Fight, fold or settle?: Modelling the reaction to FTC merger challenges.

Coate, Malcolm B. ; Kleit, Andrew N. ; Bustamante, Rene 等

I. INTRODUCTION

Scholars studying law and economics have identified four stages through which every legal dispute moves. The first stage involves an alleged injury which leads to the legal dispute. Second, the injured party decides whether to advance a claim. Once a claim is advanced, the plaintiff and defendant may negotiate to settle the claim. If the negotiation fails, the dispute moves to litigation. By subdividing the legal process into these four stages, it is possible to study the interaction at any one stage in greater detail.

This paper models the third stage of this process in an antitrust setting, the interaction between the Federal Trade Commission (FTC) and private parties interested in consummating horizontal mergers that may adversely affect competition. The FTC must decide whether to advance a claim against the merger and seek to block the transaction. If such a claim is made, the merging parties may "fight," forcing a resolution of the dispute through litigation by moving forward with the transaction. They may also "fold," abandoning the transaction, or "settle," entering into an agreement with the government to resolve the issues underlying the merger challenge. This decision to fight, fold or settle is somewhat more complicated than the standard settle or litigate question because the prospective nature of the alleged competitive injury gives the defendant an opportunity to abandon the transaction before the injury occurs.

Our basic hypotheses with respect to this process are that firms' decisions depend not only on the competitive merits of the FTC's case, but, perhaps more importantly, on the political, conglomerate and financial nature of the relevant transaction and how the nature of the case fits into the merger review process. Section II presents the background for the analysis by discussing the institutional structure of the FTC, explaining the merger review process, and then presenting an overview of a firm's options when faced with a decision by the FTC to block a merger. Section III models the decision faced by firms, defines the equations to be estimated, and surveys the data set of FTC merger challenges. A multinomial logit model of a firm's fight, fold or settle decision is estimated in section IV, followed by an interpretation of the results and an application of the model using both historical data and information from mergers proposed during fiscal 1992 and 1993. We find that the level of the competitive overlap associated with the relevant transaction is an important determinant of firms' fight, fold or settle decisions. In particular, if the competitive overlap involves a small percentage of the transaction, firms generally prefer to reach a negotiated settlement with the Commission. In such cases, the settlement decision appears unaffected by the underlying merits of the FTC's legal position. In other cases, the merits of the case may play a role in the firm's litigation decision. We also find that the potential efficiencies offered by the merger affect the firms' decision-making process, with firms more likely to fight and less likely to settle if the merger is expected to generate efficiencies.

II. ISSUES IN MERGER ENFORCEMENT

Background on the Federal Trade Commission

The FTC is a government agency charged, along with the Antitrust Division of the Department of Justice, with enforcing the antitrust laws. Much of the antitrust casework involves the evaluation of the competitive effects of proposed horizontal mergers.(1) To interdict a proposed merger, the FTC must obtain a preliminary injunction from a federal district court, with possible review from the relevant court of appeals. If the court declines to issue the injunction, the firms are free to merge, though the FTC can initiate a lengthy court process to attempt to obtain a permanent injunction against the merger.

Commission decisions are made by a majority vote of the five Commissioners, each of whom are appointed by the President and confirmed by the Senate for terms of up to seven years. In each matter, the Commissioners usually receive separate memoranda from both the staff and senior management of both the Bureau of Competition (the lawyers' bureau) and Bureau of Economics (the economists' bureau) to assist in their decision-making process. The parties also have an opportunity to present their positions to the Commissioners through submissions and meetings.

Overview of the Merger Review Process

A legal dispute over a horizontal merger fits into the standard law and economics framework, although the prospective nature of the competitive injury requires a slight modification of the model. Firms propose mergers based on their respective interests. A small fraction of proposed mergers, if consummated, may threaten competition and present a risk of antitrust injury. If the firms recognize the potential injury, they have the option to abandon the transaction. Parties, however, may have a difficult time a priori knowing how the FTC will react to their proposed transaction. (See Johnson and Parkman [1991].) Little is generally known about the decision of firms to advance a potentially anticompetitive merger because of the secrecy of the merger negotiation process.

If the firms choose to proceed, the Hart-Scott-Rodino Act would, in most cases, require them to notify the government of the proposed transaction and observe a thirty-day waiting period.(2) If the merger raises competitive concerns, the government can issue a request for additional information and further delay the process. At the end of the investigation, the government will either challenge the transaction or close the investigation. Coate, Higgins and McChesney [1990] have found that the FTC's decision to move against a merger depends on both political considerations and case-specific facts. The Commission is more likely to challenge the transaction if the Herfindahl index is over the Merger Guidelines' threshold of 1800, if barriers to entry exist, or if other factors make the market conducive to anti-competitive behavior.(3) Data generated by both the Bureau of Competition and Bureau of Economics are statistically significant in explaining the FTC's decision.

If the government decides to challenge the transaction, the parties have three choices.(4) They can "fight" by moving forward and forcing the government to obtain a court order to block the transaction, can "fold" and abandon the transaction, or "settle" by entering into a consent agreement with the government to spin off assets to eliminate the competitive concern.

The final stage in the horizontal merger enforcement process is litigation. Recent studies by Coate [1992] and Kleit and Coate [1993] of federal judicial merger decisions made after the issuance of the 1982 Merger Guidelines have found that to prevail in a merger case, the government must show high concentration in the market, as measured by the relevant Herfindahl statistic, and barriers to entry in the relevant market. Even in these cases, if the Herfindahl is not well over 1800, evidence on structural factors conducive to competition can rebut the presumption from a high Herfindahl. Thus, it appears that the merits of a case have a significant impact on the outcome of litigation.

