Measuring the Effects of Natural Gas 
Pipeline Constraints on Regional Pricing and 
Market Integration
Authors: Roger Avalos, Timothy Fitzgerald, and 
Randal R. Rucker
NOTICE: this is the author’s version of a work that was accepted for publication in Energy 
Economics. Changes resulting from the publishing process, such as peer review, editing, 
corrections, structural formatting, and other quality control mechanisms may not be reflected in 
this document. Changes may have been made to this work since it was submitted for publication. 
A definitive version was subsequently published in Energy Economics, [VOL# 60, (November 
2016)] DOI# 10.1016/j.eneco.2016.09.017
Avalos, Roger, Timothy Fitzgerald, and Randal R Rucker. "Measuring the Effects of Natural Gas 
Pipeline Constraints on Regional Pricing and Market Integration." Energy Economics 60 
(September 2016). DOI: 10.1016/j.eneco.2016.09.017.
Made available through Montana State University’s ScholarWorks 
scholarworks.montana.edu 
Measuring the Effects of Natural Gas Pipeline
Constraints on Regional Pricing and Market Integration*
Roger Avalos„
Timothy Fitzgerald…
Randal R. Rucker§
September 6, 2016
Abstract
Natural gas pipeline capacity sets physical limits on the quantity of gas that can be
moved between regions, with attendant price effects. We find support for the hypoth-
esis of integrated regional markets. Using data on daily pipeline flows and capacities
in Florida and Southern California, we estimate reduced-form price effects of capacity
constraints. We find that pipeline congestion increased realized citygate prices by at
least 11 percent over the mean in Florida and by 6 percent over the mean in Southern
California. We attribute the difference in price effects to more binding capacity con-
straints in the Florida pipeline network. Our estimates provide guidance for interstate
pipeline investments.
Keywords: natural gas; pipelines; market integration; pipeline congestion
JEL Codes: Q40, L95
*The authors would like to thank seminar participants at the University of New Hampshire, the 31st
USAEE/IAEE North American Conference, the Colorado School of Mines, and the WEAI 89th Annual
Conference for constructive comments.
„California ISO
…Texas Tech University; Contact: timothy.fitzgerald@ttu.edu
§Montana State University, Department of Agricultural Economics and Economics
1 Introduction
An extensive network of pipelines provides essential infrastructure for the operation of
the natural gas market in the United States. Sweeping regulatory reforms in recent decades
have restructured the market for the transportation services provided by these pipelines.
These reforms notwithstanding, pipeline capacity sets physical limits on the quantity of gas
that can be moved from production fields to end users. The potential impacts of pipeline
capacity constraints on natural gas markets have long been recognized (Lyon and Hackett
1993; DeVany and Walls 1996; Marmer et al. 2007; Brown and Yu¨cel 2008a). Market
participants, industry analysts, and academic researchers have often asserted that
breakdowns in market integration and periodically diverging prices result from binding
pipeline capacity between particular hubs. In this paper, we use daily data on pipeline
flows relative to capacity for markets in Florida and Southern California to obtain the first
empirical estimates of the impacts of pipeline constraints on wholesale prices in regional
markets.
The U.S. network of natural gas pipelines is extensive and constantly expanding.
Increasing supply of natural gas due to technological advances and the concomitant
increase in the quantity of natural gas demanded, combined with the regulatory lags
associated with the expansion of pipeline capacity, make understanding the impact of
pipeline capacity constraints a pressing issue. Important reserve additions from
unconventional sources have changed the natural gas map and have provided the impetus
for substantial investment in new pipeline capacity. Some capacity increases are the result
of retrofitting existing pipelines to increase flow, while other expansions involve building
new pipelines to link new producing regions with consumers. Focusing on new and
expanded interstate pipelines entering service in 2007–2012, 121 Bcfd (billion cubic feet per
day) of capacity was added along 13,163 miles, at a nominal cost of investment of $47.8
1
billion.1 Those expansions increased the state-to-state capacity of natural gas pipelines by
15 percent and have accommodated consumption increases of 11 percent.
Not only do these investments costs billions of dollars, but they also entail substantial
transaction costs. Makholm (2012) explains how obsolesence associated with asset
specificity affects the incentives to invest in new pipelines. Klein et al. (1978) describe how
the potential for opportunistic behavior arising from appropriable quasi-rents affects
transaction costs and the resulting contractual arrangements. These factors, combined with
the fact that—given current market institutions—quasi-rents created by scarce pipeline
capacity largely do not accrue to pipeline owners and investors, may reduce the incentives
to invest in additional capacity.
The flexibility and complexity of the natural gas pipeline network largely determine
whether regional natural gas markets interact to form a single smoothly functioning
national market. We conduct a series of cointegration tests and fail to reject the hypothesis
that wholesale natural gas markets in the United States generally function as a single
well-integrated national market. Our period of analysis covers a more recent period than
previous studies, and some of our results are qualitatively similar (Doane and Spulber
1994; DeVany and Walls 1993 1996; Serletis and Rangel-Ruiz 2004; Cuddington and Wang
2006). Using daily prices from 14 regional markets over the period October 2006–August
2011, our results indicate that all market pairs are cointegrated, that the panel as a whole
is cointegrated, and that possible regime shifts do not alter this result.
We then examine Florida and Southern California markets in more detail, using daily
price and flow data to examine the price effects of limited capacity. Figures 1a and 1b
display plots of regional wholesale market prices in Florida and Southern California,
respectively, and the benchmark U.S. natural gas price at Henry Hub in Louisiana. From
1These estimates are derived from EIA natural gas pipeline projects available at http://www.eia.gov/
naturalgas/data.cfm.
2
both of these plots, it is clear that the respective regional prices move together with the
benchmark prices. It is also apparent—especially in Florida—that there are periods
(typically brief) during which the regional price is substantially greater than the
benchmark price.
In Florida, we find that the price is cointegrated with the Henry Hub price. We also
estimate a reduced-form model of equilibrium Florida prices and find that, on average,
pipeline congestion increased Florida wholesale prices by $0.79/MMBtu over the sample
period, which represents an increase of 11 percent relative to the mean price.2
In Southern California, contrary to earlier studies, we find that prices are cointegrated
with prices in other regional markets, suggesting that this region is currently integrated
with the larger national market. Our estimate of the effect of binding pipeline constraints
on Southern California prices is smaller than in Florida. At about $0.32/MMBtu, the price
increase is about 6 percent above the mean. The fact that there is only one day during the
time span of our data, combined with other information presented below, suggests that
capacity constraints in the Southern California market are less binding than those in
Florida.
2 Background
Natural gas markets and pipelines in the United States have a long and important
history, much of which is pertinent to understanding the current state of affairs.3 Briefly,
prior to the 1980s, natural gas markets in the United States were regulated “from wellhead
2A price spike (or congestion rent) of $0.79 is the most directly comparable estimate to our Southern
California estimate. This estimate is also the most conservative estimate we obtain for Florida. Our other
estimates of the magnitude of the congestion rents in Florida—obtained using more restrictive definitions
of the threshold levels at which pipeline capacity becomes binding—are much larger, ranging as high as
$4.54/Mcf.
3McGrew (2009) provides a comprehensive review of the legal and regulatory environment in which
natural gas pipelines have evolved. Makholm (2012) provides an institutional perspective on the history of
pipelines. In appendix A, we describe the regulatory and contractual landscape pertinent to our study.
3
to burnertip.” Starting in the late 1970s and culminating in 1992 with FERC Order 636,
natural gas markets underwent sweeping changes, most of which were focused on
deregulating segments of the market. Today, wholesale natural gas markets are
characterized by many producers and marketers competing to provide gas to a multitude of
consumers. Spot and futures markets are available at dozens of locations and over a broad
range of time horizons, providing price discovery and allowing for indexed contracts.
Although interstate pipelines are still subject to regulation, ready access to transportation
services and the ability of shippers to buy and sell their contracted primary transportation
services on short notice in a secondary market allows the movement of gas between regions
with substantial flexibility (Dahl and Matson 1998; Leitzinger and Collette 2002).
There are two strands of literature that are related to our empirical analysis. The first
uses cointegration tests on prices to examine whether the reforms of natural gas markets
that started in the 1970s have effectively resulted in the integration of regional markets
into a single national market. Early studies in this area find that most markets became
integrated fairly quickly during the post-deregulation periods they examined. Later studies
by, for example, Marmer et al. (2007), Murry and Zhu (2008), and Brown and Yu¨cel
(2008a) find that the links between regional natural gas markets appear to have weakened
over time. A common thread in these studies is that the markets that are not fully
integrated are typically in California, the Rocky Mountains, and the Midwest. The
explanation offered for this finding is typically that segmentation of natural gas markets is
due to binding pipeline capacity constraints. One argument is that although there was
ample excess interstate pipeline capacity in the early years after restructuring, over time as
consumption grew, pipeline expansions did not keep pace. The resulting imbalance led to
bottlenecks and pipeline congestion that reduced the ability to arbitrage between some
regional markets. Our results suggest that these constraints have been relaxed, probably by
the addition of additional miles of pipelines, to such an extent that the markets are again
4
integrated.
The second strand of the literature directly examines the impacts of pipeline capacity on
natural gas prices. Theoretical models designed to describe the workings of
capacity-constrained pipeline systems are developed by Cremer and Laffont (2002) and
Cremer et al. (2003). These studies focus on the analytics of price formation in relatively
simple pipeline networks. A related study that estimates the price impacts of natural gas
pipeline constraints is Oliver et al. (2014), which develops a structural model of pipeline
constraints in a producing region. Whereas their model highlights the importance of
transportation infrastructure investments, and the buffering effect that natural gas storage
has on the market, our empirical analysis focuses on the effect of capacity constraints at
the terminal end of the transport network, where consumers bear the brunt of price effects.
3 Regional Market Model
Regional markets for natural gas depend critically on the particular pipeline network
serving the market. Weather-induced stochastic demand, especially for electricity
generation and heating, drives the equilibrium outcomes we observe. The system
engineering constraints involved in pipeline operation require a balance between the amount
of gas injected and withdrawn. An important feature of these markets is that pipeline
capacity constraints impose a limit on the quantity that can be delivered at any given time.
Pipelines share attributes with other networks subject to congestion, such as electric grids
and road networks (Joskow and Tirole 2000; Arnott et al. 1993). In the model developed
below, we borrow from the framework for natural gas pipelines suggested by Cremer and
Laffont (2002) and Cremer et al. (2003). We focus on the effects of capacity constraints in
the short run and pricing considerations when multiple pipelines serve the same market.
Assume that deliveries to the regional natural gas market can be made from any of i ∈ I
5
serving pipelines. In a single time period, the total benefits of delivering the aggregate
quantity Q =
∑
I qi is given by B(Q) =
∫ Q
0
D(X)dX, where D(·) is the regional demand
for natural gas, and the marginal value of quantity Q is D(Q). Because aggregate benefits
and marginal value depend on the aggregate quantity, deliveries of homogeneous gas from
alternative suppliers are assumed to be perfect substitutes for one another in the product
market. The costs of supplying gas are separable by the delivering pipeline, as
characterized by pSi = Si(qi). Because we focus on equilibrium price formation, the
superscripts on price are superfluous. Each pipeline has a maximum capacity, given by qci .
In the short run, a regional market has an aggregate capacity constraint of Qc =
∑
I q
c
i .
Taking pipeline capacity as exogenous and ignoring the fixed costs of that capacity for now,
the socially optimal equilibrium price will solve the constrained optimization problem:
max
qi
B(Q)−
∫ Q
0
∑
I
Si(qi) s.t. qi ≤ qci ∀ i (1)
The Kuhn-Tucker conditions for each pipeline take the form:
D(Q)− Si(qi)− λi ≤ 0 λi(qci − qi) = 0
If the capacity constraint on pipeline i does not bind, then λi = 0, where λi represents the
shadow value of an additional unit of capacity in pipeline i. We are interested in three
cases that can arise.4
Case A – System Capacity Flow. All pipelines serving the regional market have
capacity flows, which implies that the market price exceeds the marginal cost of
supply and also that D(Qc) = Si(q
c
i ) + λi ∀ i ∈ I. The values of the λi depend on the
degree to which realized demand at Qc exceeds the individual Si(q
c
i ) values.
4A graphical representation of a simple two-pipeline model is presented in appendix B.
6
Case B – Subsystem Capacity Flows. One or more, but not all, pipelines servicing
the market experience flows at capacity. Without loss of generality, consider the case
where only pipeline j ∈ I binds. Here, at the market clearing price,
D(Q) = Sj(q
c
j) + λj and Si(qi) = Sj(q
c
j) + λj ∀ i 6= j.
Case C – No Capacity Flows. None of the pipelines serving the market have flows at
capacity, so D(Q) = Si(qi) and λi = 0 ∀ i. Flows will adjust so that marginal costs
across pipelines will be equal, both to each other and to the market clearing price.
This simple framework offers two important hypotheses about price formation.
Hypothesis 1. When flow is at system capacity (Case A), the resulting regional market
price will exceed the marginal delivery cost, with the difference being a function of the
magnitude of the demand shock relative to system capacity.
Hypothesis 2. When increases in system flows cause some (but not all) pipelines to have
binding flows (Case B), the equilibrium market price will exceed the marginal cost of
delivery by an amount that increases with the difference in marginal costs between pipelines.
Figure 2 provides a graphical interpretation of the effects of a capacity constraint in a
gas-importing region.5 For clarity, focus on daily variation in weather as a demand shifter.
In periods of seasonable weather, demand is represented by DL, and the intersection of
supply and demand determines quantity and price – Q0 and p0. In periods with extreme
heat or cold, demand might increase to DH , and if there were no pipeline capacity
constraints, the equilibrium quantity and price would be Q1 and p1. Pipeline capacity is
the maximum quantity of gas that can be supplied, so the supply curve becomes vertical at
Qc. At the margin, the wholesale price must rise to pc so that the quantity demanded also
5To avoid graphical congestion, figure 2 abstracts from the fact that the regional market depicted may
be comprised of several separate pipelines.
7
equals Qc.6
We turn now to a set of issues pertaining to the scale and scope of the pipeline network.
These are long-term issues, but we find it useful to develop two additional hypotheses to
make predictions about how the network evolves over time. In figure 2, the shaded area C
represents the total gains (in a single period) to pipeline companies, shippers, and
wholesale consumers from expanding pipeline capacity to an amount greater than or equal
to Q1. This does not suggest that optimal capacity is Q1.7 The impact on price of optimal
network scale is a Ramsey price that amortizes fixed investments in capacity (Cremer et al.
2003). We are interested in the impact of changes in the scale and scope of the network.
Hypothesis 3. When increases in system flows cause all pipelines to have binding flows
(Case A), an increase in the total installed capacity (Qc) of the network will cause the
equilibrium price to fall, holding constant demand.
