COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE - Specialized Research AI at MIT

COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE
COMMITTEE, 1880–1886

Franco Mariuzzo and Patrick Paul Walsh*

Abstract—Using weekly spot and future commodity prices in Chicago and
New York, we construct expected transportation rates for grain between
these two cities, expected inventory levels in New York, and realized errors
in the expectations of such variables. We incorporate these exogenous com-
modity market dynamics into Porter’s (1983) structural modeling of the
Joint Executive Committee Railroad Cartel. As in Porter, we model mar-
ginal cost as a parametric function of (instrumented) output, among other
factors. Unlike Porter, we model pricing over marginal cost as a nonparamet-
ric function of a set of variables, which include expectations of deterministic
demand cycles and cartel stability. We estimate the pricing and demand
equation simultaneously and semiparametrically. Our estimated weekly
markups during periods of cartel stability are shown to reﬂect optimal collu-
sive pricing over deterministic business cycles, as modeled in Haltiwanger
and Harrington (1991). Periods of cartel instability are proven to be triggered
by realized mistakes in expectations of New York grain prices.

Introduction

T HE Joint Executive Committee (JEC) Railroad Cartel

acted as a legal cartel that transported grain between
Chicago and New York during the period 1880 to 1886.1
A key issue for this paper, and for most of the literature on
cartels, is to understand how a cartel sustains itself over deter-
ministic demand cycles and in the presence of unexpected
demand shocks, among other factors. Our innovation is to
construct demand cycles and shocks from weekly data on
spot and future grain prices on the Chicago and New York
stock markets. Optimal inventory management of grain in
New York during the nineteenth century, as written down
by Coleman (2009), allows us to employ commodity prices
to construct the expected rate for grain conveyance between
Chicago and New York. Inventory management over the
opening and closing of the Great Lakes is shown to generate
anticipated expected deterministic cycles for transportation
modalities in terms of expected transportation rates. We also
employ spot and future commodity prices in New York to

Received for publication June 17, 2009. Revision accepted for publication

June 7, 2012.

* Mariuzzo: University of East Anglia; Walsh: University College Dublin.
Andrew Coleman kindly gave us the weekly transportation price data that
he collated from the Aldrich (1893) Senate report and the weekly com-
modity price data that he collated from the New York Times. This includes
transportation rates on Great Lakes and canals, as well as commodity spot
and future prices in the Chicago and New York Stock Market Exchanges
from 1878 to 1886. These data allow us to model the impact of commodity
markets cycles and shocks on the price and quantity movements in the Joint
Executive Committee Railroad Cartel. This paper was presented at the IOS
conference in Boston 2006, and 2009, CEPR/IIIS productivity workshop
in Dublin 2006, and EARIE 2006 in Amsterdam. We thank Silvi Berger,
Robert Clark, Gregory Crawford, Peter Davis, John Haltiwanger, Joseph
Harrington, Mike Harrison, Julie Mortimer, Ariel Pakes, Robert Porter,
Paul Scott, John Sutton, Chad Syverson, and Ciara Whelan for helpful
comments. We are extremely grateful to referees for helpful suggestions
that have greatly improved our paper.

A supplemental appendix is available online at http://www.mitpress

journals.org/doi/suppl/10.1162/REST_a_00320.

1 The JEC was in operation before the establishment of the Interstate
Commerce Commission in 1887 and the passing of the Sherman Act in
1890.

model scheduled inventory levels. Finally, we quantify real-
ized errors in the expectations of transportation rates and
inventory levels, created by unexpected shocks to the grain
market in New York. We incorporate such previously omit-
ted data into Porter’s (1983) analysis of equilibrium pricing
to generate estimates of weekly price-cost margin (degree
of oligopoly power) dynamics that were not previously esti-
mated and documented for this cartel. The pricing equation
that Porter estimates is a generalized ﬁrst-order condition
for the log of pricing (ln GRt) in an imperfectly competitive
homogeneous goods market:

ln GRt = − ln

1 +

+ ln MCt.

(1)

(cid:2)

(cid:3)

θt
η

Our empirical strategy is to estimate this equation with ran-
dom errors. We work with precisely the same functional form
for cost employed by Porter. We impose a log-linear marginal
cost function (ln MCt) that, among other factors, depends on
the log of output (demand). Thus, a model of demand will be
simultaneously estimated with pricing. Our ﬁrst innovation is
to enrich the modeling of demand, and hence marginal cost,
by expanding its primitives. We then compute the JEC overall
elasticity of demand, η, to be made up of the own- and cross-
price elasticity, as theoretically outlined in Holmes (1989).
Thus, demand will be speciﬁed to depend on the expected
rate of the outside transportation option, among other new
controls.

Our second innovation is motivated by the work of Appel-
baum (1982). We model Porter’s “hidden regime” (omitted
variables), the log transformation of the markup, − ln(1+ θt
η ),
as a nonparametric function of a set of variables that reﬂect
expectations of demand cycles and cartel stability.2 We wish
to show how the inclusion of previously omitted data on
expected demand cycles and demand shocks into this non-
parametric function will allow us to estimate rich price-cost
margin dynamics for this cartel.3 We include the follow-
ing observables in a nonparametric modeling of what was
Porter’s unobservable markup: the expected rate of the out-
side transportation option (which has a one-period-forward

2 Theory based on repeated games suggests that Bresnahan’s (1989) θ is
not static, as the intensity of price competition (market share rivalry) can
vary over time. The way one models demand affects the trade-off between
one-shot gains and discounted losses in incentive compatibility constraints
in repeated games. This has been shown to generate very different time
paths of the conduct and equilibrium price-cost margin (see, for exam-
ple, Green & Porter, 1984; Rotemberg & Saloner, 1986; Haltiwanger &
Harrington, 1991; Fabra, 2006). Genesove and Mullin (1998) provides us
with a “sweet” overview of the empirical issues surrounding the estimation
of the generalized ﬁrst-order condition for pricing in homogeneous good
industries.

3 We provide a direct comparison of our results with Porter and for this
reason make no attempt to separate out this function in pricing from costs
using the techniques suggested in Berry, Levinsohn, and Pakes (1995),
overviewed in Mariuzzo, Walsh, and Whelan (2010).

The Review of Economics and Statistics, December 2013, 95(5): 1722–1739
© 2012 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology
Published under a Creative Commons Attribution = NonCommercial 3.0 Unported (CC BY = NC 3.0) License

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE

1723

expectation on the future of the New York commodity price);
a control for additional demand cycles created by inventory
management in New York (we expect increasing pressures to
build up inventories as the number of weeks that the Great
Lakes are open decrease or, if in Lakes Great closed periods,
New York can bargain credibly with railroads to lower prices,
as the number of weeks to Great Lakes opening decrease);
ﬁnally, we include a probability of cartel stability in the next
period, in the markup term.

Instability in this cartel is deﬁned, as documented by Porter,
as a ﬁnite period of revision to low markups to sustain the
cartel; this is modeled in Green and Porter (1984). We ﬁnd
that Porter’s estimates of such ﬁnite revision phases are very
similar to the periods of cartel instability, deﬁned in the
Aldrich Report (1893) (our new data source). Following Elli-
son (1994) we face off anticipated demand cycles against
unanticipated demand shocks (realized mistakes from a his-
torical expectation of a New York future) as triggers of ﬁnite
periods of instability, all else equal. Consistent with Ellison,
we ﬁnd that realized errors in past expectations, created by
unexpected commodity price shocks in New York, are the
reason for the JEC’s lapsing into ﬁnite periods of instability.4
Incorporating data on deterministic cycles and unexpected
demand shocks into our econometric modeling of a gen-
eralized ﬁrst-order condition for pricing, in an imperfectly
competitive homogeneous good market, gives us a richer
model of the factors driving JEC pricing over marginal cost.
This is done using an innovative semiparametric approach
for estimating pricing simultaneously with demand. While
the results are broadly consistent with Ellison in terms of our
analysis of stability and demand, we do ﬁnd that the antic-
ipated cyclical nature of demand is important for markups
(pricing). We show that the JEC was able to manage opti-
mal collusive pricing over deterministic cycles as modeled
in Haltiwanger and Harrington (1991), also reiterating the
ﬁndings in Borenstein and Shepard (1996). In section II, we
describe the data and literature. In section III, we replicate
Porter and introduce our extension. In section IV, we provide
results. Finally, we draw some conclusions in section V.

4 Several theoretical papers have discussed this problem within the JEC
from both traditional and game-theoretic frameworks. The focus of this
work has been on the causes of so-called price wars, or periods of insta-
bility, identiﬁed by Porter (1983). The key papers on this issue are Ulen
(1983), Porter (1985), Ellison (1994), Rotemberg and Saloner (1986), and
Vasconcelos (2004). The core aim of the Ellison paper is to try to under-
stand plausible trigger strategies that could send the cartel into ﬁnite periods
of punishment. His main ﬁnding is that unexpected demand shocks in the
AR(1) residual of demand triggered the price war. This is tested against
a Rotemberg and Saloner (1986) effect, where the incentive compatibility
constraint (ICC) comes under pressure when anticipated demand is high
in the current period but low in the next period. Using data on commodity
markets in New York, we intend to show that unanticipated commodity
market shocks in New York were the trigger and not the cyclical nature of
pricing as controlled by us. This is compatible with the ﬁndings of Ellison.
As in Ellison, unusual movements in the market shares of companies are
not found to be the culprit. It seems that a common external unanticipated
commodity shock was the trigger, consistent with the mechanisms in the
trigger strategy discussed in Green and Porter (1984).

II. The Literature and Data Sources

The JEC railroad cartel managed eastbound freight ship-
ments of grain, ﬂour, and provisions from Chicago to the
Atlantic Coast. Grain was by far the most important com-
modity that the cartel transported. The JEC was a legal cartel
that set ofﬁcial rates and market share allotments for compa-
nies with trafﬁc out of Chicago. In this section, we undertake
a detailed analysis of the data summarized in table 1. Some
of the data have already been used in the literature on this
famous cartel. We provide a summary of these analyses. In
the second part of table 1, we introduce our new variables.
One set of new variables comes from the Aldrich Report
(1893), which details transportation rates for all transporta-
tion modalities, including steamships, between Chicago and
New York. The report also documents detailed information
on the opening and closing of Great Lakes and canals, and
ﬁnally, reports when the JEC did not peg to a recommended
railroad rate that all railroads, including the JEC, were sup-
posed to adhere to, in the transportation of grain between
Chicago and New York. We also construct a new set of vari-
ables, described and motivated below, from weekly data on
grain commodity prices from the New York Times.

A. Porter Data

Statistics on the functioning of the JEC cartel were pub-
lished in weekly reports in the Railway Review and the
Chicago Tribune. Ulen (1979) has collated this information
into weekly time series data. Porter (1983) employed this
data set to provide empirical evidence of revisions in this
cartel to ﬁnite periods of low markups. We reproduce sum-
mary statistics for the variables that Porter used in the ﬁrst
part of table 1. The period of reference spans from January
1, 1880, to April 18, 1886, a total of 328 weeks. The key
endogenous variables for Porter, and for us, are the weekly
ofﬁcial JEC grain transportation rate, GR, and the weekly
tons of grain shipped, Q. Porter uses a Great Lakes dummy,
L, that controls for when the lakes were open and closed.
The Aldrich Report (1893) gives us more detailed data on
the timing of the opening and the closing of the ports of the
lakes and the canals. In addition to the dummies representing
the line expansion of companies (S1–S2), and the entry (S3)
and exit (S4) of companies, Porter also reports data on peri-
ods of cartel stability, measured by the binary variable PO.
PO is the variable that Ulen (1979) constructed on the basis
of internal reports of adherence within the cartel. The main
focus of the Porter’s (1983) paper was to estimate a PN binary
variable endogenously in a hidden switching regime model
that reﬂected ﬁnite switches to periods of low markups in the
JEC cartel. The Aldrich Report (1893) gives us new data to
construct a dummy, PR, on the JEC’s periods of instability.
We come back to a description of the Aldrich data below.

