Joint Executive Committee Cartel Ability to Maintain Market Power
Or how misread signals derail cooperation.

Executive Overview

The JEC cartel could be made more efficient in regards to collusion. Up to $7,587 (1886 dollars) in resources can be used to fund an enforcement entity if that entity or system is able to improve collusion by 50%. 

Introduction

In a market that naturally has few sellers, suppliers may more readily cooperate by forming a cartel to restrict output and increase their total revenue. However, each individual firm within the cartel faces incentives to cheat. Enforcing cooperation requires both a strategy to detect cheating as well as a method of punishment. We will be looking at an American railroad cartel from the 1880s to determine whether they were able to obtain market power. Instrumental variable regression will be used to estimate price elasticity of demand to determine if the cartel was successful in exerting market power. We'll see that despite some success, they went off the tracks in maintaining cooperation. We also determine the upper bound of funds available to increase cartel compliance by 50% to be $7,587 per year.

JEC Overview

The Joint Executive Committee was a railroad cartel in operation during the 1880s to set prices and quantity for grain transport from Chicago to the Eastern Seaboard. Railroads, with their high fixed costs and relatively low variable costs, are primed for cartel and monopoly formation. We will be looking at a data set from Stock & Watson covering the JEC from 1880 to 1886 based off Porter's 1983 study. Conveniently this dataset includes when the cartel self reported itself as cooperating.

Enforcing cooperation between cartel members for the JEC was difficult on two fronts. Firstly, enforcing collusion and punishing cheaters could not be done through contract law. Secondly detecting cheating, deviation from agreed upon outputs, was not an easy task. We will take a brief look at a trigger strategy that would overcome these difficulties. We will then delve into the actual data to see how the JEC fared in maintaining a trigger strategy. 

Enforcing Collusion with a Trigger - Suffering Together

While collusion was legal, contracts between colluding entities were unenforceable. The law neither supported, nor prevented, collusion. Without the support of contract law, the JEC cartel was forced to use self governance. All members were rewarded if the cooperated by obtaining revenues greater than would be had by not cooperating. Yet each individual member was rewarded greater still if the increased their output while all other members restricted their output. This secondary incentive, to be the first mover to cheat from the agreed to strategy, required a means to detect and punish cheaters.

To support collusion the JEC cartel resorted to a trigger strategy. If a cartel member is suspected of cheating, all cartel members would pull the trigger and revert to Cournot competitive levels of output. The flood of grain transport caused by resorting to competition increased market supply and thereby lowered the price below what would be had by a collusive level of output. All cartel members are harmed when the trigger is enacted as they all operate below potential total revenue that would be gained by collusion. The punishment lasts for a duration in which the estimated collusion level is greater than or equal to the estimated revenue gained by the cheating party. In theory this would create a strong incentive to refrain from cheating as all benefits from cheating would be wiped out. 

Signal & Noise

Detecting when a member company was engaged in cheating was not an easy task. Incentives for individual companies change over time as their discounting of future revenue fluctuates. An increase in quantity of grain shipped may be due to a supply shock, a demand shock, or to cheating. Signals to pull the trigger can be misread. False positives result in all cartel members be unnecessarily harmed. False negatives result in stronger incentives to cheat. While cartel members were concerned with detecting deviation of other members, we will be looking at a broader view see if all cartel members were able to, on average, maintain collusive levels of output.

Let's delve into the JEC data to try to tease out the signal from the noise. Our data set consists of 328 weeks that record price and quantity for JEC shipments between 1880 and 1886. The range for price is between 12 cents and 40 cents, and tends to jump in 5 cent increments. For reference, 5 cents in 1880 is equivalent $1.11 in 2016 dollars.  Quantity is the total recorded tonnage of grain shipped per week. We also have an ice variable that notes when the shipping operation on the Great Lakes, a primary substitute for rail transport,  were halted due to ice (ice = 1 ). Another dummy variable, cartel, signifies if the cartel was cooperating for any given week. Not shown are 12 binary seasonal variables that denote the month.

JEC Summary Table
Statistic N Mean St. Dev. Min Max
week328164.5094.831328
price3280.250.070.120.40
quantity32825,384.3911,632.774,81076,407
cartel3280.620.4901
ice3280.430.5001


The binary variable signifying cartel cooperation within the data set was self reported by the cartel, which raises concerns with regards to its fidelity. An initial timeline plot reveals cooperation deteriorating over time as the cartel resorts to pulling the trigger, competitive levels of output, with greater frequency.
 