In private litigation, the four stages of the process would be interdependent (Perloff and Rubinfeld [1988]). Here, the interdependence hinges on the ability of the FTC to exploit the peculiarities of the legal process that may lead firms not to fight the FTC, regardless of the merits of the case. For example, litigation against the FTC may force the firm to incur opportunity costs if the firm forgoes a settlement that would bring the process to a quick resolution. In effect, aspects of the transaction unrelated to the competitive concerns can be held "hostage" by the FTC. This situation would not occur in a textbook example of an anticompetitive merger involving two firms competing to sell a product in a highly concentrated market, because that scenario leaves no assets to hold hostage. In the modern world of conglomerate firms, however, a merger may involve both competitive overlaps and wide areas where the firms are not related at all. In such acquisitions, the firms would desire to consummate the noncontested portion of the merger quickly and deal with the contested portion later.

There is, however, no simple legal mechanism currently available to allow a partial acquisition to occur. The FTC has the ability to hold up the entire acquisition while seeking a preliminary injunction to block a relatively small aspect of the deal. (See Kolasky, Proger, and Englert [1985, 52].) In such situations, a firm may have a strong incentive to reach a settlement with the Commission, spin off the contested assets, and allow the remainder of the transaction to proceed. Thus, we suggest that, in a number of cases, firms would desire to settle with the FTC on the Commission's terms, even if the merits of the Commission's case are weak. (See a similar discussion along these lines in Posner [1972].) Here we seek to test this hypothesis by focusing on the share of the transaction linked to competitive concerns. We will test whether the firm's decision responds to this competitive overlap variable.

In addition, by modelling the effect of the share of the transaction linked to competitive concerns and the value of the transaction not subject to the competitive concerns, we can use the model to predict which cases would end in litigation if the FTC allowed the innocuous parts of the transaction to proceed unchallenged. If a number of such cases are found, the FTC would appear to behave more as a regulatory commission rather than as a prosecutor subject to judicial review under a "preponderance of the evidence" standard. Moreover, if various other factors, such as the value of the transaction and the size of the acquiring firm (see discussions below), enter into the decision, we can also net out these effects to allow the litigation decision to turn on the merits of the case. Again, if the model predicts a number of cases would end in litigation but for these considerations, we would suggest that the FTC functions as a regulatory agency, substituting its evaluation of the competitive merits of a transaction in place of what the caselaw would imply.

Finally, abstracting from the "hostage" issue, enforcement decisions can also be based on non-systematic factors, such as politics or incentives of FTC staffers. Thus, the FTC may desire to accept less stringent settlements to both increase its number of successful cases and to prevent potentially embarrassing losses in federal court. On the other hand, the FTC's complicated internal procedures may prevent it from taking such actions. (See McCubbins, Noll, and Weingast [1987].) Below we will test to see if evidence suggests that parties can enter into less stringent settlements with the Commission if the Commission's case is weak. We also test to determine if the general political climate affects a firm's decision-making process.

III. MODELLING THE FIGHT VS. FOLD VS. SETTLE PROCESS

Modelling the Utilities of the Choices

When faced with an FTC merger challenge, a firm has a choice of either fighting, folding, or settling. Along the lines of Landes [1971], it will choose the option that offers it the highest expected utility (profit). Therefore, it is necessary for us to model the expected utilities of each option. Arbitrarily, we assume that the relative level of utility from folding is zero.

The utility from fighting depends on a number of factors. Let X represent a vector of independent variables affecting the benefits of fighting and Y represent a vector of independent variables affecting the costs of fighting. This makes the utility of fighting

(1) U(Fighting) = F(X) - G(Y) + [[Epsilon].sub.1] .

The benefits of fighting will be a function of a number of factors. Conceptually, the benefits depend on the probability of winning in court, which is a function of the merits of the case. Benefits will also depend on the profits from realizing the acquisition. The benefits of the acquisition will have two parts, the anticompetitive benefits (which will also be a function of the merits of the case), and the available efficiencies from the transaction.(5)

The costs of fighting also have several components. The most obvious is the actual cost of litigation in federal court. Further costs arise because the FTC has a policy of always following a preliminary injunction case with a complaint to its own administrative law system. (See Lopatka and Mongoven, forthcoming.) These proceedings can take years, with hearings before an administrative law judge, an appeal to the FTC sitting as a court, and then (if the FTC rules for its own staff and against the merger) an appeal to a federal circuit court. We suggest, however, that these costs can be viewed as relatively constant across firms. The interest a firm has in incurring these costs, however, appears to depend on the general efficiencies associated with the transaction, a factor likely linked to the overall value of the transaction.

Fighting may also affect the firm's reputation with respect to the government. Large firms may expect more interactions with government regulators than small firms. Thus, their behavior in any one interaction should take into account potential effects on later interactions. For example, a large firm could be more likely to litigate to establish a reputation for toughness along the lines of Milgrom and Roberts [1982]. On the other hand, the large firm could be less likely to litigate to avoid causing problems with future government encounters.

In addition to the direct costs of litigation, a firm must contend with the opportunity costs associated with delaying the transaction. These costs could be avoided if the firm abandons the transaction and refiles for only the competitively neutral aspect of the deal. It may be difficult to isolate the competitive problems in some deals, but this opportunity cost will not exist for all transactions and will likely be a function of the conglomerate aspects of the relevant deal.