Hypothesis 4. Holding demand and system capacity constant, increasing the number of
pipelines in the system (I) will not increase the equilibrium price if the increase keeps
marginal supply costs constant.
Capacity expansion may come with higher costs, but is predicted to reduce prices. The
differences in costs as capacity increases are therefore important to understanding the
likely effect on prices—some expansions may be intensive, allowing more gas to flow
through existing lines. Extensive changes such as new pipelines will also expand total
system capacity, but may have different marginal costs (and different Ramsey prices
6Because the prices in many contracts for transportation services are negotiated in advance, purchasers
of some inframarginal units will pay a lower price, perhaps p0. Others might pay a higher price. Our focus
here is on the price of the marginal quantity of gas, and in the empirical analysis we use the average price
of gas.
7The frequency with which Case A obtains, and the extent to which demand exceeds capacity today and
is expected to exceed capacity in the future, can be translated into an expected demand versus installed
capacity. The current costs of expanding capacity must then be compared to the discounted value of the
welfare gains as measured by present and future values of C, with the socially optimal capacity expansion
being the one that maximizes the difference.
8
accounting for infrastructure costs). However, increasing the number of pipelines per se
will not increase prices.
In our model to this point, price spikes would not be observed until the demand curve in
figure 2 intersects the kink in the supply curve (where the capacity factor—the ratio of
daily flow to capacity—reaches 1.00). Other important realities suggest at least three
reasons why a price path like Sg in figure 2 might be observed as flow approaches
capacity—that is, why prices might rise before capacity constraints bind.
First, as demand shifts to the right from DL in figure 2, there will be an increasing
number of transactions in the secondary market as increasingly scarce (and valuable)
capacity is shifted from relatively low value users to higher value users through secondary
market exchanges. The transaction costs for these exchanges are higher than those in the
primary capacity market because the secondary transactions involve price negotiation.
These higher transaction costs would cause observed prices to rise above the supply curve
(S) in figure 2 along a price path like Sg. Second, and closely related, the sellers of
secondary contracts provide additional services relative to the services provided in
primary-market contracts (Reiter and Economides 1999). The associated higher costs will
also cause prices to rise along Sg in figure 2.
Third, in certain cases, a limited number of sellers may be able to exercise market power
in the secondary market. Imagine that for each pipeline, a total of m shippers subscribe
the primary capacity. As flow approaches the capacity of the line, some subset of these
shippers m− k are willing to sell capacity. In case where m− k is small or even equal to
one, the possibility of market power is very real.
Our simple model highlights the importance of both demand and supply factors in
determining prices and the transacted quantities of natural gas. Below, we extend our
regional market model to an empirical specification that explicitly incorporates those
factors.
9
3.1 Supply Factors
Many factors affect natural gas production, including levels of reserves, labor and
capital costs, historic drilling activities and production decline curves, government policy,
and stochastic factors such as major weather events. At any given point in time, a
benchmark price, such as Henry Hub, incorporates all relevant information, as well as
expectations about future natural gas demand and supply conditions. An increase in the
price at Henry Hub, insofar as it corresponds to (or causes) increased prices in other
regions, will induce an increase in the quantity of natural gas supplied to all regions. This
observation underscores the importance of the cointegration tests we conduct below.
Holding constant the price at Henry Hub, an increase in price at a particular region (that
is, an increase in the price basis differential between Henry Hub and the region) will lead to
increased shipments of gas to that region.
Regional supply at time t (QSt ) will therefore be a function of the price at Henry Hub
(pHHt ), the regional price (pt), and other factors that interrupt supply (Xt), up to the
quantity Qc in figure 2, where the capacity constraint becomes binding, as in equation (2).
QSt =

S(pt, p
HH
t ,Xt) if S(pt, p
HH
t ,Xt) < Q
c
Qc if S(pt, p
HH
t ,Xt) ≥ Qc,
(2)
where S(pt, p
HH
t ,Xt) is the regional supply function representing the aggregation of Si(qi)
from the I pipelines serving the market.8 The proxy for supply interruptions (Xt) we use
in our empirical analysis is the level of shut-in production in the Gulf of Mexico due to
hurricanes (Brown and Yu¨cel 2008b).
8An alternative way to specify the problem is as a function of regional basis differentials, or pt − pHHt .
See Oliver et al. (2014).
10
3.2 Demand Factors
A number of factors affect the demand for natural gas in both the short and long run,
including economic activity and incomes, prices of energy substitutes, and weather
variation that drives demand for heating and cooling. The wholesale demand for natural
gas is represented in equation (3).
QDt = f(pt, P
oil
t , P
coal
t , Incomet, P
E
t , P
I
t ,Weathert) (3)
The prices of oil and coal, which are long-run substitutes for natural gas in these sectors,
are denoted by P oilt and P
coal
t . Total personal income (Incomet) is included as a demand
shifter. Output prices in the electric and industrial sectors are represented by PEt and P
I
t .
Residential, commercial, and electric power demand are affected by weather conditions
because natural gas and electricity are used both for space heating and for cooling.
4 Empirical Strategy
Our empirical analysis has two parts: first, we test for integration of different regional
markets; then, we conduct reduced-form analyses of two specific regional markets.
4.1 Cointegration Tests
Testing to determine whether price series from two spatially distinct markets are
cointegrated is a widely-used approach for examining the extent to which the two markets
are integrated.9 Two non-stationary price series, Pit and Pjt, integrated of order i and j,
respectively, are cointegrated if there exists a linear relationship between them, as in
equation (4), and if the errors from the relationship are integrated of order i− j
9Pesaran (2015) comprehensively addresses cointegration issues and estimation methods.
11
(I(i− j)) (Engle and Granger 1987).
εt = Pit − β0 − β1Pjt (4)
If wholesale markets efficiently distribute natural gas across regions, the difference in
equilibrium prices will correspond to transportation and transaction costs. If those costs
are constant over time, the relationship between upstream and downstream prices will be
stationary. We are interested in the order of integration of the residuals (ε) from a
regression of the price in market i on the the price in market j. If estimation of (5) yields a
coefficient on α1 that is significantly less than one, then the series are cointegrated.
∆εt = α0 + α1εt−1 +
∑
j
∆εt−j + ut (5)
The Augmented Dickey-Fuller (ADF) test in (5) is widely-used, thanks to the availability of
appropriate critical values for the test statistic (Elliott et al. 1996). In addition, we report
results from other commonly used tests: in particular, the Phillips-Perron (PP) test and
the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test (Kwiatkowski et al. 1992). To compare
all regional prices, we test whether the whole panel of 14 market prices—rather than just
pairs of prices—is integrated using the test suggested by Nyblom and Harvey (2000).
We also employ tests developed by Gregory and Hansen (1996) and Kejriwal et al.
(2012) that test for cointegration while allowing for the possibility of regime shifts in the
price formation process. In our empirical application to capacity-constrained pipelines, it is
possible that the data-generating processes for wholesale prices are not constant
throughout the span of our data. Such regime shifts might occur, for example, due to
changes in the demand regime depicted in figure 2. Regime shifts may also arise in response
to changes in macroeconomic conditions or to structural changes in the natural gas market
12
due, for example, to changes in the ability of buyers or sellers to exercise market power.
4.2 Reduced-Form Model
Regional price differences will be arbitraged away quickly if markets are competitive and
integrated. Open access policies help ensure that pipeline access is competitive.10 With
non-binding pipeline capacity constraints, the difference between prices in the upstream
and downstream (with respect to pipeline flow) regions will be equal to the costs of
transportation plus the transaction costs associated with arbitrage. When pipeline capacity
is reached, the price difference is expected to become larger.
To identify the effects of capacity constraints, we estimate the reduced-form
specification in equation (6) by OLS. The daily regional gas price is the dependent
variable, and demand and supply factors (including one or more binary variables to
indicate various levels of pipeline congestion) are explanatory variables.11
pimt =pi0 + pi1p
HH
t + pi2P
oil
t + pi3P
coal
im + pi4Incomeim + pi5P
E
im + pi6P
I
im + pi7HDDimt
+ pi8CDDimt + +pi9Hurricane Shutinst + pi10Binding Capacityimt + θm + φy + eimt,
(6)
Here the i subscript refers to the region (Florida or Southern California), m refers to the
month of the year, and t refers to the day of the observation in our sample. To address
10As indicated in appendix A,“open access” refers to the requirement that pipeline companies offer their
transportation service to all market participants. Due to the pro-rationing of oversubscribed capacity in
open access pipelines, the assumption of competitiveness for pipelines that are likely to be at or near their
full capacity generally seems warranted. We are assuming that there are always a sufficiently large number of
shippers with subscribed capacity on the pipelines we study so that market power among primary capacity
subscribers is not a concern. Incorporating market power would be a separate, interesting, and pertinent
study.
11Our reduced form depends critically on the cointegration of the regional and benchmark prices. In
the next section, we find robust support for the notion of cointegration under a number of alternative
specifications. As a result, our regression results are consistent, and concerns associated with spurious
regression are reduced (Greene and Zhang 2011).
13
seasonality issues, we include month-of-year fixed effects represented by θs. To account for
changing macroeconomic conditions and trends, we include year fixed effects, φy.
We define the pipeline capacity factor as the ratio of daily flow to daily stated capacity.
We initially define Binding Capacity as a binary variable equal to one if the capacity factor
is 0.95 or greater, and zero otherwise 12 The estimated coefficient pi10 represents the average
increase in prices on days when pipelines are congested and provides a direct estimate of
the effect of such congestion on prices. In variants of equation (6), we examine the
possibility that binding capacity constraints have different price impacts depending on the
definition of the Binding Capacity variable. We also perform a number of robustness checks
to investigate the possibility of nonlinear price reactions consistent with the “early” price
increases depicted by the Sg curve in figure 2.
Identification in our model comes from two sources, each of which yields a testable
prediction:
Prediction 1: Consider two days, both with identical values for all of the independent
variables. Suppose that on the first day, demand in figure 2 is DL. On the second day, a
factor that affects demand, but is not included as an explanatory variable in equation (6),
is different than on the first day. For example, suppose that the second day is a day in
mid-July with the same temperature as the first day, but humidity is higher. This will
cause the demand curve to shift to the right. If the second-day demand curve intersects the
vertical portion of the supply curve in figure 2, the price of natural gas will rise by more
than it would (from the same demand shift) if the pipeline constraint did not become
binding, and the estimated value of pi10 will be positive.
Prediction 2: Consider two days on which all the independent variables are identical
except that on the first day the capacity factor is 1.00. On the second day, capacity has
12We define the Binding Capacity variable in our base model in this manner to more easily compare our
results for Florida and Southern California. See section 6.2 below for discussion of this issue.
14
been expanded so that the same flow does not fill the pipeline. The predicted effect of this
expansion is to reduce the wholesale price of natural gas.
5 Data
The natural gas price data we use for our analysis of the integration of natural gas
markets were obtained from Natural Gas Intelligence (NGI), which are published by NGI
as the Gas Daily Price Index.13 These data comprise daily prices for 14 hubs and market
points in the United States for the time period from October 1, 2006 to August 29, 2011.
The daily spot price at Henry Hub is one of the series we obtained, as are prices for the
regional wholesale markets in Florida and Southern California that we use in our reduced
form analysis of the impacts of pipeline capacity constraints on prices.
The price of electricity is a variable in equation (3) for which data are available monthly,
but not daily. Instead of using monthly electricity prices, we posit that daily variation in
electricity prices is likely to be an important determinant of daily variation in wholesale gas
demand because of its use in the production of electricity. Lacking daily electricity prices,
we capture the impacts of such daily variation in electricity prices with two weather
variables that represent important short-term demand shifters.14
As proxies for weather factors that affect the demand for natural gas, we use daily
temperature data from the National Oceanic and Atmospheric Administration to calculate
daily cooling and heating degree days. The value of the cooling degree day (CDD) variable
13See Natural Gas Intelligence at http://www.naturalgasintel.com. Table D.1 in appendix D provides
provenances for all variables used in the empirical analysis.
14Because Florida electricity markets were not restructured during the time period we examine, there is no
source of daily electricity prices for Florida as there is in California. To maintain symmetry in our analyses
between Florida and Southern California, we use the weather variables (described below) for both markets.
Our primary regression specifications are similar in spirit to those estimated by Brown and Yu¨cel (2008b);
Hartley et al. (2008); Villar and Joutz (2006); Sebastian and Thoenes (2014), all of whom use weather
variables similar to ours. Note also that the weather variables we use may reflect changes in the demand for
natural gas through residential, commercial, and power sector demand, and not just through their indirect
impact on electricity prices.
15
is obtained by subtracting 65 from the actual temperature, with a lower bound of zero. An
increase in the value of the CDD variable corresponds to days with higher temperatures
and increased demand for cooling services, which we expect to increase natural gas demand
by increasing electricity demand. The value of the heating degree day (HDD) variable is
obtained by subtracting the average temperature for each day from 65, with a lower bound
of zero. As temperature falls, the calculated value of the HDD variable rises. This induces
an increase in the demand for heating services by consumers, which increases the derived
demand for, and the price of, natural gas.
Our measures of HDD and CDD differ from the typical calculation. We obtain daily
HDD and CDD values for each weather station in Florida and Southern California and
then average these across all the relevant stations in those two regions to calculate the
variable values we use in our empirical analysis.15 With this approach, we observe some
days with positive values of both HDD and CDD because of the spatial average across the
market area. This is a more accurate indicator of natural gas demand than a simple
average of all temperatures across the entire market area. In a dramatic example, a day on
which the average temperature is 65 at every weather station across the market area is not
the same as a day on which half the market area has a temperature of 100 and the other
half has a temperature of 30. In both cases the average temperature is 65 degrees. Our
measure would record such a day as having both heating and cooling demands, not a day
with HDD and CDD values of zero.
The data on daily pipeline capacities and flows we use in our empirical analysis are
provided by Lippman Consulting Inc. (LCI), which was a major supplier of natural gas
data to industry participants. LCI collects data at receipt and delivery points, as well as at
various compressor stations, for most interstate pipelines in the United States and Canada.
15We thank Joe Atwood at Montana State University for providing these data. Most other series, such as
the NOAA series (http://www.ncdc.noaa.gov), average the mean daily temperatures from all stations and
calculate a single state-wide HDD or CDD value.
16
We use the LCI data to create a daily capacity factor measure, calculated as the daily flow
rate divided by the stated daily capacity of the pipeline.
Figure 3 shows pipeline maps for the regional pipeline systems in our empirical study.
