Valerie Y. Suslow
University of Michigan

A formal cartel organization sets up a multilateral relationship between firms in a given industry. While standard contracting literature has studied bilateral monopoly relationships in detail, there are horizontal relationships, like cartels, that also have contractual aspects. To shed light on how these contracts work, this paper takes a sample of legal formal cartel contracts and examines their structure and durability. Borrowing from industrial organization theories and transaction cost theories, the empirical model tests for the importance of demand uncertainty and cartel organizational characteristics in determining cartel contract duration. The results show that the more uncertain the environment within which the cartel operates, the shorter the expected cartel contract duration. Other industry structure characteristics and cartel organizational variables included in the empirical model have a much weaker influence on contract duration.

1. Introduction

In his 1985 book, The Economic Institutions of Capitalism, Oliver Williamson states: "Many...issues which at the outset appear to lack a contracting aspect turn out, upon scrutiny, to have an implicit contracting quality. (The cartel problem is an example.)" [p.17]. In other words, while standard contracting literature has intently studied bilateral monopoly relationships, there are multilateral relationships, like cartels, that have a contractual quality as well. In the same way that buyers and sellers must transact with each other in a repeated setting, firms in an industry striving to achieve a cooperative outcome also interact repeatedly. But there are important differences. These firms are in a horizontal, not a vertical, relationship. Cartel members do not transact with each other, but rather, they are connected in their effort to solve the "cartel problem" of intermittent price wars. In order to shed light on how these contracts work, this paper takes a sample of legal formal cartel contracts and examines their structure and durability. The data set assembled for this analysis covers international manufacturing and commodity cartels in 45 industries over the period 1920 to 1939.

Standard industrial organization theory sees cartels as infinitely lived and has focused on those factors affecting occasional price wars.1 Mainstream empirical literature on cooperative oligopoly has therefore concentrated on whether specific cartels have succeeded in charging a price greater than the competitive price, and if so, for how long.2 Any contracting theory of cartel arrangements would absorb many of the equilibrium results of this price war literature. For example, industry structure and demand uncertainty are seen as important determinants of the timing and frequency of breakdowns in pricing discipline. As the environment grows more volatile (or as the number of firms in the cartel grows) the cartel equilibrium point changes and the contract renewal process grows more complex. Also, members are more likely to make a misstep in judging production levels, perhaps triggering a price war, which may in turn terminate the contract.

In the transaction cost literature, Williamson stresses asset specificity, uncertainty, and frequency as the fundamental dimensions along which transactions differ. Although relation-specific investments and transaction frequency play little role in determining the duration of formal cartel organizations, the degree of uncertainty could well be significant. Forms of governance could also affect the longevity of the cartel organization.

Thus, the industrial organization literature and transaction cost literature suggest a list of variables which might influence contract duration. By application of a censored proportional hazards model, this paper examines the empirical relationship between cartel contract duration and aggregate economic uncertainty, while controlling for selected industry characteristics and organizational characteristics of the cartel.

The results show that the more uncertain the environment within which the cartel operates, the shorter the expected cartel contract duration. Both economic downturns and increased economic volatility have destabilizing effects on the cartel contract. Other industry structure characteristics and cartel organizational variables included in the empirical model have a much weaker influence on contract duration. In particular, the number of participants to the agreement has no strong influence on the durability of the contract.

2. The Data

The data set consists of observations on forty-five industries with cartel activity between 1920 and 1939. These manufacturing and commodity cartels operated on an international level. European firms were frequently the organizers and more powerful cartel members. Firms in the United States participated on occasion, by joining 1) illegally, 2) "legally" through the Webb-Pomerene Act, or 3) informally, by having an "understanding" with the organized cartel.3

The overt nature of these pre-WWII cartels sets this sample apart from most previous cross-section samples of cooperative agreements. The major European countries had no systematic antitrust legislation until the mid-1930's. Those few laws that were passed (by Great Britain, for example) took a very lenient stance towards cartel agreements.4 Thus, most of these cartels were explicit contractual arrangements (with the occasional exception of U.S. participation) organized in countries which tolerated horizontal cooperation.

Table 2 of the Appendix gives the citations for the data. The data were collected from a variety of secondary sources: books and articles written between the late 1920's and mid-1940's by economists and political scientists, U.S. Congressional hearings, U.S. Tariff Commission Reports, and League of Nations sponsored monographs. Books by Hexner [1945], Stocking and Watkins [1946], and Elliott, et. al. [1946] were used to build the foundation of the data set. Many earlier sources were used to verify and fill in information as needed.

3. Cartel Contract Duration

The duration of a cartel contract is defined by specifying the beginning and ending dates. Cartel contracts begin either on the date the cooperative agreement was signed by participating firms, or the effective date of the agreement when that information was available. Although the sample runs from January 1920 to September 1939, a given cartel contract could begin at any time within the sample period.