B. Ellison Data

Ellison (1994) extends the Porter analysis by imposing
a Markov structure on the transitions to ﬁnite periods of

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

1724

THE REVIEW OF ECONOMICS AND STATISTICS

Table 1.—Summary Statistics

Variable name

Deﬁnition

Observations

Mean

Minimum

Maximum

GR
L

PN
Q
S1

N
BIG1

BIG2

BIGQ

SMALL1

GRAll

GRLC

GRLR

PRN

GPCH
GPNY
GP1NY ,E

GR∗E

ER
MNCYE

Other

NWC

A. Porter

The ofﬁcial grain rate, in dollars per 100 pounds.
Great Lakes dummy, reported as 1 when lakes are

open, 0 otherwise.

Cheating dummy as reported in the Railway Review

and Chicago Tribune.
Estimated cheating dummy.
Total quantity of grain shipped in tons.
Dummy equal to 1 from week 28 in 1880 to week 10

in 1883, 0 otherwise. This period reﬂects the
opening of a new line by Grand Trunk Railway.
Dummy equal to 1 from week 11 in 1883 to week 25

in 1883, 0 otherwise. This period reﬂects the
opening of a new line by New York Central.

Dummy equal to 1 from week 26 in 1883 to week 11
in 1886, 0 otherwise. This period reﬂects the entry
of Chicago and Atlantic Railways.

Dummy equal to 1 from week 12 in 1886 to week 16

in 1886, 0 otherwise. This period reﬂects the
departure of Chicago and Atlantic Railways.

B. Ellisona

Number of ﬁrms (railroads).
Unusually high market share of one ﬁrm
(measure 1).
Unusually high market share of one ﬁrm
(measure 2).
Unusually high market share of one ﬁrm
(measure 3).
Unusually small market share of one ﬁrm
(measure 4).

328
328

328

328
328
328

328

328
327

327

Our New Variables
C. Aldrich Report

Grain rate of transport by all railroads in dollars per
100 pounds. Available for the period 1878–1891.
Grain rate of transport by Great Lakes and canals in
dollars per 100 pounds. Available for the period
1878–1883.

Grain rate of transport by Great Lakes and railroads
in dollars per 100 pounds. Available for the period
1878–1891.

Lakes and canals dummy, reported 1 when

GRLC > 0, 0 otherwise.

Lakes and railroads dummy, reported 1 when

GRLR > 0, 0 otherwise.

A dummy equal to 1 if the JEC grain rate was equal
to the Chicago–New York grain rate that railroads,
including the JEC, tried to peg to; 0 otherwise.

328

138

190

138

190

328

Estimated PR dummy (see section IIIB).

328
D. New York Times
328
328
320

Chicago spot (call) corn prices.b
New York spot (call) corn prices.b
New York future corn prices for delivery within the

month.b

Proxy for transportation rates of competitors =

(GP1NY ,E − GPCH ).

Error in expectations = (GP1NY ,E − GPNY ).
Proxy for marginal net convenience yield =

(GPNY − GP1NY ,E).

Yearly de cumulative number of weeks to the
opening of Great Lakes (0 when open).

320

320
320

328

0.246
0.573

0.619

0.750
25,384
0.424

0.067
0.495

0.486

0.434
11,632
0.495

0.046

0.209

0.433

0.496

0.015

0.123

4.351
1.072
[1.091]
1.141
[1.241]
1.971
[1.174]
1.139
[1.230]

0.254

0.153

0.627
0.548
[0.569]
0.680
[0.710]
0.945
[0.516]
0.700
[0.737]

0.059

0.046

0.125
0

0
4,810
0

3
0.130
[0.040]
0.169
[0.185]
−0.148
[0.158]
0
[0.116]

0.140

0.063

0.400
1

1
76,407
1

5
3.156
[2.971]
3.975
[4.235]
4.888
[2.973]
6.116
[5.757]

0.400

0.288

0.185

0.044

0.110

0.295

0.421

0.579

0.765

0.759

0.889
1.105
1.097

0.208

−0.008
0.010

0.494

0.424

0.428

0.209
0.187
0.186

0.085

0.058
0.020

0.603
0.817
0.815

0.010

−0.393
−0.045

4.159

6.282

9.311

10.636

1.482
1.946
1.964

0.750

0.554
0.123

25,384

6,973

10,517

41,485

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

NWO

Yearly cumulative number of weeks the lakes remain

˜Q

open (0 when closed).

Estimated output business cycle (estimated using

Hodrick and Prescott’s (1997) ﬁlter).

(cid:4)

a In square brackets the values computed by Ellison. Because of a different way of averaging over the ﬁrst twelve weeks, our reproduction of Ellison’s variables is slightly different from the original. The ﬁrst of
. With the market share of ﬁrm i in period t deﬁned as

Ellison’s variables, BIG1t , is aimed at capturing a particularly high market share for one of the ﬁrms in the cartel, and its value is computed as max
sit ≡ (ln Qit − 1
j ln Qjt ), and the term sit denoting the average market share of ﬁrm i over the previous twelve weeks (for the ﬁrst twelve weeks the average is over any previous available week). The heteroskedastic
Nt
parameter σi indicates ﬁrm-speciﬁc standard deviations. Nt labels the number of ﬁrms in the cartel in period t. BIG2t is a variant of BIG1t , with the only difference being that the market share is deﬁned with sit ≡ Qit
.
Qt
BIGQt is another variant of BIG1t , which uses sit as deﬁned in BIG1t , but computes the average sit over the allotted market shares ait . The last of Ellison’s variables is SMALL1t . Its role is to detect an unusually small
market share for one of the ﬁrms in the cartel. It is calculated each period t as the min
> 0, and thus
have SMALL1t equal to 0.

, where sit and sit are those earlier deﬁned for BIG1t . In our data, two observations have min

(sit −sit )
σi

min
i

, 0

(cid:6)

(cid:5)

bWeekly average of daily prices, where a daily price is the average between the minimum and maximum price of the day.

COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE

1725

unobservable price wars. He estimates the coefﬁcients of the
demand and the pricing model of Porter by maximizing the
joint likelihood of a system of demand, pricing, and paths of
transitions into and out of cartel instability. He imposes an
autoregressive structure for the residuals of the Porter demand
function and ﬁnds evidence of hidden regimes in demand
(omitted variables). The core aim of Ellison’s paper was to try
to understand plausible trigger strategies that could send the
cartel into ﬁnite periods of punishment. He tested, but found
little evidence of, four triggers constructed from disaggre-
gated railroad-level data, each designated to proxy for signals
of cheating by ﬁrms inside the cartel. Computing these inter-
nal ﬁrm-level variables requires having railroad-level data on
assigned quotas and actual market shares. We replicate his
variables for use in our empirical equation of cartel stability.
We document our reproduction in table 1 and detail the con-
struction of each of these variables in a footnote to the table.
Overall, Ellison found evidence that random demand-side
shocks triggered transitions to cartel instability. This will be
shown to be consistent with our results, but we model these as
actual shocks in the New York commodity market. Unlike us,
he ﬁnds no evidence that cyclical or seasonal demand cycles
had any impact on the general run of pricing.

C. Aldrich Data

The Aldrich Report (1893) provides data on weekly trans-
portation rates for shipments of grain between Chicago and
New York over the Great Lakes and canals (GRLC), over
Lakes and Railroads (GRLR), and the rate recommended
for all railroads (GRAll). Summary statistics are provided
in table 1. To understand the nature of these prices, we need
to summarize the historical context.

To facilitate the movement of grain from the Great Plains
to Europe after harvesting, there was a rush to get the com-
modity over the GRLC before they closed, to New York for
storage in grain elevators (huge warehouses). The slowest and
least expensive method was to transport to Buffalo by ship
via the Great Lakes and then on to New York along the Erie
Canal (purposely enlarged during in 1836 and 1862). This
took approximately three weeks. A faster and more expen-
sive method, taking ten days, was to ship it to Buffalo and
then transport it by rail to New York. This was useful partic-
ularly as the canals would freeze up before the ports of the
lakes. Transportation over the Great Lakes was not available
between November and late April, as both the canals and the
ports of the lakes were frozen. The fastest and most expensive
method, available all year, was to send grain over three days
by rail to New York.

Having the rates for shipments on different modalities in
the Aldrich Report (1893), we can recover more detailed
information on the timing of lake opening and closing, rele-
vant for the lake reliant methods of shipping: the lakes and
canals (LC) or the lakes and railroads (LR). Table 2 doc-
uments yearly information on the number of weeks that
Porter’s-Lakes, LC, and LR remain open. Canals freeze

Table 2.—Number of Weeks Lakes Open (L, from Porter), Lakes and
Railroads Open (LR, from Aldrich), Lakes and Canals Open (LC, from
Aldrich), and Number of Weeks of Stability based on PO (from
Porter), PR (from Aldrich), and Their Estimates

Year Number of Weeks

PRN

1880
1881
1882
1883
1884
1885
1886
Total

52
52
52
52
52
52
16
328

34
28
33
33
31
29
0
188

33
29
35
31
32
30
0
190

33
26
32
27
.a
.a
.a
118

52
15
48
47
22
12
7
203

52
26
41
52
33
26
16
246

51
26
40
49
26
24
15
251

52
26
41
52
33
29
16
249

a Not available for the period 1884–1886.

before the ports of the Great Lakes, and hence we see longer
periods of lakes open in the case of LR. Yet the duration of
Great Lakes open using L, LC, and LR dummies is very sim-
ilar, and for that reason, in our empirical analysis we stay
loyal to Porter and employ his Lakes dummy, L.

The Aldrich Report (1893) also gives us new data on peri-
ods of nonadherence of the JEC to recommend transportation
rates set for grain across all railroads. We use this to construct
a dummy, PR, on the JEC’s periods of instability, as we detail
in section IIIB, where PO (as previously deﬁned) and PR are
used as starting values to endogenously estimate the hidden
switching regimes PN and PRN. A brief overview of the
behavior of these four variables is given in tables 1 and 2.
From table 1, it emerges that PO and PR are different, as the
number of periods of stability in cartel pricing in the JEC are
greater using the PR dummy. Stability is broken down year
by year in table 2. The presence and variation of instability
is different when we compare PO and PR. Interestingly, we
see that PR is comparable to the PN cheating binary variable
estimated endogenously by Porter, and PRN has little added
value. Our use of PR as data that do not need to be estimated
proves to be important in our empirical methodology, and we
justify our position on this further in our empirical section.

D. New York Times Commodity Data

Our core ambition in this paper is to reexamine the func-
tioning of the JEC cartel with commodity market data that
control for deterministic pricing cycles between Chicago
and New York, expected inventory levels in New York, and
unexpected errors in expectations in these variables. Table 1
offers deﬁnitions and summary statistics for the Chicago and
New York (spot and future) commodity prices and the three
variables that we construct from commodity prices.

We turn ﬁrst to the construction of a variable from com-
modity prices that will control for expected pricing cycles. In
terms of transporting grain from Chicago to New York, eleva-
tors tried to stock inventories supplied by the cheap and slow,
but closed during the winter, mode of transportation. During
the period 1878 to 1890, Coleman (2009) estimates that 95%
of corn was transported in the open water season and was
shipped by the Great Lakes. This created interesting demand
cycles for the JEC to price against. Coleman writes down a

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

1726

THE REVIEW OF ECONOMICS AND STATISTICS

Figure 1.—Modes of transportation rates: Lakes and Canals (GRLC, Dashed Line), JEC Railroads (GR, Plus Sign), Expected Outside
Transportation Option (GR∗E, Line), Lakes Open Period (L = 1, Diamond at .5), Cartel Instability Period (PR = 0, Triangle at 0)

5
.