Two interstate pipelines transport natural gas into Florida: Florida Gas Transmission
Company (FGT) and Gulfstream Natural Gas System (Gulfstream). There are no
interstate pipelines used to export gas from Florida, and natural gas production in the
state accounts for less than 1 percent of in-state consumption. Moreover, there are no
major storage or liquefied natural gas import facilities in the state. Table 1 describes
capacity characteristics of the Florida pipeline system—capacity usage in each pipeline is
listed separately, as well as in the combined system. Each of the two pipelines was at
capacity on about 30 out of 961 days in our sample, and the system was full on nine days.
The system was at 95 percent or more of capacity on 144 days in our data, which
represents 15 percent of the observations we use in our analysis. The lower panel of the
table records the changes in stated pipeline capacity over time.
In contrast, Southern California is served by four major interstate pipeline systems;
Kern River, Transwestern, and two branches of the El Paso network (hereafter El Paso
North and El Paso South).16 In addition, Pacific Gas and Electric Company has an
intrastate pipeline (not shown in figure 3) that ships gas from southern to northern
California. Unlike Florida, California has several storage facilities and also has significant
marketed natural gas production. This production accounted for 12.6 percent of in-state
consumption in 2010.
Table 2 lists the four pipelines for which we have daily flow and capacity data. While
two of the pipelines (Kern River and El Paso North) were regularly at their capacity, a
third pipeline (El Paso South) was full on only 15 of the days in our sample, and the fourth
16An additional pipeline, Questar Southern Trails, also serves this market. It was excluded in part because
of its relatively small size (80 MMcfd).
17
(Transwestern) did not flow at its full capacity during the period covered by our data
(January 2007 – August 2011).17 Further, the Southern California system was only at full
capacity for one out of the 1164 observations in our data, and flow was above 95 percent of
capacity on 48 days (4.1 percent of our observations). These observations suggest that the
pipeline system capacity constraints in Southern California were considerably less binding
than in Florida during the period of our analysis. Additional details for both the Florida
and California pipeline networks are included in appendix C.
Tables 3 and 4 show summary statistics for the Florida and Southern California data.
Although we received the same time span of data for both markets, we have fewer
observations for Florida because there is a gap in the pipeline flow data for Florida. We
have a single daily price for the two-pipeline Florida system that is the focus of our
reduced form analysis and also a single price for the four-pipeline California system that we
analyze. One interesting observation is that although the Florida citygate prices are higher
than the corresponding Henry Hub prices, the Southern California citygate prices are
lower. The gas shipped to California does not come directly from Henry Hub, but from
other hubs further west. The use of the Henry Hub price as a benchmark is appropriate,
however, because Henry Hub is widely agreed to be a national price discovery point.
Moreover, as can be seen in figure 1b, the prices move together and the focus of our
analysis is not on identifying the magnitude of the effect of Henry Hub prices on California
prices, but on obtaining direct estimates of the effects of binding pipeline capacity
constraints on prices in the two markets in our data set.
Daily spot West Texas Intermediate (WTI) oil prices were downloaded from the EIA’s
website18 and converted from dollars per barrel to dollars per MMBtu using the standard
17Transwestern serves four power plant facilities in Arizona located downstream from the West compressor
station. Our flow data are from the Arizona–California state line, but capacity is at the West compressor
station. Although West is the point at which capacity binds, some portion of total flow is delivered in
Arizona and not Southern California. This contributes to the appearance of flows below the stated capacity.
18http://www.eia.gov/dnav/pet/pet_pri_spt_s1_d.htm
18
conversion factor of 5.8 MMBtu per barrel. Coal prices were obtained from EIA Form 923
Schedule 2 data on utility and non-utility generator fuel purchases, and are reported in
$/MMBtu.19 Total personal income data for Florida and California were obtained from the
Bureau of Economic Analysis. The index of industrial output prices we use is from the
Bureau of Labor Statistics, as is the producer price index (excluding energy) that we use to
convert all prices (and personal income) to 2011 dollars. Insofar as oil and coal are
(long-run) substitutes for natural gas, an increase in their prices is predicted to lead to an
increase in the price of natural gas. Increases in incomes or industrial prices are expected
to lead to increases in the demand for natural gas and in its price. Data for coal prices, the
industrial output price index, personal income, and the producer price index are only
available at monthly intervals. Because of the inconsistency in frequencies between the
daily natural gas prices and the variables available only on a monthly basis, our primary
regression specifications do not include the monthly variables. When we do include these
variables, we find that (for Florida in particular) their estimated coefficients are generally
not statistically significant. More importantly, for both Florida and Southern California,
their inclusion does not substantively affect the estimated impacts of binding capacity
constraints on natural gas prices.
A potentially important exogenous factor that can affect the supply of natural gas is
hurricanes, which are an annual threat to natural gas production in historically important
producing regions such as the Gulf of Mexico and along the Gulf Coast. To account for
these impacts, we use weekly data from the Bureau of Ocean Energy Management,
Regulation and Enforcement on the level of production shut-ins (in MMcfd) due to
hurricanes. We convert the numbers posted weekly to a daily frequency by linearly
interpolating between reports during hurricane season. Outside of hurricane season
(December 1 – May 30) we assign this variable a value of zero.
19www.eia.gov/electricity/data/eia923/index.html
19
6 Results
6.1 Cointegration Results
We conduct a battery of cointegration tests on a sample of prices from 14 U.S. pricing
points for the period from 2006 to 2010. Using ADF tests, we find that all market prices
are I(1) in levels and I(0) in differences, and that all individual price pairs are
cointegrated. We also use a Nyblom-Harvey test (Nyblom and Harvey 2000) of whether the
panel of U.S. natural gas markets is integrated. To balance the panel for this test, we drop
two price series with missing data (Kingsgate and FGT Citygate). The test on the
remaining 12 series rejects the null of no common trends among the price series with a test
statistic of 19.84 (the critical value for 10 series at the 0.01 level is 3.14).20 These results
provide evidence that the restructured U.S. wholesale natural gas market is integrated and
generally functions as one market over the time span of our data. It is worth noting that
these results do not preclude the possibility that there are binding bottlenecks along the
systems of pipelines connecting the markets in our data. Any such bottlenecks, however,
are not sufficiently strong or persistent to cause the statistical tests to suggest that natural
gas markets are not well integrated over the entire period.
As a precursor to our reduced-form model, we subject the Florida and Southern
California markets to additional analysis and present the results in table 5. In the ADF
and Phillips-Perron tests, the null hypothesis is that the series are not cointegrated, so our
results (which reject the null) indicate that both the Florida (FGT Zone 3 Citygate) prices
and the Southern California prices are cointegrated with the Henry Hub price. Note that
the estimated coefficient on the Henry Hub Price is significantly less than 1.0 for the
Southern California regression. Rather than flowing directly from Henry Hub, gas is
imported into Southern California from the Rockies, western Texas, and the Southwest.
20See tables D.3 and D.4 in appendix D for additional detail on these tests.
20
Substituting the price at the Opal Hub in Wyoming in place of Henry Hub prices yields
identical inference regarding the cointegrating relationship for the California series. We
also conduct a KPSS test, the null hypothesis of which is that the relationship is
stationary, and fail to reject the null for both Florida and Southern California.
An additional consideration is the possibility that there are structural breaks in the
pricing process that might alter our conclusion to this point that the Henry Hub price is
cointegrated with Florida and Southern California prices. To address this possibility, we
use a test developed by Gregory and Hansen (1996), who demonstrate that if the true
model includes a structural break, then the application of their test increases the power of
the ADF tests. The G-H test allows for a single break in the structure of the pricing
process and identifies the observation in the data set that yields the highest probability of
accepting the null hypothesis of no cointegration of the prices. We consider structural
breaks that involve intercept-only changes as well as intercept-and-slope changes. The
results of these G-H tests support our previous findings that Henry Hub prices are
cointegrated with Florida and Southern California. One interesting result is that the test
discloses the most likely date of a structural break; we note that dates in early- to mid-2008
are the most likely times for both markets (see the bottom entries in table 5). These dates
are in the midst of the Great Recession, suggesting that the structural breaks identified
may reflect the impacts of changing macroeconomic conditions on natural gas markets.
It is also possible that there are multiple structural breaks in the pricing process over the
period of our sample as a result of changing macroeconomic conditions or the other factors
mentioned above. Kejriwal et al. (2012) suggest a test for cointegration in the presence of
multiple structural breaks and provide the appropriate critical values. Using the days
identified by the various Gregory-Hansen tests we ran as dates where there may have been
structural breaks, we fail to reject the null of an unchanging unit root process over the
whole sample. The most conservative Florida test statistic we obtain is 0.11, whereas the
21
critical value at the 0.05 level is 7.91. The same test performed on the California series
yields a (most conservative) test statistic of 0.99 with a 0.05 critical value of 10.62.
Summarizing, the cointegration tests discussed above indicate that the regional and
benchmark prices are cointegrated. We take this as evidence of integration between the
Florida and Southern California regional markets and the national market. Based on these
results, we turn to our direct investigation of the effects of capacity constraints on regional
prices.
6.2 Reduced Form Results
We estimate several variants of the reduced-form specification in equation (6), first for
Florida, and then for Southern California.
6.2.1 Florida
The results for Florida support the predictions of the model discussed above.21 Consider
first the results in model 1 of table 6.22 All specifications include month-of-year fixed
effects to control for seasonality of natural gas demand and prices, as well as year fixed
effects to control for changing macroeconomic conditions and trends.23 The estimated
coefficient on the Henry Hub price is positive and significant, as we expected. The
21Most of our analysis below treats the two Florida pipelines as a system. We do this because (1) we have
a single natural gas price for the two Florida pipelines and (2) our examination (described below) of price
effects from a single pipeline reaching capacity suggest that the interactions between capacity conditions
in the two-pipeline system are important. The same comments and findings apply to our analysis of the
four-pipeline Southern California system.
22Durbin-Watson tests on exploratory specifications indicated the presence of autocorrelation of the resid-
uals, and White tests indicated that heteroskedasticity is also present. Accordingly, we report OLS estimates
and use the Newey-West correction of standard errors for heteroskedasticity and autocorrelation in the re-
duced form results discussed below. The number of lags in our model is six, which is informed by our
cointegration results, as well as by the rule of thumb (first proposed by Davidson and MacKinnon (1993))
of selecting the number of lags as the smallest integer that is greater than or equal to T 0.25, where T is the
number of observations.
23December is the omitted month and 2006 is the omitted year. The month-of-year fixed effects are jointly
significant, as are the year fixed effects.
22
WTIPrice is also positive and significant, consistent with the notion of gross substitution
between natural gas and oil.24
The estimated coefficients on both the HDD and CDD variables are positive and
significant, with the point estimate of HDD being significantly larger. The estimated
coefficient on Hurricane Shutins, which measures the impact on price of a one unit (Bcfd)
increase in the volume of natural gas production that was shut-in on a given day as a result
of hurricanes, is positive and significant. Gustav and Ike were the largest hurricanes during
the period of our data. At the peak of their impact on production in September of 2008,
our results suggest that the reduced production associated with these hurricanes increased
prices in Florida by about $1.77 (= 0.25×7.062).
The variable of primary interest in each of our regressions is the Binding Capacity
variable, which is a binary variable that we assign a value of one when capacity is binding
and a value of zero otherwise. In the first two models in table 6 we define binding capacity
to correspond to capacity factors of 95 percent or greater (this definition allows us to
directly compare our Florida results to the California results discussed below). In models 3
and 4, capacity is defined to be binding for capacity factors of 99 percent or greater, and in
model 5 binding capacity corresponds to capacity factors of 100 percent or more. The
estimated coefficients on the Binding Capacity variables in models 1, 3, and 5 support our
prediction that pipeline constraints increase wholesale regional prices and indicate that
binding capacity constraints have a large and statistically significant effect on FGT Zone 3
Citygate Prices. The average impact in model 1, for example, is an increase of $0.79, which
is a nontrivial 11.3 percent of the average FGT price in our sample ($7.01). Comparing
models 1, 3 and 5, one sees that as the capacity factor used to define the Binding Capacity
24The extent to which oil and natural gas are substitutes in the short run and long run is an issue that has
received considerable attention from economists. See, for example Brown and Yucel (2008), Villar and Joutz
(2006), Hartley et al. (2008), and Sebastian and Thoenes (2014). For our purposes, we note the sign and
significance of the estimated coefficient on the WTI Price variable and focus our attention on the impacts
of pipeline capacity constraints on natural gas prices.
23
variable increases, so does the estimated impact of binding pipeline capacity on prices.25
In models 2 and 4 of table 6, we examine whether (as suggested by Prediction 1 from
section 5.2 above) short-run demand shifters (specifically, HDD and CDD) have greater
impacts on prices on days when the Florida system is at or near capacity. In both models,
the estimated coefficients on the multiplicative interaction terms between the Binding
Capacity variable and the HDD and CDD variables are positive and statistically
significant.26 This suggests—consistent with our prediction—that increases in HDD
(indicating colder days) and in CDD (indicating hotter days) have significantly greater
impacts on prices when the pipelines are full or nearly so. We note that with the
introduction of the multiplicative interaction terms, the estimated coefficients on the
Binding Capacity variables in models 2 and 4 no longer represent the estimated price
impacts of binding pipeline capacity.27
To further examine how capacity constraints affect prices, we estimate a series of
regressions with flexible piecewise capacity factor variables and report the piecewise
coefficient estimates from these regressions in table 7 (estimated coefficients on the other
variables are not substantively affected and are not reported, but are available on request).
In model 1 in table 7, we include dummy variables for five bins of capacity between 75 and
100 percent in addition to the variable Total Capacity 100+% discussed in table 6. Each
coefficient measures the effect of flows at the indicated range of capacity on prices relative
to prices when the daily capacity factor is less than 75 percent; for example, when the flow
25As a further test of our identification strategy, we estimate a specification like model 1, but without
the three demand variables for which we have daily data (implying that a change in any demand variable
that results in the capacity constraint being binding may be identifying the price effects of interest). The
estimated coefficient on the capacity variable in this specification is somewhat larger, but is not statistically
different than the estimates in table 6.
26We do not estimate these impacts for the model in which binding capacity corresponds to capacity factors
of 100 percent or more because that would require estimating three coefficients with only eight observations.
27The implied estimates of the price impacts of capacity constraints from, for example, model 4 suggest
that the impacts are significantly larger on very cold days than on very hot days. Moreover, a weighted
average impact over all the days on which capacity was binding is very close to the corresponding estimate
in model 3. Details on these calculations are provided in in the final subsection of appendix E.
24
is between 85 and 90 percent of the stated pipeline capacity, prices are $0.40 higher than
when the flow is less than 75 percent of capacity. The estimated coefficients in model 1
reveal a monotonically increasing price function.
