Liu, M., and La Croix, S. (2008). A cross-country index of intellectual property rights in pharmaceutical Innovations. Unpublished manuscript

Murphy, K., and Topel, R. (2003). The economic value of medical research. In K. Murphy & R. Topel (Eds.), Measuring the Gains from Medical Research: An Economic Approach. The University of Chicago Press.

Oster, E., Shoulson, I., and Dorsey, R. (2012). Limited Life Expectancy, Human Capital and Health Investments: Evidence from Huntington Disease. NBER Working Paper No. 17931

Rosen, S. (1988). The value of changes in life expectancy. Journal of Risk and Uncertainty, 1(3), 285-304.

Soares, R.R. (2007). Health and the evolution of welfare across Brazilian municipalities. Journal of Development Economics, 84 (2), 590‐608.

Appendix A. Alternative Specifications

Table A1. Probit Estimates of Drug Launch–Alternative Specifications


Notes: Standard errors below coefficients are corrected for heteroskedasticity and clustered by country. I.P. protection, population, health expenditures and disease incidence in logs. Health expenditures divided by GDP. World-wide launch year and total drugs in class divided by 10. Significance levels: (a) 1 percent; (b) 5 percent; (c) 10 percent. Columns (1) and (2) omit drugs that have only been launched in one country.

Table A2. Probit Estimation including Fixed Effects

Probit Specification

Coefficient on IP


Country Fixed Effects


Country Fixed Effects interacted with Time Trend


Country Fixed Effects interacted with Time Trend and

Property Rights. Average over all Countries.


By Country











Czech Republic


































New Zealand








Slovak Republic










United Kingdom


United States


Notes: Country coefficient not estimated for Slovenia because of lack of observations.

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.