4
.

3
.

2
.

1
.

5
.

4
.

3
.

2
.

s
e
t
a
R

5
.

4
.

5
.

4
.

s
e
t
a
R

3
.

2
.

s
e
t
a
R

3
.

2
.

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

(a) Year 1878

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

(b) Year 1879

1
.

5
.

4
.

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

1
.

5
.

4
.

s
e
t
a
R

3
.

2
.

s
e
t
a
R

3
.

2
.

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

(d) Year 1881

1
.

5
.

4
.

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

(e) Year 1882

1
.

5
.

4
.

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

(f) Year 1883

1
.

5
.

4
.

s
e
t
a
R

3
.

2
.

s
e
t
a
R

3
.

2
.

s
e
t
a
R

3
.

2
.

1
.

1 5 9 13 17 21 25 29 33 37 41 45 49 53
weeks

(g) Year 1884

(h) Year 1885

(i) Year 1886

model and provides empirical evidence that the weekly dis-
crepancy between the spot commodity price in Chicago and
the future commodity price in New York is a good measure
of the expected transportation and storage costs of moving
grain from Chicago to New York. We use this variable in
both our demand and pricing equations. While it is common
to include the price of a competing product in the demand
side, we also include it in the pricing equation and model the
JEC as a Stackelberg price follower. Outside transportation
options over the GRLC were by far the major players in the
shipment of grain from Chicago to New York. The railroads
that made the JEC cartel were small players and hence were
followers in a global pricing system. We deﬁne the expected
outside option in transportation rates facing the JEC as5

GR∗E
t

≡ Et(GPNY

t+1) − GPCH

(2)

5 Our proxy is also valid when the Great Lakes are frozen, since compe-
tition from non-JEC railroad companies was present, as was competition
from inventories from the Great Lakes in New York.

where GPCH and GPNY are the commodity prices of grain set
in the Chicago and New York stock markets. The speed of the
delivery by the JEC cartel ensured that if any weekly top-ups
were necessary in New York to meet demand, the railroad
cartel would oblige. The +1 in the expectation of equation
(2) postulates that it takes one period to ship the grain via the
Great Lakes from Chicago to New York.

The expected weekly transportation rate between Chicago
and New York GR∗E has interesting dynamics all year. Using
data in the Aldrich Report (1893), one can ﬁnd the actual rates
for competing modes of transportation that used the LC or LR
options. This allows us to compare and validate our expected
outside option variable constructed from commodity prices.
Figure 1 relates the pattern of weekly shipment rates between
Chicago and New York in each year in terms of the JEC ofﬁ-
cial rate (Porter’s GR), the rate over the Great Lakes and
canals (GRLC), and the expected transportation rate GR∗E,
as deﬁned above. We notice that GR∗E shares the same trends
of the JEC rate, GR, and the rate for shipments over the Great

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE

1727

Table 3.—Transportation Rate Correlations

GRLC

GRLR

GR∗

Figure 2.—Demand (Q, Dashed Line), Estimated Demand Business
Cycles ( ˜Q, Bold Dashed Line), Lakes Open Period (L = 1, Diamond at
80000), Cartel Instability Period (PR = 0, Triangle at 0)

GR
GRLC
GRLR
GR∗E

1.00 (328)
0.65† (138)
0.79† (190)
0.65† (320)

1.00 (138)
0.86∗∗ (137)
0.81∗∗(135)

1.00 (190)
0.62∗∗∗ (187)

1.00 (320)

In bracket number of observations. Signiﬁcant at 95% conﬁdence level.

Lakes and canals, GRLC.6 Table 3 shows a strong positive
correlation between GR∗E and the rates of alternative modes
of conveyance. In Figure 1 we also plot LC transportation
rates and our expected transportation rate in the years before
the JEC (1878–1879) in ﬁgures 1a and 1b and in the years of
the JEC (1880–1886) in ﬁgures 1c to 1i. The most striking
feature of transportation rates occurs in the Great Lakes open
period. During Lakes open, the rates decrease up to harvest-
ing, then increase as we move toward the closing of the lakes.
This effect is created by elevators in New York building up
stocks for the winter. We also observe that during Great Lakes
closed, the rates increase and then decrease. Such downward
trends on railroad rates in the weeks before lakes opening
seem to reﬂect declining railroad bargaining power as the
time to lakes opening shortens. These cyclical movements in
the expected transportation rates before and after the forma-
tion of the JEC suggest to us that the opening and closing of
the Great Lakes had effects on demand and pricing (via the
internal incentive compatibility constraint, ICC) in the weeks
running up to their closing and opening. This should be cap-
tured by our expected weekly transportation rate between
Chicago and New York GR∗E variable, which will control
for anticipated pricing cycles, pressures outside the control
of the JEC, at least one period ahead.

The presence of deterministic cycles created by harvesting
and inventory management over the opening and closing of
the Great Lakes leads us to control for anticipated demand
cycles that look further into the future. We do this by creating
two variables that reﬂect the cumulated number of weeks that
the lakes remained open at a given point in time in a year,
NWO, and the countdown on the number of weeks until they
reopened at a given point of time in a year, NWC. NWO starts
with a value of 1 associated with the ﬁrst week the Great
Lakes are accessible to navigation and reaches its maximum
the week prior to the their freezing again. The variable is set
to 0 during the period in which the lakes are closed for nav-
igation. NWC has its maximum the ﬁrst week the lakes are
not accessible to navigation and reduces to a value of 1 the
week before the lakes reopen to navigation. The variable is set
to 0 during lakes open. Therefore, rather than having a sim-
ple lakes open and closed dummy, we have varying degrees
of pressures coming from the expectations of lakes opening
and closing represented by NWC and NWO, which together
exhibit an asymmetric sawtooth proﬁle. The latter reﬂects the

6 In order to make the ﬁgure more readable, we have purposely omitted
the lakes and railroads rate (GRLR). Its yearly trends are comparable to
those of the variable GRLC but with slightly greater values.

0
0
0
0
8

0
0
0
0
6

0
0
0
0
4

0
0
0
0
2

y
t
i
t
n
a
u
Q

104

156
weeks

208

260

312

pressure to transport grain to inventories in New York after the
harvest to avoid exposure to high transportation rates during
the winter. The former embodies grain elevators’ increased
ability to bargain with the JEC as the weeks to lakes open-
ing come nearer. We can see some evidence of these forces
at work by plotting in ﬁgure 2 smoothed cycle of the JEC
shipments of grain with our lakes open and cartel stability
dummies.7 Inventory management in New York made ship-
ment cycles for the JEC over the lakes open and closed regime
more complex. We have four periods of lakes open when the
cartel is generally stable. Shipments fall at ﬁrst in lakes open
regimes but then rise after harvesting, up to and beyond lakes
closing. Inventory management and the closing of the lakes
create an interesting U-shaped demand cycle for the cartel.8
Could controls for NWC and NWO reﬂect week effects in
current demand rather than expectations of demand cycles?
We do control for current output and the expected pricing
from outside options, but we will verify, postregression, that
for the same level of current demand, in either lakes open
or closed, estimated markups are higher going into a period
of growing demand and lower going into a slump. This will
reﬂect the broader role of expected demand cycles created by
the lakes opening and closing and, more important, optimal
cartel pricing over such.

We now introduce a control for inventory levels con-
structed from commodity prices in New York. Elevators had
the key role of storing the commodity. They solved intertem-
poral (dynamic) models and determined the optimal amount
of grain to keep in stock. A by-product of their optimiza-
tion was the demand for transportation. We relate the amount
of grain available as inventory in elevators in period t, Yt,

7 We make use of Hodrick and Prescott’s (1997) ﬁlter and estimate
smoothed business cycles, which we denote with ˜Q. We plot those against
the actual demand in ﬁgure 2.

8 We count only six Great Lakes open spells during the cartel period
because the cartel was declared illegal before the opening of the lakes in
1886.

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

1728

THE REVIEW OF ECONOMICS AND STATISTICS

Figure 3.—Marginal Net Convenience Yield Function

the depreciation commodity rate (δ) and the foregone long-
run interest rate (r), can be used to identify the expected
equilibrium level of our second key variable: the marginal
net convenience yield. Without loss of generality, in the rest
of the paper, we use a simpliﬁed version of equation (5) that
relies on setting δ = r = 0, so as to have

−
GP E GP+
(

)
1

MNCY (Y E

t ) ≡ MNCY E

= GPNY

− Et(GPNY

t+1).

(6)

-u

Y-F – u

to the amount of grain imported (mainly from Chicago) in
period t, Mt, to the contemporaneous consumption (mainly
the amount of grain shipped to Europe), Ct and to lagged
inventory, as

Yt = Yt−1 + Mt − Ct.

(3)

The volume imported is the result of transportation by ship
or train.9 Similar to Thurman (1988) and Pindyck (1994), we
formulate the cost of holding inventories as the sum of a unit
cost for the physical use of grain elevators, u, and a func-
tion of current inventories, expected future consumption and
current commodity prices, F[Yt, Et(Ct+T ), GPt]. We assume
the function F to be well behaved, strictly convex in Y , and
increasing in GP and in the expectation variable CE. Under
optimal inventory management, the negative of the marginal
cost of inventories is the net beneﬁt of an extra unit of inven-
tory and is known in the literature as marginal net convenience
yield. It is here deﬁned as10

MNCY (Y ; u) ≡ −FY − u,

(4)

where the partial derivative of F with respect to the argument
Y is assumed negative, FY < 0, and the cross-derivatives FYCE and FYH are set to be 0. We sketch the marginal net con- venience yield function against the inventory level in ﬁgure 3. The presence of future contracts with all arbitrage opportu- nities exploited is an optimal allocation of inventories across time, and we expect the following equality condition to hold: MNCY (Y E t ) = GPt − (cid:3) T (cid:2) 1 − δ 1 + r Et(GPt+T ). (5) That is, the difference between a spot commodity price and the future commodity price T periods ahead, discounted for 9 Given that railroads can convey grain within three days, we assume that they are able to provide immediate delivery (within the week). In addition, we postulate that the alternative modes of transportation meet deliveries within the following period. In this way, Mt casts the sum of transport by railroads (R) and by lakes (L), as follows: Mt ≡ M R t 10 A term originally introduced by Working (1949). + M L t−1. Because of the presence of uncertainty in future commod- ity prices, equilibrium equation (6) backs out an expected optimal level of inventories that is negatively related to the marginal net convenience yield of the period in question as documented in ﬁgure 3. As demonstrated in Coleman (2009), New York (NY) was by far the main receiving city from Chicago (CH). So we can use the difference between a spot and a future price in New York as a good measure of the (expected) marginal net convenience yield or the level of inventories. Finally, we consider our control for unexpected demand shocks. This is our third variable that we construct from com- modity prices, and it will be a key variable in our modeling of cartel stability. It is derived as the difference ERt = Et−1(GPNY t ) − GPNY t . (7) Equation (7) turns out to be negative if the realization of the commodity price in New York at time t—a spot price at time t in New York—exceeds its corresponding expectation for- mulated at time t − 1. The error in the expectation variable, equation (7), will generate a difference between the expected and realized marginal net convenience yield, as well as the outside transportation rate. A negative error in expectations at time t, ERt < 0, that is, an unexpected jump in commod- ity prices, signs the inequalities MNCY E t−1 > MNCYt−1 and
GR∗E
t−1. From the point of view of railroads in the
JEC, their shipments should have been greater (inventories
should have been higher) and transportation rates higher for
all modes of transportation. In monetary terms, the cartel lost
an opportunity to extract higher revenues and reduce costs in
the presence of economies of scale. This sort of event in the
presence of imperfect information about other ﬁrms’ actual
price and shipments can trigger a ﬁnite revision to a low
markup to maintain cooperation into the future, as modeled
in Green and Porter (1984).