Note the very large jump in price ($5.32 - $0.98 = $4.34) when the flow increases from
the 95-100% to the 100+% capacity range. In models 2–4 we include more narrowly
defined bins in the highest near-capacity flow ranges to gain additional insights into the
price increase as flow approaches capacity. The estimates in model 2 indicate that the price
increase when flow is between 99 and 100 percent of capacity is almost twice as large as
when flow is between 95 and 99 percent of capacity ($1.66 vs. $0.85).
There are nine observations in our sample with daily flow rates that are greater than or
equal to the stated total capacity of the two Florida pipelines—observations for which our
Total Capacity 100+% variable takes a value of one. We also have 35 observations for
which the flow is between 99 and 100 percent of capacity. In model 3 of table 7, we split
these 35 observations into two bins (the first with 17 observations, the second with 18).
The estimated impact of increased flow on daily prices continues to increase monotonically
in this model, and the jump in price when the pipeline network becomes completely full
remains large ($5.47 - $1.86 = $3.61). In model 4 we create three bins with
(approximately) equal numbers of observations. With this smaller number of observations
in each bin, the monotonicity of the point estimates breaks down (though the point
estimates are statistically non-decreasing), although the large increase in price as flows
increase from near-capacity to full-capacity persists.
6.2.2 Florida Capacity Expansion
The next issue we investigate with the Florida data relates to our prediction that an
increase in pipeline capacity will lead to a reduction in prices when capacity is binding.
The bottom portion of table 1 shows that there is one instance during the time span of our
25
data in which the increased capacity was large enough for us to estimate the impacts. This
is the increase in capacity from 3,557 MMcfd to 4,280 MMcfd that took place early in April
of 2011.
A straightforward approach to testing the prediction would be to estimate a regression
that allowed the price effects of being at capacity to differ before and after the capacity
increase in early April of 2011. This approach is precluded by the fact that there is only
one observation for which flows were at 100 percent of the expanded capacity.28 Our data
set includes 36 observations following the expansion in capacity early in April of 2011. For
12 of those observations, daily flows were less than the stated capacity before expansion.
The average flow on these days was 3,355 MMcfd, which is 94.3 percent of the
pre-expansion capacity and 78.4 percent of the increased capacity (see Q2 in figure 4). Our
earlier results (table 7) using capacity bins suggest that price effects increase as capacity
factors increase. We predict that the post-expansion prices for these days will be lower
than they would have been for the same flow levels prior to the expansion. Assuming the
structure of the price effects as flows approach capacity did not change with the April 2011
expansion, we can actually predict the magnitude of the price reduction.29
Based on the regression estimates in model 1 of table 7, prices on days when the average
28While we cannot conduct statistical tests with one observation, our prediction is supported in that the
observed price after the expansion is less than the price on every one of the 31 days on which the capacity
factor exceeded 99 percent in June–August of the preceding year.
29One reason for assuming that the structure of the price effects did not change is that the number of post-
expansion observations in our data set (36) is too small to be able to reliably estimate five or more distinct
post-expansion bin effects. The whole suite of cointegration tests presented in section 6.1 holds up around
the capacity expansion with only one exception. That exception is the Gregory-Hansen test for a structural
break in intercept-only after the expansion. All specifications of the test, both with and without additional
independent variables in the cointegrating relationship, yielded the same result of no cointegration. This
provides some evidence that the pipeline expansion itself may have changed the price generation process, but
the power of this test is low with only 36 observations after the expansion. Gregory-Hansen tests allowing for
either a regime break (intercept and slope), or a slope change alone, support the assumption that the pricing
process did not change, suggesting that effects could be explained by a difference in slope, or in intercept
and slope, if not by a change in intercept alone. This result suggests that our additional analysis is valid,
because our models allow for regime shifts insofar as separate slope and intercept parameters are estimated
below, and we find both are statistically significant.
26
flow is between 90 and 95 percent of capacity are $0.55 higher than when capacity is less
than 75 percent. The point estimate of the difference between prices on days when flow is
between 75 and 80 percent of capacity and days when flow is less than 75 percent of
capacity is $0.04. Thus, as illustrated in figure 4, our prediction is that prices will be $0.51
lower for this set of 12 post-expansion observations.30
The other 24 post-expansion observations all have daily flows greater than the
pre-expansion capacity level, and all but one of those has a daily flow level less than the
post-expansion capacity level. We omit from the ensuing analysis the one observation that
had a flow level of 100 percent of the new capacity. The average flow rate of the other 23
observations is 3,780 MMcfd (Q3 in figure 4) and the average of the capacity factor values
for these days is 88.3 percent. The fact that the capacity factor for these 23 observations is
lower after expansion than before (88.3 percent as compared to 100 percent or more) leads
us to predict lower prices for these observations after the expansion in capacity. A more
specific prediction can be obtained using the demand curve labeled D3 in figure 4. All 23 of
the observations under consideration would have been at capacity prior to the expansion.
Using the estimation results from model 1 in table 7, the average price on days with flows
at full capacity was $5.32 higher than days with less than 75 percent capacity, while the
price on days with flows between 85 and 90 percent of capacity was $0.40 higher. Thus, we
predict that prices for these observations will be lower than they would have been prior to
the expansion by $4.92 (= $5.32 - $0.40).31
30Although the estimated price impact when flows are between 75 and 80 percent of capacity is not
statistically different from zero, $0.04 represents the best point estimate of the impact, so we use it for our
calculations. Our conclusions are not altered if we assume this price impact is zero and then test whether
the reduction in price after the expansion is equal to $0.55.
31Note that points A and B represent two points on the same demand curve. The arc price elasticity of
demand between these points is -0.12, a value that is similar to estimates in the literature of the short-run
demand for natural gas (Al-Sahlawi 1989).
27
To test these two predictions, we estimate the following regression model
pmt = X
′pi + pi10Flow > Old Capacitymt + pi11Increased Capacitymt
+ pi12(Flow > Old Capacity × Increased Capacity)mt + θm + φy + emt (7)
where X is the vector of covariates corresponding to pi0 to pi9 in equation (6).
Flow > Old Capacity is a binary variable that we assign a value of one if the daily flow
was at or above the pre-expansion binding capacity (for days both before and after
capacity expanded). There are eight such days in our data prior to the April 2011
expansion and 24 such days after the expansion. Increased Capacity is a binary variable
we assign a value of one for the 36 post-expansion observations in our sample. We also
include a multiplicative interaction term.
We expect the Flow > Old Capacity coefficient to be positive and roughly equal to the
estimated coefficients on the 100+% Capacity variable in the regressions discussed above.
The preceding discussion suggests the estimated coefficients pi11 and pi12 will be negative.
Moreover, insofar as the structure of price effects does not change following the capacity
expansion, we predict that (1) the Increased Capacity effect will be -$0.51 and (2) the
interaction term will be -$4.92.
Because the estimated coefficients on other covariates are largely unchanged from
previous specifications, table 8 presents the key coefficient estimates from a specification
with the same covariates as in table 6. All three of the estimated coefficients have the
predicted signs. The estimated coefficient for the Flow > Old Capacity variable is $5.24,
which is statistically indistinguishable from the coefficients on the 100+% Capacity
variable in earlier specifications. The estimated coefficient on the Increased Capacity
variable is -$0.51, which is exactly equal to its predicted value. The estimated coefficient
on the interactive term between these two variables is -$5.67. Although this estimate is
28
somewhat larger in absolute value than our predicted value of -$4.92, the two values do not
differ statistically. These results provide strong support for our predictions regarding the
price impacts of increases in pipeline capacity.
Appendix E contains results from our analysis of two other issues related to our Florida
data. First, is an examination of the impacts of including the “monthly” variables (coal
price, income, and the industrial price index) in several of the specifications in table 6. We
find that the inclusion of these variables does not substantively affect our estimates of the
price effects of pipeline capacity constraints. Second, is an investigation of the price effects
of capacity constraints if one looks only at the individual pipelines, rather than at the
system as a whole. We find that (1) the estimated price effects are substantially smaller if
one focuses on capacity constraints in individual pipelines than if the focus is on the
system’s capacity constraints, and (2) there are substantial price effects due to interactions
between flows in the two pipelines, but generally only when both pipelines are flowing at
levels very close to capacity.
6.2.3 California
The specification we employ for California is very similar to the Florida
specification—we use a binary indicator of days with average flow above 95 percent of
stated capacity as a measure of pipeline congestion. As in the earlier Florida regressions,
we include year and month-of-year fixed effects to control for trends and seasonality.
Starting with model 1 in table 9, the Henry Hub price is statistically significant, which
is consistent with the results reported in table 5, and the discussion of those results applies
here. The estimated coefficient on WTI Price is also positive and significant, as it was in
Florida. The estimated coefficients on HDD and CDD are positive and statistically
significant. Their estimated values—0.028 and 0.041—are considerably smaller than for
Florida and are not significantly different from each other.
29
The Southern California pipeline system for which we have data rarely reaches its stated
capacity. From table 2, 48 of the 1164 days in the California sample have a capacity at or
above 95 percent, so our coefficient estimates are averages over those days. The estimated
value of the coefficient on the capacity variable indicates that prices are $0.32 higher on
days with flows that are more than 95 percent capacity than on days with lower flows. It is
notable that the estimated coefficients on the capacity variable are virtually the same if the
binary capacity variable is defined using 96 or 97 percent (results available on request from
the authors). In table 6 we reported the estimated coefficient on the same capacity level in
Florida capacity variable as $0.79, which is 2.5 times larger than the coefficient for
Southern California. This result supports hypothesis 1 in section 3. From tables 1 and 2, it
can be inferred that pipeline capacity (or near capacity) constraints in the Florida system
are more often binding than in the CA system.32 Under plausible assumptions, an
implication of this is that in figure 2, on those days when the demand curve in California
intersects the vertical portion of the supply curve, it is (on average) at a lower price than in
the Florida system, which is consistent with our empirical results.33
In model 2 of table 9, we report the results from a regression with piecewise capacity
factor variables designed to provide more information on how Southern California natural
gas prices change as available capacity approaches zero. We include binary variables for
three ranges (or bins) of capacity between 85 and 100 percent. The addition of two bins for
capacity factor levels from 85 to 95 percent results in a statistically insignificant change in
the value of the estimated coefficient on the 95+% capacity variable. The estimated
32See, in particular, the top panels in tables 1 and 2. There is only a single observation in our Southern
California data where the system of pipelines is 100 percent full compared with nine if Florida. There are
44 observations in Florida where the system of pipelines has flows in excess of 99 percent of capacity, but
only two in California. There are analogous differences between the two systems for other “near capacity”
flow levels as far down as the “> 90%” rows in tables 1 and 2.
33As with our Florida results, the addition of monthly variables to model 1 in table 2 does not substantively
affect our estimate of the impact of binding capacity. In contrast to our Florida results, interactive terms
between weather variables (HDD and CDD) and Binding Capacity 95+% are not statistically significant.
These results are available on request from the authors.
30
coefficient on the 90 to 95 percent variable is $0.15 and is statistically significant, while the
coefficient on the 85 to 90 percent variable is not significant. So, while there is some
evidence to support the presence of “early” price increases in Southern California, these
only occur above 90 percent of the stated system capacity, and the magnitude of the
increases are much smaller than for Florida.34 These results are consistent with capacity
constraints being considerably less binding in Southern California than in Florida.
To this point in our analysis of California prices, we have investigated how prices change
when the pipeline system as a whole approaches its capacity. In model 3, we examine
whether prices are affected when the individual pipelines in the Southern California system
have flows at or near their capacities. To do this, we construct the four 0-1 variables listed
in table 9. We assign the variable Kern Capacity 100% a value of one when the capacity
factor for Kern pipeline is 100 percent or greater and a value of zero otherwise. Similarly
for the other three variables for the individual pipelines. We also include the variable
Total Capacity 95 + % that was included in models 1 and 2. The estimated coefficients on
the individual pipeline variables indicate that when any of the individual pipelines reach
(or gets close to) their capacity, there is no significant effect on prices, but again, when the
system as a whole gets close to its capacity, there is a significant increase (of $0.33) in price.
In our analysis of Florida prices, we examined the impacts of an expansion in the
capacity of the pipeline system in Florida. Table 2 indicates that there was an expansion in
the capacity of the Kern River pipeline on April 9, 2010 from 2,042 to 2,917 MMcfd. Prior
to that date, the Kern River pipeline was at capacity on 22 percent of the days in our
sample. After that date, the pipeline was at capacity on only 5.3 percent of the days.
Unfortunately, beyond this, we can say little about the impacts of the increase in capacity.
As a result of declining production in the fields that provided gas for the El Paso North
34We also estimate several specifications with further breakdowns of the 48 observations with capacity
factors greater than 0.95 and (in contrast to Florida) find no indication of systematically increasing prices
within this range.
31
pipeline, flows were 22 percent lower in that pipeline after the Kern River pipeline was
expanded (2,062 vs. 1,599 MMcfd). So, even though flows through Kern River increased
after the expansion (from 1,998 to 2,084 MMcfd), total flow through the Southern
California pipeline system decreased after the expansion.35 There were, in fact, no days
after the Kern River expansion on which flows exceeded the initial capacity of the system
(7,732 MMcfd). These circumstances preclude further analysis of the price impacts of the
Kern River capacity expansion.
7 Conclusion
We find support for earlier research showing that restructuring in wholesale natural gas
markets and open access pipelines have created arbitrage opportunities and integration
between regional wholesale natural gas markets. Previous work identified regions where
prices were not integrated with a national market, with several studies noting the
balkanization of the Southern California market (DeVany and Walls 1996; Cuddington and
Wang 2006; Marmer et al. 2007). In our analysis of a later time period, the Southern
California market appears to be integrated into the larger national market, and gas flows
into the region are rarely constrained by pipeline capacity. Our sample also indicates that
prices at Kern River and Opal Hub in the Rockies are cointegrated with national prices,
again in contrast to results from earlier periods. We find that regional markets in Florida
and Southern California are integrated with a national market, even though substantial
price increases occur in periods of high demand.
We conduct a reduced-form investigation of regional market pricing. In both Florida and
Southern California, pipeline capacity constraints have immediate and significant absolute
and relative price effects. We estimate effects in Florida of $0.79/MMBtu —which is about
35Average flows for the system before and after the Kern River expansion were 6,735 and 6,358 MMcfd.
32
12 percent of the mean price—when flows exceed 95 percent of stated pipeline capacity. As
we increase the threshold to 99 and then 100 percent of capacity, we estimate much larger
price effects of $1.71 and $4.54, with the latter increase corresponding to 65 percent of the
average observed price. In California the effects are more moderate, about $0.32/MMBtu
—or 6 percent of the mean price—for 95 percent (or higher) capacity flow thresholds.
