The second peculiarity of the entry process is the existence of a matching stage where upstream entrants and downstream entrants form vertical relationships and decide on API trade. While in principle such a stage can be built into the econometric model, the incremental benefit from doing so is unlikely to compensate for the computational difficulty that it entails. Therefore, I assume that matching in the API market takes the following simplified form: every upstream entrant, including the upstream plants of vertically integrated entrants, is matched with every downstream entrant including units belonging to vertically integrated firms. In other words, every downstream unit is entitled – from a regulatory point of view – to use the API produced by any of the upstream units. Once we ignore patent challenges and allow every upstream entrant to be matched with every downstream entrant, the generic entry process can be characterized as a simple simultaneous-move game with discrete actions. Potential entrants choose their actions and receive payoffs according to the oligopoly game played out in the resulting vertical market structure. I do not explicitly model how firms trade the intermediate good and compete against each other in the post-entry vertical oligopoly; the effects of such interactions among firms are summarized into reduced-form payoff equations that have the actions of rival potential entrants as arguments. For a firm that is a potential entrant in the downstream segment but not in the upstream segment, the action set is {Not Enter, Unintegrated Downstream Entry}. Similarly, a firm that is a potential upstream entrant but not a potential downstream entrant has the action set {Not Enter, Unintegrated Upstream Entry}. Finally, grow vertical the choice set of a firm that is a potential entrant in both segments is {Not Enter, Unintegrated Downstream Entry, Unintegrated Upstream Entry, Vertically Integrated Entry}.
It is assumed that only pure strategy Nash equilibria of the entry game are played. Therefore, market structure outcomes that are not pure strategy Nash equilibria – for example, one unintegrated downstream entry and no other entries, or one independent upstream entrant and no other entrants – are ruled out. Elberfeld shows that vertical entry games are characterized by multiple equilibria even if entry decisions in one vertical segment are made prior to those in the other segment. This implies that the existence of multiple equilibria is an unavoidable aspect of simultaneous-move vertical entry games. Before moving on to the empirical analysis, let us briefly consider the motives for, and effects of, vertical integration in the generics industry. A former executive at Sandoz, one of the largest firms, mentions lower API costs, earlier access to APIs, and stability of supply as the advantages of vertical integration . Others have mentioned the possibility that vertical integration allows better control over the information flow between segments as well as better risk-sharing , which would presumably lead to higher levels of productive investment. These point to the existence of efficiency effects generated through vertical integration. Such effects are likely to benefit final consumers through lower prices, more reliable supply of drugs, or higher product quality. On the other hand, recent antitrust cases suggest that vertical integration can generate anticompetitive foreclosure effects. In FTC v. Mylan et al. , the Federal Trade Commission claimed that an exclusive dealing contract, signed between a finished formulation manufacturer and its upstream supplier, regarding the APIs for lorazepam and clorazepate tablets contributed to price increases of between 1,900 and 3,200 percent for the downstream product.7 Unlike exclusive dealing contracts, vertically integrated entry in the generic drug industry is not subject to antitrust scrutiny.
As a result, such anecdotal evidence on foreclosure effects is not readily available in the case of vertical integration. There is, however, no reason to assume that vertical integration cannot have similar anticompetitive effects.8 Estimation of the model is based on a maximum likelihood framework. At each iteration of the parameter search, I calculate the probability that the market structure observed in each of the sample markets is an equilibrium of the entry game. Loosely speaking, the likelihood contribution of a market is calculated by integrating over the region of the error term vector where the observed market structure is predicted as an equilibrium. Thus, the first tool we need is a mapping from the error term vector to equilibrium market structures. I implement the mapping by programming an equilibrium finding algorithm. The error term vector in my model has fairly high dimensionality and the region of integration is expected to have a complex shape. As a result, we can expect no closed-form solution to exist for the market structure probability. I therefore approximate the integral using a simulation-based method in which draws of the error term vector are taken . For each draw, the equilibrium finding algorithm is run to see if the market structure observed in the data is predicted as an equilibrium . The market structure probability is approximated by calculating the proportion of the draws for which the observed market structure is an equilibrium. In general, vertical entry games are characterized by multiple equilibria . Therefore, the equilibrium finding algorithm often indicates more than one market structure as a possible outcome of the entry game in a given market. When such equilibrium multiplicity occurs, an additional assumption is required in order to assign a unique value to the probability that the observed market structure is generated by the model.18 One possibility is to assume that the equilibrium with the highest joint profits is always realized.. Alternatively, one can specify an equilibrium selection function whose arguments consist of the characteristics of each equilibrium, such as whether or not it is Pareto dominant . Following Bjorn and Vuong , I employ the simpler rule that when there are multiple pure strategy equilibria, each market structure has the same probability of occurring. The entry game for each generic drug market involves a large number of heterogeneous players, which makes the equilibrium finding algorithm time-consuming. Even if one has a fast algorithm, the large number of evaluations that are required during estimation suggests the need to economize on the number of runs of the algorithm. To this end, I employ the method of importance sampling with change-of-variables . The advantage of this method is that the equilibrium finding algorithm needs to be run only once during estimation. In the remainder of this section, I describe the equilibrium finding algorithm, the simulation-based approximation of the market structure probability, and the importance sampling with change-of-variables method.As mentioned at the beginning of Section 2.5, there are 128 generic drug markets that satisfy the following criteria: the market opened up to generic competition during 1993-2005; the downstream product is the first one, among all single-ingredient products using the same API, to become generic; the downstream product is an oral solid, injectable, or topical formulation; and data on market characteristics are available for the product. These are the markets where we are likely to see upstream and downstream entry decisions being made at around the same time. 43 of the 128 markets are subject to a patent challenge by one or more of the generic entrants. These are identified by the Food and Drug Administration’s list of drug markets where one or more Abbreviated New Drug Applications containing a paragraph IV certification have been filed. As discussed in Section 3.3, the existence of a patent challenge changes the market structure formation process from a simultaneous-move entry game to a race to be the first-to- file entrant. Meanwhile, my econometric model is only designed to estimate the parameters of a simultaneous entry game. Therefore, the 43 markets with paragraph IV certification are dropped from the analysis. This leaves a sample of 85 markets that are not subject to patent challenge by any of the entrants. The exclusion of paragraph IV markets raises concerns of sample selection.