t−1 < GR∗ The analysis in this section has summarized the approach and data Porter and Ellison used. We have also noted our use of new data constructed from weekly commodity price data available in the New York Times. This includes our expected transportation rate (outside option to the JEC), the expected level of inventories in New York (negatively related to the marginal net convenience yield), and a real- ized error in the expectations of commodity prices in New York leading to mistakes in shipments and rate setting across all modes of transportation. These new variables allow us to control for the anticipated demand cycles and for realized mistakes created by unexpected demand shocks. We hope l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE 1729 to get new insights into the determinants of JEC pricing, demand, and stability. The data on transportation prices on different modalities between Chicago and New York in the Aldrich Report (1893) were very useful in conﬁrming the value of using the expected transportation rate, constructed from commodity prices. The Aldrich Report provides us with new data on JEC cartel adherence, our PR variable. This will be important in our empirical methodology, as we treat cartel stability as observable. We justify this further in the next section, where we extend the structural model in Porter. III. The Model In this section we ﬁrst reproduce the model and results of Porter. Step by step, we introduce our extensions and estima- tion procedure. Like Ellison, our extensions are done to tackle the instrumentation of marginal cost and to explore how best to model the markup term (cartel stability, among other factors) and the two components of the estimated pricing equation. Unlike Ellison, we employ a two-step semipara- metric procedure that is largely motivated by the use of our commodity market data and our focus on the estimation of the time-varying markup driven by demand cycles, all else equal. A. Porter Porter speciﬁes the following pricing equation, where lowercase letters denote natural logs of the variables: grt = β0[t] + β1qt + β2St + β3It + U1t, (8) where U1t is the error term, assumed NID with mean zero and variance σ2 1. The ﬁrst part of the pricing equation, β0[t]+β1qt +β2St, is the marginal cost associated with an aver- age ﬁrm in the market at time t, whose particular speciﬁcation is the result of the assumption of log linearity in each ﬁrm’s heterogeneous cost functions. Porter denotes with St a set of structural dummies that accommodate entry and exit and with β0[t] a constant augmented with month dummies. Mar- ginal cost is modeled to depend on output, which needs to be instrumented. Porter estimates the following output equation, qt = α0[t] + α1grt + α3Lt + U2t, (9) where grt is the natural log of the JEC grain rate per bushel shipped in week t and qt the natural log of the total quantity of grain shipped by railroad that week.11 Lt is a dummy equal to 1 when the Great Lakes are open to shipping and 0 oth- erwise. This is the variable (instrument) that is included in demand but not in the pricing equation. U2t is the error term, assumed NID with mean zero and variance σ2 2. The demand and pricing equation error terms are allowed to be correlated, and their covariance is denoted with the parameter σ12. The coefﬁcient α1 is expected to be negative. The time-varying vector of coefﬁcients α0[t] encompasses a constant and month dummies. The second component remaining in the pricing equation is the markup. Following the ﬁrst-order condition for pricing, derived in Porter, the markup is equal to − ln(1 + θt/α1). In the Porter pricing equation, β3It = − ln(1 + θt/α1), where It is a dummy that equals 1 during a collusive regime, and 0 otherwise. This is motivated by Green and Porter (1984) to capture a time-varying average conduct parameter θt. The market elasticity of demand is constant over time. We should expect β3 to be positive in the term, β3It = − ln(1 + θt/α1), as α1 is expected to be negative. The variation in the markup comes from the varying average conduct parameter. The key assumption is that there are only two discrete regimes in the markup—one that is collusive and one that is revisionary. Theory predicts that θt is higher during collusive regimes. We intend to model the markup as a nonparametric function of external demand cycles that drive week-on-week dynamics. This is a key departure from both Porter and Ellison in our modeling of markups, although Ellison does interact It with indices of anticipated cycles, which moves a step toward our approach. When It is a known exogenous variable, using the PO cheating variable collected by Ulen (1979), Porter estimates equations (8) and (9) using 2SLS. He estimates demand and pricing simultaneously and uses the exclusion restrictions observed in the deﬁned pricing and output equations. We see that the Great Lakes dummy is excluded from the pricing equation, and the structural dummies and cartel breakdown dummy are excluded from the demand equation. Identiﬁca- tion of the markup comes from an explicit functional form for both marginal cost and the markup term: β3It. Porter felt that PO was not very reliable and treated It as an unknown. Hence, he estimated an endogenous switching (hidden) regime in the pricing equation over and above marginal cost following the procedure that originated in Lee and Porter (1984).12 This is how he constructed his estimated PN binary variable. B. Aldrich Data on Cartel Stability As we mentioned in section II, we prefer the PR binary variable, recovered from the Aldrich Report (1893), to reﬂect cartel stability or not. We wish to treat PR as data, which will allow us to employ a two-step estimation procedure in our modeling of the markup. The expected probability of stability in the next period will be just one additional component of the markup. We intend to estimate this probability in a ﬁrst stage and then condition the markup on it in a second step that estimates prices and output in a simultaneous equation. The two-stage least squares (2SLS) and maximum likeli- hood (ML) estimates reported in columns 1 to 4 of table 4 11 Given the possibility of secret price cutting, one can interpret grt as a weighted average of list prices. 12 In this case, the demand and pricing equation error terms are assumed to be normally distributed. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 1730 THE REVIEW OF ECONOMICS AND STATISTICS ∗ ∗ ∗ ) 7 5 0 . 0 ( 6 0 4 . 0 − ∗ ∗ ∗ ) 1 4 0 . 0 ( 5 9 2 . 0 − ) 6 9 0 . 0 ( ∗ ∗ ∗ 4 3 4 . 0 ) 9 2 0 . 0 ( 2 5 0 . 0 ) 0 5 0 . 0 ( s e Y o N 8 2 3 6 1 r g ) 0 1 9 . 0 ( 2 6 3 . 1 − ) 4 3 0 . 0 ( 5 4 0 . 0 ) 8 5 0 . 0 ( 2 0 0 . 0 ) 9 5 0 . 0 ( 3 6 0 . 0 ) 7 8 0 . 0 ( 5 2 0 . 0 − ∗ ∗ ∗ ) 8 9 0 . 0 ( 4 0 8 . 0 − ) 7 7 0 . 0 ( ∗ ∗ ∗ 2 3 2 . 0 − ∗ ∗ ∗ ) 9 3 0 . 0 ( 2 6 1 . 0 − ∗ ∗ ∗ ) 4 0 1 . 0 ( 5 2 9 . 0 − ∗ ∗ ∗ ) 2 2 1 . 0 ( 0 0 4 . 0 − ) 4 1 5 . 0 ( ∗ ∗ ∗ ) 6 3 2 . 0 ( 4 3 4 . 0 − ) 8 8 0 . 1 ( ) 1 4 1 . 0 ( ) 4 8 0 . 0 ( 4 2 0 . 0 ) 1 4 0 . 0 ( 6 3 0 . 0 ) 2 7 0 . 0 ( 3 3 0 . 0 − ) 3 7 0 . 0 ( ∗ 4 9 1 . 0 − ) 1 1 1 . 0 ( ∗ ∗ ∗ 8 8 6 . 0 ) 4 5 0 . 0 ( ∗ ∗ ∗ ) 4 2 1 . 0 ( 4 4 3 . 0 ) 4 3 0 . 0 ( ∗ ∗ ∗ 9 8 5 . 0 ) 7 4 0 . 0 ( 7 2 0 . 0 − ) 7 8 0 . 0 ( s e Y o N 8 2 3 s e Y s e Y 8 2 3 s e Y s e Y 8 2 3 ∗ 8 8 1 . 0 ) 4 0 1 . 0 ( s e Y o N 8 2 3 s e Y s e Y 8 2 3 ∗ ∗ 8 5 1 . 0 ) 0 7 0 . 0 ( s e Y o N 8 2 3 s e Y s e Y 8 2 3 ∗ ∗ ) 0 5 0 . 0 ( 2 7 1 . 0 − ∗ ∗ ∗ ) 4 7 0 . 0 ( 2 0 3 . 0 − ∗ ∗ ) 4 5 0 . 0 ( 3 7 2 0 . 0 − ∗ ∗ ∗ ) 8 2 0 . 0 ( 3 9 1 . 0 − ∗ ∗ ∗ ) 8 3 0 . 0 ( 3 6 1 . 0 − ∗ ∗ ∗ ) 7 3 0 . 0 ( 4 3 2 . 0 − ∗ ∗ ∗ ) 5 5 0 . 0 ( 3 7 1 . 0 − ∗ ∗ ∗ ) 6 3 0 . 0 ( 4 0 3 . 0 − ∗ ∗ ∗ ) 3 3 0 . 0 ( 2 1 2 . 0 − ∗ ∗ ∗ ) 6 5 0 . 0 ( 5 9 3 . 0 − ∗ ∗ ∗ ) 7 4 0 . 0 ( 5 2 3 . 0 − ∗ ∗ ∗ ) 7 4 0 . 0 ( 9 6 3 . 0 − ∗ ∗ ∗ ) 8 4 0 . 0 ( 6 3 3 . 0 − ∗ ∗ ∗ ) 1 8 0 . 0 ( 9 1 3 . 0 − ) 5 6 0 . 0 ( 8 0 2 . 0 − ∗ ∗ ∗ 1 0 2 . 0 − ) 1 9 1 . 0 ( ∗ ∗ ∗ 1 6 1 . 0 − ) 8 8 0 . 0 ( ∗ ∗ ∗ 0 3 2 . 0 − ) 5 0 1 . 0 ( ∗ ∗ ∗ 8 9 1 . 0 − ) 0 9 0 . 0 ( ∗ ∗ ∗ 1 0 2 . 0 − ) 9 1 1 . 0 ( ) 8 5 0 . 0 ( ) 3 1 1 . 0 ( ) 6 7 0 . 0 ( ∗ ∗ ∗ ) 2 7 1 . 0 ( 8 6 3 . 0 ∗ ∗ ∗ 6 2 5 . 0 ) 5 3 0 . 0 ( 7 0 0 . 0 ) 9 7 0 . 0 ( s e Y o N 8 2 3 s e Y o N 8 2 3 ∗ ∗ ∗ 0 2 4 . 0 ) 0 4 0 . 0 ( 1 1 0 . 0 ) 1 0 1 . 0 ( s e Y o N 8 2 3 ∗ ∗ ∗ 7 2 6 . 0 ) 6 5 0 . 0 ( ) 4 5 0 . 0 ( s e Y o N 8 2 3 ∗ ∗ 2 4 2 . 0 ) 2 1 1 . 0 ( s e Y o N 8 2 3 s e Y o N 8 2 3 3 5 2 . 0 ) 3 7 1 . 0 ( s e Y o N 8 2 3 s e Y o N 8 2 3 a ) N R P ( L M ) R P ( S L S 2 a ) N P ( L M ) O P ( S L S 2 a ) N R P ( L M ) R P ( S L S 2 a ) N P ( L M ) O P ( S L S 2 s e i m m u D r a e Y s e i m m u D r a e Y o N y t i l i b a t S l e t r a C f o s e i r o e h T s u o n e g o d n E d n a s u o n e g o x E g n i s U r e t r o P g n i t a c i l p e R — . 4 e l b a T 5 1 q 4 1 r g 3 1 q 2 1 r g 1 1 q 0 1 r g 9 q ∗ ∗ ∗ 0 0 2 . 9 ∗ ∗ ∗ 3 8 8 . 1 − ∗ ∗ ∗ 2 1 3 . 8 ∗ ∗ ∗ 5 8 7 . 3 − ∗ ∗ ∗ 1 1 3 . 9 ∗ ∗ ∗ 0 0 0 . 3 − ∗ ∗ ∗ 5 4 7 . 8 ∗ ∗ ∗ ∗ ∗ ∗ ) 9 3 1 . 0 ( ) 3 0 1 . 9 4 5 . 0 4 7 . 0 − 0 ( 0 − ) 5 2 7 . 0 ( ∗ ∗ ∗ ) 2 0 3 . 0 ( 5 1 4 . 0 − ∗ ∗ ∗ ) 4 0 1 . 0 ( 9 2 6 . 0 − 8 r g ∗ ∗ 6 8 5 . 1 − ) 6 0 8 . 0 ( 7 q ∗ ∗ ∗ 3 0 1 . 9 ∗ ∗ ∗ ) 5 4 1 . 0 ( 9 4 4 . 0 − ∗ ∗ ∗ ) 8 1 1 . 0 ( 5 0 8 . 0 − 6 r g ) 3 3 0 . 1 ( 8 5 4 . 1 − 5 q 4 r g 3 q ∗ ∗ ∗ 0 4 1 . 9 ∗ ∗ ∗ 3 9 0 . 4 − ∗ ∗ ∗ 7 8 9 . 8 ∗ ∗ ∗ ) 0 2 1 . 0 ( 2 6 7 . 0 − ∗ ∗ ∗ ) 6 6 1 . 0 ( 4 3 4 . 0 − ) 2 5 1 . 1 ( ∗ ∗ ∗ ) 0 5 1 . 0 ( 9 5 4 . 0 − ∗ ∗ ∗ ) 8 1 1 . 0 ( 1 8 8 . 0 − 2 r g ∗ ∗ 5 7 9 . 3 − ) 8 7 7 . 1 ( ∗ ∗ ∗ ) 4 8 1 . 0 ( 7 3 4 . 0 − ∗ ∗ ∗ ) 0 2 1 . 0 ( 5 3 7 . 0 − 1 q ∗ ∗ ∗ 7 7 1 . 9 s e l b a i r a V s n o C L r g 1 S 2 S 3 S 4 S O P N P R P N R P q s e i m m u d h t n o M s e i m m u d r a e Y . s b O 2 R 5 6 8 . 0 7 1 5 . 0 4 5 6 . 0 5 1 5 . 0 5 1 7 . 0 8 9 4 . 0 1 3 4 . 0 8 1 5 . 0 4 5 8 . 0 8 0 3 . 0 0 7 6 . 0 8 0 3 . 0 0 8 7 . 0 2 0 3 . 0 6 1 3 . 0 0 1 3 . 0 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 . ) 4 8 9 1 ( r e t r o P d n a e e L n i d e b i r c s e d l e d o m n o i s s e r g e r g n i h c t i w s e h t s e s U a . 1 . 0 < p ∗ d n a , . 5 0 0 < p ∗ ∗ , 1 0 . 0 < p ∗ ∗ ∗ : s e s e h t n e r a p n i s r o r r e d r a d n a t S COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE 1731 reproduce the results of Porter’s table 3. Columns 5 to 8 repeat the estimations, but replace his variable PO with our observed cheating variable, PR. The main aim of this exercise is to see whether PR estimates the effect of the incidence of cheating in a similar way to PN, estimated endogenously in Porter. The employment of PR, instead of PO, raises the R2 in the pricing equation from 0.32 in column 2 to 0.67 in column 6, for the 2SLS results. This is a lower value than the 0.78 in column 4, when the hidden regime is endogenously esti- mated using PO as the original variable, yet the ML estimator can increase the R2 by even more to 0.85 in column 8 when using PR as the initial variable to generate PRN. We run an independent two-sample Student’s t-test and ﬁnd that while the estimated constant elasticities of demand are similar, the estimated coefﬁcients on the cartel stability dummy in the 2SLS and ML estimators are statistically different from one another, at the 5% signiﬁcance level. The subsequent columns, 9 to 16, in table 4 are a rerun of the estimations in columns 1 to 8, with the addition of year dummies as controls in the demand equation. Porter did not have year dummies in his speciﬁcation. As expected, control- ling for year dummies in demand strengthens the explanatory power of the demand side. Again we run an independent two- sample Student’s t-test and ﬁnd the estimated coefﬁcients on the cartel stability dummy in the 2SLS and ML estimators are statistically different from one another at the 5% signiﬁcance level when using PR instead of PO. We now ﬁnd that the estimated constant elasticities of demand are different when PR is used instead of PO. When one compares columns 3 and 4 to columns 5 and 6, or columns 11 and 12 to columns 13 and 14, our 2SLS estimates using data on PR, our new variable, perform just as well as Porter’s estimated PN. Although there will be slight differences in the incidence of price wars, we have shown now that these are similar variables. In section II, we reported basic summary statistics for PO, PN, PR, and PRN. While PO is very different from the others, PN, PR, and PRN are very similar. This is conﬁrmed by ﬁgure 4, where we graph the cumulated periods of cartel stability represented by these alternative measures of cartel stability. Hence, our use of PR and inclusion of year dummies in the demand equation does not change the key Porter result. There is a hidden regime in the pricing equation that is clearly linked to the price wars that occurred intermittently in this cartel. Therefore, when we estimate our pricing equations, we choose to condition our modeling of the markup term on being in a regime of cartel stability or not as deﬁned by PR, which we treat as an observable variable that does not need to be estimated. C. Further Extensions of Porter We extend Porter’s pricing equation (8) in the following way: grt = β0[t] + β1qt + β2St + Ω1(gr∗E t , NWCt, NWOt, Et(PRt+1)) + U1t, (10) Figure 4.