Similarly, the utility of settling can be specified as

(2) U(Settle) = H(Z) - K(W) + [[Epsilon].sub.2],

where Z represents a vector of benefits of settling and W represents a vector of costs of settling. We see the benefits of settling as being related to the value of the acquisition excluding the assets to be divested. Moreover, both the benefits and the costs of a settlement are likely to be a function of the available efficiencies. Although divestitures are likely to require the firm to give up most efficiencies, it is possible that a settlement can be crafted to retain some of the efficiencies. On the other hand, the purchase price may implicitly include the value of a significant portion of the efficiencies, so a divestiture would be costly to the firm. Other costs of the settlement depend on the burden of complying with the FTC's order. FTC settlements, with their procedures for divestiture and "hold separate" provisions, impose both legal and operational costs on firms. It is possible, however, that the FTC may offer less costly settlements to firms where the merits of the FTC's case are weaker. Thus, the cost of settlement could be negatively related to the merits of the FTC's position.

Data and Specification

We reviewed all seventy-eight of the Commission's attempts to enjoin horizontal transactions for fiscal years 1984 to 1991. The memos presented a wealth of background data such as the sales of the acquiring firm, the acquired entity and the parent company, the price of the transaction, the magnitude of the competitive overlap, measures of the potential anti-competitive effect of the acquisition and the date of the acquisition. We include those cases in which the parties have offered a settlement, but exclude the transactions that were abandoned before Commission action.(6) Cases were deleted from the sample when the merger involved a partial stock acquisition (one case), when the Commission did not have sufficient notice to complete a full investigation before moving to enjoin the merger in court (two cases) or when the transaction was characterized as a joint venture (three cases). This left a total of seventy-two cases in which the Commission voted for a complaint against a prospective merger.

To represent the value of the transaction, we use the variable VALUE, which is the estimated acquisition price of the transaction (in 1982 dollars using the GDP deflator) taken from the Bureau of Competition's memos. To model the ability of the FTC to hold up non-contested portions of the acquisition, we calculate the overlap figure (OVERLAP) by estimating the percentage of the deal subject to the competitive concerns, usually by dividing the sales in the market(s) under investigation by the total sales of the target entity.(7) When sales data were not available, we used other proxies such as the ratio of the number of stores in the geographic market(s) of concern to the total number of stores sold. We also calculate a variable SETTLE VALUE = (1 - OVERLAP)VALUE, which is equal to the value of the uncontested part of the transaction that potentially could be held up by the Commission. Reputation effects associated with the size of the acquiring firm are modelled with SALES defined as the total sales of the acquiring firm, measured in 1982 dollars.

Each memo also contained an analysis of the basic facts of the case, such as the Herfindahl index, ease of entry and competitive effects (e.g., ability to exploit market power). The Bureau of Economics data showed much more variance than the Bureau of Competition data. Generally, the Bureau of Competition's memos noted the Herfindahl was high, barriers to entry were present and anticompetitive behavior was possible.(8)

The Bureau of Economics's memos were significantly different, with some case evaluations having relatively low Herfindahls, other case discussions noting no barriers to entry and still other case reviews finding no theory of anticompetitive effects. We believe that the Bureau of Economics's data represent the best available proxy for the competitive potential of the merger. We will measure the merits of the FTC's case with the variable STRUCTURE, which is equal to the Bureau of Economics's estimate of the post-acquisition Herfindahl index multiplied by a dummy variable which takes on the value one if the bureau's staff found barriers to entry in the relevant market and equal to zero if the Bureau of Economics found no barriers. In the few cases involving multiple overlaps, the data from the most anti-competitive overlap is recorded. Thus, the STRUCTURE variable measures both the firm's probability of winning in court (negatively related to the probability of fighting) and the potential anticompetitive gains from the acquisition (positively related).

A number of the memoranda also contained limited information on merger-related efficiencies. Although these analyses did not always generate a clear recommendation as to the relevance or the level of efficiencies, we believe it is reasonable to measure the strength of the efficiency concerns by a variable, EFFICIENCIES, equal to the number of pages dedicated to dealing with the efficiency question in both the Bureau of Economics's and Bureau of Competition's analyses.(9) In effect, the strength of the parties' efficiency defense is measured by the number of pages in the staff memos needed to fully explain these issues to the Commission. It is also important to note that lengthy efficiency discussions in the staff memos are in very large part a response to the parties' efficiency claims. Finally, the FTC (and Department of Justice) policy of requiring evidence to support efficiency claims gives parties incentives to make presentations when they have such evidence and not to make such presentations when they do not. (Another way of viewing this variable would be to consider it a measure of how seriously the Commission's staff took the parties' efficiency arguments.)

To measure unspecified efficiencies, which we model as proportional to the size of the transaction, we use the variable VALUE. This in turn implies that larger transactions will have larger efficiencies associated with them, which in turn implies that parties with larger acquisitions are more likely to incur the transactions costs necessary to fight the Commission. We also include in our specification a variable, TIME, equal to the number of months after the first case in the sample when the FTC made its relevant enforcement decision. (The first case in the sample is January 1984.)

The final variable attempts to measure the political considerations facing a firm. We use a political index variable, POLITICS, which averages the percentage of Democratic representatives in the House and Senate after the biannual elections. As Congress becomes more Democratic, we expect firms to bear higher costs from resisting enforcement action, hence litigation is less likely. It may be that the more Democratics in Congress, the more likely a firm is to face retribution from the legislature as a result of resisting the FTC.

The theoretical benefit model thus includes both the structural and efficiency variables. Adding the time index yields

(3) F(X) = [[Alpha].sub.0] + [[Alpha].sub.1]STRUCTURE

+ [[Alpha].sub.2]EFFICIENCIES + [[Alpha].sub.3]VALUE + [[Alpha].sub.4]TIME.