We are able to directly examine the short-run price effects of capacity expansions in
Florida, and we find that following expansion the network reacted exactly as we would
predict—days with binding capacity and price spikes became rare, and prices generally fell.
However, in other markets (such as Boston) price spikes and congestion rents remain an
important issue.36
If capacity expansions are effective in reducing the incidence of large price spikes, a
natural question is why constraints persist. Our research on the particular market
institutions, combined with our empirical analysis, lead us to three insights. First, the
prices we observe are a mix of prices for primary capacity and prices in a secondary
market. Gains from price increases in the secondary market due to constrained capacity do
not accrue to the pipeline owners. Instead, the gains are captured by primary capacity
owners (shippers), despite the intention of FERC Order 736.
Second, because many of the purchasers of wholesale gas are bound by guaranteed
service contracts, their demand elasticity is small. However, these same purchasers are
often protected by regulated rates that allow for averaging of acquisition costs. This mutes
the signal to final consumers of gas, who might make different decisions if confronted with
the full cost of consuming gas and scarce transportation capacity. Davis and Muehlegger
(2010) have proposed an alternative pricing structure to address this second issue.
Third, despite substantial regulatory hurdles, over the time span of our data, we do
observe non-trivial increases in pipeline capacity, especially in the Florida system we
36See, for example, Ailworth (2014); EIA (2013)
33
examine. In the first three-and-a-half years of our Florida data, we observe one day on
which the Florida system capacity factor was at 99 percent or more. Between June and
October of 2010, we observe 34 days at this level of congestion, with the concomitant price
increases that we estimate above. By early April of 2011, the capacity of the Florida
system had been expanded by roughly 20 percent, after which there is only one day in our
data on which the system reached its new capacity. Despite the particular institutions and
the regulatory barriers in U.S. natural gas markets, there were sufficient incentives for
pipeline owners in Florida to expand capacity when it became binding in 2010.
From our analysis, with the assistance of figures 2 and 4, we can provide insights into
the magnitude of the costs and benefits from the 2011 capacity expansion in Florida. As
discussed in section 6.2, our empirical estimates of the price effects of binding capacity are
used to calibrate figure 4. From this, we can estimate one component of the net social
gains from capacity expansion—the portion of area C above p1 in figure 2 (or triangle ABC
in figure 4). Realizing that prices are measured in units of $/Mcf, and quantities are in
units of MMcf/day, this component of net social gains is $4.92×1,000×(3,780 - 3,557)×0.5
= $548,580 per day. Our reduced form approach provides no information on the size of the
portion of triangle C below p1 in figure 2—the other component of the net social gains from
capacity expansion. Depending on the inelasticity of the short-run supply of natural gas
relative to demand, this area may be larger or smaller than the upper portion of the
triangle. Another critical component of a benefit-cost analysis is the cost of the 2011 FGT
pipeline expansion, which was almost $2.5 billion(EIA 2011 2015).
Distributional issues are also important and the rectangle to the left of Qc between p1
and pc, which represents a transfer from “producers” to consumers, is large —
$4.92×1,000×(3,557) = $17.5 million per day— relative to the upper portion of triangle C.
Our data provide no information on how that rectangle of congestion rents is divided
between the pipeline owners and those shippers who were gaining from selling their
34
capacity at high prices in the secondary market. That division, of course, has important
implications for pipeline owners’ incentives to invest in costly pipeline expansions.
With natural gas consumption expected to continue to increase in the future, decisions
on whether or not to expand pipeline capacity will significantly affect regional market
integration and pricing. Current transportation infrastructure was built to meet the
requirements of past and current market conditions and policies. Increased production,
especially from new regions, requires investment in additional pipeline capacity. Policy
proposals that broadly affect natural gas markets, particularly those that lead to increased
gas demand (e.g., stricter environmental standards that result in further displacement of
coal-fired electric generation with natural gas), require an understanding of the impacts of
pipeline constraints. Our analysis in this paper contributes to that understanding. More
focused proposals might include changes to the regulatory structure of natural gas markets
to improve transmission of price signals to firms making pipeline investment decisions.
35
References
Ailworth, Erin, “Natural Gas Prices Spike,” Boston Globe, January 10, 2014. Available
at https://www.bostonglobe.com/business/2014/01/10/natural-gas-prices-
spike/H8d7cjyAbnE8cnNliWzEHN/story.html.
Al-Sahlawi, Mohammed A., “The demand for natural gas: a survey of price and
income elasticities,” The Energy Journal, 1989, 10 (1), 77–90.
Arnott, Richard, Andre De Palma, and Robin Lindsey, “A structural model of
peak-period congestion: A traffic bottleneck with elastic demand,” The American
Economic Review, 1993, 83 (1), 161–179.
Brown, Stephen P. A. and Mine K. Yu¨cel, “Deliverability and Regional Pricing in
US Natural Gas Markets,” Energy Economics, 2008, 30 (5), 2441–2453.
and , “What Drives Natural Gas Prices?,” The Energy Journal, 2008, 29 (2), pp.
45–60.
Cremer, Helmuth and Jean-Jacques Laffont, “Competition in gas markets,”
European Economic Review, 2002, 46 (4), 928–935.
, Farid Gasmi, and Jean-Jacques Laffont, “Access to pipelines in competitive gas
markets,” Journal of Regulatory Economics, 2003, 24 (1), 5–33.
Cuddington, J.T. and Z. Wang, “Assessing the Degree of Spot Market Integration for
US Natural Gas: Evidence from Daily Price Data,” Journal of Regulatory Economics,
2006, 29 (2), 195–210.
Dahl, C.A. and T.K. Matson, “Evolution of the US Natural Gas Industry in Response
to Changes in Transaction Costs,” Land Economics, 1998, 74 (3), 390–408.
Davidson, Russell and James G MacKinnon, Estimation and Inference in
Econometrics, New York, NY.: Oxford University Press, 1993.
Davis, L.W. and E. Muehlegger, “Do Americans Consume Too Little Natural Gas?
An Empirical Test of Marginal Cost Pricing,” The RAND Journal of Economics, 2010,
41 (4), 791–810.
DeVany, A. and W.D. Walls, “Pipeline Access and Market Integration in the Natural
Gas Industry: Evidence from Cointegration Tests,” Energy Journal, 1993, 14, 1–19.
and , “The Law of One Price in a Network: Arbitrage and Price Dynamics in
Natural Gas City Gate Markets,” Journal of Regional Science, 1996, 36 (4), 555–570.
Doane, Michael J. and Daniel F. Spulber, “Open Access and the Evolution of the
U.S. Spot Market for Natural Gas,” Journal of Law and Economics, October 1994, 37
(2), 477–517.
36
Doane, M.J., R.P. McAfee, A. Nayyar, and M.A. Williams, “Interpreting
Concentration Indices in the Secondary Market for Natural Gas Transportation: The
Implication of Pipeline Residual Rights,” Energy Economics, 2008, 30 (3), 807–817.
EIA, “New Natural Gas Pipeline Capacity Adds Service in Florida,” Today in Energy, May
25, 2011. Available at http://www.eia.gov/todayinenergy/detail.cfm?id=1530.
, “December Natural Gas Prices Spike in Boston,” Today in Energy, December 6, 2013.
Available at http://www.eia.gov/todayinenergy/detail.cfm?id=14071.
, U.S. Natural Gas Pipeline Projects, Excel File
Name:EIA-NaturalGasPipelineProjects.xls, U.S. Department of Energy, October 1, 2015.
Available at http://www.eia.gov/naturalgas/data.cfm.
Elliott, Graham, Thomas J. Rothenberg, and James H. Stock, “Efficient Tests for
an Autoregressive Unit Root,” Econometrica, 1996, 64 (4), pp. 813–836.
Engle, R.F. and C.W.J. Granger, “Co-integration and Error Correction:
Representation, Estimation, and Testing,” Econometrica, 1987, pp. 251–276.
Greene, W.H. and C. Zhang, Econometric Analysis, Vol. 7, Upper Saddle River, NJ:
Prentice Hall, 2011.
Gregory, Allan W. and Bruce E. Hansen, “Residual-based tests for cointegration in
models with regime shifts,” Journal of Econometrics, 1996, 70 (1), 99–126.
Hartley, Peter R, Kenneth B Medlock III, and Jennifer E Rosthal, “The
relationship of natural gas to oil prices,” The Energy Journal, 2008, pp. 47–65.
Joskow, Paul L. and Jean Tirole, “Transmission rights and market power on electric
power networks,” The Rand Journal of Economics, 2000, 31 (3), 450–487.
Kejriwal, M., P. Perron, and J. Zhou, “Wald tests for detecting multiple structural
changes in persistence,” Econometric Theory, 2012, 29 (1), 1–35.
Klein, Benjamin, Robert G. Crawford, and Armen A. Alchian, “Vertical
Integration, Appropriable Rents, and the Competitive Contracting Process,” Journal of
Law and Economics, 1978, 21 (2), 297–326.
Kwiatkowski, Denis, Peter CB Phillips, Peter Schmidt, and Yongcheol Shin,
“Testing the null hypothesis of stationarity against the alternative of a unit root: How
sure are we that economic time series have a unit root?,” Journal of Econometrics, 1992,
54 (1), 159–178.
Leitzinger, J. and M. Collette, “A Retrospective Look at Wholesale Gas: Industry
Restructuring,” Journal of Regulatory Economics, 2002, 21 (1), 79–101.
37
Lyon, Thomas P. and Steven C. Hackett, “Bottlenecks and Governance Structures:
Open Access and Long-Term Contracting in Natural Gas,” Journal Law Economics &
Organization, 1993, 9 (2), 380–398.
Makholm, Jeff D., The Political Economy of Pipelines: A Century of Comparative
Institutional Development, Chicago: University of Chicago Press, 2012.
Marmer, V., D. Shapiro, and P. MacAvoy, “Bottlenecks in Regional Markets for
Natural Gas Transmission Services,” Energy Economics, 2007, 29 (1), 37–45.
McGrew, J.H., FERC: Federal Energy Regulatory Commission, American Bar
Association, 2009.
Murry, D. and Z. Zhu, “Asymmetric Price Responses, Market Integration and Market
Power: A Study of the US Natural Gas Market,” Energy Economics, 2008, 30 (3),
748–765.
Nyblom, J. and A. Harvey, “Tests of common stochastic trends,” Econometric Theory,
2000, 16 (2), 176–199.
Oliver, Matthew E, Charles F Mason, and David Finnoff, “Pipeline congestion and
basis differentials,” Journal of Regulatory Economics, 2014, 46 (3), 261–291.
Pesaran, M. Hashem, Time Series and Panel Data Econometrics, New York, NY:
Oxford University Press, 2015.
Reiter, Doris F. and Michael J. Economides, “Structure and economics of the
natural gas industry following deregulation,” Advances in the Economics of Energy and
Resources, 1999, 11, 161–186.
Sebastian, Nick and Stefan Thoenes, “What drives natural gas prices?A structural
VAR approach,” Energy Economics, 2014, 45, 517–527.
Serletis, A. and R. Rangel-Ruiz, “Testing for Common Features in North American
Energy Markets,” Energy Economics, 2004, 26 (3), 401–414.
Villar, Jose A and Frederick L Joutz, “The relationship between crude oil and natural
gas prices,” Energy Information Administration, Office of Oil and Gas, 2006, pp. 1–43.
38
8 Tables
Table 1: Florida Pipeline Capacity Figures: Estimation Sample
FGT Gulfstream Total
Days at Specified Capacity Factors
Capacity Factor Days Days Days
100% 29 31 9
>99% 43 116 44
>98% 56 147 59
>97% 66 176 78
>96% 78 203 116
>95% 105 227 144
>90% 228 357 246
Pipeline Capacities
As Of MMcf/d MMcf/d MMcf/d
10/1/2006 2,250 1,114 3,364
1/7/2009 2,250 1,118 3,368
1/12/2009 2,250 1,253 3,503
4/1/2009 2,250 1,263 3,513
4/5/2010 2,250 1,273 3,523
5/21/2010 2,250 1,277 3,527
8/13/2010 2,250 1,307 3,557
4/2/2011 2,250 1,330 3,580
4/4/2011 2,950 1,330 4,280
7/12/2011 3,050 1,330 4,380
Note: We calculate the daily capacity factor as the daily
flow divided by the stated daily pipeline capacity, and then
convert to percent.
39
Table 2: Southern California Pipeline Capacity Figures: Estimation Sample
Kern River El Paso El Paso Transwestern Total
Veyo North South West SoCal
Days at Specified Capacity Factors
Cap. Factor Days Days Days Days Days
100% 197 298 15 0 1
>99% 575 375 21 0 2
>98% 714 418 27 0 8
>97% 818 454 32 0 16
>96% 889 487 41 3 30
>95% 949 524 48 25 48
>90% 1092 676 102 419 282
Pipeline Capacities
As Of MMcf/d MMcf/d MMcf/d MMcf/d MMcf/d
10/1/06 2,042 2,190 2,200 1300 7,732
4/9/10 2,197 2,190 2,200 1300 7,887
Note: See notes from table 1.
Table 3: Florida Summary Statistics: Oct. 3, 2006 – Aug. 29, 2011
Variable Units Source Mean Std. Dev. Min. Max. N
Florida Price $/MMBtu NGI 7.011 2.861 2.333 18.081 961
Henry Hub Price $/MMBtu NGI 6.15 2.548 2.127 13.879 961
WTI Price $/MMBtu EIA 14.375 3.87 5.912 26.125 961
Eastern Coal Price $/MMBtu EIA 3.173 0.358 2.534 3.732 961
FL Industrial Price Index BLS 96.282 5.359 88.409 109.12 961
FL Personal Income Billion $ BEA 0.755 0.031 0.713 0.802 961
Heating Degree Days HDD NOAA 2.522 4.825 0 27.57 961
Cooling Degree Days CDD NOAA 9.247 6.725 0 22.312 961
Hurricane Shutins Bcfd BOEMRE 0.214 0.884 0 7.062 961
Total System Flow MMcf/d LCI 2730.516 533.503 1563.141 4380.003 961
Total System Capacity MMcf/d LCI 3482.303 179.396 3364 4380 961
Capacity Factor Proportion 0.782 0.136 0.465 1.073 961
Binding Capacity 95+% Indicator 0.0?? 0.0?? 0 1 961
Binding Capacity 99+% Indicator 0.0?? 0.0?? 0 1 961
Total Capacity 100+% Indicator 0.009 0.096 0 1 961
Notes: All prices in real 2011 dollars. The capacity factors and binary indicators for binding system capacity
are calculated from the LCI data. Binding Capacity 95+% is a binary variable assigned a value of one if
daily flow is 95 percent or more of stated system capacity, and a value of zero otherwise. Binding Capacity
99+% and 100+% are defined similarly. The heating and cooling degree day measures are calculated from
NOAA weather station data—see discussion and illustrative example in the text.