—Cumulated Periods of Cartel Stability: PO, Dashed Line; PN, Dotted Line; PR, Solid Line; PRN, Dashed-Dotted Line s k e e w 2 1 3 0 6 2 8 0 2 6 5 1 4 0 1 2 5 0 0 52 104 156 weeks 208 260 312 where β0[t] is the constant augmented with month dummies, deﬁned earlier in section IIA. As in Porter, we specify as determinants of the marginal costs the sum of the error term and β0[t] + β1qt + β2St. Yet we instrument his marginal cost function by including additional previously omitted exoge- nous variables in the demand equation. We extend Porter’s demand equation (9) in two ways. First, we expand the set of variables in the linear structure by introducing the expected price of the outside transportation mode. Second, we control for expected demand cycles using a nonpara- metric function. Ellison allows for hidden omitted variable regimes in demand and serial correlation in the demand residuals. We address these issues by including new con- trol variables and a lagged dependent variable in demand to allow for partial adjustments. The demand equation that we estimate is qt = α0[t] + α1grt + α2gr∗E + Ω2(MNCY E t t , NWCt, NWOt) + U2t. + α3Lt + α4qt−1 (11) Equation (11) depends on the own grain transportation rate and the expected rate of the outside transportation. U2t is an independently distributed mean zero error term. Here the parameter α0[t] includes a constant, month dummies and year dummies. To control for deterministic demand cycles, not captured by our price variables, we also include the covari- ates’ number of weeks to the Great Lakes opening and number of weeks lakes reopen and a functional form of the expected marginal net convenience yield (which we have shown to be theoretically related to the level of inventories in New York) in the unspeciﬁed Ω2 function. An additional clear point of departure is the way that we model the deterministic pricing component Ω1(·), Porter’s hidden regime. Having an estimate of Ω1(·) we can back out the price-cost margin, − θt η , from the relation Ω1t = − ln(1 + θt/η), where η is the total market constant elasticity l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 1732 THE REVIEW OF ECONOMICS AND STATISTICS deﬁned as, η ≡ α1 + α2.13 We model Porter’s unobservable with observables that control for expected demand cycles and cartel stability: the expected pricing cycles from exter- nal competition, gr∗E, anticipated future demand cycles as we come towards Great Lakes closing and opening (NWC and NWO) and expectations of the cartel being stable one week ahead, Et(PRt+1). We account for the effect on the markup of movements in anticipated future demand, as moti- vated by Haltiwanger and Harrington, through gr∗E, NWC and NWO.14 We believe that our controls are approximations to the way that Gallet and Schroeter (1995) empirically relate the markup at time t to the discounted expected value of future JEC collusive proﬁts. This is a key ingredient that prevents cartel breakdown and allows the cartel to sustain a price over marginal cost. Unlike Porter, we do not model the entire markup, Ω1t in pricing equation (10) as a revision to and from cartel stability but rather as a function of anticipated demand cycles whose effect the markup needs to be conditioned on, whether the cartel is stable or not. As a ﬁrst step, we need to describe our model of the probability of cartel stability one period ahead. To allow for persistence in regimes and unobservable behavior, we model cartel stability as an ARMA(1,1) linear probability model, PRt+1 = γ0 + γ1ELt + γ2ERt + γ3Lt + γ4Nt + γ5NWCt + γ6NWOt + γ7PRt + U3,t+1, U3,t+1 = γ8Vt + Vt+1. (12) The variable Vt+1 denotes a mean zero independently dis- tributed innovation term. PR set to 0 represents ﬁnite periods of revisions to noncollusive pricing. The interesting issue relates to what is triggering these periods of revisions, and that has been a focus of many empirical papers. We compete alternative theories against our new variables. As in Porter (1985) and Vasconcelos (2004), we account for the number of ﬁrms in the cartel, N, and for the opening and closing of the Great Lakes as internal factors that may affect the ICC. We add external controls for the countdown on number of weeks until the lakes reopen (NWC) and the cumulated number of 13 The Ω1(·) function relates to the price-cost margin in the following way. Ω1t is the value of the nonparametric function deﬁned in our log-log pricing equation (10), and U1t is the mean zero additive error term part of the same equation. The expected value of a transformation of equation (10) η ) ≡ Ω1t. Once gives us our relation of interest: E(ln GRt Ct = ln(1 + MUt we recognize that ln GRt ), where MUt denotes the markup Ct Ct and Cˆt the marginal cost, we can simplify the notation further and deﬁne . Now we recall that the Taylor series of the function ln(1 + Wt) Wt ≡ MUt Ct gives ln(1 + Wt) = Wt − W 2 + O(W 4 t ). For low absolute values of Wt, the left-hand side can be approximated by the term Wt itself. In our context, this means that the nonparametric function Ω1 is an approximation of the price-cost margin normalized to the marginal cost. This will be the way we interpret it. ) = − ln(1 + θt + W 3 t 3 t 2 14 We capture exogenous movements in current demand using the price gr∗E, month dummies, and endogenous movements in current demand are accounted for through the marginal cost component. NWC and NWO may reﬂect just week effects in current demand, but the controls are consistent with and do control for the effects of expected future demand. weeks that lakes remain open (NWO) to control for the effect of anticipated deterministic cycles modeled in Haltiwanger and Harrington. We also include, one at a time, the set of Ellison’s triggers (EL) that reﬂect internal conﬂict, thus con- trolling for unusual movements in ﬁrm-level market shares. Finally, we introduce our new variable that captures realized errors in past expectations of corn prices in New York (ER), as deﬁned in equation (7). Negative errors reveal a situation where the elevators in New York should have been transport- ing more at a higher rate over all modes of transportation. This creates a scene where the JEC is likely to have simultaneously lost the opportunity to transport more and price higher in the last period. The ﬁrms in the railroad cartel observe a boom in New York commodity prices but realize that they did not get the reward. In the presence of imperfect observation of the output and prices in other ﬁrms, this could force an unstable ICC and a period of ﬁnite competitive prices to discipline the cartel, as modeled in Green and Porter (1984). We put for- ward realized and unexpected past mistakes in commodity market price forecasts as the key factor that can threaten the cartel’s stability. As displayed in equation (10), we have conditioned the Ω1(·) function in the pricing equation on the expected cartel stability one period ahead. The expectation of cartel stability one period ahead raises concern over a possible correlation between the error terms in the system of pricing and output equations and the components of the expectation of cartel sta- bility one period ahead. These components include Ellison’s variables, the lagged error term in the cartel equation, Vt, the current cartel stability variable PRt, and our new variable that captures realized errors in past expectations of grain prices in New York (ER). It will turn out that the forward expectations on cartel stability one period ahead will depend strongly on the presence of backward-looking realized mistakes in the expectations of New York grain commodity prices. Mistakes in the past trigger a negative forward look that is modeled to persist and fade out over a ﬁnite spell. Such mistakes are externally driven by realized differences in yesterday’s future and today’s spot grain prices in New York. These mis- takes are compelled by conditions external to the cartel, such as an unexpected boom in the European market. These are fully exogenous and are not expected to affect current pric- ing and output setting in the JEC, but they can destabilize the JEC in the next period and affect future pricing. The other variables turn out not to be signiﬁcant in modeling cartel stability. We hope to model our ﬁrst step with a set of factors that are orthogonal to the errors in the pricing and output system of equations. The markup Ω1 function will be computed in a second step where we estimate the demand and pricing sys- tem together. Ω1 is modeled to be the deterministic part of pricing over and above our imposed linear model of the nat- ural log of marginal cost. Markups are easily derived: they are a nonparametric deterministic residual over and above the Porter natural log of marginal cost parametric speciﬁ- cation. Thus, the only identiﬁcation that we need to focus l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE 1733 on is that of a classical simultaneous demand and supply relation. The exclusion restrictions that we employ to identify the endogenous (natural log of the) rate of transportation in demand are cost shifters, such as the structural dummies used in Porter, and a markup shifter, which differs from Porter’s, as we employ the predicted probability of the cartel being sta- ble one period ahead rather than its observed current binary realization. As mentioned above, the latter is a function of the set of exogenous variables displayed in equation (12). Both types of shifters are correlated with the rate of trans- portation. To deal with the endogeneity of (the natural log of) quantity in the pricing equation, we employ as exclu- sion restrictions the demand shifters such as year dummies and the expectation of the marginal net convenience yield. We estimate our model with both a 2SLS estimator and an adjusted (to deal with endogeneity) Robinson’s (1988) esti- mator. The former methodology relies on a linearization of the Ω1 and Ω2 nonparametric functions. We have deliberately chosen to estimate a linearized version of the nonparamet- ric functions to better inspect and understand the partial contributions of each variable. Yet unspeciﬁed nonlinear interactions lead to a better modeling of our endogenous vari- ables and greater conﬁdence, statistical and otherwise, in our estimated markup cycles that will be shown to have interest- ing properties. Here we have described the identiﬁcation for the simpliﬁed linearized case. (Refer to the online appendix for the econometric identiﬁcation of the more cumbersome semiparametric approach.) IV. Results We present our results in three steps. In step 1, we esti- mate the weekly forward look on the probability of cartel stability in the next period. In step 2, while conditioning the markup on the probability of cartel stability in the next period, we estimate the weekly price-cost and proﬁt cycles for the JEC in a simultaneous equation model for pricing and demand. Finally, in step 3, we model the cyclical nature of our estimated markups. A. Step 1: Modeling Cartel Instability Table 5 reports the estimates of the ARMA (1,1) linear probability speciﬁcation that we have modeled in equation (12). The table highlights persistence in the state of the dependent variable, as indicated by the high value of the γ7 parameter. Once a shock brings the dependent variable from a state of stability to one of instability, it will take a number of weeks before it goes back to stability, showing evidence of Green and Porter’s (1984) theory of ﬁnite periods of punish- ments. The span of instability is embedded in the intensity of the shock that has caused the drifting away and in realizations of other opposite (in sign) future shocks. Another key variable that is signiﬁcant in explaining cartel stability one period ahead is the realized mistake in expec- tations of the commodity price in New York introduced in Variables Cons N L NWO NWC ER BIG1 BIG2 BIGQ SMALL1 PR V Table 5.—ARMA (1,1) Cartel Stability Estimation PR+1 (= 1Stability) 1 2 3 4 5 1.141∗∗∗ (0.310) −0.051 (0.072) −0.311 (0.181) 0.005 (0.008) −0.005 (0.009) 0.570∗∗∗ (0.218) 1.127∗∗∗ 1.120∗∗∗ (0.311) (0.309) −0.055 −0.058 (0.074) (0.073) −0.307∗ −0.301 (0.184) (0.181) 0.006 0.005 (0.008) (0.008) −0.003 −0.004 (0.009) (0.009) 0.572∗∗∗ 0.561∗∗∗ (0.219) (0.215) 0.026 (0.034) 0.030 (0.023) 1.143∗∗∗ 1.103∗∗∗ (0.310) (0.311) −0.046 −0.054 (0.075) (0.071) −0.310∗ −0.314∗ (0.180) (0.182) 0.006 0.006 (0.008) (0.008) −0.004 −0.004 (0.009) (0.009) 0.565∗∗∗ 0.570∗∗∗ (0.214) (0.218) −0.016 (0.024) 0.042 (0.026) 0.794∗∗∗ (0.051) 0.023 (0.066) No No 318 295/318 1/318 58/318 0.787∗∗∗ (0.051) 0.046 (0.062) No No 319 296/319 1/319 45/319 0.789∗∗∗ (0.051) 0.040 (0.064) No No 318 295/318 1/318 53/318 0.786∗∗∗ (0.051) 0.043 (0.064) No No 318 295/318 1/318 45/318 0.785∗∗∗ (0.053) 0.045 (0.063) No No 318 295/318 1/318 51/318 Month dummies Year dummies Observations R2 Obs. Obs. ll Robust standard error in parentheses. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. We employ an optimization −11.691 −11.620 −11.029 −11.863 −10.232 ˆPR+1 < 0 ˆPR+1 > 1