For the costs of fighting, we assume SALES proxies the effect of reputation costs. In addition, we use OVERLAP to proxy the probability that the opportunity costs of delay must be considered, because the deal can be restructured to avoid the competitive concerns. As OVERLAP approaches one, the probability of restructuring the deal falls to zero and the opportunity costs of delay become moot. The political variable POLITICS is also included to reflect the possibility that the firms' perception of costs is affected by the political environment. Again adding a time parameter, the costs of litigation are

(4) (4) G(Y) = [[Alpha].sub.5] + [[Alpha].sub.6]SALES

+ [[Alpha].sub.7]OVERLAP + [[Alpha].sub.8]POLITICS + [[Alpha].sub.9]TIME.

Let [[Alpha].sub.10] = [[Alpha].sub.0] - [[Alpha].sub.5] and [[Alpha].sub.11] = [[Alpha].sub.4] - [[Alpha].sub.9]. Equation (1) now becomes

(5) U(Fight) = [[Alpha].sub.10] + [[Alpha].sub.1]STRUCTURE

+ [[Alpha].sub.2]EFFICIENCIES + [[Alpha].sub.3]VALUE

- [[Alpha].sub.6]SALES - [[Alpha].sub.7]OVERLAP

- [[Alpha].sub.8]POLITICS + [[Alpha].sub.11]TIME + [[Epsilon].sub.1]

with the signs on [[Alpha].sub.2], [[Alpha].sub.3] and [[-Alpha].sub.7] expected to be positive, the sign on [[-Alpha].sub.8] expected to be negative and no priors for the other coefficients.

For the benefit of settling, we first take into account the value of the firm that would be obtained through acquisition after a (potential) divestiture of the offending assets, SETTLE VALUE. We also consider any efficiencies that might be available, using our EFFICIENCIES variable as discussed above, as well as the impact of TIME. This generates

(6) H(z) = [[Beta].sub.0] + [[Beta].sub.1]SETTLE VALUE

+ [[Beta].sub.2]EFFICIENCIES + [[Beta].sub.3]TIME.

We view the cost of settling as having two attributes, the stringency of the FTC settlement and the efficiencies taken away by the settlement. We suggest that the weaker the merits of the FTC's case (as measured by STRUCTURE), the less costly the settlement required by the Commission. Also, the cost of the settlement could depend on political factors, although we hesitate to predict a sign. The cost of settling may also be a function of time. This generates

(7) K(W) = [[Beta].sub.4] + [[Beta].sub.5]STRUCTURE

+ [[Beta].sub.6]EFFICIENCIES + [[Beta].sub.7]POLITICS

+ [[Beta].sub.8]TIME.

Let [[Beta].sub.9] = [[Beta].sub.0] - [[Beta].sub.4], [[Beta].sub.10] = [[Beta].sub.2] - [[Beta].sub.6], [[Beta].sub.11] = [[Beta].sub.3] - [[Beta].sub.8], all three unsigned. The utility of settling now becomes

(8) U(Settle) = [[Beta].sub.9] + [[Beta].sub.1]SETTLE VALUE

+ [[Beta].sub.10]EFFICIENCIES - [[Beta].sub.5]STRUCTURE

- [[Beta].sub.7]POLITICS + [[Beta].sub.11]TIME + [[Epsilon].sub.2],

with [[Beta].sub.1] expected to have a positive sign, [[-Beta].sub.5] taking on a negative sign and [[Beta].sub.7] unsigned.

Let us now assume that the error terms are distributed logistically and that independence of irrelevant alternatives holds. Let

A = [[Alpha].sub.10] + [[Alpha].sub.1]STRUCTURE

+ [[Alpha].sub.2]EFFICIENCIES + [[Alpha].sub.3]VALUE

+ [[Alpha].sub.6]SALES + [[Alpha].sub.7]OVERLAP

+ [[Alpha].sub.8]POLITICS + [[Alpha].sub.11]TIME,

and

B = [[Beta].sub.9] + [[Beta].sub.1]SETTLE VALUE

+ [[Beta].sub.10]EFFICIENCIES + [[Beta].sub.5]STRUCTURE + [[Beta].sub.7]POLITICS + [[Beta].sub.11]TIME.

This implies that the probability of fighting, settling, and folding respectively equal

(9) [P.sub.1] = [e.sup.A] / (1 + [e.sup.A] + [ e.sup.B])

(10) [P.sub.2] = [e.sup.B]/(1 + [e.sup.A] + [e.sup.B]),

(11) [P.sub.3] = 1 - [P.sub.1] - [P.sub.2].

Using standard methods of maximum likelihood and multinomial logit analysis, we can now estimate the parameters in equations (5) and (8).

We first examine the data before presenting the regression results. Table I presents the average values of the variables for the given fight, fold or settle outcome. The average OVERLAP was 82.7 percent in the litigated cases, 62.7 percent in the abandoned cases and only 17.3 percent in the settled cases. In particular, overlaps of 10 percent or less almost always (94 percent in the sample) result in settlements, while overlaps of up to 30 percent usually (77 percent) end in settlements. On the other hand, large overlaps of over 50 percent almost always (94 percent) lead to either litigation or the abandonment of the deal. Similarly, the mean of the page-based efficiency variable is 22.5 pages for the litigated cases, but only 11.5 for the abandoned cases and 5.3 for the settled cases. Both the differences in the average overlap and efficiencies are statistically significant. Also of interest is the price of the transaction (VALUE) and the value of the acquisition not related to the competitive overlap (SETTLE VALUE). Both numbers are relatively low for the litigated and abandoned cases, and significantly higher for the settled transactions. Finally, the average size of the acquiring firm (SALES) is significantly higher if the transaction ends in a settlement rather than litigation.