40
Table 4: California Summary Statistics: Jan. 2, 2007 – Aug. 29, 2011
Variable Units Source Mean Std. Dev. Min. Max. N
SoCal Price $/MMBtu NGI 5.637 2.264 2.292 13.3 1164
Henry Hub Price $/MMBtu NGI 6.047 2.487 1.9 13.879 1164
WTI Price $/MMBtu EIA 14.445 3.826 5.397 26.125 1164
Rockies Coal Price $/MMBtu EIA 1.441 0.12 1.219 1.833 1164
CA Industrial Price Index BLS 96.803 8.132 82.337 115.657 1164
CA Personal Income Billion $ BEA 1.643 0.05 1.571 1.72 1164
Heating Degree Days HDD NOAA 8.572 7.343 0.115 30.029 1164
Cooling Degree Days CDD NOAA 3.271 4.082 0 16.471 1164
Total System Flow MMcfd LCI 5836.85 845.864 3418.423 7087.747 1164
Total System Capacity MMcfd LCI 6409.547 898.024 4166.932 7341.928 1164
Capacity Factor Proportion 0.912 0.058 0.553 1.032 1164
Binding Capacity 95+% Indicator 0.041 0.199 0 1 1164
Notes: See notes to table 3.
41
Table 5: Cointegration Tests for Florida and Southern California Prices
with the Henry Hub Price
Dependent Variable Florida Price SoCal Price
Henry Hub Price 1.011*** 0.885***
(0.016) (0.006)
Constant 0.785*** 0.286***
(0.105) (0.041)
Number of Observations 960 1163
R-squared 0.812 0.945
Test Statistics
ADF -11.211*** -5.423***
Phillips-Perron -11.127*** -4.211***
KPSS Test
Test Statistic 0.352 0.423
Lags 18 21
Gregory Hansen Test
Test Statistic -8.55*** -5.29***
Break Date 2/8/08 6/18/08
Notes: Both regional markets are tested for cointegration
with Henry Hub. California cointegration test results are
comparable when Opal prices is substituted for Henry Hub
prices. ADF and PP tests are performed on levels without
trend. 1% Critical Value: -3.43. The KPSS test is conducted
on levels without trend. KPSS Critical Value: 5% : 0.463.
Gregory-Hansen values and dates are reported including ca-
pacity factor and six fixed lags, determined using the ADF
test. *** p<0.01
42
Table 6: Florida: Determinants of Natural Gas Prices - All Pipelines
Binding Capacity 95+% Binding Capacity 99+% Binding Capacity 100+%
(1) (2) (3) (4) (5)
Constant -1.5399*** -1.0758*** -1.3189*** -1.0375*** -1.1360***
(0.4380) (0.3868) (0.3520) (0.3213) (0.3093)
Real HH Price 1.0629*** 1.0696*** 1.0151*** 1.0117*** 1.0113***
(0.0632) (0.0562) (0.0528) (0.0497) (0.0526)
Real WTI Price 0.0599** 0.0404** 0.0683*** 0.0582*** 0.0561**
(0.0242) (0.0200) (0.0214) (0.0202) (0.0226)
Heating Degree Days 0.1781*** 0.1043*** 0.1879*** 0.1618*** 0.1749***
(0.0365) (0.0211) (0.0399) (0.0426) (0.0388)
Cooling Degree Days 0.1183*** 0.0962*** 0.1256*** 0.1212*** 0.1364***
(0.0164) (0.0153) (0.0198) (0.0204) (0.0193)
Hurricane Shutins, Bcfd 0.3760*** 0.3287*** 0.3134*** 0.3018*** 0.3124***
(0.0932) (0.0984) (0.0902) (0.0933) (0.0911)
Binding Capacity 95+% 0.7853*** -2.6523**
(0.2257) (1.1553)
HDD × Binding Capacity 95+% 0.2870***
(0.0851)
CDD × Binding Capacity 95+% 0.1735**
(0.0695)
Binding Capacity 99+% 1.7130*** -1.1257
(0.4807) (2.1845)
HDD × Binding Capacity 99+% 0.2563**
(0.1266)
CDD × Binding Capacity 99+% 0.1203
(0.1212)
Binding Capacity 100+% 4.5361***
(1.0133)
N 961 961 961 961 961
Notes: The dependent variable in all regressions is the daily natural gas price for Florida for the period
Oct. 6, 2006 - Aug. 29, 2011. Binding Capacity 95+% is a binary variable assigned a value of one
if daily flow is 95 percent or more of stated system capacity, and a value of zero otherwise. Binding
Capacity 99+% and 100+% are defined similarly. Newey-West standard errors are in parentheses. All
specifications include month-of-year fixed effects (with December as the excluded month), as well as year
fixed effects (with 2006 as the excluded year). All prices are expressed in real 2011 dollars. *** p<0.01,
** p<0.05, * p<0.1. All p-values are for two-tailed tests.
43
Table 7: Determinants of Natural Gas Prices with Piecewise Capacity Measures:
Florida - All Pipelines
(1) (2) (3) (4)
Binding Capacity 100+% 5.3213*** 5.4498*** 5.4664*** 5.4593***
(1.0002) (1.0207) (1.0205) (1.0224)
Capacity Factor 95–100% 0.9804***
(0.2016)
Capacity Factor 99–100% 1.6584***
(0.3101)
Capacity Factor 99.84–100% 1.8639***
(0.3611)
Capacity Factor 99–99.84% 1.4827***
(0.3163)
Capacity Factor 99.939–100% 1.8243***
(0.4520)
Capacity Factor 99.52–99.939% 2.1165***
(0.3439)
Capacity Factor 99–99.52% 1.0354***
(0.2892)
Capacity Factor 95–99% 0.8458*** 0.8529*** 0.8562***
(0.2137) (0.2135) (0.2135)
Capacity Factor 90–95% 0.5456*** 0.5929*** 0.5995*** 0.5994***
(0.1596) (0.1551) (0.1544) (0.1547)
Capacity Factor 85–90% 0.3978*** 0.4472*** 0.4520*** 0.4498***
(0.1296) (0.1232) (0.1234) (0.1231)
Capacity Factor 80–85% 0.2335** 0.2607** 0.2638** 0.2633**
(0.1065) (0.1058) (0.1059) (0.1060)
Capacity Factor 75–80% 0.0425 0.0608 0.0624 0.0637
(0.0775) (0.0756) (0.0756) (0.0756)
N 961 961 961 961
Notes: The dependent variable in all regressions is the daily natural gas price for Florida
for the period Oct. 6, 2006 - Aug. 29, 2011. Binding Capacity 100+% is a binary variable
assigned a value of one if daily flow is 100 percent or more of stated system capacity, and
a value of zero otherwise. Each Capacity Factor variable is a binary variable assigned
a value of one if the flow level is in the indicated range and a value of zero otherwise.
Newey-West standard errors are in parentheses. All specifications include month-of-year
fixed effects (with December as the excluded month), as well as year fixed effects (with 2006
as the excluded year) and other covariates from table 6. The piecewise capacity variables
for capacity factors between 99 and 100 percent in models 3 and 4 are defined to have
approximately the same number of observations in each binary bin. *** p<0.01, ** p<0.05.
All p-values are for two-tailed tests.
44
Table 8: Florida Capacity Expansion Impacts
(1)
Flow> Old Capacity 5.2356***
(0.8807)
Increased Capacity -0.5122**
(0.2005)
Flow> Old Capacity × Increased Capacity -5.6673***
(0.8854)
N 961
Notes: The dependent variable is the daily natural gas price for Florida
for the period Oct. 6, 2006 - Aug. 29, 2011. Flow> Old Capacity is a
binary variable assigned a value of one if the daily flow is greater than
the pre-expansion capacity level and a value of zero otherwise. Increased
Capacity is a binary variable assigned a value of one for days after the
capacity expansion and a value of zero otherwise. Newey-West standard
errors are in parentheses. The regression specification includes month-
of-year fixed effects (with December as the excluded month), as well as
year fixed effects (with 2006 as the excluded year). Other covariates
from table 6 are also included and estimated coefficients are available
on request. *** p<0.01, ** p<0.05 (two-tailed test)
45
Table 9: California: Determinants of Natural Gas Prices - All Pipelines
(1) (2) (3)
Constant -0.6887** -0.7029** -0.6802**
(0.2989) (0.2960) (0.2969)
Real HH Price 0.8613*** 0.8657*** 0.8639***
(0.0326) (0.0334) (0.0334)
Real WTI Price 0.0786*** 0.0742*** 0.0783***
(0.0155) (0.0156) (0.0156)
Heating Degree Days 0.0284*** 0.0280*** 0.0277***
(0.0060) (0.0058) (0.0059)
Cooling Degree Days 0.0412*** 0.0372*** 0.0406***
(0.0094) (0.0096) (0.0094)
Binding Capacity 95+% 0.3201*** 0.4060*** 0.3258***
(0.0943) (0.1107) (0.0878)
Capacity Factor 90–95% 0.1477**
(0.0732)
Capacity Factor 85–90% 0.0232
(0.0510)
Kern at Capacity 100+% -0.0121
(0.0419)
El Paso N at Capacity 100+% -0.0374
(0.0648)
El Paso S at Capacity 100+% 0.0371
(0.0747)
Transwestern at Capacity ≥ 95% 0.0991
(0.0965)
N 1164 1164 1164
Notes: The dependent variable in all regressions is the daily natural
gas price for Southern California for the period Jan. 2, 2007 - Aug. 29,
2011. Binding Capacity 95+% is a binary variable assigned a value of
one if daily flow is 95 percent or more of stated system capacity, and a
value of zero otherwise. Each Capacity Factor variable is a binary vari-
able assigned a value of one if the flow level is in the indicated range and
a value of zero otherwise. Each of the final four binary Capacity vari-
ables in the table is assigned a value of one when the respective individ-
ual pipeline is at the level indicated in the variable name. Newey-West
standard errors are in parentheses. All specifications include month-
of-year fixed effects (with December as the excluded month), as well
as year fixed effects (with 2007 as the excluded year). All prices are
expressed in real 2011 dollars. *** p<0.01, ** p<0.05, * p<0.1. All
p-values are for two-tailed tests.
46
9 Figures
Figure 1: Hub and Regional Market Prices
(a) Henry Hub and Florida Daily Gas Prices
(b) Henry Hub and Southern California Daily Gas
Prices
47
Figure 2: Representation of the Wholesale Natural Gas Market
DH
DL
S
Capacity
Q
p
p0
Q0
pc
Qc
p1
Q1
Sg
C
48
Figure 3: Stylized Map of Florida and Southern California Pipelines
49
Figure 4: Florida Capacity Expansion Effects
50
10 Appendixes
A Regulatory Context
Prior to the 1980s, producers did not sell their gas directly to gas consumers, or to local
distribution companies (LDCs), who are important wholesale buyers.37 Instead, gas was
initially sold to interstate pipeline companies at prices that, until 1989, were regulated by
the Federal Energy Regulatory Commission (FERC). Pipeline companies then offered the
entire range of merchant functions by providing the commodity, as well as transportation,
storage, and portfolio management services to downstream gas buyers. Pipelines charged a
single regulated rate for these bundled services, with their primary buyers being LDCs and
industrial consumers. After buying gas from pipeline companies, LDCs sold gas to
residential and commercial end-users at regulated rates.
Between 1978 and 1992, natural gas production and transportation markets were
dramatically changed by a series of regulatory changes that allowed market forces greater
sway. In the natural gas market, wellhead price controls were removed in 1984 by Order
436. In the market for gas transportation services, the owners of pipelines were required to
decouple the transportation services they offered from commodity sales and other merchant
functions. As a result, pipeline companies no longer take ownership of natural gas. They
are now limited to providing transportation services for a (regulated) fee, and they are
required to offer their transportation services to all market participants. This requirement
of “open access” to transportation services has resulted in wholesale natural gas markets
that, in the United States today, are characterized by many producers and marketers
competing to provide gas to a multitude of customers through bilateral contracts covering
a broad range of time horizons. In addition, spot and futures markets provide price
discovery and allow for indexed contracts.
There are both primary and secondary markets for pipeline transportation services. As
a rudimentary example of how these two markets function and relate to one another,
suppose an LDC (the shipper in this scenario) purchases the primary right to ship a stated
daily volume of natural gas from a producing region to the LDC’s location during a
specified time period, typically a month. This primary right can be firm or it can be
interruptible. A shipper that purchases firm primary rights is guaranteed access to its
37McGrew (2009) and Makholm (2012) are good sources for additional details on the institutional infor-
mation provided below.
51
contracted daily transportation capacity for the duration of the contract.38 Interruptible
service, on the other hand, allows the LDC to move a specified volume of gas within a
specified time period, but the actual timing of the contracted gas transportation is at the
discretion of the pipeline company.
A shipper may choose to purchase primary capacity, but due to fluctuating demand,
there may be times when additional deliveries are less valuable than anticipated. In this
case, two options are available to the shipper. First, the primary capacity can be released
back to the pipeline company, which can then resell the capacity to another buyer. It is
possible, though not guaranteed, that the original shipper could be reimbursed for up-front
payments in this case. Alternatively, the LDC has the option of selling its primary
contracted transportation services into a secondary market—an option it is most likely to
choose to exercise during a period when there is a price spike (due, for example, to a
binding pipeline capacity constraint). It is noteworthy that in this scenario, the party that
gains from the price spike is the shipper, not the pipeline company. This feature of the
secondary markets mutes a valuable signal for decisions regarding long-run pipeline
investments.39
The contract between the shipper and the transporter specifies a maximum daily
quantity the shipper can transport. Pipeline transportation contracts typically consist of
three parts. The first component—the reservation fee—is fixed with respect to both the
quantity of gas actually shipped and the price of natural gas. The shipper is responsible for
payment of the product of the reservation fee and the maximum daily quantity. The second
component—the usage fee—varies with the quantity of natural gas actually shipped, but
not with the price of natural gas. The shipper is also responsible for usage payments equal
to the volume of natural gas actually shipped multiplied by the usage fee. The third
component—the fuel reimbursement charge—varies with both the quantity of gas shipped
and the price of natural gas. Because natural gas is used to operate the pipeline, the
purpose of the variable charge is to compensate the pipeline for these costs. The buyer of
the transportation service thus pays the fuel cost by providing this gas to the pipeline.
38As with many other types of contracts, force majuere is an exception. Further detail about the structure
of transportation contracts is provided below.