method that switches between the BHHH and the BFGS optimization algorithms.

equation (7). If the error ends up being negative, the spot
price at time t in New York would prove to be higher than
that expected at time t − 1. We showed in section II how that
relates to a realized mistake in the expected marginal net con-
venience yield and the expected outside transportation rate
that the cartel was optimizing against in the previous period.
The JEC railroad should have been setting higher prices and
transporting more volume than it actually was. Setting a rate
below the optimal level under certainty can be seen as a desta-
bilizing force, particularly if each ﬁrm observes a boom but
makes losses and is uncertain about the pricing and output
of other ﬁrms in the cartel. We quantify the intensity of the
effect as follows: 1 standard deviation reduction in the error in
expectation will increase the cartel instability by over 3%, in
the short run and by over 15% in the long run, all else equal.
We ﬁnd clear evidence that unexpected demand shocks in
corn markets in New York, which lead to mistakes in a his-
torical price setting, triggered instability in the JEC. This is
consistent with Green and Porter (1984) and is similar to a
ﬁnding in Ellison (1994), who worked with the random part
of the demand residual to model unexpected demand shocks.
We have gone a step further and linked it to realized mistakes
in the expectations of corn market prices in New York, lead-
ing to suboptimal inventory management and price setting in
the whole transportation sector.

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

1734

THE REVIEW OF ECONOMICS AND STATISTICS

Anticipated demand cycles, the countdown on the num-
ber of weeks until the Great Lakes reopen, NWC, and the
cumulated number of weeks that the lakes remain open,
NWO, which control for the effect of anticipated determin-
istic cycles modeled in Haltiwanger and Harrington, are
insigniﬁcant. This is also consistent with the ﬁndings of Elli-
son, who used different endogenous indices to control for
expected demand, including the autocorrelated demand resid-
uals, among other components. The same important result,
that demand shocks and unanticipated cycles were the key
drivers of cartel instability, emerges. This supports the theory
of Green and Porter (1984).

As in Porter (1985) and Vasconcelos (2004), we account
for the number of ﬁrms in the cartel and for the opening
and closing of the Great Lakes as factors affecting the ICC
constraint. In contrast to them, the number of ﬁrms is never
signiﬁcant in our model, and the probability of cartel instabil-
ity is more likely to happen in the lakes open regime. A feature
of the lakes open regime is that the demand tends to be low
for the JEC up to the harvesting and then increases rapidly
up to lakes closing. We estimate that proﬁts for the JEC were
normally highest in the latest weeks of lakes open, when both
prices and shipments increased. Finally, we include the set of
triggers used in Ellison and ﬁnd, as he did, that unusual move-
ments in market shares inside the JEC were not as important
as the common external demand shocks that came from New
York, which all ﬁrms faced.

We argued in the previous section that an unexpected boom
in the New York commodity market is likely to be driven by
conditions external to the cartel. Advancing this notion, our
modeling of the probability of cartel breakdown is mainly
driven by an exogenous shock that should not affect current
pricing and output setting in the JEC but can destabilize the
JEC in the next period and beyond. We turn to estimating
the markup where we condition its Ω1 function on the esti-
mated probability of cartel stability one period ahead, which
is driven by an external demand shock to commodity prices
in New York and a strong path dependency.

B. Step 2: Estimating Markup and Proﬁt

We estimate markup and proﬁt dynamics for the JEC using
a standard simultaneous equations model for demand and
pricing equations. We instrument output as a component of
the marginal cost in pricing. In addition, the markup compo-
nent of pricing is conditioned on the probability of the cartel’s
remaining in a collusive regime in the next time period. We
use the speciﬁcation displayed in column 5 of table 5 to com-
pute the predicted probability of cartel stability ˆPRt+1, which
we incorporate in our modeling of Ω1(·) in the regressions
documented in table 6.

In the 2SLS columns for the baseline model in table 6,
we linearize the Ω1(·) function in the pricing equation and
the Ω2(·) function in the demand equation. We can think
of this as a polynomial of order 1 in the variables that enter
the two functions. This has the advantage that it simpliﬁes the

estimator to a 2SLS approach that, given the use of year dum-
mies, can be compared to the results documented in columns
13 and 14 in table 4. We can judge whether these new con-
trol variables have interesting partial effects in terms of sign,
magnitude, and signiﬁcance. The downside is that we are
missing out on potentially interesting interactions between
the variables in these functions. For example, our controls
for deterministic cycles in Ω1(·), which are gr∗E, NWC, and
NWO, could have an impact on JEC rate setting very differ-
ently when interacted with the predicted probability of cartel
ˆPRt+1. Hence we also estimate and document the
stability,
results of using a semiparametric estimation for this baseline
model.

We now discuss the results for the baseline model, begin-
ning with the demand equation. Our variable, gr∗E, which
controls for the expected rate on alternative modes of trans-
portation, is signiﬁcant in demand. An increase in the grain
rate of the alternative modes of transportation increases
demand for shipments by the cartel. Of the variables in the
Ω2(·) part of the demand speciﬁcation, only positive spikes
in the marginal convenience yield have a signiﬁcant par-
tial effect. We interpret this particular effect as evidence of
the urgent need of grain elevators to top up their invento-
ries. Normally, demand cycles created by elevators should
be reﬂected in expected transportation prices. When inven-
tories fall below a designated threshold, the demand for the
commodity can spike and lead to higher demand for the rail-
road to transport grain immediately to New York. With the
contribution of the partial effect of DMNCYH E, we do not
reject at the 5% level the joint signiﬁcance of the coefﬁcients
of variables that make Ω2(·). However, Ω2(·) explains on its
own only about 15% of the variation in demand.