IV. ESTIMATION OF THE MODEL

Presentation of the Results

Table II presents the estimated results of our econometric model. The first column defines the fight vs. fold equation, while the second column identifies the settle vs. fight equation.(10) As implied above in equations (5) and (8), coefficients on VALUE, OVERLAP and SALES are only estimated for the fight equation, while a coefficient on SETTLE VALUE is only estimated for the settle equation.(11)

There is marginal evidence in Table II that, on net, the merits of a case affect the fight vs. fold decision, as the t-statistic on STRUCTURE equals only 1.59.(12) Thus, the statistical analysis gives only weak support to the hypothesis that the opportunity to succeed on the merits outweighs the incentive to capture anticompetitive profits related to a high structural index in litigation decisions. We also found no evidence that the Commission offers less burdensome settlements when it is pursuing a weaker case on the merits, as the t-statistic on the merits variable in the settle vs. fold equation is only 0.06.

TABLE I

Variables Means by Fight, Fold and Settle Status

(Standard Deviations In Parentheses)

Variable Fight Fold Settle Overall
 N=13 N=30 N=29 N=72

OVERLAP(123) 82.69 62.73 17.34 48.06
 (26.54) (32.33) (20.65) (37.55)

EFFICIENCIES(123) 22.42 11.51 5.328 10.99
 (19.15) (11.95) (6.776) (13.23)

STRUCTURE 2985 3766 3527 3529
 (2313) (2396) (3861) (3034)

VALUE(23) 134.0 88.95 1969 854.5
 (203.2) (117.7) (4385) (2906)

SETTLE VALUE(23) 24.36 36.53 1780 736.5
 (64.59) (74.92) (4131) (2735)

SALES(2) 2598 5266 7592 5721
 (4123) (7743) (1184) (8944)

TIME 44.23 55.63 51.62 51.96
 (29.3) (23.15) (27.8) (26.19)

POLITICS 55.45 56.57 56.28 56.25
 (2.41) (1.99) (2.09) (2.12)

1 indicates that the mean of the variable for fight outcomes is
statistically different from the mean for fold outcomes at the 10
percent level, 2 indicates that the mean of the variable for fight
outcomes is statistically different from the mean for settle
outcomes at the 10 percent level, and 3 indicates that the mean of
the variable for fold outcomes is statistically different from the
mean for settle outcomes at the 10 percent level.

Other control variables have the anticipated effects. The level of efficiencies has the predicted effects on fighting relative to folding, with higher efficiencies making fighting more likely. In addition, the presence of efficiencies also makes settling less likely relative to folding.(13) A higher overlap variable increases the odds on litigation vs. folding. The time variable has no significant impact on the odds of settlement and only a very weak effect on the odds of litigation. The level of a firm's sales is negatively related to the fight vs. fold decision, indicating that large firms seek to generate a reputation for "good behavior" with the government. The settle value of the transaction (after adjustment for divestiture) also serves to increase the probability of a settlement, while the value of the transaction has a weakly positive impact on the ratio of probability of litigation to folding. Finally, the political index significantly reduces the likelihood of litigation relative to folding, while having no effect on the probability of a settlement.

TABLE II

Multinomial Logit Model

 Fight vs. Fold Settle vs. Fold
 Equation Equation

VALUE .8575 -
 (1.64)

OVERLAP .05083(**) -
 (2.26)

SALES -1.280(**) -
 (-2.04)

SETTLE VALUE - 1.260(**)
 (3.45)

EFFICIENCIES 1.901(**) -1.112(**)
 (2.37) (-2.24)

STRUCTURE -.0005153 .00000806
 (-1.59) (0.06)

TIME .08136 .02290
 (1.50) (0.68)

POLITICS -1.244(**) .1081
 (-1.95) (0.27)

Constant -.9498 -5.086(**)
 (-0.24) (-2.11)

Pseudo R-square: 0.5598 Log Likelihood: -32.97

t-statistics in parentheses, * indicates significance at the 10
percent level and ** indicates a 5 percent level, using two-tailed
tests. The Pseudo R-square is calculated by taking one minus the log
likelihood of the equation divided by the log likelihood estimated
using only the constant terms.

Given these results, we can conclude that the size of the overlap (and the derived level of SETTLE VALUE) affects decisions, with firms involved in deals with small overlaps almost certain to settle and firms involved in deals with large overlaps more likely to fight. Efficiencies are also important. If efficiencies are relatively large, the firm is more likely to fight to save the proposed merger. Settlements are also affected by efficiencies, with firms tending to abandon, rather than settle, transactions with larger efficiencies. The merits of a case appear to play a relatively small role in the litigation decision. The value of the transaction also plays a role in the decision, with large deals tending to be more likely to end in litigation. Likewise, political considerations affect the litigation decision, with firms less likely to litigate in periods of strong Democratic political control.

TABLE III

Prediction Table for Historical Data

 Predicted Outcome

 Fight Fold Settle Total % Correct

Actual Fight 10 2 1 13 77

Outcome Fold 3 24 3 30 80
 Settle 0 3 26 29 90

Total 13 29 30 72 83

Application of the Model: Predictions vs. Actual Outcomes

Table III examines how well the model predicted the actual outcome of the merger cases. Fitted values were computed for each data point and the result with the largest probability is taken as the predicted outcome.(14) The model predicts correctly 77 percent of the cases that end in litigation, 90 percent of the settlements and 80 percent of the abandoned mergers. Overall, the model predicts 83 percent of the cases correctly.