39In 2010 FERC Order 736 was issued to allow pipeline companies to profit from released capacity, and
thereby receive an important signal regarding investments in pipeline capacity expansions. Implementation as
intended would have eliminated the secondary market and channeled all unused capacity back to the pipeline
company for resale to other shippers. A substantial secondary market in capacity still exists, however. See
Doane et al. (2008) for a discussion of the competitiveness of the secondary market prior to Order 736. We
are not aware of any similar studies of this market since Order 736 went into effect.
52
As an illustration, consider the rate schedule of Florida Gas Transmission Company
(FGT), one of the pipeline companies (or transporters) whose operations are relevant for
our empirical analysis.40 For primary firm transportation contracts the shipper pays all
three components. The currently effective maximum rate for the reservation fee is 46.94
¢/MMBtu.41 The specified usage fee is 2.56 ¢/MMBtu and the maximum fuel
reimbursement charge is 3.17 percent of the value of the natural gas shipped.
A shipper who has a contract for interruptible primary services does not pay the
reservation fee (because the transporter has the right to determine when the contracted
volumes will be shipped), but does pay the usage fee and the fuel reimbursement charge. In
what appears to be an unusual case, on FGT the usage fee for interruptible transportation
services is larger than the total tariff for firm services. For FGT the maximum usage fee for
interruptible services that corresponds to the firm contract usage fee above is 61.04
¢/MMBtu.
Actual contracts can vary along a number of margins from the three part charges
described above. For example, different classes of customers can be charged different prices,
usage rates may vary with the distance the natural gas is shipped, and fuel reimbursement
charges may be determined by the number of compressor stations through which
contracted gas is shipped. In addition, the rates discussed above represent maximum rates,
and negotiated rates with individual customers may be lower than these maximum rates.
40See http://fgttransfer.energytransfer.com/InfoPost/tariff/fgt/FGTTariff.pdf.
41FGT has different rate schedules for different classes of customers. The rates in the text are for FTS-1
customers.
53
B Graphical Representation of the Regional Market
Model
In this appendix, we provide and briefly discuss a graphical representation of a simple
two-pipeline natural gas model. Even with only two pipelines, there are many options for
how to represent the mathematical model presented in the text. For graphical tractability,
as can be seen in figures B.1a, B.1b, and B.1c, we depict the two pipelines in the system
as having different marginal costs, MC1 = S1 and MC2 = S2, and the same capacities, q
c
1
= qc2. The two marginal cost (supply) curves are drawn to be parallel, with different
intercepts S1(0) and S2(0). The supply of natural gas from this system is the horizontal
summation of the two marginal cost curves, subject to the constraint that the flow through
each pipeline may not exceed its capacity. This yields the bold, kinked supply curve
labeled S1 + S2 = SMKT in each of the three graphs below. For prices between S1(0) and
S2(0), natural gas will be supplied only from pipeline 1. For prices between S2(0) and
MCC1 , gas will be shipped through both pipelines. For prices greater than MC
C
1 , the
capacity constraint for the first pipeline is binding. If price is between MCC1 and MC
C
2
additional gas will be provided (only) by the second pipeline. If price exceeds MCC2 , both
pipeline constraints are binding and the supply curve for the system is vertical.
Figure B.1a illustrates Case A from the text. Both pipelines are at capacity, so total
production is QC , the market price is P ∗A, and the shadow values of capacity are λ1 = P
∗
A -
MCC1 and λ2 = P
∗
A - MC
C
2 .
Figure B.1b corresponds to Case B from the text. The market clearing price is P ∗B and
the quantity is Q∗T . The quantity q
C
1 is shipped through pipeline 1, and q
∗
2 is shipped
through pipeline 2. The shadow value of capacity is λ1 = P
∗
B - MC
C
1 for pipeline 1 and
λ2 = 0 for pipeline 2.
Case C is depicted in figure B.1c. With the demand curve DC , the market clearing price
and quantity are P ∗C and Q
∗
T . The quantity q
∗
1 is shipped through pipeline 1, with q
∗
2 being
shipped through the second pipeline. Note that as the demand curve shifts further to the
left, the market price will fall to a level less than S2(0) and the market clearing quantity
will all be provided by pipeline 1.
54
F
ig
u
re
B
.1
:
G
ra
p
h
ic
al
D
ep
ic
ti
on
of
a
T
w
o-
P
ip
el
in
e
S
y
st
em
(a
)
G
ra
p
h
of
C
as
e
A
(b
)
G
ra
p
h
of
C
as
e
B
(c
)
G
ra
p
h
o
f
C
a
se
C
55
C Florida and Southern California Pipelines
Natural gas consumption in Florida is dominated by the electric power sector, which
accounts for over 80 percent of all consumption. The industrial sector consumes nearly 9
percent, while residential and commercial sectors combine for about 8 percent. Mt.
Vernon, a compressor station near the border of Mississippi and Alabama, is a point where
capacity often binds on the FGT trunkline. A similar point on Gulfstream is Station 200
(St. 200 in figure 3) off the Florida Coast. The daily capacity and flow data for these two
points can be seen in figures C.1 and C.2. It is notable that there are numerous instances
in the two figures of daily flows reaching capacity.
As can be seen in figure 3, if a pipeline transporting natural gas into Southern
California nears capacity during a period of high demand, there are three alternative
pipelines through which adjustments might be made. Storage withdrawals and increased
in-state production provide additional alternative supply sources. The availability of these
alternative supply sources suggests the price responses to binding pipeline constraints will
be lower in California than in the Florida markets we study. The responsiveness of storage
withdrawals and in-state production to increased natural gas prices are not precisely
known, but we expect them to be positive. In 2010, 32.4 percent of natural gas
consumption in Southern California was in the electric power sector, 30.9 percent was in
the industrial sector, 21.8 percent was in the residential sector, and 10.9 percent was in the
commercial sector.
The Veyo compressor station, in the southwest corner of Utah, is a point where capacity
sometimes binds along the Kern River pipeline. Potential bottlenecks on the El Paso
system are in transporting gas west along the North system in Arizona, and in moving gas
west from Texas into New Mexico. Along the Transwestern system, which transports gas
west through Arizona, there is also a potential bottleneck at the West compressor station.
The daily capacity and flow data for these points are displayed in figures C.3, C.4, C.5,
and C.6. Because both Kern River and the El Paso North pipelines are above ground, their
capacity fluctuates due to temperature changes. Falling flows in recent years through El
Paso North have been attributed to falling production in the mature San Juan Basin fields
that feed the pipeline.
56
Figure C.1: FGT Mt. Vernon Throughput Capacity and Flows
Figure C.2: Gulfstream Station 200 Throughput Capacity and Flows
57
Figure C.3: Kern River Veyo Station Throughput Capacity and Flows
Figure C.4: Transwestern West Throughput Capacity and Flows
58
Figure C.5: El Paso North Throughput Capacity and Flows
Figure C.6: El Paso South Throughput Capacity and Flows
59
D Sources, Summary Statistics, and Cointegration
Analysis for 14 Regional Markets
In this appendix, we provide our data sources, as well as additional detail on our
examination of the 14 regional prices for evidence regarding market integration. Table D.1
reports the sources for all the data we use in our empirical analysis and table D.2 presents
summary statistics for the price data we obtained for 14 regions for the period from
October 1, 2006 - August 29, 2011. Table D.3 presents the results of tests for the
stationarity properties of each of the 14 regional price series in our data. The ADF tests we
run indicate that each of the 14 price series is I(1) in levels and I(0) in differences.
Table D.4 presents the results from our tests for pairwise cointegration between each of the
14 price series. As indicated in section 4.1 in the text of the paper, we use the errors from
each regression of one regional price on another regional price to conduct ADF tests for
whether the coefficient on α1 in equation (5) is significantly less than 1.0. The null
hypothesis in this test is that α1 is equal to 1.0, or equivalently that the prices are not
cointegrated. In table D.4 it is seen that the null hypothesis is rejected at standard
significance levels for every pair of prices—that is, every pair of prices is cointegrated.
Table D.1: List of Data
Data Units Source Frequency
Pipeline Capacities MMcfd LCI Inc. Daily
Pipeline Flows MMcfd LCI Inc. Daily
Natural Gas Prices $/MMBtu NGI Gas Daily Daily
WTI Oil Spot Prices $/MMBtu EIA Daily
Coal Prices $/MMBtu EIA Monthly
Heating Degree Days HDD MSU-NOAA Daily
Cooling Degree Days CDD MSU-NOAA Daily
Industrial Output Prices Index BLS Monthly
Producer Price Index Index BLS Monthly
State Personal Income Billion $ BEA Monthly
Hurricane Shut-ins MMcfD BOEMRE Incident Reports
60
Table D.2: Summary Statistics for 14 Regional Prices: October 1, 2006 – August 29, 2011
Variable Region Mean Std. Dev. Min. Max. N
Henry Hub Price South 5.79 2.23 1.84 13.31 1250
Malin Price Northwest 5.48 2.03 2.23 12.31 1252
PG&E Citygate Price West 5.84 2.07 2.63 12.82 1252
So Cal Avg Price West 5.42 2.02 2.22 12.68 1252
Katy Hub Price Texas 5.60 2.13 1.95 12.98 1252
FGT Z3 Citygate Price Southeast 6.73 2.65 2.26 18.00 1013
Chicago Citygate Price Midwest 5.81 2.20 2.00 13.15 1252
NE Columbia Price Northeast 6.01 2.32 1.94 13.85 1252
Kingsgate Price Northwest 5.29 1.95 2.05 11.94 1221
Kern River Hub Price Rockies 4.38 1.793 0.12 10.29 1252
Opal Hub Price Rockies 4.39 1.794 0.12 10.26 1252
Stanfield Price Northwest 5.40 2.00 2.14 12.08 1252
FGT Z3 AL-MS Price Southeast 5.98 2.36 1.86 14.39 1252
Waha Hub Price Texas 5.29 1.99 1.99 12.33 1252
Notes: All prices are in real 2011 dollars.
61
Table D.3: Results of ADF Unit Root Tests on Regional Prices
ADF on Levels ADF on First Difference
Price Series Test Statistic p-Value Test Statistic p-Value
Henry Hub -0.98 0.7623 -30.05 0.0000
Malin -1.39 0.5867 -33.19 0.0000
PG&E Citygate -1.26 0.6494 -34.86 0.0000
SoCal Avg. -0.88 0.7943 -38.52 0.0000
Katy Hub -0.92 0.7822 -27.08 0.0000
FGT Zone 3 Citygate -2.13 0.2332 -19.51 0.0000
Chicago Citygate -1.42 0.5746 -28.25 0.0000
NE Columbia -1.46 0.5531 -28.90 0.0000
Kingsgate -2.02 0.2759 -31.45 0.0000
Kern River -2.21 0.2024 -38.17 0.0000
Opal Hub -2.24 0.1928 -37.80 0.0000
Stanfield -1.60 0.4822 -32.87 0.0000
FGT Zone 3 AL-MS -1.25 0.6530 -29.22 0.0000
Waha Hub -0.93 0.7775 -31.04 0.0000
Notes: All prices are in real 2011 dollars. Augmented Dickey-Fuller (ADF) tests for unit
roots are run on each of the 14 price series. The null hypothesis for the ADF test is that
the series in question is non-stationary (i.e., it has a unit root). Test results in the two
columns under the heading “ADF on Levels” indicate uniform failure to reject the null,
or equivalently that all 14 price series are non-stationary in levels. Test results in the last
two columns indicate uniform rejection of the null that each series is non-stationary in first
differences, or equivalently that all 14 series are stationary in first differences.
62
T
ab
le
D
.4
:
t-
st
at
is
ti
cs
fr
om
P
ai
rw
is
e
A
D
F
T
es
ts
fo
r
C
oi
n
te
gr
at
io
n
of
R
eg
io
n
al
P
ri
ce
s
P
r
ic
e
S
e
r
ie
s
H
en
ry
H
u
b
M
a
li
n
P
G
&
E
S
o
C
a
l
K
a
ty
F
G
T
C
G
C
h
ic
a
g
o
C
o
lu
m
b
ia
K
in
g
sg
a
te
K
er
n
O
p
a
l
S
ta
n
fi
el
d
F
G
T
A
L
-M
S
W
a
h
a
H
en
ry
H
u
b
–
-5
.7
3
-5
.7
6
-1
4
.2
3
-7
.4
0
-9
.9
4
-1
4
.4
9
-8
.0
2
-5
.2
5
-5
.9
4
-5
.8
6
-5
.5
0
-7
.7
5
-1
0
.7
8
M
a
li
n
-5
.6
5
–
-1
0
.5
2
-1
8
.6
0
-5
.0
3
-9
.2
3
-9
.0
4
-3
.9
6
-1
1
.7
9
-6
.3
4
-6
.2
7
-1
0
.6
6
-6
.3
8
-1
1
.1
7
P
G
&
E
C
it
y
g
a
te
-5
.7
2
-1
0
.5
5
–
-1
7
.8
1
-4
.6
1
-9
.1
1
-8
.8
4
-4
.2
9
-1
0
.2
0
-6
.2
7
-6
.2
0
-1
0
.4
8
-6
.1
2
-1
0
.2
2
S
o
C
a
l
A
v
g
.