In the pricing equation, all our new variables in Ω1(·) are
signiﬁcant. All things equal, the accumulation of weeks to
lakes closing, NWO, puts upward pressure on pricing, while
the loss of weeks to lakes reopening, NWC, puts downward
pressure on prices. The expected rate of the outside trans-
portation option, gr∗E, creates an upward pressure on price.
We see the JEC as a price follower in this optimal response
ˆPRt+1, is
function. The estimated probability of stability,
signiﬁcant and has an important positive effect on pricing.

This baseline model is also estimated using a semipara-
metric 2SLS estimation method to allow Ω1(·) in pricing and
Ω2(·) in demand to be estimated nonparametrically. We adjust
Robinson’s (1988) difference estimator in order to account for
the endogeneity in the system of simultaneous equations (see
the online appendix for details on the estimator). The results
from this semiparametric estimation for the baseline model
are presented in columns 3 and 4 of table 6. The sum of the
own- and cross-price elasticities is now estimated to be lower
than those computed with the linear 2SLS estimator. The
Ω1(·) function is computed to be a bigger deterministic com-
ponent as the overall explanatory power of the supply relation
has increased from 0.685 to 0.870 due to implicit interactions
between the variables in the Ω1(·) function. Comparing the
coefﬁcient of determination with and without Ω1(·), we note

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
/
r
e
s
t
/

a
r
t
i
c
e
–
p
d

f
/

9
5
5
1
7
2
2
1
9
7
4
5
4
0
/
r
e
s
t
_
a
_
0
0
3
2
0
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE

1735

Table 6.—Demand System Estimations

Baseline Model

Partial Adjustment Modela

Linear 2SLS

Semiparametric 2SLS

Linear 2SLS

Semiparametric 2SLS

1
q

8.389∗∗∗
(0.421)
−0.529∗∗∗
(0.123)
−0.013
(0.013)
0.011
(0.009)
−1.768
(1.256)
0.580∗∗
(0.237)
0.057
(0.214)
−1.108∗∗∗
(0.196)
0.117∗∗
(0.059)

Variables

Cons

NWC

NWO

MNCYE

DMNCYHEb

DMNCYLEc

gr∗E

q−1

ˆPR+1

3
q

4
gr

−0.106
(0.286)

1
gr

−2.346∗∗∗
(0.479)

0.023∗∗∗
(0.006)
0.018∗∗∗
(0.004)

−0.916∗∗∗
(0.186)
0.105∗∗
(0.052)

0.112∗∗∗
(0.024)
0.067
(0.044)

0.434∗∗∗
(0.035)
−0.228∗∗∗
(0.042)
−0.142∗∗
(0.057)
−0.370∗∗∗
(0.045)
−0.245∗∗∗
(0.091)
Yes
No
312
0.685
−0.041

−0.021
(0.042)

−0.017
(0.019)
0.036
(0.026)
−0.056∗∗∗
(0.020)
−0.047
(0.049)
Yes
No
312
0.870
0.051

5
q

2.815∗∗∗
(0.468)
−0.260∗∗∗
(0.093)
−0.005
(0.009)
0.013∗
(0.007)
−0.304
(0.918)
0.168
(0.176)
−0.043
(0.157)
−0.553∗∗∗
(0.162)
0.085∗∗
(0.043)

0.638∗∗∗
(0.050)

Yes
Yes
312
0.741
0.631
0.196

6
gr

−1.127∗∗∗
(0.326)

0.022∗∗∗
(0.006)
0.016∗∗∗
(0.004)

0.107∗∗∗
(0.023)
−0.049∗
(0.029)

0.395∗∗∗
(0.032)
−0.227∗∗∗
(0.040)
−0.143∗∗∗
(0.054)
−0.344∗∗∗
(0.042)
−0.331∗∗∗
(0.085)
Yes
No
312
0.712
0.113
0.000∗∗∗

7
q

8
gr

−0.102
(0.216)

−0.419∗∗∗
(0.162)
0.056∗
(0.035)

0.645∗∗∗
(0.059)

Yes
Yes
312
0.748
0.701

−0.067∗∗
(0.032)

−0.019
(0.020)
0.047∗
(0.027)
−0.051∗∗
(0.021)
−0.083∗
(0.046)
Yes
No
312
0.875
0.061