Increasing frequency of the JEC engaging in competitive levels of output.

The worsening levels of cooperation is noteworthy, but we shouldn't trust the data at first sight. We can check to see if the rest of the data complies with a cartel status. Collusion will lead to a higher prices than would be obtained through competition for a given output, else there would be no incentive to cooperate. This naturally works itself into the hypothesis: 

  • H0 Null - On average non-cartel prices are equal to or greater than cartel prices for a given quantity. Equal or less total revenue.
  • H1 Alternative  - On average cartel prices are higher than competitive prices for any given level of output. 

A visual check of the data is enough to satisfy a rejection of the null hypothesis, as would a simple linear regression. Non-cartel prices are significantly lower than cartel prices in both statistical and economic sense. Cartel status is associated with roughly a $500 ($11,000 in 2016 prices) increase in weekly revenue. We reject the null hypothesis that Cartel status lacks any change, and if change is associated with lower total revenue.

A quick visual check of the data points reveals that when the cartel status is noted as cooperating, the prices do indeed tend to be higher for a given level of output.

Qualifier for Cartel Success

A cartel is successful if the members are able to sustain collusion and obtain market power. Market power is the ability to profitably raise price above marginal costs. We unfortunately do not have details on the marginal costs for the cartel members. Rather than trying deduce members costs, we will look at the extreme end of market power to see if the cartel is operating at its fullest potential. The goal of a cartel is to operate in such a fashion as to obtain monopoly rents. In essence, if the cartel is able cooperate to a level that they appear to act as one unified company.

A qualifier of a successful monopoly, or very successful cartel, is that it operates within the elastic range of demand. Within the elastic range of demand, a percentage increase in price is meet with a greater percentage decrease in quantity. Similarly, a percentage decrease in quantity supplied is meet with a lesser percentage increase in market price. In this range any increase to price or reduction in quantity supplied is meet with reduced total revenue for the producers.

While odd at first glance, the counter point makes this clear. Within the inelastic range of demand a monopoly can either increase price, or reduce quantity, resulting in increased total revenue. It is safe to assume increasing price or decreasing output has little significant effect on operating costs, and as such profits increase. Without competition to check them why would a monopoly, much less any other company or collusive group, not seek to increase their profit? 

Therefore the cartel is successful when the Price Elasticity of Demand is greater than 1 in absolute terms, within the elastic range.    | E | > 1

Price Elasticity of Demand

Market Prices and Instruments

The issue with price and quantity data that we posses is that they are determined by a simultaneous equation. Price and quantity are determined by both supply and demand. A change to price and quantity may be result of a change to demand, or it may a result of a change to supply, or it may be both. If we are able to isolate demand, we can determine whether the JEC cartel operated along the elastic or inelastic range of the demand curve. This simultaneous equation poses a problem with OLS regression in teasing out the demand curve. A simple log-log multivariate regression model produces significant results, but neglects the structural problem of endogeneity with suppl and demand.

log(Quantity) = log(Price) + Cartel + Ice + Seasons + e

JEC Transport Demand
Dependent variable:
log(Quantity)
OLS
log(Price)-0.488*** (0.106)
Cartel0.135** (0.060)
Ice0.410*** (0.120)
Season 1-0.146 (0.110)
Season 20.039 (0.111)
Season 30.064 (0.113)
Season 40.119 (0.111)
Season 50.067 (0.130)
Season 60.004 (0.160)
Season 70.086 (0.160)
Season 8-0.280* (0.160)
Season 9-0.056 (0.161)
Season 100.124 (0.161)
Season 110.185 (0.160)
Season 120.174 (0.159)
Constant9.079*** (0.196)
Observations328
R20.324
Adjusted R20.291
Residual Std. Error0.395 (df = 312)
F Statistic9.958*** (df = 15; 312)
Note:*p<0.1; **p<0.05; ***p<0.01

The simple log-log model shows that the JEC cartel is operating in the inelastic range of demand ( 0 < | PEd | < 1 ) and on average unsuccessful at collusive behavior. However, as mentioned this would be a naive conclusion as the model does not address the endogeneity of price caused by both supply and demand. 

IV Solution - Two Stage Least Squares

In order to tease out the price elasticity of demand we need an instrument that can account for the supply curve.  There are two key qualifications that a instrumental variable must posses.

  1. Relevance; The instrument must be correlated with variable in question.
  2. Exclusion; The instrument is not correlated with any other determinant of the dependent variable. 