Of the twelve cases where the model did not predict the actual outcome, three cases (one of which was a fight) had actual outcomes with probabilities over 40 percent. In two other cases, the third highest probability represented the actual choice. The first involved an unusual settlement mechanism, and the second involved a case that ended in litigation under peculiar circumstances. In the seven remaining cases, one represented a situation in which the parties chose fighting (predicted probability 17 percent) over folding (82 percent), three were cases where folding (24 percent, 23 percent and 28 percent) were chosen over settling (76 percent, 77 percent and 72 percent), two involved folding (31 percent and 16 percent) over fighting (69 percent and 84 percent) and in one case the parties decided to settle (12 percent) instead of fold (87 percent).

Application of the Model: Sensitivity

Some indication of the impact of particular variables can be gained by reviewing what effects they have on the probability of each outcome, given our basic model of Table II. For example, assuming a case has the attributes described by the mean of the estimated sample, the basic model shows that firms will fold with a 55.6 percent probability, settle with a 42.8 percent probability, and fight with only a 1.6 percent probability.(15) Holding all other variables constant and raising the overlap variable one standard deviation to 85.6 percent (with the accompanying change in SETTLE VALUE) generates an 18.3 percent increase in the probability of folding to 73.9 percent, a 30.9 percent decrease in the settle rate to 11.9 percent, and an 12.6 percent rise in the fight probability to 14.2 percent. Similarly, holding all other variables constant and reducing the overlap variable one standard deviation to 10.5 percent reduces the fold probability 15.9 percent to 39.7 percent, raises the settle rate 17.4 percent to 60.2 percent, and lowers the fight rate 1.5 percent to an unlikely 0.1 percent.

Changing the level of efficiencies has similar, but smaller effects. Raising the efficiency variable one standard deviation while holding all other variables at their sample means generates an increase in the fold rate to 69.1 percent, a decrease in the settle rate to 16.8 percent, and an increase in the fight rate to 14.1 percent. Lowering the efficiency variable to zero reduces the probability of folding to 12.8 percent, raises the settle rate to 87.3 percent, and lowers the fight rate to a tiny 0.01 percent.

Raising the STRUCTURE variable one standard deviation while holding all other variables at their sample means has little effect on outcomes, raising the probability of folding 0.2 percent, the probability of settling 1.0 percent, while reducing the chances of fighting 1.3 percent. Lowering the STRUCTURE variable 1.13 standard deviations (to 0, representing a case with no barriers to entry perceived by the Bureau of Economics's staff), however, has a slightly larger impact. In this scenario, the probability of folding falls 3.6 percent to 52 percent, the probability of settling declines 4.0 percent to 38.8 percent, while the probability of fighting rises 7.6 percent to 9.2 percent.

Changes in the political index (POLITICS) also affect the likely outcome. A one standard deviation increase in the political variable would increase the probability of a settlement to 49.1 percent, while lowering the likelihood of abandoning the transaction to 50.8 percent and almost eliminating the chance of litigation at 0.1 percent. On the other hand, a one standard deviation reduction in the political variable raises the likelihood of litigation to 19.8 percent with reductions to 49.7 percent for the fold rate and 30.5 percent for the settle rate.

Application of the Model: Predictions

One can also apply the model to proposed mergers to predict the likely response to a Commission merger challenge. In the two-year period immediately following September 1991 (the end of the data set for the regression in Table II), the Commission challenged twelve proposed horizontal transactions. Nine cases ended in settlements, with the parties selling off offending assets or technology. In six of these cases, the model predicted a probability of settlement of more than 90 percent, while in the others the settlement probability exceeded 50 percent. The model also correctly predicted the one abandoned merger, as well as both the cases that ended in litigation. Thus, the model has successfully predicted the outcomes of all of the FTC merger challenges for these two years.

Finally, as discussed above, the FTC does not allow firms to consummate part of an acquisition while litigating the remainder. The model, however, can predict the outcome of the various cases if the FTC changed policies and allowed the consummation of the competitively neutral part of the transaction by lowering VALUE to the value of the contested part of the transaction, assigning SETTLE VALUE to zero (because there would be nothing left to settle) and replacing OVERLAP with 100 to reflect the fact that the remaining part of the deal would be subject to competitive concerns. Under these conditions, our results indicate that twenty of the cases in the data set would have been litigated, including eleven of the thirteen that actually were litigated and eight cases that were actually abandoned. Five of the twenty-four transactions correctly predicted to be abandoned would have ended in litigation without this "hostage" effect, while one of the twenty-six settlements would have gone to court. Thus, the model suggests that the Commission's refusal to allow partial acquisitions precluded a litigated outcome of six of the fifty successfully modelled cases.

In addition, it is also possible to exploit other hostage effects related to the size of the transaction. As our model shows, both small deals and transactions by large corporations are unlikely to be litigated. We can use our model to estimate the impact of these factors. For example, starting after the FTC allows the noncontroversial aspects of a transaction to settle, one can replace the actual value of the potentially anticompetitive transaction with the sample mean of $118 million for all cases with contested values under $118 million.(16) Calculating expected probabilities for each data point suggests that twenty-eight cases would end in litigation, including all of the thirteen actual litigations. Moreover, the model predicts that in these circumstances ten of the twenty-four folds and two of the twenty-six settlements predicted by the method used for Table III would go to litigation. The procedure can be taken one step further to minimize the reputation effect by setting the acquiring firm's sales to $100 million for all transactions assigned a value of $118 million which currently show sales over $100 million.(17) The model would then predict fifty-seven litigated cases, including all thirteen actual litigations. Transactions would rarely fold, with only one of the twenty-four abandoned cases and eleven of the twenty-six settlements predicted in Table III now being dropped before litigation. Thus, when taken together, the three "hostage" effects tend to induce firms to either drop or settle the competitive problems when faced with FTC action.