If the incidence of paragraph IV certification is correlated with any of the error terms of the model, vertical grow systems the removal of paragraph IV markets from the sample may lead to biased estimates. While acknowledging the importance of such concerns, the estimation conducted in this chapter does not take them into account. The main reason is the difficulty of incorporating selection into the econometric model. As in Chapter 2, sample selection can be modeled by specifying a dichotomous choice process for the determination of paragraph IV status at the market level, and allowing the error term in the paragraph IV equation to be correlated with the remaining error terms of the model.27 In practice,however, I find that the parameter estimates fail to converge when joint estimation of the vertical entry model and the paragraph IV equation is attempted.28 Another factor that may allow us to ignore the sample selection problem is that, according to the results in Chapter 2, the error term of the paragraph IV equation is not likely to be strongly correlated with the error terms of the firm-level payoff equations.29 The definition of downstream and upstream market entry follows that in Chapter 2. Downstream entry into market m is observed when a potential entrant’s ANDA for that market is approved by the FDA. Upstream entry is observed when the potential entrant’s submission of a DMF for that market is publicized by the FDA. Vertical entry occurs when the potential entrant receives ANDA approval and submits a DMF in the same market. The definition of potential entrant status also follows from Chapter 2. A firm is a potential downstream entrant of market m if its previous entry into the downstream segment of another market was less than five years before the market opening date of market m. 30 Similarly, a firm is a potential upstream entrant if its previous upstream entry in another market was not more than seven years before the market opening date. If a firm is both a potential downstream entrant and a potential upstream entrant, it is considered to be a potential vertically integrated entrant. Table 3.1 shows summary statistics for the number of potential entrants and the number of actual entrants in each market. The number of potential upstream entrants is greater than the number of downstream entrants. On average, there are 75.859 potential upstream entrants in a market while the average number of potential downstream entrants is 30.635. 18.929 of those entrants, on average, are potential vertically integrated entrants. The actual number of entrants is much smaller: an average market structure contains 3.365 unintegrated downstream entrants, 3.565 unintegrated upstream entrants, and 0.647 vertically integrated entrants. The estimated coefficients on the two continuous market characteristics, User Population and Per-User Expenditure, are largely within expectations. User Population has a significantly positive coefficient in the unintegrated downstream payoff equation, while in the other equations its coeffi- cient is not significantly different from zero. The Per-User Expenditure variable is not significant in the unintegrated downstream and vertically integrated payoff equations, but it has a significantly positive coefficient in the unintegrated upstream payoff equation. These results support the notion that larger market size and higher willingness-to-pay raise firms’ entry incentives. The dichotomous market characteristic variables have contrasting effects in the three equations. The Gastrointestinal/Endocrine-Metabolic dummy variable has a significantly positive impact on unintegrated downstream payoffs but its effect on vertically integrated payoffs is significantly negative. This conforms to the finding in Chapter 2 that vertical integration probabilities are lower in markets belonging to the gastrointestinal and endocrine-metabolic classes. It may be because the tighter control over upstream manufacturing processes afforded by vertical integration is less important for such drugs than for other drugs. Another finding that agrees with the results in Chapter 2 is that the Injectable dummy variable has a significantly negative coefficient in the unintegrated downstream and upstream equations, while its coefficient in the vertically integrated equation is significantly positive. This confirms the intuition that tighter manufacturing controls through vertical integration are more important for injectables than for other dosage forms. The Post-2000 dummy variable follows the same pattern as the Injectable dummy: it is significantly negative in the unintegrated downstream and upstream equations while being significantly positive in the vertically integrated payoff equation. One possible explanation comes from the finding in Chapter 2 that vertically integrated entry became more common after the year 2000 .