A cartel contract ends either when the cartel dissolves endogenously (through the defection of an important member, for example) or because of an exogenous event. It is common when working with economic data sets that some of the observed durations will be incomplete. The outbreak of World War II is the primary reason for a premature ending of a cartel in this sample. These observations are right-censored, and can be accommodated in standard duration analysis.

Determining the approximate ending dates for non-censored contracts is a formidable problem. The question is whether to use the actual contract termination date, as specified in the original contract or subsequent renewal, or to use price and cost data to measure the economic dissolution of the cartel. The first option clearly ignores information. The second option simply is not feasible for this particular sample, nor does it directly address the question of contract duration.

A compromise approach that is feasible is to date the approximate end of an alliance by linking it to some specific event affecting the firms or the market. The process used for determining the ending dates thus relies heavily on a careful reading of statements made by the secondary source documents cited above and listed in the Appendix, Table 2. Cartel contract durations in this study are often measured as shorter than the originally stated contract length. Only for one cartel agreement is it the case that there is no information on the date of termination of the cartel other than the initially stated contract period. In this way the other observables, such as the choice of cartel organization, can be linked to a more meaningful measure of the lifespan of the formal organization.

The data therefore consist of a contract length together with information on whether the contract duration was censored. These contract durations serve as the dependent variable. For censored observations the observed random variable is the contract duration at the time of censoring. Table 1 of the Appendix lists some of the information on the cartels in the sample: the industry or product(s) involved, approximate dates of cartel operation, identities of participating countries, and the apparent reason for the end of the contractual relationship.

4. Empirical Results

The probability that a cartel contract ends this month, conditional on the cartel being intact last month is known as the hazard function. The empirical results summarized here are based on a proportional hazards model where the hazard rate is modelled as a function of time and explanatory variables.

Most industrial organization theories and theories of contracting would predict first and foremost that the larger the number of participants, the harder the organizational problem. For this sample of international cartels the sign is as predicted, but the magnitude of the effect is much less than anticipated.

There are a few organizational variables that are consistently significantly different from zero. For example, cartels which have formal stipulations for enforcement mechanisms have a lower probability of failure. Cartels built around patent or cross-licensing agreements also have a reduced probability of failure, due most likely to the substantially limited threat of entry. (It may also be that we are seeing the effect of a permanent cross-licensing contract although the price-fixing conspiracy itself may not be.) There are also indications that that agreements attempting to cover a wide variety of products have a lower expected duration. The enforcement problems undoubtedly multiply rapidly as the number of products regulated by the contract increases.

Regarding the cyclical indicators, the empirical evidence indicates an inverse relationship between cartel contract duration and low levels of demand or downward movements in economic activity near the ending date of the cartel. In addition, the data support the hypothesis that both positive and negative "surprises" matter. That is, uncertainty in any direction affects the stable workings of a cartel organization. It is also possible that upswings in economic activity make cooperation (and formal cartel contracts) unnecessary, yielding an inverse relationship between positive volatility and contract duration.5

Given the inductive nature of this paper it may be best to turn back to the facts to shed some light on this result. Using the NBER business cycle reference dates for either the U.S., the U.K., or France, roughly sixty percent of the uncensored cartel contracts in this sample ended during a peak-to-trough period. The majority of these ended during the Great Depression. We know that a severe downturn will affect cartel contract duration. And yet, some of the agreements that ended during the early 1930's had survived an earlier downturn. There is no clear cut case.

5. Discussion of Results

The goals of the empirical analysis in this paper are twofold. First, it is a useful exercise to learn more about the structure and organization of cartels as an input to industrial organization theory. Second, an empirical model of the determinants of contract duration can shed light, indirectly, on those factors affecting the costs and benefits of organizing a formal cooperative agreement.

There are two classes of variables of interest, those describing economic uncertainty and those describing the organization of the contract. The empirical results presented lend support to the hypothesis that the more uncertain the environment, the more likely the cartel contract will end. The organizational variables explain much less of the variance in contract duration. While the number of products covered by the agreement and the use of penalties to enforce the agreement seem to matter, other structural characteristics, including the number of participants to the contract, have little effect on the length of formal cartel agreements. The most important contract provision in determining contract length is also the most enforceable, i.e. whether there is a cross-licensing or patent agreement as part of the cooperative contract.

These cartel contracts are conspiracies against the social good. If policymakers want to spur on their demise, the empirical results above imply either that there are no policy instruments or that the only choices high cost. A highly volatile economy will cause cartels to collapse, but minor remedies to industry structure will not. Laws against price fixing may be the most reasonable approach to society's "cartel problem."

References, appendixes, tables and footnotes available at the session.