اختلاف نظر در چانه زنی: تجزیه و تحلیل تجربی از سازمان اوپک

We consider a stylised model in which two cartel members bargain over the aggregate-production quota in a world of asymmetric information. We show that when the two cartel members are sufficiently different, the probability of agreement depends on both the current state of demand and initial production. Specifically, the probability of agreement is much lower when demand is low (and initial production is relatively high) than when demand is high (and initial production is relatively low). We also find that, regardless of the current demand state, the more extreme is initial production, the higher is the probability of agreement. Using an event study, where we take as events OPEC production quota announcements, we demonstrate empirically that the predictions of the model are borne out.

In this paper, we study whether agreements are easier to achieve when times are good or when times are bad. In the industrial organisation literature, this is an issue that has received careful attention starting with the work of Rotemberg and Saloner (1986) who, more precisely, ask when is it easier for firms to collude. In a world of i.i.d. demand shocks, Rotemberg and Saloner show that collusion is most difficult when demand is high. Subsequent work has provided conditions for which it is hardest to collude during recessions (see e.g., Athey, Bagwell and Sanchirico – henceforth ABS – (2004), Bagwell and Staiger (1997), Haltiwanger and Harrington (1991) and Staiger and Wolak (1992)). There is also some debate at the empirical level as to whether collusion is easier or harder to sustain during good times. Scherer and Ross (1990, Ch. 8) have argued that collusion is more difficult to sustain during recessions. Wilson and Reynolds (2005) provide empirical evidence consistent with the view that successful collusion is more difficult during booms, though they caution that other macroeconomic factors may be at play which are not captured by their oligopoly model (see p. 165).1 We contribute to this debate by providing a model in which the probability of agreement depends on the current state of nature, and by providing conditions under which agreements are more likely in good times than in bad times. Our model eschews the traditional oligopoly models in favour of a bargaining approach. This is because our interest lies in examining the behaviour of the Organization of Petroleum Exporting Countries (OPEC) and we feel that the bargaining problem its members confront at each meeting cannot be ignored. OPEC is obviously a cartel that restricts output in order to obtain super-competitive profits and must be concerned with the incentives each of its members has to overproduce. Given the many folk theorems present in the literature, there is not a unique way to split the gains from cartelisation of the oil market, and the problem of splitting these gains involves a great deal of closed-door negotiations. At times individual members' posturing for market share leads to extended periods of inaction, causing lower (though still super-competitive) profits for all. Before discussing the precise details of our bargaining model, we first motivate the necessary departure from the standard oligopoly models of collusion. Consider ABS (2004) who study optimal collusive behaviour when firms' marginal costs are subject to random shocks.2 While they show that collusion is more difficult to sustain during bad times, their model makes other predictions which are not borne out in our data. In particular, market share instability is a key feature of ABS (and other models of collusion), since low-cost firms are allowed to undercut the monopoly price without fear of punishment in the low-demand state. In contrast, at least since the early 1980s, OPEC has been playing a quantity game, setting shares of an aggregate quota and allowing prices to fluctuate. If one looks at Fig. 1, which displays each member's share of the aggregate quota over time, it appears that the shares of most countries have been relatively stable over time. There are a couple of exceptions but we feel that they have little to do with OPEC. For example, during the first Persian Gulf War, Saudi Arabia's share of the aggregate quota increased dramatically, while Iraq's and Kuwait's shares dropped to zero. Following the war, Kuwait's share recovered to its pre-war level, while Iraq's share has fluctuated widely for obvious political reasons. To a lesser extent Indonesia's share has declined, while Qatar and Venezuela have seen their shares increase over time. For Indonesia, this is due mainly to declining reserves, while the increased prominence of Qatar and Venezuela can partly be explained by the drop in Iraq's production after 1998. Hence, our point is not that OPEC shares have not changed, but that they are much more stable than received theory would predict.3 Full-size image (70 K) Fig. 1. OPEC-member quota shares. Figure options Another important stylised fact which motivates our bargaining approach concerns the size of shocks and the probability of agreement. In Section 3.4, we show that for large shocks (positive or negative) OPEC is more likely to reach agreement than for small shocks. In the model of Staiger and Wolak (1992), for small negative shocks, the maximal level of collusion decreases in a continuous manner, while for large negative shocks, collusion breaks down and players employ mixed strategies in the quantity-setting subgame. As with ABS, market shares are unstable but, beyond that, large shocks lead to unsuccessful collusion. In our bargaining model, the presence of private information creates a wedge between the interests of the proposer and the responder. Importantly, the size of this wedge is independent of the size of the demand shock. Therefore, for a larger demand shock, the private information is relatively less burdensome, leading to an increased likelihood of agreement. We are not the first to discuss collusion and cartels in a bargaining framework. For example, Ray and Vohra (1997) use their model to characterise stable cartels, while Seidmann and Winter (1998) provide a brief discussion of cartels in the context of gradual coalition formation. In papers more closely related to ours, Cramton and Palfrey, 1990 and Cramton and Palfrey, 1995 discuss cartel formation through the lens of mechanism design. The connection to bargaining is particularly strong in the latter paper where the authors include an explicit ratification stage. Finally, and perhaps most valuable to us, there is a survey by Levenstein and Suslow (2002) on what determines cartel success. In it, they argue that, “[b]argaining problems were much more likely to undermine collusion than was secret cheating” (p. 16). In addition they state that, “bargaining issues may arise as a result of a decline in demand” (p. 18). Scott Morton (1997) provides further evidence along these lines in her study of British shipping cartels arguing that, “It was much easier for a [cartel] to allocate six sailings a year to an entrant if the original members could keep their current schedules… Hence, increasing trade on a route made negotiating entry easier” (p. 702). In Section 2 we provide our model of a quota-setting cartel and develop some testable implications, which we then take to our empirical study of OPEC. Our model completely abstracts from the problem of sustaining collusion, assuming that agreements between agents are binding and enforceable. In the model's static incarnation, there are two cartel members who negotiate over the aggregate quota, leaving individual shares fixed. The basic ingredients of the model are as follows: given demand and initial production, the proposer (firm 1) offers a potentially different aggregate quota, which is either accepted or rejected by the responder (firm 2). If the proposal is accepted, the new production level is implemented, while if it is rejected, the status quo obtains. In order to introduce the possibility of disagreement, we assume that there is some uncertainty over the respondent's ideal point. This uncertainty may arise, for example, due to randomness in its marginal cost of production, or other economic or political considerations that influence the respondent's ideal point.4 Firms are also asymmetric in the sense that one of them, on average, has a higher ideal point than the other. We provide conditions under which disagreements are more likely when initial production is high relative to demand (and so should be reduced) than when initial production is low relative to demand (and so should be increased). We also show that the more extreme is initial production (either too high or too low), the more likely is an agreement to be reached. Of course, the real-world in which OPEC operates is not static — demand fluctuates over time, as do costs and other economic and political variables. In Appendix A we show that the model's two main predictions remain intact in a dynamic world with Markovian demand shocks. In Section 3 we turn to our empirical study of OPEC. Our approach is at first indirect. In particular, we conduct an event study in order to determine how oil prices react to various OPEC production quota announcements. As is standard with this kind of analysis, we measure the impact of announcements by examining the pattern of abnormal returns in the days surrounding each event. More specifically, we divide OPEC announcements into three classes: (1) increases in the aggregate quota, (2) decreases in the aggregate quota and (3) no change in the aggregate quota (or status quo). We then examine how the price of oil responds to each announcement type and argue that the results imply that OPEC members find it more difficult to agree during bad times than during good times.5 In analysing how oil prices react to the three kinds of announcements, Section 3.3 presents three stylised facts. First, we find no evidence of abnormal returns, either positive or negative, when OPEC announces an increase in the aggregate quota. Second, we find significantly positive abnormal returns following an announced reduction in the aggregate quota. Third, we find significantly negative abnormal returns following status quo announcements. This suggests that the market is surprised by status quo announcements and aggregate quota reductions, but not by announcements of increases in the aggregate quota. All three stylised facts hold for both the spot and two-month forward price of oil. Our take on this is as follows: status quo announcements represent a failure to reduce the quota during bad times. That is, during bad times, the market holds intermediate beliefs and so when the aggregate quota is reduced, a positive reaction occurs, while when no change is announced a negative reaction occurs. In contrast, during good times, OPEC always increases the aggregate quota and so the market fully expects this, leading to no reaction to such announcements. The remainder of Section 3 takes pains to show that the initial results and our interpretation of them are robust. In particular, we explicitly correlate OPEC announcements with variables pointing to strength or weakness in the oil market (specifically inventory growth by western nations and residual demand growth). Beyond merely reinforcing our earlier interpretation, these results support the model's second main prediction: the larger is the size of the shock, the more likely is OPEC to reach an agreement. Finally, in Section 4 we provide some concluding remarks.