Relevance can be tested, but exclusion only stands on the grounds of a logical argument that often cannot be empirically confirmed. In our case we can make the claim that cartel status has no effect on demand for grain on the Eastern Seaboard.  A New Yorker's increased preference for grain for hot dog buns, due to baseball, is not relevant to whether a cartel members in Chicago are cheating one another. This change in demand should not be confused with a change in quantity demanded. Surely the increased price from constricted supply will result in decreased quantity demanded. But this is movement along the demand curve, as opposed to a shift of the entire demand curve caused by a change in preference.

To test for relevance we can regress price upon cartel status. We find this relationship to be statistically and economically, roughly one standard deviation difference, significant. The previous data checks to see if the self reported cartel variable was reasonable revealed already established this relationship.  We will run several checks on the cartel instrument to see if we can reject it as a valid instrument.

Diagnostic tests:

Test df1 df2 Statistic P-value  
Weak Instruments 2 313 191.3749 <2e-16 ***
Wu-Hausman 1 312 4.769 0.0297 *

The weak instruments test's null hypothesis is that the instrument, cartel, has a low correlation with the endogenous variable. We significantly reject the hypothesis that cartel is a weak instrument.  Further, we weakly reject the Wu-Hausman hypothesis that the OLS and IV models are equally consistent. If they were equally consistent, it would be more efficient to make use of the OLS model.

JEC Transport Demand Equations
Dependent variable:
log(quantity)pricelog(quantity)
OLSInstrument Test2SLS
log(Price) -0.488*** (0.106)-0.867*** (0.132)
Cartel-0.135** (0.060)0.077*** (0.006)
Ice0.410*** (0.120)0.009 (0.015)0.423*** (0.122)
Season 1-0.146 (0.110)0.016 (0.014)-0.131 (0.112)
Season 20.039 (0.111)0.034** (0.014)0.091 (0.113)
Season 30.064 (0.113)0.043*** (0.014)0.136 (0.113)
Season 40.119 (0.111)0.020 (0.014)0.153 (0.112)
Season 50.067 (0.130)0.001 (0.017)0.074 (0.132)
Season 60.004 (0.160)-0.006 (0.020)-0.006 (0.163)
Season 70.086 (0.160)-0.017 (0.020)0.060 (0.164)
Season 8-0.280* (0.160)-0.011 (0.020)-0.294* (0.164)
Season 9-0.056 (0.161)-0.004 (0.020)-0.058 (0.164)
Season 100.124 (0.161)-0.023 (0.020)0.086 (0.168)
Season 110.185 (0.160)-0.020 (0.020)0.152 (0.165)
Season 120.174 (0.159)-0.002 (0.020)0.179 (0.162)
Constant9.215*** (0.231)0.191*** (0.019)8.574*** (0.216)
Observations328328328
R20.3240.4600.296
Adjusted R20.2910.4360.264
Residual Std. Error0.395 (df = 312)0.050 (df = 313)0.402 (df = 313)
F Statistic9.958*** (df = 15; 312)19.031*** (df = 14; 313)
Note:*p<0.1; **p<0.05; ***p<0.01

Results - Unable to Sustain Cooperation

The IV models reveals a price elasticity of .867, within the inelastic range of the demand curve. The cartel, on average, was not successful in their collusive efforts. It should be noted that if one used an OLS model, ignoring the endogeneity of price, one would incorrectly believe the collusive efforts of the cartel to be less successful.

If we were a JEC cartel member, how is this information useful? Firstly we know that we are leaving money on the table. Our current self governance may not be strong enough to sustain collusion. Secondly, we know that on average weekly total revenue increases by $488.31 when the cartel is colluding. The cartel pulls the trigger, either do to cheating or perceived cheating, on average 38% of the time for a given week of the year. As such, the cartel can dedicate resources up to $15,715 ($348,500 in 2016 dollars) for improving detection and enforcement to sustain collusion. This should be treated as a naive upper bound estimate as the current strategies are approaching the elastic portion of demand. If we believe an analyst can improve collusion by 50%, we should engage them at a cost up to $7,587 per year ($174,241 in 2016 dollars) for their time and needed resources. While this estimate for resource dedication may be a bit simplistic, it is a good starting point to frame a discussion with cartel members.

Of course ideally we, as humans with well intentions, would prefer the cartel member to not exert market power by means of collusion.

 R code is available on Github under the JEC repo.