V. CONCLUSION

The results of this study have clear implications for our understanding of antitrust policy. The econometric model shows that firms are very unlikely to litigate when the overlap is small, the value of the transaction is low and/or the acquiring firm is large. These conditions encompass a significant number of FTC cases. Thus, it appears that the nature of the merger process, not the merits of individual cases, is the driving force in the outcomes of these cases. Therefore, the FTC (and likely its sister agency, the Department of Justice) must be seen in large part as a regulator, not a law enforcer, because there is no viable independent review of a large number of its bureaucratic decisions. To retain a judicial standard in antitrust cases, the decision makers must therefore act as courts and assign the litigation staff the burden of proof when making enforcement decisions.

The model also sheds light on the firm's decision-making process. The significance of the efficiency variable appears to suggest that some mergers are efficiency motivated, because firms are more willing to incur litigation costs to complete a transaction when efficiencies are high. Similarly, firms appear less willing to enter into settlements in markets where significant efficiencies are likely to be available. The structure of the market may have an effect, with firms more willing to litigate weak cases, though the evidence did not strongly confirm this result. Finally, political considerations also appear to have some influence on the firm's fight, fold or settle decision.

1. The FTC's merger enforcement also includes cases against consummated horizontal mergers, vertical transactions and conglomerate mergers when issues of potential competition exist. Such cases, however, are relatively rare and therefore do not present the opportunity for systematic analysis.

2. Transactions are covered if a firm with assets or net sales of $100 million or more acquires an ownership interest in a firm with assets or net sales of at least $10 million, or a firm with assets or net sales of $10 million or more acquires a firm with assets or net sales of at least $100 million (15 U.S.C. sec. 18A(a) (2)). Reporting requirements are triggered if, as a result of the transaction, the acquiring firm would hold at least 15 percent of the stock in the target or more than $15 million of the target's voting securities (15 U.S.C. sec. 18A(a) (3)). For further details, see Johnson and Parkman [1991]. The government often learns of small uncovered transactions through the news media or informal channels. An investigation may then result, although the government lacks the formal waiting period requirements. Because the legal cost of defending a small transaction can easily dwarf the efficiency benefits, firms in this situation almost always cooperate with the government's investigation.

3. For a model of Canadian merger enforcement policy, see Khemani and Shapiro [1993]. They find that high market shares and barriers to entry make enforcement more likely, while import penetration makes enforcement less likely.

4. The actual negotiation process may begin before the Commission has taken a formal vote to move against a merger. A vote, however, must be taken to issue a complaint against a transaction and accept a consent for public comment. While the Commission can reject consents after the public comment period, this rarely occurs. Minor changes, however, may be made to account for technical problems overlooked in the initial negotiations.

5. It is unlikely that courts recognized efficiencies as much of an antitrust defense during our period of study. Even if efficiencies were thought to have a significant impact on the outcome of the case, the 1984 Guidelines stated that the defense should be based on "clear and convincing" evidence. Appeals Court Judge and former Assistant Attorney General for Antitrust Ginsburg [1991, 97] has called reaching this standard "well-nigh impossible." Further, even if defendants can meet this level of proof, only efficiencies in the relevant anticompetitive market may count. See also FTC v. University Health, Inc., 938 E2d 1206 (11th Cir. 1991).

6. By checking the records on second requests issued, FTC enforcement actions and agency files, we found that very few transactions were abandoned after the staff arrived at an enforcement recommendation, but before the Commission decision. In some of these cases, withdrawn deals were refiled later and the Commission moved to enjoin the transaction.

7. All the overlaps subject to merger challenge are counted in the overlap variable, because the Commission tends to obtain full settlements. It would make little sense for a party to settle one overlap, merely to go through a trial on the merits of other overlaps.

8. This observation does not imply that the Bureau of Competition's analyses are biased. The sample of seventy-two cases were all recommended by the Bureau of Competition's staff, so one may expect the bureau's staff would believe that they are all meritorious cases. Numerous other investigations were closed, with the bureau's analysis showing low Herfindahls, low barriers or no anticompetitive story. For a further discussion, see Coate, Higgins, and McChesney [1990] and Coate and McChesney [1992].

9. We found no evidence that staff or management at the Commission engages in explicit trade-offs of anticompetitive and efficiency effects of mergers, contrary to the analysis of Comanor and White [1992].

10. If the FTC used OVERLAP as a proxy for the probability of an easy settlement and allowed this factor to influence its enforcement decision, we would have to adjust for sample selection bias with respect to the cases where the parties chose to settle. (Under this scenario, the low overlap cases would be over-represented in the sample.) Following Hughes and Snyder [1989] we test for this problem by estimating a bivariate probit model, with one equation for the FTC enforcement decision (with a model similar to Coate, Higgins and McChesney [1990]) and the other for the firm's decision to settle (with a model similar to the one in this paper). (See Snyder [1990, 451-2] for a similar approach in a setting with two available choices.) The coefficient relating the error terms of the two equations is insignificant, with a t-statistic of -0.32, indicating no relationship between the equations. Moreover, unlike results reported in Hughes and Snyder, the coefficients generated by the bivariate estimation process are almost identical to the coefficients generated by independent univariate probits for the coefficients that were significantly different from zero. Thus, we are unable to reject the null hypotheses of no sample selection bias in the data. A test of the means of the OVERLAP variable tends to confirm this result, as no statistical difference is observed between the mean value of OVERLAP from the sample of closed cases and the mean value from the sample of enforcement actions.