In this paper we have presented the results of an event study analysis of OPEC behaviour and its effects on both oil prices and stock returns in the oil industry. The results are quite clear. When OPEC reduces the aggregate quota, positive and often significant abnormal returns accrue to oil prices and in the three sub-sectors considered. Next, when the aggregate quota is increased, there is no significant pattern of abnormal returns. Finally, when OPEC takes no action negative and significant abnormal return accrue across all sectors. Moreover, these abnormal returns appear to be economically meaningful, accumulating in some cases to ± 5% by the end of the event window. We believe strongly that these results offer us an interesting insight into OPEC's decision-making behaviour. Specifically, the results indicate an asymmetric ability of OPEC to secure agreements. That is, during good times, OPEC is far more likely to increase the aggregate quota than they are to decrease it during bad times. This interpretation is given further credence when we correlate the outcome of OPEC meetings with changes in residual demand for oil and changes in inventories. In addition, that analysis also showed that OPEC is more likely to agree the larger is the shock (either positive or negative). Finally, our interpretation is further supported when one examines the news reports surrounding these status quo announcements. As we argued in the Introduction, the standard oligopoly models must be cast aside. Therefore, to explain our results we adopt a bargaining approach and develop a simple model of bargaining with one-sided private information that neatly captures our empirical results. In the model, the private information creates a wedge between the interests of the proposer and the responder. Disagreement then occurs for two distinct reasons: symmetric uncertainty, as captured by the parameter α, and asymmetric uncertainty, as captured by the parameter β. The former simply captures the idea that cartel members face shocks to their cost of production or to other economic and political factors which affect their ideal quantity, while the latter introduces an asymmetry between the two cartel members such that one prefers, on average, higher production than the other. When β > 0, disagreements are much more likely during periods of low demand than during periods of high demand.