Yes
Yes
312
0.515
0.367
0.043∗∗

Month dummies
Year dummies
Observations
R2
R2d
−Ω
P-value F-test coefﬁcients in Ω
Standard error in parentheses. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. a Coefﬁcients and standard errors have to be divided by bDummy set to 1 for highest 1% MNCY values; 0 otherwise. cDummy set to 1 for lowest 1% MNCY values; 0 otherwise. dThe coefﬁcient of determination for the semiparametric demand equation has been computed as 0.000∗∗∗ 1 Yes Yes 312 0.554 0.318 (1−ˆα4 ) to be comparable to those of the baseline model, where ˆα4 is the estimated coefﬁcient of q−1. (cid:4) [ (cid:4) [ (qt −¯qt )(ˆqt −¯qt )]2 (cid:4) (qt −¯qt )2 (ˆqt −¯qt )2 ] and similarly for the pricing equation. the dominant role of the markup function, which explains on its own almost all of the pricing equation.15 Ellison allows for serial correlation in the demand equa- tion. To improve the overall explanatory power of the demand equation, instrument marginal cost, and get better estimates of price elasticities, we factor in demand as a partial adjust- ment model. The structure in the demand equation suggests that we do not control for differences in actual rates (which change daily) and the ofﬁcial rate (set weekly) well enough, and we should allow for a one-week partial adjustment. Our results for this model using a 2SLS estimator of our linear 15 Note that the R2 can be negative in the 2SLS estimator, for some of the regressors enter in the model as instruments for the endogenous variables. However, when the model is ﬁtted, the actual values of the endogenous variables, not the instruments, are used. The implication is that the sum of squares is no longer constrained to be smaller than the total sum of squares. modeling are presented in columns 5 and 6 of table 6 and its semiparametric version in columns 7 and 8. The parame- ters and standard errors have to be divided by (1 − ˆα4) to be comparable to those of the baseline model. The explanatory power of the demand model is now 0.74. More important, we see that the sums of the adjusted own- and cross-price elas- ticities are (in absolute value) higher when compared to the baseline model. The explanatory power of Ω2 now accounts for only 11% of total variation in demand. We also observe some changes in our pricing side. The linear modeling of Ω1(·) shows that the expected outside transportation rate and the estimated cheating probability has a bigger effect. Output in marginal cost is displaying some economies of scale. Unlike the earlier literature, having controlled effec- tively for the omitted variables and partial adjustment in our modeling of demand, we now observe economies of scale l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 1736 THE REVIEW OF ECONOMICS AND STATISTICS Figure 5.—Estimated Price-Cost Margin and Profit (Dashed Line), and Estimated Price-Cost Margin and Profit Business Cycles (Bold Curved Line), Cartel Instability Period (PR = 0, Triangle at 0), Lakes Open (L = 1, Diamond at .5 or 1) 5 . M C P 0 5 . − 1 5 . 0 5 . − t i f o r P 5 . M C P 0 5 . − 0 52 104 156 208 260 312 0 52 104 156 208 260 312 weeks [Linear with PA] weeks [Semiparametric with PA] t i f o r P 1 5 . 0 5 . − 0 52 104 156 208 260 312 0 52 104 156 208 260 312 weeks [Linear with PA] weeks [Semiparametric with PA] in marginal costs.16 While Fabra (2006) shows us that the results of Haltiwanger and Harrington would be less likely to hold in industries with capacity constraints, economies of scale theoretically reinforce the mechanisms in Haltiwanger and Harrington, since when demand is expected to be high, marginal costs are expected to be low. A threat of a revision to a zero proﬁt becomes more binding as expected demand rises and less binding as expected demand falls. The explanatory power of the pricing equation increases to 0.87 when esti- mated semiparametrically. Our Ω1(·) function maintains its high explanatory power, explaining about 80% of the pricing equation in the semiparametric approach. Our estimated price-cost margin dynamics are calculated using the estimated ˆΩ1(·) in the pricing equation. The top two graphs of ﬁgure 5 are plots of the estimated price-cost margins for the partial adjustment model, − ˆθt ˆη , overlapped by their smoothed cycles (computed using Hodrick and Prescott’s, 1997, ﬁlter). The estimates of the price-cost margin are plot- ted over the lakes open and closed periods and against periods of cartel instability as deﬁned by our PR = 0 variable. The bottom two graphs are the corresponding plots of estimated proﬁt (overlapped by their smoothed cycles).17 Given that the graphs of the baseline model are very similar to those of the partial adjustment model we do not include them in ﬁgures 5, 6, and 7. The estimated cycles for the Linear and Semi-parametric model are reasonably similar in trends but differ in levels. Clearly, the weeks of cartel instability are associated with unusually low mark-ups and proﬁts for the cartel in both mod- els. Railroads made losses over some spells when PR was zero. Our semi-parametric estimation of the partial adjust- ment model of demand with supply relation suggests that the cartel actually incurred sustained losses during these periods. In periods of stability we do see some interesting “stylized” cycles emerging. In lakes closed regimes we see price-cost margins drop as lakes reopening approaches. While mark-ups are low at the start of lakes open they consistently rise over the period. More importantly, looking at proﬁt cycles, the periods coming to the end of lakes open normally generated the highest weekly proﬁts for the cartel. The race against the clock in inventory management normally induced increases on the outside transportation rates (captured by gr∗E), hence 16 Walters (1967) has surveyed estimates of cost functions in 34 industries 17 Cartel proﬁt is computed as markup times quantity and is expressed in and found evidence of constant or increasing returns to scale. tens of thousands of dollars. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE 1737 Figure 6.—Normalized Quantity Business Cycles (Dashed Line) and Estimated Price-Cost Margin and Profit Business Cycles (Solid Lines). The Triangle at 0 Denotes Cartel Instability; the Diamond at .5 Denotes Lakes Open l s e c y C M C P d e t a m i t s E d n a y t i t i t n a u Q l s e c y C t i f o r P d e t a m i t s E d n a y t i t i t n a u Q 5 . 4 . 3 . 2 . 1 . 0 5 . 4 . 3 . 2 . 1 . 0 l s e c y C M C P d e t a m i t s E d n a y t i t i t n a u Q 5 . 4 . 3 . 2 . 1 . 0 0 52 104 208 156 weeks [Linear with PA] 260 312 0 52 5 . 4 . 3 . 2 . 1 . 0 l s e c y C t i f o r P d e t a m i t s E d n a y t i t i t n a u Q 0 52 104 208 156 weeks [Linear with PA] 260 312 0 52 104 156 weeks [Semiparametric with PA] 208 104 156 weeks [Semiparametric with PA] 208 260 312 260 312 Figure 7.—Estimated price-cost margin booms (solid lines) and price-cost margin recessions (dashed lines) by quantity business cycle and number of firms l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . 6 . 5 . 4 . 3 . 2 . 1 . 6 . 5 . 4 . 3 . 2 . 1 . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 i s n o s s e c e R d n a s m o o B M C P d e a m t i t s E i s n o s s e c e R d n a s m o o B M C P d e a m t i t s E 10000 20000 30000 40000 10000 20000 30000 40000 Business Cycles Q [Linear with PA] Business Cycles Q [Semiparametric with PA] the volume of trade and the grain rate for the cartel can both increase. Given that the JEC had higher monopolistic power during the lakes closed regime, it is interesting to see proﬁts peaking at the end of the lakes open periods. This highlights the role of, and the need to have, data on the actual exter- nal demand cycles that come from inventory management in New York, which attempts to mitigate the potential high costs of transportation created by the lakes closing. The use of inventories over the seasons creates interesting deterministic demand and pricing cycles. The end of the lakes open period is now the expected boom period for the JEC. This is at ﬁrst sight surprising given that their competitors on the lakes and 1738 THE REVIEW OF ECONOMICS AND STATISTICS Table 7.—Postregression Estimation of PCM during Boom and Recession PCM Business Cycle Baseline Model Partial Adjustment Model 1 Linear 2SLS 2 Semiparametric 2SLS 3 Linear 2SLS 4 Semiparametric 2SLS Phase Boom Variables Cons N ˜Q 10000 (cid:7) ˜Q 10000 (cid:7) (cid:7) ˜Q 10000 ˜Q 10000 (cid:8) 2 (cid:8) 3 (cid:8) 4 Recession Observations Adjusted R2 Cons N ˜Q 10000 (cid:7) ˜Q 10000 (cid:7) ˜Q 10000 (cid:8) 2 (cid:8) 3 Observations Adjusted R2 −1.027∗∗ (0.436) 0.056∗∗∗ (0.011) 2.073∗∗ (0.804) −1.160∗∗ (0.519) 0.264∗ (0.141) −0.022 (0.014) 135 0.497 −0.131 (0.166) −0.010 (0.010) 0.704∗∗∗ (0.214) −0.305∗∗∗ (0.090) 0.039∗∗∗ (0.018) 116 0.274 −0.013 (0.318) −0.106∗∗∗ (0.008) 1.173∗∗ (0.586) −0.553 (0.378) 0.113 (0.103) 0.009 (0.010) 135 0.751 −0.091 (0.212) −0.107∗∗∗ (0.013) 1.280∗∗∗ (0.273) −0.561∗∗∗ (0.115) 0.074∗∗∗ (0.002) 116 0.549 −1.035∗∗ (0.419) 0.048∗∗∗ (0.010) 2.067∗∗∗ (0.773) −1.160∗∗ (0.498) 0.266∗ (0.136) −0.022∗ (0.013) 135 0.474 −0.153 (0.156) −0.010 (0.010) 0.693∗∗∗ (0.200) −0.300∗∗∗ (0.084) 0.038∗∗∗ (0.011) 116 0.280 −0.134 (0.334) −0.113∗∗∗ (0.008) 1.354∗∗ (0.616) −0.659∗ (0.397) 0.141 (0.108) −0.011 (0.011) 135 0.739 −0.184 (0.221) −0.112∗∗∗ (0.014) 1.380∗∗∗ (0.284) −0.601∗∗∗ (0.119) 0.079∗∗∗ (0.016) 116 0.536 canals are open for business. One must realize, however, that inventories offer even more aggressive competitive pressures when the lakes are closed. C. Step 3: Cyclical Nature of Markups Having estimated the component of pricing that reﬂects a markup over marginal cost, we can now analyze some of its behavioral patterns. In particular, we provide evidence of strategic pricing by a legal cartel over the deterministic demand cycles that were created by inventory management, as necessitated by the seasonal opening and closing of trans- portation over the Great Lakes. We represent the smoothed price-cost margin and proﬁt cycles against rescaled output cycles in ﬁgure 6. Can we see obvious counter cyclical or pro- cyclical movements of markups with output? What emerges is that during periods of cartel stability, we observe four lakes reopen episodes where output and price-cost margins move up together as we move toward lakes closing. This generates rising proﬁts that peak just before the lakes close. We also see that during periods of cartel stability, we have ﬁve lakes closed episodes, where output is rising but price-cost margins are falling, as we move closer to lakes opening. Periods of instability are less clear-cut but look countercyclical. To test the theory of Haltiwanger and Harrington during the collusive periods, we regress price-cost margins during upswings and price-cost margins during downturns on the number of ﬁrms and on a polynomial function of the demand business cycle. The results from the regression are reported in table 7 and plotted in ﬁgure 7.18 The Haltiwanger and Harrington theory seems to be validated: for the same level of grain demanded, the price (markup) is lower when the JEC is in a period of pro- longed decline in demand, compared to coming into a period of prolonged increase in demand. This gap is narrowed when the JEC expands the number of ﬁrms in the cartel, which is again consistent with the theory. Higher markups coming in to an expected prolonged increase in demand are harder to sustain when the JEC has more members. This is pow- erful evidence that during periods of stability, the JEC did price optimally over deterministic demand cycles, creating interesting dynamics in markups that reiterate those found in Borenstein and Shepard (1996). This is strong evidence that pricing in cartels reacts in a predictable way to antici- pated demand cycles. It is not just the current level of demand that matters but the expectation of a continued expansion in demand that leads to sustainable high price-cost margins. 18 The spikes in the ﬁgures are due to the number of ﬁrms varying from three to ﬁve during the period. We have no spikes when there are four ﬁrms, decreasing spikes when we have ﬁve ﬁrms, and increasing spikes in case of three ﬁrms. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 COMMODITY MARKET DYNAMICS AND THE JOINT EXECUTIVE COMMITTEE 1739 V. Conclusion The use of commodity prices in Chicago and New York, spot and future, through ﬁnance theory allows us to construct the expected transportation rates across the dominant modes of transportation between Chicago and New York. We also construct the expected marginal net convenience yield of holding inventories in New York. Finally, we measure the unexpected demand shocks in New York commodity markets that lead to realized mistakes in pricing and volume (expec- tations of prices and inventories) setting by transportation modalities. These variables are estimated to have a tremen- dous impact on the modeling of equilibrium price-cost margin movements in the JEC. In terms of modeling the linear marginal cost component of pricing, our model of output is greatly enhanced with the addition of our new data. The market elasticity of demand is now made up of an important cross-price, as well as an own-price, effect. We also include year dummies, controls for spikes in demand by elevators in New York and partial adjustment effects. This model of demand allows us to instru- ment and model marginal cost, and hence pricing, relatively better. In terms of modeling the second component of pricing, the markup, we model it as a function of deterministic demand cycles and condition it on expected cartel stability one period ahead. Cartel instability is found to be triggered by real- ized errors in forecasts of commodity price shocks in New York rather than demand cycles. An unanticipated boom in New York can make the cartel unstable, as ﬁrms with imper- fect monitoring of other ﬁrms realize that price and output should have been higher last period leading to ﬁnite periods of breakdown, as suggested in Green and Porter (1984). In stable periods, the markup is modeled to depend on antici- pated demand cycles. Due to the short window of opportunity between harvesting and lakes closing, inventory management over the Great Lakes and canals leads to distinctive deter- ministic demand cycles for the railroad cartel to set prices against. We estimate the equilibrium price path semipara- metrically (linear marginal cost and the nonlinear markup function) simultaneously to the demand. Linear estimates of the parameters of marginal cost allow us to backout the non- parametric function to estimate rich weekly markup and proﬁt cycles. We test for optimal collusive pricing over the deter- ministic cycles and ﬁnd, for the same volume of sales, that the markup in a prolonged boom tends to be higher when compared to the markup coming into a prolonged recession. In other words, the JEC set prices over deterministic demand cycles in a way that supports the theoretical considerations in Haltiwanger and Harrington (1991). As highlighted in Boren- stein and Shepard (1996), such cyclical pricing may be a way of detecting illegal pricing behavior in a modern cartel. REFERENCES Aldrich Report, U.S. 52nd Congress 2nd Session, Senate Report 1394. Wholesale Prices, Wages and Transportation Report by Mr. Aldrich from the Committee on Finance. Part 1 (Washington, DC: Govern- ment Printing Ofﬁce, 1893). Appelbaum, E., “The Estimation of the Degree of Oligopoly Power,” Journal of Econometrics 19 (1982), 287–299. Berry, S., J. Levinsohn, and A. Pakes, “Automobile Prices in Market Equilibrium,” Econometrica 63 (1995), 841–890. Borenstein, S., and A. Shepard, “Dynamic Pricing in Retail Gaso- line Markets,” Rand Journal of Economics 27 (1996), 429– 451. Bresnahan, T., “Empirical Studies of Industries with Market Power” (pp. 1011–1058), in Richard Schmalensee and Robert Willig (eds.), Handbook of Industrial Organization (Amsterdam: North-Holland, 1989). Coleman, A., “Storage, Slow Transport and the Law of One Price: Theory with Evidence from Nineteenth Century U.S. Corn Markets,” this review, 91 (2009), 332–350. Ellison, G., “Theories of Cartel Stability and the Joint Executive Commit- tee,” Rand Journal of Economics 25 (1994), 37–57. Fabra, N., “Collusion with Capacity Constraints over the Business Cycle,” International Journal of Industrial Organization 24 (2006), 69–81. Gallet C. A., and J. R., Schroeter, “The Effects of the Business Cycle on Oligopoly Coordination: Evidence from the U.S. Rayon Industry,” Review of Industrial Organization 10 (1995), 181–196. Genesove, D., and W. Mullin, “Testing Static Oligopoly Models: Con- duct and Cost in the Sugar Industry, 1890–1914,” Rand Journal of Economics 9 (1998), 355–377. Green, E., and R. Porter, “Noncooperative Collusion under Imperfect Price Information,” Econometrica 52 (1984), 87–100. Haltiwanger, J., and J. Harrington, “The Impact of Cyclical Demand Move- ment on Collusive Behavior,” Rand Journal of Economics 22 (1991), 89–106. Hodrick, R., and E. Prescott, “Post-War U.S. Business Cycles: An Empirical Investigation,” Journal of Money, Credit and Banking 29 (1997), 1–16. Holmes, T. J., “The Effects of Third-Degree Price Discrimination in Oligopoly,” American Economic Review 79 (1989), 244– 250. Lee, L. F., and R. H. Porter, “Switching Regression Models with Imper- fect Sample Separation Information: With an Application on Cartel Stability,” Econometrica 52 (1984), 391–418. Mariuzzo, F., P. P. Walsh, and C. Whelan, “Coverage of Retail Stores and Discrete Choice Models of Demand: Estimating Price Elas- ticities and Welfare Effects,” International Journal of Industrial Organization 28 (2010), 555–578. Pindyck, R. S., “Inventories and the Short-Run Dynamics of Commodity Prices,” Rand Journal of Economics 25 (1994), 141–159. Porter, R., “A Study of Cartel Stability: The Joint Executive Commit- tee, 1880–1886,” Bell Journal of Economics 14 (1983), 301– 314. ——— “On the Incidence and Duration of Price Wars,” Journal of Industrial Economics 33 (1985), 415–426. Robinson, P. M., “Root-N-Consistent Semiparametric Regression,” Econo- metrica 56 (1988), 931–954. Rotemberg, J., and G. Saloner, “A Supergame Theoretic Model of Price Wars during Booms,” American Economic Review 76 (1986), 390– 407. Thurman, W. N., “Speculative Carryover: An Empirical Examination of the U.S. Reﬁned Copper Market,” Rand Journal of Economics 19 (1988), 420–437. Ulen, T. S., “Cartels and Regulation: Late Nineteenth Century Railroad Col- lusion and the Creation of the Interstate Commerce Commission,” PhD dissertation, Stanford University (1979). ——— “Railroad Cartels before 1887: The Effectiveness of Private Enforcement of Collusion,” Research in Economic History 8 (1983), 125–144. Vasconcelos, H., “Entry Effects on Cartel Stability and the Joint Execu- tive Committee,” Review of Industrial Organization 24 (2004), 219– 241. Walters, A. A., “Production and Cost Functions: An Econometric Survey,” Econometrica 31 (1967), 1–66. Working, H., “The Theory of Price of Storage,” American Economic Review 39 (1949), 1254–1262. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u / r e s t / l a r t i c e - p d f / / / / 9 5 5 1 7 2 2 1 9 7 4 5 4 0 / r e s t _ a _ 0 0 3 2 0 p d . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3
Download pdf