11. A logarithmic transformation was applied to the VALUE, SALES and EFFICIENCY variables before the model was estimated.

12. Models not reported here with different merits variables have lower t-statistics on these variables. We also ran models with [STRUCTURE.sup.2], but failed to obtain a significant coefficient for that variable.

13. We test for the independence of irrelevant alternatives by estimating the parameters without either the fight or settle cases along the lines of Hausman and McFadden [1984]. In either situation, the coefficient estimates were almost identical, with Hausman statistics of 1.34 and 0.0002, respectively. Neither statistic is significant, indicating that the hypothesis of equal coefficients cannot be rejected. For a more complete robustness analysis of a related model see Coate, Kleit, and Bustamante [1993]. The results in that paper generally carry over to the slightly different specification presented here.

14. The smallest highest predicted value in the data set was 44.2 percent.

15. For a case that has attributes equal to the sample mean of those cases that folded, the model predicts an 80 percent chance of folding, with an 18 percent chance of a settlement and a 2 percent chance of litigation. For a case with attributes equal to the sample mean of those cases where the firms chose to settle, settling has a 98 percent chance of occurring, while folding is expected to occur with a 2 percent likelihood. Finally, for a case where the attributes are equal to that of the sample mean for the cases that ended in litigation, fighting is expected to occur with a 66 percent probability. Folding is the primary alternative with almost a 32 percent chance.

16. This is calculated as the difference between the average value for the overall transaction and the mean settle value, both taken from Table I.

17. This number is chosen arbitrarily to be similar to the $118 million value of the transaction used in the analysis. One could use a larger figure for sales and find fewer firms litigate their cases.

REFERENCES

Coate, Malcolm B. "Economics, the Guidelines and the Evolution of Merger Policy." Antitrust Bulletin, Winter 1992, 997-1024.

Coate, Malcolm B., Richard S. Higgins, and Fred S. McChesney. "Bureaucracy and Politics in FTC Merger Challenges." Journal of Law and Economics, October 1990, 463-82.

Coate, Malcolm B., Andrew N. Kleit, and Rene Bustamante. "Fight, Fold, or Settle?: Modelling the Reaction to FTC Merger Challenges." Federal Trade Commission Working Paper No. 200, February 1993.

Coate, Malcolm B., and Fred McChesney. "Empirical Evidence on FTC Enforcement of the Merger Guidelines." Economic Inquiry, April 1992, 277-293.

Comanor, William S., and Lawrence J. White. "Market Power or Efficiency?: A Review of Antitrust Standards." Review of Industrial Organization, 7(2), 1992, 105-16.

Ginsburg, Douglas H. "Antitrust as Antimonopoly." Regulation, Summer 1991, 91-100.

Hausman, Jerry, and Daniel McFadden. "Specification Test for the Multinomial Logit Model." Econometrica, September 1984, 1219-40.

Hughes, James, and Edward Snyder. "Policy Analysis of Medical Malpractice Reforms: What Can We Learn From Claims Data?" Journal of Business and Economic Statistics, October 1989, 423-30.

Johnson, Ronald, and Allen Parkman. "Premerger Notification and the Incentive to Merge and Litigate." Journal of Law, Economics & Organization, April 1991, 145-62.

Khemani, R. S., and D. M. Shapiro. "An Empirical Analysis of Canadian Merger Policy." Journal of Industrial Economics, June 1993, 161-77.

Kleit, Andrew N., and Malcolm B. Coate. "Are Judges Leading Economic Theory? Sunk Costs, the Threat of Entry and the Competitive Process." Southern Economic Journal, July 1993, 103-18.

Kolasky, William J., Jr., Phillip A. Proger, and Roy T. Englert. "Anticompetitive Mergers: Prevention and Cure," in Antitrust and Regulation, edited by Franklin M. Fisher. Cambridge, Mass.: MIT Press, 1985, 49-84.

Landes, William M. "An Economic Analysis of the Courts." Journal of Law and Economics, April 1971, 61-108.

Lopatka, John E., and James F. Mongoven. "After Preliminary Relief in Merger Cases is Denied, What Then?" Research in Law and Economics, 17, 1995 forthcoming.

McCubbins, Matthew D., Roger G. Noll, and Barry R. Weingast. "Administrative Procedures as Instruments of Political Control." Journal of Law, Economics, and Organization, Fall 1987, 243-77.

Milgrom, Paul, and John Roberts. "Predation, Reputation, and Entry Deterrence." Journal of Economic Theory, 27(2), 1982, 280-312.

Perloff, Jeffrey M., and Daniel L. Rubinfeld. "Settlements in Private Antitrust Litigation," in Private Antitrust Litigation: New Evidence, New Learning, edited by Lawrence White. Cambridge, Mass.: MIT Press, 1988.

Posner, Richard A. "The Behavior of Administrative Agencies." Journal of Legal Studies, June 1972, 305-47.

Snyder, Edward A. "The Effect of Higher Criminal Penalties on Antitrust Enforcement." Journal of Law and Economics, October 1990, 439-62.

U.S. Department of Justice. "Department of Justice and Federal Trade Commission Horizontal Merger Guidelines." Antitrust Trade and Regulation Report, No. 1559, April 2, 1992.