-1
4
.6
9
-1
9
.1
1
-1
8
.2
9
–
-1
6
.3
8
-9
.8
4
-1
6
.0
8
-1
3
.9
1
-1
6
.3
1
-5
.4
4
-5
.3
1
-1
8
.1
6
-1
1
.7
1
-8
.9
9
K
a
ty
H
u
b
-7
.4
7
-5
.1
6
-4
.7
0
-1
6
.0
2
–
-1
0
.3
1
-1
0
.7
8
-4
.9
3
-4
.3
0
-6
.0
4
-5
.9
9
-4
.9
3
-7
.3
6
-1
2
.8
1
F
G
T
Z
o
n
e
3
C
it
y
g
a
te
-9
.1
0
-8
.2
5
-8
.0
8
-8
.6
6
-9
.3
4
–
-9
.1
6
-9
.1
1
-7
.9
8
-5
.5
3
-5
.4
1
-8
.1
0
-9
.8
4
-9
.0
0
C
h
ic
a
g
o
C
it
y
g
a
te
-1
4
.4
5
-9
.0
2
-8
.8
0
-1
5
.6
1
-1
0
.6
6
-1
0
.1
5
–
-1
2
.5
3
-8
.0
8
-5
.9
4
-5
.8
5
-8
.3
8
-8
.8
6
-1
1
.0
7
N
E
C
o
lu
m
b
ia
-7
.9
5
-3
.9
3
-4
.2
3
-1
3
.3
6
-4
.8
0
-1
0
.0
0
-1
2
.4
8
–
-3
.4
4
-5
.8
7
-5
.8
1
-3
.7
4
-7
.5
2
-9
.5
9
K
in
g
sg
a
te
-5
.0
1
-1
1
.7
4
-1
0
.1
2
-1
6
.0
0
-4
.0
5
-8
.8
8
-8
.0
2
-3
.3
4
–
-6
.3
1
-6
.2
3
-1
8
.9
9
-5
.3
1
-9
.7
0
K
er
n
R
iv
er
-5
.0
8
-5
.5
2
-5
.4
0
-4
.2
0
-5
.0
5
-6
.4
8
-5
.0
5
-5
.0
1
-5
.6
3
–
-3
2
.6
5
-5
.7
7
-4
.8
2
-4
.3
5
O
p
a
l
H
u
b
-5
.0
2
-5
.4
9
-5
.3
8
-4
.0
9
-5
.0
2
-6
.4
4
-4
.9
8
-5
.0
0
-5
.6
0
-3
2
.6
8
–
-5
.7
5
-4
.8
1
-4
.2
9
S
ta
n
fi
el
d
-5
.3
6
-1
0
.6
2
-1
0
.4
1
-1
7
.6
0
-4
.7
3
-9
.0
7
-8
.3
4
-3
.7
0
-1
9
.0
1
-6
.5
0
-6
.4
4
–
-5
.8
1
-1
0
.7
6
F
G
T
Z
o
n
e
3
A
L
-M
S
-7
.7
5
-6
.4
1
-6
.1
2
-1
1
.4
3
-7
.2
9
-1
0
.7
4
-8
.9
2
-7
.5
8
-5
.5
4
-5
.7
8
-5
.7
3
-5
.9
0
–
-9
.1
3
W
a
h
a
-1
1
.1
6
-1
1
.5
3
-1
0
.5
4
-8
.9
5
-1
3
.1
4
-1
0
.0
7
-1
1
.4
4
-1
0
.0
4
-9
.9
0
-5
.5
3
-5
.4
4
-1
1
.1
9
-9
.3
3
–
N
o
te
s:
A
D
F
cr
it
ic
a
l
v
a
lu
es
:
5
%
:
-2
.8
6
;
1
%
:
-3
.4
3
.
T
h
e
p
ri
ce
in
co
lu
m
n
i
is
re
g
re
ss
ed
o
n
th
e
p
ri
ce
in
co
lu
m
n
j
a
n
d
th
e
er
ro
r
te
rm
fr
o
m
ea
ch
o
f
th
es
e
re
g
re
ss
io
n
s
(s
ee
ε t
in
eq
u
a
ti
o
n
(4
)
in
th
e
te
x
t)
is
u
se
d
to
ru
n
A
u
g
m
en
te
d
D
ic
k
ey
-F
u
ll
er
(A
D
F
)
te
st
s
u
si
n
g
eq
u
a
ti
o
n
(5
).
T
h
e
re
p
o
rt
ed
t-
v
a
lu
es
a
b
o
v
e
a
re
fo
r
th
e
co
effi
ci
en
t
es
ti
m
a
te
o
f
α
1
fr
o
m
ea
ch
o
f
th
es
e
p
a
ir
w
is
e
re
g
re
ss
io
n
s.
A
s
ca
n
b
e
se
en
,
th
e
n
u
ll
h
y
p
o
th
es
is
(w
h
ic
h
fo
r
th
is
A
D
F
te
st
is
th
a
t
th
e
tw
o
p
ri
ce
se
ri
es
a
re
n
o
t
co
in
te
g
ra
te
d
),
is
re
je
ct
ed
fo
r
ev
er
y
p
ri
ce
p
a
ir
a
t
th
e
0
.0
5
le
v
el
.
63
E Supplementary Florida Regression Analyses
In this appendix, we present and briefly discuss alternative specifications for the Florida
regressions. We also provide detail on the price effects resulting from capacity constraints
that are implied by the coefficients estimates in model 4 of table 6 in the text.
Adding the Monthly Variables to Specifications in Table 6
In table E.1, we present results from the same regression specifications as in table 6,
models 1, 3, and 5, with the addition of the three variables that are available only at the
monthly level—the price of coal, personal income, and the price index for industrial output.
In table E.1, the difference between the three models is the definition of the capacity factor
at which the pipeline capacity constraint becomes binding. As is evident, the three monthly
variables are generally not individually significant, and their addition to the specifications
in table 6 in the text does not have substantive impacts on the estimated coefficients of the
other variables in the model. Most importantly, the addition of the three monthly variables
has no qualitative impact on the values of the Binding Capacity variables.
Individual Pipelines
To this point, we have pooled all the flows and the stated capacities of the FGT and
Gulfstream pipelines. In table E.2, we investigate the pipelines separately, as well as the
price effects of interactions between flow levels in the two pipelines in the system. In model
1, we define three binary capacity measures. The first two variables are assigned a value of
one when one of the two pipelines is full to stated capacity and the other is not, with both
variables assigned a value of zero if neither pipeline is full or if both pipelines are full. The
third variable is assigned a value of one when both pipelines are full to their stated
capacities. These three variables have a value of one on 22, 24, and seven days, respectively.
The estimated coefficients on these variables indicate the difference between prices on days
when these conditions hold relative to day when neither of the two pipelines is full. The
separate treatment of the capacity variables in the two pipelines has little impact on the
estimated coefficients on the other variables in the model, so we report only the coefficients
for the three variables described above. As can be seen, the estimated coefficients on each
of the three variables is significantly greater than zero. The estimated price impact when
FGT is full and Gulfstream is not, is roughly half as large as when the reverse holds ($1.10
as compared to $2.04), although these two values are not statistically different. When both
64
pipelines are full on the same day, prices are $4.66 greater than on days when neither of the
two pipelines is full. This price impact is statistically greater than each of the other two
individual price impacts and is not statistically different from the estimated coefficients on
the Binding Capacity 100+% variable in the final column of table 6 and in table 7.
In model 2 of table E.2, we explore whether the price effects of capacity constraints in
one pipeline are affected by flow levels in the other. The variable names are intuitive—for
example, the zero-one variable FGT 95-99, GS 90-95 is assigned a value of one on days
when the flow in the FGT pipeline is between 95 and 99 percent of stated capacity and the
flow in the Gulfstream pipeline is between 90 and 95 percent of its capacity. The estimated
coefficient on that variable measures the average difference between prices on days when
flows are in these ranges and days on which flows in one or both of the pipelines are less
than 90 percent of their stated capacities. The results suggest that the price effects of
capacity or near-capacity flows in each individual pipeline depends on the flow level in the
other pipeline.
Figures E.1a and E.1b display the coefficient estimates from model 2 as
two-dimensional depictions of the relationships between price and capacity factors in the
two pipelines. In figure E.1a, each of the four lines shows the impacts on price of an
increase in the FGT capacity factor for one of four different levels of the Gulfstream
capacity factor. The top line in the figure shows the impact on prices of each of four
different levels of daily FGT flows on days when the Gulfstream pipeline is at full capacity.
When the Gulfstream pipeline is full and the FGT pipeline capacity factor is between 99
and 100 percent, the price is $5.62 greater than on days when one or the other or both of
the pipelines are less than 90 percent full. If the price impacts of increasing flow in the
FGT pipeline were independent of the flow levels in the Gulfstream pipeline, then all four
of the lines in this figure would be coincident. The fact that they are not suggests the
presence of interactions. Moreover, most of the impacts of increases in the FGT capacity
factor are about $1 (and are mostly statistically insignificant) except when the Gulfstream
pipeline is full. In those instances, the price impacts of all four FGT levels shown are
statistically greater than zero, and when the FGT pipeline is flowing at 99 percent or more
of its stated capacity, prices are more than $5 greater than when flows are at less than 90
percent of capacity in one or the other, or both, of the pipelines.
In figure E.1b, each of the four lines shows the price impacts of an increase in the
Gulfstream capacity factor for one of four different levels of the FGT capacity factor. As in
the previous figure, the fact that the four lines are not coincident indicates important
65
interactions. For any given level of the Gulfstream capacity factor, the two lines for when
the FGT capacity factor is from 99-100 percent and 100 percent or greater lie above the
lines for the two lower level of FGT capacity factors (although several of the price effects
are not statistically different from each other). For all four of the FGT capacity factor
levels, the jump in price comes predominately when the Gulfstream pipeline is at full
capacity, and not before. Moreover, the increase in price when Gulfstream becomes full is
much greater when the FGT flow is at 99 percent or more of its capacity.
This analysis of the individual pipeline effects offers two insights. First, although our
results in model 1 of table E.2 suggest that there are substantial price effects when either
of the pipelines is flowing at full capacity, they also suggest that the price impacts are
substantially greater when both the pipelines are at full capacity. This is consistent with
our earlier results, which indicate that price effects are largely a result of no pathways
being available to get more gas to the regional market. Second, our results in model 2
suggest there are substantial price effects due to interactions between flows in the two
pipelines, but only when both pipelines are flowing at levels very close to capacity. This is
also consistent with our earlier results reported in table 7. These insights taken together
suggest that treating the two Florida pipelines individually reveals interesting and
important price effects related to pipeline capacity constraints. This approach also,
however, causes one to overlook important interactions between flow levels in the system of
pipelines that result in substantially larger price effects related to capacity constraints.
Calculating Capacity Impacts with Multiplicative Interactive Terms
In models 2 and 4 of table 6 in the text, changes in the short-run demand shifters, HDD
and CDD, have greater impacts when the pipelines are full, or nearly so, than on other
days. It is informative to compare the impacts of capacity being binding from the
specification in models 2 and 4 of table 6 with those in models 1 and 3 in table 6. Here, we
work through these calculations for model 4, where the threshold for the Binding Capacity
variable is 99 percent. The estimate of this impact in model 3 is $1.71.
To estimate comparable values for model 4 in table 6, we have to calculate marginal
effects taking all interaction terms into account. We make this calculation first on an
average HDD day and then on an average CDD day, where the average HDD value is taken
over the eight days (out of the 44 days with flow greater than 99 percent of capacity) with
positive HDD values, and the average CDD value is taken over the 36 days with positive
CDD values. Using this approach, our estimates of the impacts of binding capacity
66
constraints on HDD and CDD days are $4.31 and $1.13, respectively. Both these estimates
are significantly different from zero at the 0.01 level. The HDD marginal effect is also
significantly greater than the CDD marginal impact. Given that Florida is a market with
hot and humid summers and mild winters, we might expect more adaptations have been
made to accommodate hot rather than cold days. These two estimates can be compared
with the $1.71 estimate of binding capacity mentioned above by calculating the weighted
average of the estimates, where the weights are the proportions of HDD and CDD days out
of the 44 days defined to have binding capacity for this exercise. This weighted average is
$1.71 in model 4, the same as in model 3 (the estimates are not identical when calculated
to further precision levels).
67
Table E.1: Florida Regression Results with Monthly Variables - All Pipelines
(1) (2) (3)
Constant -16.4429 -11.7859 -21.5055*
(14.5473) (13.5640) (11.0299)
Real HH Price 1.0885*** 1.0342*** 1.0335***
(0.0710) (0.0574) (0.0560)
Real WTI Price 0.0775** 0.0798** 0.0670**
(0.0348) (0.0317) (0.0307)
Real Coal Price 0.5452 0.1895 0.6803
(0.7587) (0.6848) (0.5315)
Monthly FL Personal Income 22.9468 16.6940 29.1404**
(16.2284) (15.4124) (12.2571)
Real Ind. Price Index–FL -0.0589 -0.0408 -0.0559*
(0.0419) (0.0338) (0.0339)
Heating Degree Days 0.1798*** 0.1893*** 0.1747***
(0.0364) (0.0395) (0.0367)
Cooling Degree Days 0.1120*** 0.1206*** 0.1263***
(0.0167) (0.0192) (0.0178)
Hurricane Shutins, Bcfd 0.5232*** 0.4330*** 0.4841***
(0.1036) (0.0955) (0.0900)
Binding Capacity 95+% 0.7347***
(0.2187)
Binding Capacity 99+% 1.6138***
(0.4877)
Binding Capacity 100+% 4.6149***
(1.0264)
N 961 961 961
Notes: The dependent variable in all regressions is the daily natural gas price for
Florida for the period Oct. 6, 2006 - Aug. 29, 2011. Binding Capacity 95+% is a
binary variable assigned a value of one if daily flow is 95 percent or more of stated
system capacity, and a value of zero otherwise. Binding Capacity 99+% and
100+% are defined similarly. Newey-West standard errors are in parentheses. All
specifications include month-of-year fixed effects (with December as the excluded
month), as well as year fixed effects (with 2006 as the excluded year). All prices
are expressed in real 2011 dollars. *** p<0.01, ** p<0.05, * p<0.1. All p-values
are for two-tailed tests.
68
Table E.2: Price Effects of Individual Pipelines and Interactions in Florida
(1) (2)
Both Capacity 100+% 4.6593*** 4.9197***
(1.0178) (0.9978)
FGT 100+%, GS < 100% 1.0952***
(0.4036)
GS 100+%, FGT < 100% 2.0386**
(0.8579)
FGT 90–95, GS 100+ 1.4627***
(0.4675)
FGT 95–99, GS 90–95 1.2547
(0.8270)
FGT 95–99, GS 95–99 0.3038*
(0.1727)
FGT 95–99, GS 99–100 0.2250
(0.1676)
FGT 95–99, GS 100+ 2.1806
(1.3766)
FGT 99–100, GS 95–99 1.1498
(0.9444)
FGT 99–100, GS 99–100 1.0953***
(0.2717)
FGT 99-100, GS 100+ 5.7783*
(3.0951)
FGT 100+, GS 95–99 0.5612**
(0.2609)
FGT 100+, GS 99–100 1.5772***
(0.4515)
FGT 90–95, GS 90-95 0.9354**
(0.4009)
FGT 90–95, GS 95–99 0.1167
(0.0922)
FGT 90–95, GS 99–100 0.1537
(0.2076)
N 961 961
Notes: The dependent variable in all regressions is the daily natural gas price for
Florida for the period Oct. 6, 2006 - Aug. 29, 2011. Both Capacity 100+% is
a binary variable assigned a value of one if daily flow is 100 percent or more of
stated capacity for both pipelines, and a value of zero otherwise. Other binary
flow variables are defined similarly. Newey-West standard errors are in parenthe-
ses. All specifications include month-of-year fixed effects (with December as the
excluded month), year fixed effects (with 2006 as the excluded year), as well as
the additional controls in table 6 . All prices are expressed in real 2011 dollars.
*** p<0.01, ** p<0.05, * p<0.1. All p-values are for two-tailed tests.
69
Figure E.1: Florida Interactive Pipeline Effects
(a) Price Impacts of Changes in FGT Capacity
Levels for Given Levels of Gulfstream Capacity
Factor
(b) Price Impacts of Changes in Gulfstream Ca-
pacity Levels for Given Levels of FGT Capacity
Factor
70